x _ .
"gfiflm‘gfl .. s
3-1: 112%...
an: Hflémufiuun
.émfina .. «.3 .
agia

5:
035.27%}!
£- 1.33»—
...w .. j...
> T! ‘. 5 .~
.«lhifin.
kits;

9
. 3......

xv... id: x.
u i . .1 3
ha. .1 an i z
. t 3.

1‘1...r.!:)
livvsr:lau5,

1y. 1‘9. 2!", 1&3...
(Git-17$!) 12K

:2.

“1:14;...1...‘
531.1. 24.111311...
f;£alatii.lutiltm€
iv??:r§§lhb.rr\i
it!!! k
E
.S ‘ V a
31;... El 2.
Eras-:‘S
i. ‘ {3.35:7

 

mu..v\~.r¢..|f .1! pile. . a. [Itxu
.7; ufighés

BRARISE

lllllllllllllllllllllllllllllllllllllllllll Hill

3 1293 02079 934-

 

 

 

 

This is to certify that the

thesis entitled

Breast Aid: Results of a Multivariate
Model to Predict for Breast Cancer
in Women with a Breast Mass

presented by

Janet Rose Osuch, M.D.

has been accepted towards fulfillment
of the requirements for

Master's degfiwin Epidemiology

 

 

Major professor

 

Date May 1, 2000

 

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

     
     

     

IJgRARY
Michigan State

UHSUQI’QH'I:

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

DATE DUE DATE DUE DATE DUE
Aug 1 3 2006

 

 

BREAST AID: RESULTS OF A MULTIVARIATE MODEL TO PREDICT FOR
BREAST CANCER IN WOMEN WITH A BREAST MASS

By

Janet Rose Osuch M.D.

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE,
Department of Epidemiology

2000

ABSTRACT

BREAST AID: RESULTS OF A MULTIVARIATE MODEL TO PREDICT FOR
BREAST CANCER IN WOMEN WITH A BREAST MASS

By

Janet Rose Osuch M.D.

Most surgeons feel obligated to perform open biopsy of a palpable breast mass,
citing the potential consequences of diagnostic delay of breast cancer.
However, most breast masses are benign. BREAST AID (Breast Risk
Evaluation and Scoring System to Aid in the Diagnosis of Mammary Masses) is
a mathematical model designed to address this problem. It used multivariate
logistic regression analysis of 32 epidemiologic variables and 8 multi-Ievel
clinical variables to arrive at a final probability of breast cancer. Of the three
hundred twenty lesions evaluated, BREAST AID triaged 76 into open biopsy
and 244 into follow-up. Thirty-five of the 36 cancers were present in the
masses triaged to open biopsy, for a cancer rate of 46%. One cancer was
present in the 244 cases triaged to follow-up for a cancer rate of 0.41%. This
patient was diagnosed on 6 week follow-up. Before widespread application, a
multi-institutional study will be necessary to test the validity and generalizability
of the model. If generalizable, BREAST AID has the potential to conserve
health care dollars while simultaneously assuring both physicians and patients

that follow-up is a safe alternative to open biopsy.

Copyright by
JANET ROSE OSUCH

2000

DEDICATION

To all of the women who sought counsel in the Comprehensive Breast Health
Clinic at Michigan State University during this project, and who gave so freely of

themselves. Earning their trust was an unparalleled privilege.

ACKNOWLEDGEMENTS

This project was an idea first explored with Matt Romans MD, a resident in the
Department of Surgery at Michigan State University, whose enthusiasm for it
inspired me to continue after his graduation. It was subsequently funded by the
MSU Institute for Managed Care and Blue Care Network - Health Central,
whose support allowed for two graduate assistants to be employed, Jennifer
Fusco, an Epidemiology graduate student, and Tosca Kinchelow, who is soon
to graduate with an M.D. degree from the College of Human Medicine at MSU.
In particular, the long and dedicated hours of Tosca, who has become a dear
friend of mine, contributed greatly to the successful outcome of this project. To
acquire the skills necessary to pursue the project further, Dorothy Pathak Ph.D.,
MS. mentored me into the Masters Degree program in Epidemiology at MSU,
an endeavor made possible through the generous funding of the US.
Department of Defense. Her unending enthusiasm fueled me in the early
stages of the development of BREAST AID and her friendship, inspiration and
support served a pivotal role for me. Last but not least, I would like to
acknowledge the countless hours of work and mentoring from my major
advisor, Mat Reeves BVSc. Ph.D. His previous work with a similar model in
veterinary practice and his experience in Epidemiology provided the needed
tools to develop BREAST AID to its current level. More importantly, however,
his patience, persistence, and energetic and creative input into this project

makes it just as much his as mine. To him I am deeply grateful.

TABLE OF CONTENTS

LIST OF TABLES ..................................................................
LIST OF FIGURES ...............................................................
LIST OF SYMBOLS AND ABBREVIATIONS ..............................
INTRODUCTION ...................................................................

LITERATURE REVIEW ..........................................................

Decision-Making in the Diagnosis of the Palpable Breast Mass:
Who Needs Open Biopsy? .............................................

The Malignant: Benign Ratio on Open Breast Biopsy ...............

Risk Factors for Breast Carcinoma .......................................
Age ...........................................................................
Reproductive Factors/ Endogenous Hormones ..................
Exogenous Hormone Use ..............................................
Personal History of Breast Cancer ..................................
Family History of Breast Cancer ......................................
Environmental Risk Factors ............................................

Variables Collected during the History of the Individual Patient...
Asymptomatic Women ...................................................
Symptomatic Women - Risk Factors .................................
Symptomatic Women - The Presenting Complaint ...............

Physical Examination of the Breasts ......................................
Asymptomatic Women ...................................................
Symptomatic Women who Present with a Breast Mass .........

Components Used in Physical Examination Evaluation ......
Size of the Palpable Mass and Detection on CBE ............
Classification Schemes for Results of CBE .....................
Test Characteristics of CBE in Women

with Palpable Masses ...........................................

Mammography in Symptomatic Women who Present

with a Breast Mass ................................................
Classification of Mammography Findings ............................

vi

xii

xiii

28
33

 

Test Characteristics of Mammography in Women
with a Palpable Mass ......................................................... 34

History and Clinical Breast Exam in Combination with Mammography... 35

Fine Needle Aspiration Consistency .............................................. 36
Fine Needle Aspiration Cytology .................................................. 38
Classification of Fine Needle Aspiration Biopsy (FNAB) Findings... 38
Test Characteristics of Fine Needle Aspiration ............................ 39
Pitfalls of F NAB ................................................................... 40
Triple Diagnosis ....................................................................... 49
Classification of Triple Test Findings ........................................ 49
Test Characteristics of the Triple Test ...................................... 53
Follow-up after Triple Diagnosis ............................................. 59
Previous Use of Likelihood Ratios in Women with a Breast Mass ........ 61
Summary of Literature Review ..................................................... 61
MATERIALS AND METHODS ......................................................... 65
Quantitative Scoring System: Preliminary Work .............................. 65
Study Design .......................................................................... 68
Data Collection ................................................................... 68
Components of Clinical Impression .................................. 69
Clinical Breast Examination (CBE) ................................... 70
Mammography ............................................................ 70
Fine Needle Aspiration — Internal Consistency and Cytology. 71
Follow-Up ................................................................... 72
Study Setting ...................................................................... 73
Period for Data Collection ...................................................... 73
Selection of Study Population ................................................ 73
Abstraction of Data from the Medical Record and
Database Creation ....................................................... 74
Definitions ......................................................................... 75
Epidemiologic Predictor Variables ................................... 75
Clinical Predictor Variables ............................................ 75
Outcome Variables ...................................................... 76
Data Analysis and Statistics .................................................. 76
Staged Application of Bayes” Theorem ................................ 76
Multiple Logistic Regression Modeling ................................. 77
Staged Approach to Model Building .................................... 77

Final Adjustment of the Model based on Disease Prevalence... 79

vii

 

Model Goodness — of — Fit ........................................................ 80

BREAST AID Evaluation .......................................................... 81
RESULTS ..................................................................................... 81
Population Studied ..................................................................... 81
Population Characteristics and Results of Bivariate Logistic Regression 82
Biopsy and Follow-up Rates .................................................. 82
Descriptive and Bivariate Logistic Regression Results .................. 84
Epidemiologic Variables ................................................ 84

Age .................................................................. 84

Body Mass Index (BMI) ........................................ 85

Personal History of Breast Cancer .......................... 85

Age at Menarche ................................................ 85

Menopausal Status .............................................. 86

Age at Menopause ............................................... 86

Reproductive Factors - Pregnancy Status, Number of
Pregnancies, Number of Deliveries, and Age of

First Delivery ................................................. 87
Oral Contraceptive Use ........................................ 88
Estrogen Replacement Use .................................. 88
Progestin Replacement Use .................................. 89
Family History of Breast Cancer, Age at Diagnosis,
and Laterality of Disease .................................. 89
History of Radiation Exposure ............................... 90
Alcohol and Tobacco Exposure .............................. 91
Clinical Variables ......................................................... 91
Mass Size .......................................................... 91
Mass Shape ....................................................... 92
Mass Mobility ...................................................... 92
Mass Consistency ................................................ 92
Mass External Texture .......................................... 93
Mammogram Results ............................................ 93
Internal Consistency on FNAB .............................. 93
FNAB Cytology Results ........................................ 94
Results of Multivariate Logistic Regression Analysis ..................... 95
Stage 1 ...................................................................... 95
Premenopausal Women ....................................... 95
Postmenopausal Women ...................................... 96
Results of Hosmer-Lemeshow Goodness-of-F it
Chi Square: Stage 1 ................................... 97
Final Step in Stage 1: Evaluation of Pre-test
Probabilities .............................................. 98

viii

Stage 2 ..................................................................... 98
Results of Hosmer-Lemeshow Goodness-of-F it
Chi Square: Stage 2 ................................... 100
Merging of Data and Modification of Final Post-Test
Probability and Likelihood Ratio for Stage 2... 101
Test Characteristics for Merged Dataset Based on

Stages 1 and 2 ......................................... 101
Stage 3 ..................................................................... 102
Results of Hosmer-Lemeshow Goodness-of-F it
Chi Square: Stage 3 ................................... 103
Merging of Data and Generation of Final Post-Test
Probabilities ............................................. 1 04
Test Characteristics for BREAST AID: Final
Merged Dataset ......................................... 105
Results of Individual Diagnostic Tests Compared with Results of
BREAST AID ............................................................... 107
DISCUSSION ................................................................................ 109
Potential Study Biases: Patient Selection and
Population Characteristics .................................................. 1 11
Selection Criteria ......................................................... 112
Population Characteristics .............................................. 1 15
SUMMARY AND CONCLUSIONS .................................................... 119
APPENDIX ................................................................................... 121
REFERENCES ............................................................................. 150

10.

11.

12.

13.

LIST OF TABLES

Probability of developing invasive breast cancer in SEER areas,
women, all races, 1987-88 ................................................ 122

Physical Exam in Women with Palpable Breast Masses - Test
Characteristics from the Literature ......................................... 123

Mammography in Women with Palpable Breast Masses - Test
Characteristics from the Literature ......................................... 125

Fine Needle Aspiration Cytology in Women with Palpable Breast
Masses — Test Characteristics from the Literature ..................... 126

Triple Diagnosis in Women with Palpable Breast Masses - Test
Characteristics from the Literature .......................................... 128

Malignant vs. Benign Results by Score from BREAST AID
Preliminary Study ................................................................ 129

Definition of Test Characteristics Used in BREAST AlD ............... 130

Biopsy Results of 135 Lesions from 336 Breast Masses Evaluated
between November 1994 and February 1998 ............................ 131

Comparison of Current Study Histology Results with Prior
Publications ........................................................................ 1 32

Prevalence of Cancer and Bivariate Logistic Regression Results for
Epidemiologic Categorical Predictor Variables in 336 Breast Lesions
Evaluated between November 1994 and February 1998 ................ 133

Prevalence of Cancer and Bivariate Logistic Regression Results for
Clinical Categorical Predictor Variables in 336 Breast Lesions
Evaluated between November 1994 and February 1998 ................ 141

Prevalence of Cancer and Bivariate Logistic Regression Results for
Continuous Epidemiologic and Clinical Predictor Variables in 336

Breast Lesions Evaluated between November 1994

and February 1998 ................................................................. 144

Logistic Regression Model of the Risk for Breast Cancer for 112
Postmenopausal Women based on Epidemiologic Variables .......... 97

 

 

 

14.

15.

16.

17.

18.

19.

20.

21.

Observed and Expected FrequencieswithinQuartiles of Risk:
Stage 1 ..............................................................................

Logistic Regression Model of the Risk for Breast Cancer in 330
Breast Masses based on CBE and Mammography Results ............

Observed and Expected Frequencies within Groups of Risk:
Stage 2 ..............................................................................

Logistic Regression Model of the Risk for Breast Cancer in 336
Breast Masses Evaluated with FNAB ........................................

Observed and Expected Frequencies within Groups of Risk:
Stage 3 ..............................................................................

Test Characteristics for BREAST AID by Final Probability of Cancer

Sensitivity, Specificity and Predictive Values for Mammography
Alone with Comparison to Test Characteristics of BREAST AID ......

Sensitivity, Specificity and Predictive Values for Fine Needle
Aspiration Cytology Alone with Comparison to Test Characteristics
of BREAST AID .....................................................................

Results of Indeterminate or Less Findings on Mammography and

FNAB in Ten Cases of Breast Cancer: Comparison with Final
Probability of Cancer Predicted by BREAST AID ..........................

xi

97

99

100

103

104

105

108

109

146

LIST OF FIGURES

1. Example of 2 X 2 Table for Calculating Test Characteristics ............ 147

2. Best Test / Worst Test ROC Curve ............................................. 148

3. BREAST AID ROC Curve .................................................... 149
xii

Bl-RADSTM

BREAST AID

CBE
cm

Cl
cont’d

DCIS

ERT
FN
FP
FNAB
HRT
LCIS
malig
md
mm

mn

n/a

NPV

LIST OF SYMBOLS OR ABBREVIATIONS

Breast Imaging Reporting and Diagnostic System

Breast Bisk Evaluation and Scoring System to Aid in the
Diagnosis of Mammary Masses

Clinical breast examination
Centimeters

Confidence Intervals
Continued

Ductal carcinoma in-situ
Degrees of freedom

Estrogen replacement therapy
False Negative

False Positive

Fine needle aspiration biopsy
Hormone replacement therapy
Lobular carcinoma in-situ
Malignancy

Median

Millimeters

Mean

Number

Not applicable

Negative Predictive Value

xiii

 

0C
OR
PAR
PPV
PRT
ROC

SEER

SENS, Se
SPEC, Sp
TN
TP

U.S.

Oral contraceptive

Odds Ratio

Population attributable risk

Positive Predictive Value
Progesterone Replacement Therapy
Receiver Operating Characteristic

Surveillance Epidemiology and End Results Registry of the
National Cancer Institute

Sensitivity
Specificity
True Negative
True Positive

United States

xiv

INTRODUCTION
Breast cancer is the most common cancer diagnosed in women, accounting for
one-third of all female cancer. The American Cancer Society estimates that
182,800 cases of invasive and 22,000 cases of in-situ breast cancer will be

diagnosed in 2000 in the United States, and that an estimated 42,000 women
will lose their lives to the disease during that same year 1. In 1970, 90% of

patients who had breast cancer presented with a palpable breast mass 2.
Since then, screening mammography use has increased steadily, and breast
cancer is being diagnosed in asymptomatic women with increasing frequency.

Despite this, 55% - 68% of women with breast cancer presented with a clinically

palpable mass in several series published in the 19908 3'6.

The optimal clinical approach to the management of palpable breast masses
has not been established. Evaluations have included the analysis of risk factor
data, historical facts associated with symptoms, physical examination results,
assessment with mammography and ultrasound, and application of fine needle
aspiration cytology results. Each of these has limitations as a single test and

historically, open biopsy has been considered necessary to establish a

definitive diagnosis of a palpable mass by many authors 2: 3: 7. Some health-
care analysts have criticized the estimated 800,000 to 1,000,000 open

diagnostic breast biopsies performed in the US. each year as “unnecessary",

and some of this criticism is justified 8'10. There is inherent emotional,
physical, and financial morbidity associated with open biopsy, a societal cost of

over $4 billion dollars annually in the US. alone, and the average

 

malignant: benign yield is low — usually no better than 1:4 or 1:5 5’ 6. 11.
Complicating these issues is the reality that failure to diagnose breast cancer

based on “failure to be impressed with clinical findings” is the most common

reason for malpractice litigation in the US. 12. Most surgeons feel obligated to
perform open biopsy of palpable breast masses, fearing the profound emotional
and potential biological and legal consequences of diagnostic delay of breast
cancen

There has been a trend over the past 25 years to apply the principles of
“triple diagnosis” to the evaluation of a breast mass, especially in diagnostic
breast centers or other settings that ensure large volumes of patients. Triple
diagnosis refers to the application of a combination of 3 diagnostic tests to
evaluate a breast mass — clinical breast examination (CBE), mammography,
and fine needle aspiration biopsy (FNAB). A strict principle of this approach is
that if any one of the 3 diagnostic test results is other than benign, then open
biopsy is mandatory. One goal of triple diagnosis is to decrease the number of
open biopsies done for benign disease. Traditional evaluation methods for
diagnostic tests, referred to as test characteristics, measure sensitivity,
specificity, and positive and negative predictive values using 2 X 2 tables and a
classification of results that divides them into benign vs. not benign categories.
This dichotomous classification scheme, although necessary in order to analyze
the accuracy of any diagnostic test, ignores the considerable degree of
diagnostic uncertainty that exists for each component of the triple test and the

resulting indeterminate categories in diagnostic ratings. As a result, although

the application of the principles of triple diagnosis allows some women to be
triaged into clinical follow-up, the open biopsy rate may still be excessive.

The goal of this study is to better define diagnostic classification schemes
contributing to triple diagnosis, and to develop a mathematical prediction model
to help distinguish between malignant and benign disease in a symptomatic
group of women who present with a breast mass. A multivariate model to
predict for malignancy in women who present with a breast mass has not been
previously reported. Application of the results of the diagnostic prediction, if
accurate, could reduce the need for open biopsy and its associated morbidity
without compromising the timely diagnosis of breast cancer, and simultaneously
offer both physician and patient assurance that clinical follow-up is an
acceptable alternative to open breast biopsy.

LITERATURE REVIEW

Decision-Making in the Diagnosis of the Palmble Breast Mass: Whg

 

Needs Omn Biogy?

A clinician considers a host of variables in making a decision about the etiology
of a breast mass, including data collected during the history, results of the
physical exam, and results of diagnostic testing. Many publications exist
examining the value of risk factor analysis, patient history and physical
examination findings, mammography, ultrasound, fine needle aspiration
cytology and a combination of these in the evaluation of breast masses. The

pertinent literature for each of these will be discussed in the following sections.

 

The Malignant: Benign Ratio on Open Breast Biogy

The literature addressing the malignant: benign ratio on open breast biopsy is
published almost exclusively by surgeons practicing in breast clinics, academic
centers, or whom have a major interest in breast disease. By its very nature,
then, the reports reflect management of a select referral population by a
specialized group of physicians whose practices may not reflect those of the
general surgery community whom are responsible for the bulk of breast mass
management in the United States. The malignant: benign ratio quoted in the
literature may therefore be an overestimate of what occurs in actual practice. In
addition, the medical malpractice climate in the United States almost insists on
100% diagnostic accuracy in the work-up of a breast problem, despite the well-
known false-negative characteristics of the diagnostic tests commonly used in
the work-up. This makes interpretation of the literature addressing the
malignant: benign breast biopsy ratio in other countries less applicable than it
would othenrvise be. Keeping these issues in mind will be helpful during this
review.

As an illustration of worldwide differences, Ciatto and colleagues published
a multi—center study of 6 institutions in Italy, their malignant: benign breast

biopsy ratios, and reasons for benign biopsies. There was a wide range of age-

adjusted ratios reported, from 2.9:1 to 121.67 13. Two-thirds of the cases had
an open biopsy for benign disease because one diagnostic component of the
triple test was indeterminate or suspicious for malignancy. Interestingly, they

explored other reasons for biopsy in this study and found that 11.3% were done

 

 

for an increasing size of the palpable mass on follow-up, 3% were done for
cosmetic reasons, 8.2% were done for psychological reasons (patient request),
and 0.7% were done for high-risk situations. Despite the range of malignant to
benign results, they are much better than any series published in the United
States, where the malignant: benign biopsy rate is usually quoted between 1:4
and 1:5 5: 6: 11.

A large study by Leis documented all of the biopsies done in the decade
1960-1969 at New York Medical College and Affiliated Hospitals prior to the use

of screening mammography. Of 3,287 breast biopsies performed, 923 (28%)

were for malignant disease 2. Egan’s data on breast biopsies done between
1963 and 1973 of 2793 biopsies done at Emory Clinic in Atlanta documented a
25% malignancy rate that was highly age-dependent, with women less than 38,
39-62, and 63-100 years of age having malignancy rates of 6.2%, 26.3%, and
62% respectively. The biopsy rate, however, did not vary by age group;

between 14 and 16% of breast exams in each age group resulted in surgical

biopsy 11. Clinicians are aware of the differential malignancy rate by age
group, and the lack of variance in the biopsy rate by age reflects the fact that
breast cancer can affect any age group in the adult women, coupled with the
prevalence of benign disease that exists in premenopausal women, which
peaks in the 5th decade of life. Because of the high degree of diagnostic
uncertainty in the presence of a breast mass, clinicians will tolerate a high
number of false positive results, especially young women, in order to achieve

the lowest number of false negatives. In a more modern study, in 1992 Seltzer

 

reported a breast biopsy cancer rate of 20.4%. This was also influenced by
age, with 34.7% of the population of biopsied patients 50 years of age or older
being diagnosed with cancer, compared to less than 12% in those less than age
50. In 1997, the same author published a new series in a different population -
2247 patients who underwent open biopsy for a variety of reasons. In this

series, 55% of the biopsies were done for a palpable breast lump (21% cancer

rate), and 35% for an abnormal mammogram (25% cancer rate) 6. The rate of
cancer in the overall population was 22%; 12% for those less than 50 years of
age and 41% for those 50 years of age and older. Lay documented a lower

malignancy rate of 15% in women biopsied for palpable abnormalities between

1983 and 1987 at University Hospital in Jacksonville Florida 14.
Selzer reported on the calculation of a “fear factor” indicator for breast
biopsies, that is, the difference between the surgeon’s positive predictive value

and the actual incidence of malignancy on biopsy. He calculated a “fear factor”

of 24% for women less than 50 years of age and 15% for those 50 and over 6.
Entering into this fear factor are considerations that affect both the physician
and the patient. The patient expects that a diagnosis be made with 100%
diagnostic certainty. The surgeon understands that no test can ensure 100%
diagnostic certainty short of open biopsy (even open biopsy has its pitfalls), and
that medicine is a science of probabilities. The surgeon, however, is vulnerable
to the medical-legal climate and in the United States, and there seems to be

little tolerance for anything less than 100% diagnostic certainty. The result is

that many more open biopsies are performed in the United States than in other
parts of the world and many more than may be necessary.

Risk Factors for Breast gapinoma

Epidemiologic factors identified to confer risk in large populations have been
widely studied from a public health perspective, but are of varying degrees of
usefulness in the clinical setting when evaluating an individual.

Misinterpretation of risk information is common, and many studies report over-

estimation of risk by patients and their physicians 15'17. Conversely, some
individuals are lulled into complacency by assuming that if they possess no risk
factors that they are at no risk for the disease. By far the highest risk for breast

cancer is the female gender, with a ratio of women: men affected of

approximately 180:1 18. All women are at risk for breast cancer, but the
probability of disease varies by a variety of additional factors. Systematic study
of these factors has provided important clues to the etiology of breast cancer.
The most important of these variables will be briefly described.

Age

Unlike most cancers, breast cancer affects the entire age spectrum of adult
women, with increasing frequency as a woman ages. The cumulative risk of
developing breast cancer by a given age, in five-year increments beginning with
age group 20-24 through age 95+, is given in Table 1. If a cohort of women is
followed for 85 years, one in 9, or 11.1% will develop invasive breast cancer at

sometime in her lifetime, and if followed for 95 years, one in 8, or 12.5% will be

diagnosed 19.

When the ageincidence curve for breast cancer is plotted on a log-log scale,
the curve produced is not a straight line throughout a woman’s lifetime as it is
with other cancers. Instead, the curve assumes a straight line from age 25 to
50, after which a slowing effect in the incidence slope is noted. The curve

assumes a straight line at this slowed rate for the remainder of women’s lives

20. This important observation provides clues to the etiology of breast cancer
and its dependence on female hormones for the induction and promotion of
carcinogenesis. It implies that premenopausal women possess the etiologic

components for breast cancer development and that cessation of ovulation

decreases, but does not eliminate them 20. Many of the known risk factors for
breast cancer relate to the specifics of ovulation, and in particular, to the
hormonal milieu to which the breast is exposed. This will be discussed in more
detail in the following sections.

Regroductive Factors/Endogenous Hormones

There is a strong correlation between the length of exposure to endogenous
hormones and the risk of breast cancer. The earlier the age at menarche, the
higher the risk. Girls who begin menses at age 11 or before have a relative risk

for breast cancer of 1.3 as compared with those with onset at or after age 16

21. Age at menopause also influences risk. Studies demonstrate an average
risk elevation of 1.5 between those women who underwent menopause at or
after age 55 as compared with the less than 45-year age group. Even more
provocative, it was found that those women who had a surgically induced

menopause through bilateral oophorectomy and before the use of estrogen

replacement therapy had a reduction in breast cancer risk to 0.4 22. It has

been estimated that for every 12-month increase in age at menopause, there is

an approximate increased risk of breast cancer of 3% 21.
Carrying a pregnancy to the third trimester confers a protective effect on risk
in most women. This is inversely correlated with age at first delivery. Russo

and Russo explain the biological mechanism for this as differentiation of the

terminal duct lobular unit of the breast during pregnancy 23. However, if a
woman delivers her first child at or after the age of 30, her risk is actually about
the same as if she was nulliparous. The risk ratio for this group is 1.9

compared to a reference group of women whose first delivery was before age

20 24. The biological mechanism for this observed phenomenon has not been
explained. Parity as a risk factor disappeared after adjusting for age at first
delivery in early studies, but there is some evidence currently that parity exerts

a paradoxical influence, or crossover effect, on risk. Parity increases risk in

premenopausal women, and decreases it in women over the age of 40 25.
Obesity (weight above ideal of 20 kg or more) increases the risk of breast

cancer in postmenopausal women to 1.8, but in premenopausal women, the

risk is actually inversely correlated with weight 26. The mechanism for the latter
has not been explained, but for the former, the hypothesis is that obesity
causes a rise in estrone, the major postmenopausal estrogen, through
aromatization of androstenedione in the excess adipose tissue of obese

women. Estrone is then converted to estradiol, increasing the hormonal milieu

of breast tissue to this potent estrogen 27.

§x_ggenou_s Hormone Use

Oral contraceptive (00) use and its possible relationship to breast cancer risk
has generated debate over several decades. A recent meta-analysis of the
relationship pooled the data from over 150,000 women in more than 50 studies.

A 24% increased risk was observed in current OC users that returned to

baseline five years after discontinuing use 28.

A meta-analysis of hormone replacement therapy (HRT) use and breast
cancer risk has also recently been done. Over 50 studies in more than 160,000
women were examined; a 35% increased risk for breast cancer was observed

in current users of HRT, that returned to baseline five years after discontinuing

use 29. Most of the studies examined the use of estrogen replacement alone
rather than combined therapy with progestins. This is important because
progestins are currently given to most postmenopausal women on HRT who

have not had a hysterectomy to decrease the proliferative effect of estrogen on

the uterus and subsequent risk of endometrial cancer 30. It has been shown
that progestins do not have a similar effect on the breast and may actually
increase cellular proliferation at that site 31.

Personal History of Breast Cancer

It has been known for some time that a history of unilateral breast cancer
predisposes a women to a greater risk of developing contralateral breast cancer

over the same time period compared to a woman without that history. This risk

is predictable at a steady rate of 0.7% per year of follow-up 32.

10

 

Family Histom of Breast Cancer

A family history of breast cancer can increase a woman’s risk anywhere from

20% to 85-90% depending on the relative(s) affected and their menopausal

status at diagnosis 33. A recent meta-analysis of 74 published studies has

provided pooled estimates of the relative risks associated with various family

histories 34. First degree relatives are most important, increasing a woman’s
relative risk to 2.1. A history of breast cancer in a second-degree relative
increases the relative risk to 1.5. The woman at the greatest risk is one who
has both a mother and a sister with breast cancer. The relative risk in this
group of women in the pooled analysis was 3.6, but could be as high as 13.6.
This high odds ratio reflects the stronger likelihood that a woman with two first-
degree relatives with breast cancer has inherited a genetic mutation increasing
her susceptibility to the disease. Recent advances in genetic research have

identified 2 genes, BRCA-1 and BRCA-2, that when abnormal are associated

with up to an 80 - 90% lifetime risk of breast cancer 35. Women who have a

mutation of one of these genes also have a higher likelihood of a family history
of premenopausal breast and/or ovarian cancer in multiple relatives, and either
disease is more likely to be bilateral. It has been estimated that approximately

5—10% of women diagnosed with breast cancer have inherited a mutation in one

of these genes 33.

Overall, the proportion of women in the general population with a family

history of breast cancer in a first degree relative is low, about 8% 36. Of those

women diagnosed with breast cancer, family history estimates range between

11

10 and 22%, depending on the characteristics of those assessed and the

number and familial proximity of those included 37’ 38.

Mmentgl Risk Factors

Two environmental risk factors have been unequivocally identified - radiation
exposure and alcohol intake. The association between radiation exposure and

breast cancer risk has been known for many years and has been studied in the
context of repeated fluoroscopic treatments for tuberculosis 39. 40, post-partum
mastitis treatments with ionizing radiation 41, and in the atomic bomb survivors

of Hiroshima and Nagasaki 42’ 43. The studies demonstrate a two to four-fold
increase in the incidence of breast cancer for women exposed before age 25,
which occurs after an average latency period of 15 years.

Associations between alcohol intake and breast cancer risk have been
studied in earnest over the past several years. A pooled analysis of cohort
studies has recently been published which examined over 50 studies. Most of
these demonstrate a 30-40% increased risk in current users of 1-2 alcoholic

drinks/day compared to non-users. This is thought to be associated with the

increased metabolism of estrogen observed in alcohol users 44. Based on this

data, an analysis has recently been done estimating the age-adjusted

population attributable risk (PAR) for alcohol and breast cancer 45. The finding
of a PAR of 2.1% caused the authors to conclude that widespread efforts to
reduce alcohol consumption would not have a substantial impact on breast

cancer rates.

12

Three major studies have addressed the issue of PAR in breast cancer
examining multiple variables of risk. Seidman examined 10 common risk
factors and found that only 29% of cases of breast cancer occurred in women

aged 55-84 with any risk factors, and only 21% of cases in women aged 30-54

had any risk factors 46. Madigan found that 47% of cases of breast cancer

occurred in women with at least one risk factor, but one of the most important,

family history, was found in only 9.1% of the population studied 47. Rockhill

and colleagues recently examined the risk factors of menarche, parity status,

age at first delivery, family history, and previous history of breast biopsies 48.
The authors estimate a PAR of 15%-25% based on these factors, depending on
the definition of risk. They estimate that 62%-98% of the female population has
been exposed to at least one of the factors studied. They conclude that the
established risk factors for breast cancer have little public health value because

of the high proportions of the population exposed, and the unmodifiability of

most factors 48. The magnitude of the association of risk for most risk factors is
also low, between 1.2 and 2.0, which contributes to the view that they are not of
much practical use. These views do not take into account the important
insights that result from the study of risk factors from an etiologic standpoint,
and the effect of risk assessment in terms of chemoprevention studies. It does,

however, reflect the perspective that clinicians have about epidemiology as it

relates to the assessment of an individual woman 49» 5°.

13

 

Variables Collected during the History of t_he Individual Patient
mmmatic Women

It is important to appreciate that risk factor analysis is complicated and
dependent on considerations of confounding and interactions between
variables. As a result, the useful application of breast cancer risk data, which is
multifactorial, has proved to be difficult in the clinical setting.

Several authors have attempted to quantitatively predict the absolute risk for
breast cancer in an individual patient based on one or more risk factors. The
earliest of these models apply only to women with a family history of the
disease. Anderson studied a cohort of hospitalized women in Houston with a

family history of breast cancer and reported results as absolute risks
categorized according to frequently observed family histories 51» 52. Ottman

used the same approach in a population-based cohort in Los Angeles 53.
These approaches were further refined by Claus, who modeled the absolute
risk of breast cancer in women with a family history in relatives from the same

lineage based on the probabilities of autosomal dominant inheritance

patterns 54.

Although family history is an important consideration of risk, lack of this risk
factor does not protect a woman from development of the disease. The great
majority of women who are diagnosed with breast cancer have no known family
history. Gail and colleagues developed a quantitative model to estimate

absolute risk for breast cancer that uses multiple risk factors in the analysis,

rather than family history alone 55. These included age, age at menarche, age

14

at first delivery, number of breast biopsies, and presence of atypical hyperplasia
on breast biopsy as well as family history in first degree relatives. The data
analyzed to construct the Gail model was taken from the Breast Cancer
Detection and Demonstration Project, a volunteer-based observational study of
women who were willing to undergo annual screening with physical exam and
mammography. The model has been validated in the Texas Breast Screening
Project and in the Nurses Health Study and has been used in modified form to

predict risk in the Tamoxifen Breast Cancer Prevention Trial of the National

Surgical Adjuvant Breast Project 56’ 57. It has been determined to be a
relatively accurate predictor of risk, although there have been concerns

expressed about underestimation of risk in some subgroups, especially those

with an extended family history 57. The Gail model is also documented to

overpredict risk by 33% among women 60 years of age and younger who do

not participate in annual screening 50. In addition, the use of breast biopsies as
an independent predictor of risk has been criticized for its inclusion in the
model. Many consider this to be a confounding variable, especially since
women at increased risk for breast cancer are likely to have breast biopsies
recommended more often than those who do not.

Symptomatic Women — Risk Factors

Accurate knowledge and appropriate application of risk factors in the clinical
setting alter the probability of disease in an individual patient. It is more likely,
for example, that a woman with a family history of breast cancer in 2 sisters

who presents with a breast mass has cancer as compared with a woman

15

without a family history. The magnitude of the risk is dependent on the patient’s
age; other risk factors present or absent in the patient’s history, and the clinical
characteristics of the mass.

No attempt has been made to predict risk for breast cancer in symptomatic
women based on history of risk factors alone in any quantitative or even non-
quantitative way. One short descriptive study published in 1977 examined 1059
symptomatic patients from one surgeon’s practice collected over a 6-year
period. The rate of breast cancer in the studied population was 13%. The
women with breast cancer were compared with those who did not have breast
cancer. No statistical analysis was done and no frequency tables were
provided. The author examined age, family history, menstrual status, a history
of benign breast disease, parity, use of hormones, and duration of symptoms,

concluding that only age and family history in first degree relatives were useful

in the history for predicting the presence of malignancy 58.

Practice guidelines addressing breast cancer and its diagnosis have recently
been written by a variety of organizations. Although recommendations are
made to collect risk factor information, physicians are cautioned to remember

that“ the majority of women who develop breast cancer have no risk factors

beyond being female and aging” 59. One guideline additionally states that “risk
factors for breast cancer should be noted, but their presence or absence should
not influence the decision to investigate a lump further” and that “although

known risk factors, including aging, all increase the risk of breast cancer, they

16

 

do not substantially influence the probability that any particular lump will be

malignant” 60.
Symptomatic Women — The Presenting Complaint
Breast complaints are extremely common. A study in the United Kingdom

documented a frequency of primary care visits for breast complaints of 18 visits

per 1000 person-years 61. A similar study published by Barton and colleagues
followed 2400 women between July 1983 and June 1995 in a large health
maintenance organization in New England. They found that 16% of the
population of women between the ages of 40 and 69 presented with a breast

symptom during the follow-up period, for a rate of 22.8 presentations per 1000

person-years 62. Women younger than 50 presented twice as often as older
women.

Most studies of the frequency of breast complaints in general practice have
been done in the United Kingdom. It has been estimated that the average

general practitioner will see 13 - 34 patients for new breast complaints per year

63'65. Breast complaints fall into 5 categories: mass, abnormal mammogram,
pain, nipple discharge, and skin changes, and the complaint offers some
information about the likelihood of cancer. In Barton’s study in primary care

physician practices, the most common symptom was pain (47%), followed

closely by a breast mass (42%) 62. Breast cancer was diagnosed in 23 of the
372 women who presented with symptoms (6.2%). A mass, which was the

most common sign of malignancy, had a 10.7% chance of representing breast

17

 

cancer and had a likelihood ratio of 65. Age was not significantly associated

with the likelihood of cancer in this study 62.

The studies of referral clinics reflect a different population than has been
discussed to this point. Primary care physicians, for example, are not likely to
refer the majority of patients with breast pain, but manage this symptom
themselves. In a study of referrals to the breast clinic at the City Hospital in
Nottingham published in 1988, 1445 patients were seen over one year, 65% of
whom presented with symptoms of a mass. This series did not include
abnormal mammogram referrals. Twenty-five percent of the patients with
biopsy confirmed masses had cancer, and a women less than 50 years of age

had a 1 in 20 chance of having malignancy, whereas the ratio was 1 in 2 in

women 50 years of age and older 66. In a study of 1,000 consecutive patients
referred to a general surgeon in London, similar findings were reported by Ellis
et al in 1984. There was an overall cancer rate of 12% in their population, and
26% for biopsied patients 67.

A more current study by Seltzer, also in a p0pulation of surgical referrals,

reported on 2,000 consecutive new patients evaluated in a surgical clinic

between January 1987 and February 1989 5. The most common presentation
to the clinic was a breast mass (50% frequency), followed by an abnormal
mammogram (31.9%). Less frequent were complaints of mastalgia (6.5%) and
nipple discharge (3.6%). Miscellaneous symptoms accounted for the reminder.
Of the 4 symptoms specified, the cancer rates were 8.2%, 9.1 %, 2.3%, and

5.9%, respectively. These malignancy rates are only slightly lower than those

18

 

reported from the breast clinic at Ninewalls Hospital in Dundee, UK, where the

malignancy rate was 11% 68. These rates are consistent with the fact that
cancer rates for overall clinic referral populations are much lower than quoted
cancer rates for patients with open biopsy, as would be expected.

Age is a strong predictor of the likelihood of malignancy. Kleinfeld reviewed
1,019 breast biopsies done at the North Miami General Hospital between 1964
and 1968. In that study, the malignancy rate was 29% overall. In the decades
of life from 10-19, 20-29, 30-39. 40-49. 50-59, 60-69, 70-79, 80-89, and 90+,

the malignancy rates were 3%, 2%, 9%, 19%, 59%, 70%, 77%, 100% and

100% respectively 59. Specifically, the pcak age at which breast biopsy
occurred was 40-49, but the malignancy rate was only 19%. This rose
incrementally in postmenopausal women so that the rates ranged between 59%
and 100% past the age of 50.

The combination of the risk factor history, the patient’s age, and the
presenting complaint evaluation begin to form a pretest likelihood of disease in
the clinician’s mind. However, results of the clinical breast exam and diagnostic
tests add to the assessment. How much each component contributes and to
what degree each diagnostic test is necessary is a matter of debate. What are
the necessary components of breast mass evaluation beyond the history? If a
palpable mass is confirmed on clinical examination, some clinicians demand
open biopsy for definitive diagnosis at this point, while others are willing to
consider follow-up under some circumstances. Mat are these circumstances?

What is the role of mammography in women with palpable masses? Is it

19

 

acceptable to stop the work-up at the point of CBE and mammogram
evaluation? How many cancers, if any, will be missed is biopsy is not done at
his point? Is cytological or histological sampling necessary in every patient with
a breast mass? What is the false negative rate of the diagnostic work-up short
of open biopsy? The next sections will address these issues.

Physical Examination of the Breasts

Asymptomatic Women

Despite numerous publications addressing the techniques of screening physical

examination of the breasts 2. 49. 70'72, the test performance characteristics of
CBE have not been studied until recently. The Canadian National Breast
Screening Study (NBSS) published data detailing the sensitivity, specificity, and
positive predictive values of CBE in 19,965 women aged 50-59 years and
25,620 women aged 40-49. On visit one, the respective values were 83%,
88%, and 3% in the older age group and 71%, 84%, and 1.5% in the younger
age group. A separate analysis of test characteristics between the study
surgeons (who also evaluated the symptomatic patients) and the nurse
examiners (who did much more of the screening exams) demonstrated
interesting differences. The sensitivity, specificity, and positive predictive value
at visit one for the nurses for screened patients aged 40-49 was 68.5%, 95.5%,
and 5.0% respectively. For the screened patients examined by nurses aged
50-59 the values were 80%, 96.8%, and 10.9% respectively. In comparison,
the same values for the surgeons in the examined 40-49 age group were

96.8%, 70.4%, and 5.0% and in women 50-59, the respective values were

20

 

92.7%, 75.6%, and 10.3% at visit one 73. This difference between the nurses
and surgeons reflects the different roles of the examiners. The nurses were
more likely to refer patients with any breast abnormality for evaluation by a
surgeon, but surgeons were likely to remove only those lesions for which biopsy
was clinically indicated. For these reasons, one would expect surgeons to have
a higher test sensitivity than nurses and for nurses to have a higher test
specificity than surgeons.

In 1999, a meta-analysis was done of the sensitivity and specificity of CBE

in asymptomatic women 62. A pooled analysis of the clinical trial data
demonstrates lower sensitivity and higher specificity for CBE than in the overall
results of the N388; values of 54% and 94% respectively. These numbers
again reflect the screening role of CBE in these studies.

Symptomatic Women Who Present with a Breast Mass

Commnents Used in Physical Examination Evaluation

Surprisingly few publications detail the components that enter into a clinician’s
assessment of whether a mass on CBE represents benign disease or cancer.
Most mention the physical examination signs of advanced malignancy and early
studies discuss these characteristics as criteria for designation of a mass as
malignant on CBE. If a mass is hard, fixed to the skin or chest wall, or irregular
in shape or texture on palpation, it is likely to be classified as malignant. The
same is true if skin puckering, nipple retraction or nipple erosion is present, or if
regional lymph nodes are palpable. In 1967, Zajicek et al, in one of the first

publications of FNAB findings in women with breast masses makes the

21

 

following statement: “It seems pertinent that of the 485 cases in which cancer
was cytologically diagnosed from aspiration biopsy, 30 percent lacked such

obvious clinical signs of advanced mammary carcinoma as fixation of skin and

muscle and retraction of the nipple 74.” Kruezer and Boquoi, in their publication
in 1976, specifically state that they used these as the criteria for a malignant or
suspicious finding on physical examination. They found the sensitivity,

specificity, and positive and negative predictive values of clinical exam to be

90%, 78%, 74% and 92% respectively 75. Grobler et al applied similar criteria

in a publication in 1990 in a series reported from the University of the Orange

Free State, Bloemfontein 76. Sixty-six percent of their patients had malignancy
and the average tumor size was 3 cm. The sensitivity, specificity, positive and
negative predictive values of physical exam were 96%, 76%, 88% and 90%.
The test characteristics published in these two studies suggests, like the
population studied by Zajicek et al, that many the patients examined had
advanced disease at the time of evaluation.

If these were the main criteria for ruling in breast cancer on CBE today,
many more false negative results would occur. Current awareness of women in
the importance of breast self examination, and the increased emphasis on the
importance of CBE and mammography in asymptomatic women, has resulted in
the decreasing size of clinically detectable tumors. In 1985, it was reported that

the average breast mass size discovered on breast self-examination was 2.5

cm 77, and in the initial screens with physical examination in the Ontario Breast

Screening Program between 1990-1995, the average invasive cancer size

22

 

found at palpation was 1.8 cm 78. Cady et aI studied 1,001 cancers diagnosed
between 1989 and 1993 at a tertiary center and a community hospital and
found a median maximal diameter for invasive cancer of 1.5 and 1.7 cm
respectively 79.

In addition to the decreasing size of tumors at presentation, the
characteristics of the mass on CBE are also changing. In 1971, Venet and
colleagues estimated the sensitivity of 3 characteristics of physical examination
conducted in asymptomatic women who were found to have palpable masses
during an early screening study, the Breast Cancer Detection and
Demonstration Project. They evaluated mass consistency, texture, and

mobility. They estimated sensitivities of 62%, 60%, and 40% respectively for

each of the three characteristics 80. This means that 4 of 10 malignant breast
masses would be either soft or cystic, that 4 of 10 would not have irregular
borders, and that 6 of 10 would be freely movable. The sensitivity for individual
physical examination characteristics is likely to be even lower than it was almost
30 years ago, since along with decreasing size of palpable masses and their
more subtle CBE characteristics, the associated advanced signs of breast
cancer that clinicians relied upon in the past have disappeared. That “failure to

be impressed with clinical findings” is the most common reason for diagnostic

delay of breast cancer underscores this point 12. To illustrate it further, Mushlin
used the data from Venet and calculated the probability of cancer based on the
clinical finding of fixation, or lack of mobility. Based on an estimated pretest

likelihood of cancer in the presence of a breast mass of 20% and a sensitivity

23

 

and specificity of 90%, he estimated a post-test likelihood of malignancy of 50%

for a positive finding and 14% for a negative one 81. Clearly, greater diagnostic
accuracy is required than that offered by this finding (or lack of it) on CBE.

Lesions that are round, smooth, and circumscribed on palpation have a

higher likelihood of being benign than malignant 60. However, the reliability of
this assessment is dependent on the experience of the examiner, and even
when evaluated as clinically benign, the false-negative rate of physical exam is
too high to depend on it alone to determine the etiology of a breast mass under
any circumstances. Rimsten found a true positive rate for physical exam of

70% — 90% and a true negative rate of 70% in one study of experienced

examiners 82. In the largest study of examiner-dependent sensitivity of CBE,
Ciatto et al found a range between 69% and 90% among 41 examiners, with a

statistically significant association between examiner experience as measured

by number of cancers palpated (<1oo, 100-299, >300 83. In another study,

Boyd et al in 1981 reported on 100 patients with breast masses examined by 4

experienced surgeons 84. Forty-one of these patients had been admitted to the
hospital for an open breast biopsy of a palpable lump, while the remaining 59 i
had no known breast disease. Masses were palpated in 32 - 42 % of the
patients. Agreement between the 4 surgeons of the finding of a mass on CBE
was 100% among only 16 of the 41 cases in which a mass was present, and full
agreement that the mass represented malignancy occurred in only 5 of the 15

patients with cancer. In this study, 7% - 20% of the patients who had cancer

had no recommendation for biopsy on the part of at least one surgeon 84.

24

 

Size of the Palpable Mass and Detecjm on CBE

Few studies have reported mass size on CBE, and those that do often fail to
distinguish whether the size of the mass reported refers to the findings at CBE
or on histopathology. This is important because CBE tends to overestimate the
size of a mass, and if the size reported refers only to the histopathology
findings, the PE test characteristics may appear better than they actually are.

Boerner and Sneige found that lesions measured as just under 3 cm by CBE

were an average of 7 mm smaller on histopathologic exam 85. In addition,
some studies report only the size of malignant lesions on histopathology, which
does not allow for a full evaluation of the population characteristics studied
unless all of the lesions were malignant. One study from Germany
retrospectively examined 160 breast carcinomas between 1988 and 1990. The
average tumor size histologically was 25.7 mm. The smallest tumor detected

by palpation was 3 mm. CBE detected 87% of the tumors overall; 77% that

were 2 cm or less, and 47% that were 1 cm or less in size 86. The others were
detected by mammography. Hermansen et al studied a series of patients
between 1980 and 1982, and they found that 33% of the tumors were less than

1 cm in size, 50% were between 1 cm and 3 cm, and 17% were greater than 3

cm. The malignancy rates were 7%, 31%, and 72% respectively 87. DiPietro
reported on 631 solid breast masses that did not have advanced signs of
malignancy. In 105 patients, the mass spontaneously regressed and these
were followed to ensure persistence of this finding. They do not discuss the

average mass size but comment that 66.3% of the benign cases were not larger

25

 

than 20 mm, and that 53.7% of the malignant cases were 20 mm or less in size.
In addition, 39.2% of masses 20 mm or less in size were malignant and that
54.5% of those larger were malignant.

The question of the sensitivity of CBE to detect a mass in relation to its size
is more directly answered in a study of screened women by F leg at al reported
in 1979. In it, 183 cancers were diagnosed in 17,000 women between the ages
of 45 and 64. The mammogram techniques were state-of-the-art for the time,
but not as technically sophisticated as they are today. Nonetheless,
comparison between CBE and mammographic detection of cancer is possible.
In this study, tumors were divided into size groups of 0—0.5 cm, 0.51-1.0 cm,
1.1-2.0 cm, 2.1-3.0 cm, and over 3 cm. Mammography detected 92%, 70%, 80,
65%, and 80% of lesions in the respective size categories. CBE performed

much worse when the lesion was 2 cm or less in size, detecting 20%, 40%,

50%, 80%, and 90% of the cases in each size category 88. A large study of 2,
648 palpable malignant tumors diagnosed between 1979 and 1986 in Italy also
examined sensitivity of physical examination in relation to histologically
measured tumor size. Physical examination identified the palpable mass as

cancer with the following frequency: ln-situ cancer: 48%, masses 2 cm or less:

70%; and masses > 2.1: 89 — 93% 83.

Classification Schemes for Results of CBE

Classification schemes for findings on CBE typically fall into 3 categories:
benign, indeterminate, and suspicious/malignant. Some investigators use a

classification that includes suspicious classifications in the indeterminate

26

 

 

category. None of these has been validated and few publications even describe
their criteria for inclusion into one classification category vs. another. In an
impressive attempt to delineate physical exam characteristics, DiPietro et al
used a “Clinical Diagnostic Index” that used the algebraic sum of 9 different
characteristics on CBE. They reported the test characteristics of the physical
exam according to the overall diagnostic index score. This score included the
following characteristics: (1) changing size of mass; (2) consistency (soft, equal
to surrounding tissue, or hard); (3) shape and borders (roundish or irregular);

(4) surface characteristics (smooth, intermediate, or irregular); (5)

 

intramammary mobility (marked, intermediate, or scarce); (6) prominence (not
protruding or protruding at palpation); (7) pain with palpation (present, scarce,
absent); (8) nipple secretion (absent or serous or hematic present); and (9)

presence of axillary lymph nodes (not palpable or suspicious). Even with this

elaborate diagnostic index, 10-23% of the malignant masses were undervalued

on clinical examination, and the test sensitivity was only 63% 89.

Test Characteristics of CBEli Women with Palgble Masses

The subjectivity of CBE interpretation and the lack of a standardized reporting
system have inhibited studies of CBE accuracy. Table 2 demonstrates the test
characteristics of physical exam from the literature. Many of these values were
derived from the data quoted in the papers, as not every report calculated the
pertinent test characteristics. The categorization into positive vs. negative tests
was based on a benign vs. not benign categorization as discussed above. Most

of the studies in table 2 were done in the context of triple diagnosis evaluation,

27

and CBE test characteristics analyzed as a separate component of this
diagnostic evaluation 7. 75, 76, 87: 89'98. One was done as a direct evaluation

of CBE 99. and three others as an evaluation of CBE in combination with other

tests 82, 100, 101.
The sensitivity of CBE has been demonstrated to be age-dependent. In

women 60 and over, values of 90% or greater are the rule 83, 96, 99: 100. In
contrast, CBE sensitivity is low in women 40 or less, ranging between 25% and
77% in those same studies. In women between the ages of 40 and 60, CBE
sensitivity rises incrementally above that in the younger group, but does not

achieve levels as high as in the older age category, ranging between 68% and

90% 83, 96: 99: 100. The reason for this observation is likely to be twofold.
One is explained by the increasing incidence of cancer as a woman ages,
knowledge incorporated into the assessment made by clinicians that increases
the index of suspicion of cancer in the older woman. In fact, there is a clinical

caveat that a postmenopausal woman who has a breast mass has cancer until

proven otherwise 102. The other is related to the ease of CBE interpretation in
postmenopausal relative to premenopausal women, the latter of whose CBE

interpretation may be challenged by the influence of ovarian hormone levels

and fluctuations 103. How hormone replacement use in postmenopausal
women will affect the sensitivity of CBE has not been studied to date.
Mammography in Symptoma_tlc Women who Present with a Breast Mass
Mammography is capable of detecting carcinoma before it can be palpated, and

numerous clinical screening trials throughout the world have demonstrated its

28

 

ability to reduce mortality from breast cancer. The sensitivity of screening

mammography is expected to be 85% or above 104. When referring to
screening mammography, a false-negative case is defined as any case of

symptomatic cancer diagnosed within 12 months of a normal screening

examination 105. Since the false-negative rate is defined as 1- sensitivity, the
usual false negative rate of mammography in the screening setting is expected
to be about 10 - 15%. This number is often erroneously applied to the
diagnostic setting. In a diagnostic setting, such as this one, the false-negative
rate refers to the number of missed cancers at the time of biopsy or if biopsy
was not done, at the time when a palpable mass was clinically evident.

Unfortunately, the false-negative rate can be far higher in the diagnostic than in

the screening setting, especially in young women with dense breasts 106. In

early publications, the false-negative rates in symptomatic populations reached

as high as 44% 107. Egan published a series of 2,793 breast biopsies done
between 1963 and 1973, 886 of which were malignant. He reported a
false-negative mammography rate of 21.9% overall in women with solitary
nodules over age 38 years, and this varied by age group. In the age categories

39-62 years the false-negative rate was 19.3% and in those 63-100 years it was

41.2% 11. This is somewhat unexpected, because most literature quotes a
higher false-negative rate in women with dense breasts, who are more likely to
be in young age groups. However, this study was limited by the technically

inferior mammogram characteristics compared with modern times.

29

 

Edeiken reviewed 499 malignant breast biopsies performed at Jersey Shore
Medical Center for palpable masses between 1976 and 1985. He found an
overall false negative rate for mammography of 22% that did not vary by study
year. When broken down by age, the false-negative rate was 44% for women

less than 51 years of age who presented with palpable masses and 13% for

those aged greater than 51 108. A study from Copenhagen during that same
time period published more favorable performance characteristics for
mammography but the results also differed by age, with false-negative
mammograms in women with palpable masses in 17%, 9%, 3% and 3%
respectively for age groups 40 or less, 41-50, 51-60, and greater than 60,
respectively 96.

Chew and colleagues, in 1996, published a series of 349 patients with breast
masses who were evaluated at the Strathfield Breast Centre in Sydney,

Australia. They found an overall false-negative mammography rate of 9% in

patients with an average age of 60 years 109. In women under the age of 50,
the false-negative rate was 17%, which was statistically significant. The
average size of the cancer was 20.3 mm in mammographically-negative breast
cancer and 23.8 mm in mammographicaIIy-positive breast cancer, which was
not significantly different. This study is especially relevant because it was
published during a time when state-of-the-art dedicated film-screen
mammography units were used. Its results were corroborated by a study
published in 1998 by Dawson and colleagues that documented a 35% false

negative rate of mammography for women 35 years of age and lower who had

30

palpable masses, 110 and by Coveney et al, who found a 30% false-negative

mammography rate for women with palpable masses 40 years of age or less

106. That study documented sensitivity rates for mammography by age in

women with palpable masses of 70%, 75%, 85%, 92% and 90% for age groups

less than 40, 41-50, 51-60, 61-70, and over 70 years of age, respectively 106.

There are reasons for false-negative mammograms beyond the signs of
malignancy being hidden by dense breasts, thus resulting in a normal
mammogram reading. Coveney et al examined 291 cases of malignant
palpable breast masses diagnosed between 1986 and 1991 at St. Vincent’s
Hospital in Dublin, and documented a false-negative rate of 16.5%. Reasons
cited for the 48 false-negative cases included misinterpretation of overt signs of
malignancy on the films (20%), lack of recognition of subtle signs of malignancy
(34%), and mammographic findings mimicking benign disease (16%). The
mammograms were actually normal in only 30% of cases 106.

In one large study of the litigation literature, 68% or patients that successfully
filed suits against their physicians had either negative (49%) or equivocal (19%)
mammogram reports. Several independent authors have reported on the effect

of false negative mammograms on the diagnostic delay of breast cancer 107-

111'113, and cautions regarding pursuit of the diagnostic work-up of a breast
mass no matter what the mammogram result are standard.

In an interesting study by Lister et al published in 1998, they compared the
test characteristics of mammography and high-frequency ultrasound in a

population of 205 women with clinically benign masses who underwent triple

31

 

test evaluation and in whom the cancer prevalence was 7%. The mean age in
the study was 40 years but mass size was not discussed, and no biopsy rates
or follow-up durations were reported. However, one can compare test
characteristics of the two radiological tests on initial evaluation of the breast
mass. Sensitivity, specificity, and positive and negative predictive values for

mammography were 57%, 93%, 61%, and 92% and for ultrasound, 93%, 97%,

68% and 99% 114. These authors now use ultrasound as the radiological study
in the triple test. They limit mammography use in women with clinically benign
lumps to those with confirmed malignancy (to screen for occult mammographic
abnormalities prior to cancer treatment), to those with suspicious ultrasound
findings, or to those in whom a screening mammogram is appropriate based on
age and timing of the previous film. They comment, however, that ultrasound is
extremely operator and machine dependent, and caution that this change in
policy is only appropriate if local test characteristics confirm the improved
accuracy of the procedure. The use of ultrasound as a substitute to
mammography, or in addition to it, in the work-up of a palpable breast mass, is
beginning to be reported with increasing frequency. Reinikainen et al reported
on the use of high-frequency ultrasound in the evaluation of 84 palpable
lesions, 63% that were malignant, evaluated between 1996 and 1997. The
average patient age was 48 years and the average mass size on ultrasound
was 20 mm for malignant lesions and 18 mm for benign masses. In a blinded
analysis, ultrasound and mammography performance were compared.

Ultrasound and mammography performed equally well, with 97% sensitivity for

32

 

both modalities. In both cases, 1 of 37 malignancies had normal imaging
studies, but they were in different patients, both of whom were diagnosed with
lobular carcinoma. One of the greatest benefits of ultrasound in this study was
the increase in radiological suspicion of malignancy that occurred; 12 cancers
that were indeterminate on mammography were clearly malignant on
ultrasound. In addition, 57 cases were evaluated with FNAB, and 3 were false—

negative. Radiological imaging with either one or both modalities correctly

identified these malignancies 115.

Classification of Mammography Findings

The classification of findings in mammography has been reported in a variety of
ways, and a standard reporting system was adapted in the United States in
1992. This system, referred to as the Breast Imaging Reporting and Data
System (Bl-RADST”) was developed by the American College of Radiology 116.
lt assigns a numerical score to a range of mammographic categories between 1
and 5. BI-RADS is an accepted standard among radiologists dedicated to
mammography, predicting for malignancy as the categorical designation
increases numerically. Exceptionally reproducible results on mammogram
interpretation of 300 test cases has been documented, with almost identical
ROC curves among 3 different expert radiologists 117. Some of the literature
published after BI- RADSTM was released uses the terminology in its
classification of mammogram readings, while other systems have been used as
well. Most that do not use the Bl-RADSTM system classify mammogram findings

into categories of benign, suspicious, and malignant.

33

 

Test Characteristics of Mammography in Women with a Palpable Mass

The test characteristics of several studies that examined mammography
performance in the presence of palpable masses are listed in Table 3. As with
CBE test characteristics, many of the quoted values were derived from the data
given in the papers, as not every report calculated the pertinent test
characteristics. The categorization into positive vs. negative tests was based
on a benign (normal or benign mammogram) vs. not benign categorization
(indeterminate, suspicious or malignant reading) as discussed above. Some
papers categorized the test values differently from the guidelines outlined above
in the calculation of the test characteristics. For purposes of comparison , all
values given in Table 3 were recalculated and checked using raw data and
consistent categorization to ensure comparability.

One study in Table 3 specifically examined mammogram results in the context
of palpable masses 108, while some did so in the context of other diagnostic
combinations, but not as a component of triple test 100, 101. However, like
CBE evaluation, most of the studies in table 3 were done in the context of triple
diagnosis evaluation, and mammogram test characteristics analyzed as a
separate component of this diagnostic evaluation 7, 75’ 76, 87, 89'98. The
papers by DiPietro from 1985 and 1987 and Martelli from 1990 deserve special
comment. These 3 papers are from the same institution. The authors
considered several mammograms unsatisfactory, because the x—ray did not
visualize the lesion that was palpable because of breast density. However, they

appropriately included the readings as “negative” in the calculation of test

34

l./_

 

characteristics (false negative in the calculation of sensitivity and true negative
in the calculation of specificity). Most papers do not make the distinction of
unsatisfactory as a criterion for mammography findings in women with a breast
mass, but instead simply include the results in the calculations of test
performance as “negative”. Although for purposes of the table the results were
recalculated to include the unsatisfactory mammogram readings into the
negative results, the reports mentioned above provide helpful information
regarding reasons for false-negative mammograms.

Histo_ry and Clinical Breast Exam in Combination with Mammography

In 1997, Seltzer published a prospective study of 6,787 consecutive patients
whom he had evaluated as part of his surgical breast practice between
September 1987 and December 1993. He used the combination of history,
CBE, and mammography and committed himself to a clinical impression of
benign or malignant disease prior to intervention. Biopsy was performed in
2,247 patients, 1240 of whom presented with a breast mass. Two hundred and
sixty-six of the patients with a breast mass had cancer for a malignancy rate of
21%. The positive predictive value for the combination of tests in this setting for

this experienced surgeon was 0.68; 0.45 for women less than 50 years and

0.89 for those 50 years of age or more 6. The negative predictive value in
women with a mass for the combination of tests was 0.95 for those less than 50
and 0.89 for those 50 and older. In a similar study reported by Stems, the
positive predictive value for malignancy was different depending on the

combination of test results and the age of the patient. For women less than 50

35

years of age, the PPV was 88.5% for those with positive CBE and
mammography, 49.6% for positive mammography but benign CBE, and 42.1%
for positive CBE and benign mammography. When both tests indicated benign
disease, the malignancy rate was still 12.6%. For women 50 years or greater,
the respective values were higher at 92.2%, 66.7%, and 36.5%. The

malignancy rate when both tests were negative in the older women was only

slightly lower than in the younger women, at 12.2% 118.

The false-negative rate of the diagnostic work-up of a breast mass is too
high to stop the evaluation at the point of a benign CBE and a benign
mammogram. Although this point has been stressed repeatedly in the litigation
literature, the clinical literature also reflects unacceptably high missed cancer
rates if cytological or histological steps are not taken.

Fine Needle Aspiration Consistency

Experienced cytologists have written about the physical sensation
experienced by the aspirator when the tip of a needle enters a solid breast
mass. This internal consistency has been described as soft, rubbery, hard,

and/or gritty. Gritty lesions, especially those that are hard and gritty, are more

likely to be malignant 119» 120. A few authors have used this feature as part of
the evaluation of whether or not a palpable breast mass represents malignancy.
Somers and colleagues examined 105 lesions that they classified as
“subsuspicious” at the Albert Einstein Medical Center between 1984 and 1987.
They specifically excluded any solid, dominant masses and examined instead

lesions which they described as areas of prominence, thickening or tenderness

36

on CBE. Unfortunately, they do not break down the frequency of each
presentation and it is difficult to discern from their publication how many of
these lesions actually represented cysts that disappeared following aspiration.
All of the cases had “subsuspicious” CBE, mammogram and internal
consistency results. They described the needle sensation upon aspiration as
“rubbery firm, cystic or soft” as opposed to “gritty or indeterminate”. Twenty-five
percent of the lesions had a scant or insufficient FNAB; that was accepted by
the authors because they used the F NAB results as a confirmatory rather than
a diagnostic test in this study. However, the study design dictated that any
case of cancer that was identified utilized FNAB cytology or open biopsy in
doing so. One cancer was identified by aspiration of suspicious atypical cells
and infiltrating ductal carcinoma was confirmed on open biopsy. Six other
biopsies were performed that were benign. Twelve percent of the patients were

lost to follow-up; the mean length of follow-up for those followed was 61

months. No further cases of cancer were identified 121. Given that we have
very little knowledge of the characteristics of the masses in the study
population, the outcome of hypocellular FNAB results for breast masses under
any circumstances, and the 25% insufficiency rate and 12% lost to follow-up
rate of this study, conclusions regarding subsuspicious breast mass
management from this study are preliminary, at best. Nonetheless, it would be
safe to conclude that the malignancy rate of subsuspicious lesions is lower than

for otherwise dominant breast masses.

37

Fine Needle Aspiration Cytology
Classification of Fine Needle Aspiration Biopsy (FNAB) Findings

Different classifications of cytology findings have been reported in the
literature, but most use a 5-level scheme as follows: benign, atypical,
suspicious, malignant, and insufficient. There is debate, however, regarding the
atypical classification, how it should be reported, and how it should be

managed. One large multi-institutional study classified a finding of “few atypical

cells” as benign 122, while others consider any finding of atypia as an indication
for open biopsy. In the latter instance, however, the atypical findings are
usually classified as having characteristics compatible with indeterminate
findings. In the practice guidelines recently published as part of the 1996
National Cancer Institute sponsored workshop on fine needle aspiration biopsy,
recommendations for classification were made as follows: benign (no evidence
of malignancy); atypical/indeterminate (correlate with imaging characteristics
and clinical impression); suspicious/probably malignant (biopsy); malignant; and
unsatisfactory. This classification scheme, by a subcommittee of national
experts, appears to allow for clinical correlation in the management of atypical
findings. Two pages prior to this classification in the same report, however, is
this statement: “The presence of atypia on a cytologic or histologic preparation,

the degree to which is to be determined by the pathologist, warrants an

excisional biopsY' 123. In the calculation of test characteristics for F NAB, most

studies include the finding of atypia in the “positive” interpretation of the test,

38

 

unless the atypical cells are consistent with benign disease, as in the study by
Zarbo et al 122.

Test Characteristics of Fine Needle Aspiration

Many studies of the diagnostic efficacy of FNAB have been reported evaluating
its role in the vvork-up of palpable breast abnormalities, and the largest have
been selected for reporting and summarizing in Table 4. Some have been
done in the context of triple diagnosis, but most have not. Most studies
document greater test sensitivity than with either CBE or mammography, and

unlike either of these two tests, the sensitivity of the procedure does not appear

to vary by age 100. In an important multi-institutional study conducted in 1988
in which 294 institutions and 988 pathologists participated throughout the
United States, 13,066 breast FNAB specimens were examined, 33% that had
histopathologic correlation to form the basis for determining test characteristics,
and 41% that demonstrated malignancy. The sensitivity, specificity, and
positive and negative predictive values were 82%, 97%, 95%, and 86%
respectively 122. A second multicenter study from Italy reported on ten sites
and 23,063 FNAB cases observed between 1987 and 1993. Histologic
confirmation was available for 7,763 cases and a minimum of 6 month follow-up
for another 13,144 cases, with a malignancy rate of 26% for the reported
population. The sensitivity, specificity, and positive and negative predictive

values for adequate aspirations was 92%, 95%, 90%, and 97% respectively

124. In another large study from 6 sites in the United Kingdom, 1010 patients

39

with adequate F NAB readings and a 61% malignancy rate demonstrated similar

test characteristics of 92%, 86%, 91%, and 88% respectively 125.

Of the numerous studies of the test characteristics of FNAB that have been
reported, few report the lesion size that was aspirated or the age of the
population, and most compare the results with open biopsy rather than all
patients who underwent F NAB during the study period. Many studies combine
results of cyst aspirations with solid mass evaluation, and some studies include
readings with atypical cells in the “positive” category while others do not. In
addition, specimen inadequacy is handled in different ways by different authors.
Some include it in the calculations as a negative reading while others do not.
Table 4 presents a representative sample of published studies that presented
raw values for recalculation of test characteristics. For purposes of comparison,
atypical findings were calculated in the positive category, cyst aspirations were
eliminated, and insufficient specimens were dropped from the calculations.
Pitfalls of F NAB

When calculating test characteristics, most authors include both atypical
readings and suspicious readings in the positive category, and approach was
taken when calculating test characteristics for Table 4. The clinical application
of a positive F NA result, however, differs depending on the reading. For
atypical or suspicious readings, open biopsy is always done, whereas many
surgeons will perform a definitive cancer operation for a cytology reading which
is definitely malignant, considering confirmatory open biopsy unnecessary. In

fact. this was one of the first applications of the procedure in the category of

40

avoiding open biopsy. False positive results of “definitely malignant” readings
on FNAB are uncommon, but do occur, usually in predictable settings. These
include the hyperplastic and atypical findings of pregnancy and lactation, the
presence of apocrine cells simulating carcinoma, and the cellular features of

some fibroadenomas. The usual rate for false positive results of definitely

malignant readings is less than 1% 126.

False-negative results on FNAB is a well-recognized problem and occur
much more frequently than false-positive results. There are two major
categories of false-negative readings: those with adequate cellular
representation, and those without it. The National Institutes of Health guideline

on FNAB recommends that aspirators have an insufficiency rate of less than

20% to maintain credentialing privileges 123. It is not clear, however, what
literature was reviewed to arrive at this figure, as there is a wide range of
insufficient rates quoted in the literature. Few studies quote insufficiency rates
for consecutive aspirations, and the possibility of selection bias is high. For
instance, many studies site an insufficiency rate that applies only to those
specimens that were subjected to open biopsy. The insufficiency rate in this
group would be expected to be lower than in a consecutive series of patients.

The explanation for this prediction is based on the finding that masses that are

malignant have lower insufficiency rates than masses that are benign 124, and

benign masses are less likely to undergo biopsy. As an example, it is

instructive to examine a series reported by Park et al 127 that examined 669

consecutive cases of F NAB specimens. They reported a 25.3% inadequacy

41

 

rate in the entire series of patients. However, for calculation of test
characteristics, the authors had no follow-up mechanisms and relied on histo-
cytologic correlation. The inadequacy rate in the 198 patients who were
selected for biopsy was 15.6% 127.

Lazda et al report on three different audit periods between 1988-89, 1991-
92, and 1993 in one cytopathology laboratory in the United Kingdom. In audit
period I, 408 aspirations were done, 46.8% of which were read as inadequate.

An FNAB clinic was instituted between audit periods I and II and the

inadequacy rate dropped to 20%. However, it rose to 30.6% in 1993 128. The
malignancy rates following an inadequate result were 37.5%, 13.2%, and 7.1%;

the latter number reflects the usual rate, which ranges between 4% and 9%

128. It is higher, however, in some reports. In a series of 1708 cases by

Martelli at the Tumor Institute of Milan, the insufficiency rate was 407/1708

(24%), and of these, 34% were malignant 7. In an earlier study from the same

institution, there was a 31% insufficiency rate overall, and 16.1% of these cases
were malignant 89. This was similar to an inadequacy rate of 26% in a

publication by Watson et al from Galavvay, Ireland 129. The insufficiency rate

has been quoted as lower by other authors. Vifinchester et al quoted a 20%
insufficiency rate in their series 98. Horgan et al documented a 12.9%

insufficiency rate, and 4 of the 258 specimens were malignant 130. Frable had

a 6.0% insufficiency rate, and 7 of 42 of these lesions were malignant on open

biopsy 131. Kaufman et al had only a 3.0% insufficiency rate, and all of these

42

lesions were benign on open biopsy 93. In Zarbo and colleagues’ more

definitive multi-institutional study in the United States, the insufficiency rate of

294 institutions and 13,066 specimens was 18% 122. It was 19% in the multi-

institutional study in Italy, with 6.5% rate of inadequate readings for the

malignant specimens and 23.2% for the benign lesions 124. Other authors

have documented higher inadequacy rates with benign lesions than with

malignant ones as well 129. 130.
Watson et al have also demonstrated that age predicts for inadequacy

rates, with women less than 45 years of age having insufficiency rates of 14 -

26%, 48 - 55 years of age 13%, and over 55 years of age 1 — 5% 129. This is
probably a reflection of the fact that young women are more likely to have
benign disease, and therefore have a greater likelihood of inadequate
aspirations.

Reasons for insufficient specimens are multifactorial. Some aspirated
lesions simply do not contain epithelial cells. This is entirely appropriate for
nonepithelial lesions of the breast. In these cases, the aspiration may
accurately diagnose the lesion as benign. At times, insufficient aspirations can
be correlated with clinical and mammographic findings to arrive at a diagnosis

of adipose tissue, scar tissue, radiation changes, or other reasons for epithelial

paucity 132. To illustrate this point, Abele and colleagues demonstrated in a
unique study in 1983 that direct aspiration of surgical specimens that were

nonproliferative yielded a paucity of epithelial cells, despite repeated aspirations

133'1 35. In another interesting study, Wolberg et al found 66 of 464 FNAB

43

results (14.2%) to demonstrate a paucity of or no epithelial cells and to
represent nonepithelial benign conditions after correlation with clinical and
radiographic findings. All of these were followed for 12 months with no
malignancies identified. However, he also documented 20 cases of inadequate

specimens (4.4% of the sample) that were not considered definitely benign.

Fourteen of these were benign and 6 were malignant 136. Concluding that
insufficient results represent benign findings can therefore be a dangerous
assumption, especially in inexperienced settings 137. Recognizing this, some
authors consider an unsatisfactory or insufficient specimen to be the equivalent

of no FNAB and believe that a reading of insufficient cells should not be used at

all in medical decision-making 137: 138. This attitude was adopted in the
construction of the test characteristics of FNAB presented in Table 4, and
insufficient specimen results were not used in the calculations presented.
Insufficient specimens that do not obtain true cellular representation of
the components of a lesion represent true false negative results. The most
common reason for this is technical, most often due to inexperience of the
aspirator. lnexperience may be related to inadequate technique of aspiration,

geographic misses of the target lesion due to small size, lack of circumscription,

or lesion location 139. Barrows and colleagues demonstrated an insufficiency
rate of between 6 and 21% among aspirators attempting to obtain

representative cells from malignant lesions on FNAB if the physician had

aspirated fewer than 70 cancers 125. Zarbo and colleagues, in the multi-

institutional study conducted in the United States, documented a 17%

44

inadequacy rate overall, but only 7.2% in specimens obtained by a pathologist
in contrast to a clinician, which they correlated with experience. However, they

also pointed out that only 5% of the over 200 institutions in the study exclusively

used pathologists for FNAB performance 122. Layfield also documented

greater FNAB accuracy rates when a cytopathologist performed the aspiration,

as discussed above 139. One must consider that the spectrum of disease in a
population, and the age of that population, are Important considerations in all
test interpretation characteristics, including inadequacy rates of F NAB. A series
with a high malignant: benign ratio should predict for a lower insufficient
specimen rate than a series with a low benign: malignant ratio. Does this
explain some of the discrepancy that is found between series that report
insufficiency rates of aspirations by surgeons as opposed to cytopathologists?
lnvariably, surgeons have an inadequacy rate of around 20%, as opposed to 5-
7% for cytopathologists. Surgeons, however, are more likely to perform
aspiration on consecutive patients, and therefore to include a greater number of
“subsuspicious” lesions than might be referred to the cytopathologist for
aspiration. This raises the possibility of a possible selection bias accounting for
the differential rates of inadequate specimens reported in the literature.
Investigation of this possibility awaits a future study.

Rather than the chosen specialty of the aspirator, the amount of clinical

experience and confidence of the aspirator have been shown to relate to

aspirated material representing the lesion in question 100. 125. Dixon reported

on FNAB inadequacy rated during two time periods, one during which multiple

45

 

aspirators were used to perform F NAB and the second during which only one

aspirator was used. The inadequacy rate dropped from 22% to 10%, and the

sensitivity improved from 81% to 99% 100.
The actual criteria for specimen adequacy, or inadequacy, is open to

debate. Layfield has suggested that a minimum of 6 clusters of epithelial cells
per case, composed of at least 5 cells each, be present 137, while others have

suggested 4-6 clusters of 5-10 cells each 140’ 141. The recently published
guidelines from the National Institutes of Health regarding fine needle aspiration

specifically state that there is no national standard requiring that a given

number of cells be present for specimen adequacy 123.

Salami et al reported on the outcomes of 61 patients who had inadequate
FNAB readings done for palpable findings at the University of California at Los
Angeles in 1995. These inadequate results represented a specimen
inadequacy rate of 23%, 21% attributed to misdirected biopsies and 1.6% to
misinterpretation. The mean size of the mass on CBE for the inadequate
results was 1.3 cm and the average age of the patient was 46 years. Twenty
cases had a histologic diagnosis and the rest were followed for a minimum of
24 months. A retrospective review confirmed the lack of epithelial cells in 34
cases, 2 of which were malignant. Seven cases actually fulfilled the criteria for
adequacy. and one of these was malignant. Twenty cases had some epithelial
components, but less than six clusters required for the study. These were all

benign. The study is unique because it is the first to examine the outcome of

inadequate FNAB results in patients with palpable abnormalities 142. It

46

 

 

confirms that in the 3 cases of a false-negative F NAB, that open biopsy was
performed on the basis of an abnormal CBE or mammogram.

There are other reasons for false-negative F NAB results. Lesion size is
one. A small target lesion can be missed by the aspirator, resulting in a
sampling error. Barrows et al reported on 689 breast cancers divided by 1 cm
size increments. They found that the percent of positive or suspicious aspirates

rose incrementally from 70% for 1 cm lesions, to 80%, 85%, 90% for 2 cm, 3

cm, and 4 cm and over sized lesions 125. Less dramatic but consistent findings

were reported by Ciatto, with sensitivities of 92% for lesions up to 2 cm, 95% up
to 5 cm, and 96% for over 5 cm 124. Paradoxically, some investigators have
found an increased false-negative rate when the lesion size is over 4 cm 139’

143. This is attributed to the presence of large areas of necrosis, hemorrhage
or fibrosis within a lesion.

There is also well-documented difficulty interpreting some aspirates,
even if the material is adequate and representative. Reasons for false negative
interpretations include specific pathology such as low-grade in-situ malignant
lesions, well-differentiated invasive ductal carcinoma, tubular carcinoma,

papillary carcinoma, comedocarcinoma, colloid carcinoma, and invasive lobular

carcinoma 123. 130- 139. It is interesting that young women are more likely to
present with smaller masses than older ones, and there is a correlation
between mass size and tumor differentiation. The probability of a false-
negative reading is correlated with tumor differentiation such that the more

differentiated the tumor, the higher the likelihood of a false-negative F NAB

47

reading 126. Therefore, a higher proportion of young women will have a false-
negative F NAB result because of this reason than older women. Because the
overall proportion of women with this profile will be low, however, test
characteristics are not likely to be affected to any great extent.

Zarbo et all was able to quantitate the false-negative rate of FNAB due to
sampling errors (7%) vs. interpretation errors (0.8%). Although this may appear
to be a relatively low rate, absolute numbers bring it into perspective. In their

nationwide study done over a 6 month period, there were 261 occurrences of

patients with breast cancer who had false negative FNAB results 122. That
false-negative FNAB can lead to a delayed diagnosis of breast cancer is
underscored in a study by Troxel et al that identified false-negative FNAB
cytology of breast lesions to account for 10% of pathology-related malpractice
claims 144.

False-positive results can also occur. False-positive interpretive errors are
usually due to poor cell fixation or inadequate staining, or misinterpretation of
findings, usually of proliferative benign lesions, some with atypia 139. This is
especially true of fibroadenomas, and this can influence clinical management.
For instance, few would insist on the removal of a fibroadenoma with a negative
FNAB unless the patient requested it. Dixon et al have shown that over 33% of
fibroadenomas become smaller or resolve when followed and that conservative
management of fibroadenomas is appropriate in 90% of cases 145. However,
the cytologic features of fibroadenomas on FNAB can be confusing 139: 146.

The lesion is moderately cellular and can have atypical features, and in some

48

cases this will require biopsy to exclude malignancy 139’ 147. The clinical and
mammographic features of fibroadenomas are generally easily differentiated
from those of carcinoma by virtue of their well-circumscribed, hypermobile
clinical characteristics, but sufficient overlap exists between these features and

medullary carcinoma and small invasive ductal carcinomas to require biopsy in

the presence of atypia 147. 148. Maygarden et al reported on 48 FNAB results
in women 30 years of age and under, 8 of which had atypical features. Two of

these represented fibrocystic changes, 2 were fibroadenomas, 3 were

infiltrating ductal carcinomas, as 1 was a medullary carcinoma 149. Clinically, 2
of the 4 malignant lesions were clinically benign, underlining the need for open
biopsy in cases of atypia.

In conclusion, even if specimen insufficiency is eliminated, FNAB has a
false-negative rate that precludes its use as the only method of ruling out
malignancy and a false-positive rate that results in excessive biopsy rates. This
has lead investigators to use a combination of methods to achieve greater
accuracy in the diagnostic evaluation of a breast mass.

Triple Diagnosis

Classification of Triple Test F Mpga

Triple diagnosis has traditionally referred to the application of the combination
of three diagnostic tests in the work-up of a breast mass: clinical impression,
mammography, and fine needle aspiration cytology results. The results of each
individual test have been classified in the same ways as has previously been

discussed. Its modern application involves clinical decision-making for test of

49

concordance both for malignant and for benign results. That is, definitive
cancer operations are often done for malignant results of the triple test, and
follow-up has been recommended for benign results. Discordant findings
between tests when the triple test is applied indicate the need for open biopsy.
Some authors have attempted to devise a scoring system for triple diagnosis
in order to increase diagnostic accuracy. In the first such attempt, Smallwood
et al conducted a study of triple diagnosis in the breast clinic at Southampton
General Hospital in 1981. Their system assigned a numerical score of -2 for a
definitely benign mass, -1 for possibly benign, 0 for equivocal or inadequate
assessment, +1 for probably malignant and +2 for definitely malignant for a
scoring range of -6 to +6. The scoring system was applied to the results of
clinical examination, mammography, and FNAB, but clinical examination was
not reported. The inadequacy rate of F NAB in the 244 lesions assessed was
34% (these lesions were assigned a 0 in the scoring system) and the
malignancy rate in the population studied was 40%. The authors report the
results of the scoring system but do not indicate a cut-off point for h'iage
purposes as all patients underwent open biopsy at that time. However, if the
score was +4 or above, definitive surgery was conducted without frozen section
and was accurate for malignancy in all of the 72 cases. No patients with scores

of -4 or below had malignancy; 65 patients fell into this group. Of the remaining

60 patients with scores between -3 and +3, 27 (45%) had malignancy 150.
No further publications regarding scoring systems appeared in the literature

until a follow-up series on Vetto’s original paper was presented at the Pacific

50

Coast Surgical Association in February 1998. In it, Morris et al departed from
the attitude that open biopsy was necessary whenever there was a discordant
result of the components of triple diagnosis. In a series of 261 breast masses
studied between 1991 and 1997, they assigned a scoring system between 1
and 3 to each of three components of triple diagnosis, corresponding to benign,
suspicious, or malignant results. This allowed for a possible score between 3
and 9. The mean age of the patient was 49 years. The authors did not
comment on mass size. One-hundred twenty-five masses scored a 3,
indicating concordance for benign results. Twenty-three of these underwent
open biopsy and 112 were followed for a mean duration of 18 months. All were
benign. Seventeen cases had a score of 4, indicating discordance with one of
the components of triple diagnosis. These all underwent open biopsy and all
were benign. A score between 6 and 9 was assigned to 88 masses, and all of
these were malignant. A score of 5 was assigned to 21 masses (2 of the 3
components of triple diagnosis suspicious and 1 benign, or 2 benign and one

overtly malignant). Of these, 8 were malignant on open biopsy and 13 were

benign 94. The authors concluded that scores of 4 or less could be
appropriately managed by follow-up. They acknowledged that FNAB cytology
has greater diagnostic accuracy than the other components of triple diagnosis,
and commented that they did not want to develop a “weighted” system of
scoring for the sake of simplicity. They justified their recommendations of
avoiding open biopsy, even in the case of a suspicious FNAB reading, by citing

that the 17 cases that fell into this category in their series were all benign, and

51

by discussing the reasons for false~positive FNAB results. In a provocative
discussion following presentation of this paper, Butler questioned the
recommended approach for scores of 4, especially if the abnormal result was a
FNAB. He stated that in 18 patients studied with these findings at his
institution, 1 case was malignant. He also questioned the advisability of
proceeding with definitive therapy in cases of scores of 6 (patients with
suspicious results for all 3 components of triple diagnosis, or in those with
malignant CBE or mammography results but benign FNAB findings). Other
discussants questioned the criteria for interpretation of CBE and whether the
results of the study were generalizable, certainly valid concerns that had been
raised by other studies as well. In the reply. Dr. Vetto addressed the question
of CBE by commenting that the evaluation is attempted blindly, “sometimes
without even looking at the mammogram”. Although this reply may not have
fully answered the question, it does reflect the reality of most surgical practices.
That is, that when a woman is evaluated for a breast mass, it is difficult to
separate the mammogram findings form those of physical exam when arriving
at a clinical impression. In fact, this is openly acknowledged in a paper by
Cardona and colleagues, who state that “PE (physical examination) sensitivity
will be overestimated when PE is performed after other diagnostic tests, such
as mammography, because the clinician may be influenced and alerted by
radiologic evidence of a suspect lesion.” This study has attempted to address
this problem, as well as to specifically define the characteristics of malignancy

used during clinical examination.

52

Test Characteristics oftIhe Triple Test

Numerous authors have reported the results of the application of triple
diagnosis, and Table 5 documents the test characteristics of a representative
group of studies. The same criteria for calculation of results from raw data was
applied as in the previous tables. Some studies only included results for

malignant diagnoses, and in these cases, only the sensitivity for the triple test is
indicated. Two studies had purposely low malignancy rates of 0.4% 151 and

1% 121. These two studies specifically examined lesions that they had
assessed as not requiring open biopsy. The positive predictive value for these
studies would be expected to be much lower than in those that had higher
disease prevalence, and this expected result was verified in the calculations.
However, the other test results are comparable with the rest of the literature.

It is important to recognize that the usefulness of triple diagnosis has
evolved over time, and its results were applied in Europe well before the United
States. Initially, its major strength was recognized as its ability to increase the
sensitivity (that is, decrease the false-negative rate) of the diagnosis of

malignancy and to reduce the need for frozen section confirmation of

malignancy prior to definitive surgery 150. In one of the first studies done in
1976 in Berlin, Germany, Kruezer and colleagues viewed the triple diagnosis
test as a three-way test of concordance. They recommended definitive surgery
in cases when all 3 test components were definitely malignant. In addition, they
advised that if all three components were definitely benign, that close follow-up

was acceptable to open biopsy. They did not define the parameters for follow-

53

 

up. Lastly, they advised that if any of the components diverged or results were

suspicious, that open biopsy was warranted. Using these guidelines, they

found a diagnostic error of less than 1% 75. It should be pointed out that the
criteria for benign findings on CBE were any mass 3 cm in size or less that
lacked advanced signs of breast cancer. The population in which their study
was done was probably representative of many in the early 1970s, but the
presentation of a breast mass in 1999 is far different than 25 years ago, as
discussed above. The recognition of this fact, as well as the high malignancy
rate in their population (41 %) obligates one to question the generalizability of
the findings of that report. In a study done in Denmark in the early 1980s,
Hermansen reported on 22 solid masses, 61% that were less than 2 cm. Their
malignancy rate was 26%. They had a 100% sensitivity with triple diagnosis,

but cautiously concluded that before open biopsy could be avoided, that a

larger study would be necessary 37.

A follow-up study was published 3 years later by the same authors. It
included 458 palpable solid breast masses with a malignancy rate of 31%. In
this study, the diagnostic accuracy for malignancy was enhanced over that of
any one of the test components alone, as has been found by many authors.
For example, 13 lesions had a false-negative mammogram report, and 10 of

these had either suspicious or malignant findings on FNAB. The remaining 3

cases were accounted for by suspicious findings on CBE 92. This again
resulted in a test with 100% sensitivity. They concluded at this point that if all 3

test components were benign, that open biopsy “was superfluous”. They also

54

 

acknowledged that it was inevitable that a false-negative result of the triple
diagnosis would occur, and that clinical follow-up was warranted. Again,
however, no guidelines regarding follow—up were provided.

Despite the initial enthusiasm for triple diagnosis, some authors have not
achieved 100% sensitivity with application of the technique. This leads to
debate regarding its use. DiPietro reported a series of 631 solid breast masses

without obvious signs of malignancy on CBE from Milan, Italy in the early 1980s

and found a 7% false negative rate with triple diagnosis 89. In another series
from the same institution reported in 1990, Martelli reported a 4.8% false
negative rate using the triple test. This lead that institution to adopt the policy

that open biopsy of any solid mass that persists on 1-2 month follow-up be done

in women over 30 years of age 7. Butler et al also recommended that open

biopsy be done on persistent breast masses, even though their study

documented a 100% sensitivity for triple diagnosis 90. Steinberg et al use
similar three-way concordance criteria for biopsy or follow-up in a more modern
study, otherwise recommending open biopsy. However, they fail to define the
components of CBE evaluation and do not comment on mass size or population
age. They found a 1% false-negative rate with a benign triple test and

emphasize the importance of follow-up, again providing no guidelines for what

is meant by this recommendation 95. Vetto et al, in a modern series in the
United States, found 100% diagnostic accuracy when all three components of
the triple diagnosis were concordant. They recommended definitive therapy for

those that were malignant, avoiding the need for diagnostic biopsy in these

55

cases. They also recommended that if all three results were benign, that follow-

up should be done rather than biopsy. For discordant results, biopsy was

recommended 97. Most authors agree with this recommendation. Layfield has

estimated that using this approach, that 50% of open biopsies could be

eliminated, and follow-up in compliant patients be done instead 152. He
recommended 3 — 6 month follow-up but does not specify a duration of time
over which this should occur.

As would be expected considering that both CBE and mammography are
components of triple diagnosis, results are influenced by age. Dawson and
colleagues studied 68 patients 35 years of age or less diagnosed with breast
carcinoma between 1984-1996. Sixty-five patients had both mammograms and
FNAB performed in addition to CBE. Twenty-three of these (35%) had benign
CBE results and mammogram findings. The diagnosis of cancer was made by
FNAB in each of these 23 cases, with readings of malignancy, suspicious for
malignancy, or atypical. There was one false-negative F NAB in this study, but

on retrospective review, the authors comment that the side was misinterpreted

as benign and should have been read as atypical 110. In another series
reported by Yelland et al, 81 patients under the age of 36 were identified with
breast cancer between 1971 and 1989. The average tumor size was 3 cm.
The masses were evaluated as malignant in only 58% of the clinical exams by
the surgeons. In a unique analysis, the authors were able to identify that only
25% of the general practitioners who initially examined the patients considered

the mass malignant. Mammography was positive in only 24% of the clinically

56

benign masses and 80% of the clinically malignant ones. FNAB results were
positive in 56% of the clinically benign masses and 80% of the clinically
malignant ones. These dismal statistics were improved upon when triple
diagnosis was applied, but the sensitivity was still only 84%, and 10 of the 63
patients evaluated had benign results on all three tests. Open biopsy was done
in these 10 patients based on inability to follow the patient. One might question
the experience of the clinicians who published this report until the institution in
which it was conducted is examined. The series of cases were derived from the
Combined Breast Clinic at St. George’s Hospital in London, which had been in
existence for 19 years prior to patient identification. The results might be
explained by the technological failures of mammography at the time that the
study was conducted. Mammography techniques have certainly improved over
the years, but are unlikely to ever achieve near 100% sensitivity in young
women. It is easy to see why some surgeons believe that open biopsy should
be performed on all women who present with a breast mass. The authors of
this paper discussed the fact that 9,768 patients less than 36 years of age
consulted the breast clinic during the same time period as this study. Assuming
no false negative rates in this population, 99.2% of the women seen in this clinic
under the age of 36 will be diagnosed with benign disease. If they all had
masses and open biopsy was required on them all, the operative burden would
have been 10 cases per week. The clinic currently is using ultrasound to
assess masses in women who have negative mammograms, as are other

centers 86- 109: 153.

57

The effect of small tumor size on the individual components of triple
diagnosis was investigated in a series of 21 invasive carcinomas from Osaka,
Japan that were less than 1 cm in size. All of the tumors were palpable,
although 57% were thought to be benign on CBE. The mammogram was
benign in 17% of the cases and FNAB benign in 4 cases (19%). Of the 4 that
were benign, CBE was suspicious in 2, equivocal in 1, and negative in the

other. Mammography was suspicious in this latter case. No false negatives

were found 154. The authors comment that breast size in Japanese women is
on average smaller than in women in the United States and CBE and FNAB
may therefore be more accurate in Japan than in other settings. The
performance of triple test on women with small breast tumors needs further
study. The only other publication that was found addressing this issue at all
was that of Di Pietro et al from 1985 in a large study of Italian women. In this
study, 24 malignant tumors measured 1 cm or less on physical examination.
The authors comment in the discussion of the paper that the positive predictive
value for fliese small tumors was found to be between 30% and 40% for CBE
and mammography, and 80% for FNAB. No data was given, however, and the
outcome of the triple test for this subgroup was not discussed.

Breast cancer is more challenging to diagnose as the size of the tumor
decreases, especially on CBE and on mammography in women with dense
breasts. In fact, FNAB has been cited by some authors to be especially helpful
in these settings. Very few studies have done comparisons of test performance

according to tumor size. Langmuir provided some insight in her study by

58

analyzing the patients who had cancer and providing data on test performance.
In that study, CBE was abnormal in 75%, 90%, and 100% of patients with
cancer
Follow-up after Triple Diagnosis

Although most authors who have published about the use of triple diagnosis
to triage patients from open biopsy to follow-up have emphasized the
importance of follow-up, few have provided guidelines for what constitutes
follow-up. The large experience of DiPietro et al, documenting false negative

rates of triple diagnosis between 4.8% and 7% make it logical that follow-up will

be required for identification of these cases 89. Follow-up recommendations

and/or practices have ranged from routine screening through the primary care
provider 94, to open biopsy of all persistent masses at 1-2 month follow-up 7-
89. Some authors recommend open biopsy if the mass persists, but allow for
longer, but often unstated, periods of observation 90’ 98. Abele and colleagues
published a series of benign FNAB results that they confirmed as benign using
the standards of benign biopsy or no change on 6 month follow-up 133. Ciatto
reported a modern series using the same criteria for confirmation of benignity of
FNAB 124. No additional malignancies were reported in either study. Barrows
et al used similar criteria in their study of triple diagnosis with the exception of a
12 month follow-up time period 125. They also reported no follow-up malignant
lesions. Brown et al reported on 526 patients with breast masses assessed by
triple diagnosis between 1986 and 1990, one eventual cancer was initially

benign on all three components of triple diagnosis. This patient was diagnosed

59

on follow-up 3 months after the original presentation by an increase in the size

of the mass. No other malignant cases were identified on follow-up, but the

number of cases of follow—up were not reported 155. Boerner and Sneige
reported on 2 false-negative cases of triple diagnosis. One woman requested
an open biopsy and a 7 mm invasive ductal carcinoma was found. The second
was followed and at 5 months, a dramatic increase in the size of the original

mass was found. It proved to represent a case of a 4 cm ductal carcinoma in-

situ 85. The authors comment that unbiopsied masses were followed for two
years, but do not comment on the results beyond the two cases mentioned
above. Rimsten reported on one false negative case by palpation and FNAB

that was diagnosed 5 months later because of an increase in the size of the

mass 82. In summary, lesions that are followed after initially judged benign by
CBE, mammography and FNAB are usually benign, with about a 1% false
negative rate in the literature that is available from modern series. The follow—
up that is deemed necessary to capture the 1% of cases is less well defined,
but it appears that the cases identified on follow-up have all been found within 6
months. It is still possible that a slovv-grovving cancer may take longer to
increase in size than 6 months, but if there is no change in 12 months, it would
appear that follow-up would be complete. This definition is used in calculating
false-negative rates for screening mammography, and it is standard to consider

symptomatic cancer diagnosed within a 12-month period after a normal

screening mammogram as evidence of a false-negative result 105

60

The most specific guideline for follow-up that is published corroborates this
conclusion. It represents a 1996 expert panel consensus from a National
Cancer Institute sponsored workshop on fine needle aspiration biopsy. A 3 to
6 month follow-up after a benign triple test is recommended, “followed by repeat
examination within an additional 12-month period, depending on subsequent
clinical findings”.

Previous Use of Likelihood Ratios in Women wig a _Byeast Mesa
Barton studied the likelihood ratio of CBE in the presence of a breast mass.
When a mass is palpated on CBE, the likelihood ratio for malignancy was

demonstrated to be 10.6 in this study, compared with a negative likelihood ratio

of 0.4762. This relatively high negative likelihood ratio in the face of a normal
CBE reflects the fact that breast cancer is often diagnosed on the basis of an
abnormal mammogram. Giard and Hermans studied the likelihood ratios for 4
different values of FNAB in 996 patients. They defined the values as follows:
Infinity for those patients diagnosed as definitely malignant; 4.81 for those

diagnosed as suspicious; 0.11 for those diagnosed as benign, and 0.41 for

those diagnosed as insufficient 155.

Sum ma_ry of Literature Review

The influence of the risk factor profile on the probability of disease in a
symptomatic population has not been studied to date. Numerous studies
testing the sensitivity, specificity, and positive and negative predictive values of
CBE, mammography, and fine needle aspiration cytology in the work-up of a

breast mass have been published, and none of these alone will result in

61

 

sufficient rates of breast cancer diagnosis to be used exclusively. Improved
accuracy of breast mass diagnosis using the principles of triple diagnosis have
been demonstrated by a number of large academic centers throughout the
world. However, these centers often have dedicated surgeons, radiologists,
and cytologists to contribute a multidisciplinary approach to the diagnosis of
breast cancer. This raises questions regarding the generalizability of the
published results. In addition, Giard and Hermans have commented, in a study

of the results of 29 studies of F NAB test performance, that any test that is

operator-dependent is likely to suffer from publication bias 157. That is, that
study results are unlikely to be submitted for publication if they do not reflect
good, or at least perceived adequate, test performance. This insight raises
questions about the applicability of FNAB in the typical general setting to which
most patients with breast masses present, and these and other authors

recommend the development of individual performance characteristics for

persons involved in FNAB 136: 157. This recommendation could easily be
extended to all of the components of triple diagnosis.

The published studies also suffer from lack of population characterization;
most fail to report average mass size or age of the population studied. When
mass size is quoted, the results from histopathology examination rather than
CBE results are often used, and sometimes only the size of the malignant
results are reported. The spectrum of disease, whether presenting in early or
advanced forms and in what proportions, can influence the test characteristics

of any diagnostic test, and age has an influence on the prevalence and test

62

characteristics as well. Studies of CBE, mammography, FNAB, or the triple
test, with few exceptions, include extremely high prevalence for cancer of
between 26 and 66%. Most of these rates do not reflect the prevalence for
cancer in most practices, even in most breast centers, and indicate a probable
selection bias among the patients to whom the tests were applied. If a
population is selected that includes a greater proportion of malignant cases
than is reflective of the patients with breast masses in general, this selection
bias can effect the measured test characteristics. Positive predictive value, for
example, will be highly dependent on the prevalence of disease; as the
prevalence decreases, so will the positive predictive value (and conversely, the
negative predictive value will increase). To illustrate this point, it is instructive to
examine a report by Logan-Young et al that selected a large population of
women with breast masses to study whom had characteristics of benign
disease, rather than malignancy, on CBE and mammography. In this study, the
prevalence for cancer was only 0.4%. These patients all had FNAB as part of
their vvork-up, and had biopsies or follow-up for 12 months. Nine cases of
cancer were found in a population of 2,247 women, eight after suspicious or

malignant FNAB results and 1 after 6 month follow-up revealed an enlarging

mass 151. The positive predictive value for triple diagnosis in this study was
calculated to be 32.0%. Given this, one might conclude that open biopsy is
necessary in all patients with a breast mass because only 32% of the cases

with a positive test actually had malignant disease. This conclusion would have

63

 

resulted in the open biopsy of 2, 238 women who did not have malignancy. The
implications to the health care community and society at large are obvious.

A comment also needs to be applied to the subject of the “gold standard” in
measurement of test accuracy. In breast mass vvork-up, there is no question
that the gold standard is open biopsy. This presents a problem because unless
every mass that presents is biopsied, test specificity will be slightly
underestimated, and sensitivity overestimated. In other words, those patients
that do not have the “gold standard” diagnostic test will be dropped from the
analysis. Therefore, the studies that report test characteristic data only on
patients who had biopsies do not reflect the true test characteristics for the
population that presents with a breast mass. A large series by Rimsten et al will
illustrate this point, and give another example of the effect of prevalence on
positive predictive value. These authors reported their data for FNAB two
different ways. One examined test characteristics in women who underwent
open biopsy (N = 381, prevalence of malignancy 28%), and the other examined
the test characteristics in the entire population that underwent FNAB (N = 986,
prevalence of malignancy 11%). The women who did not undergo open biopsy
were followed for 24 months. One cancer was found at follow-up 7 months
after the negative FNAB. The sensitivity, specificity, and positive and negative
predictive values for FNAB in the population that underwent open biopsy was
96%, 85%, 71 %, and 98% respectively. The test characteristics for the
population of women who underwent open biopsy with the addition of the

followed patients were 95%, 91%, 58% and 99%, respectively. There are no

64

well-defined criteria for a substitute gold standard short of biopsy. Most authors
agree that a diagnosis of cancer can be ruled out (therefore eliminating false-
negative test results) after a reasonable period of follow-up. The follow-up
period varies by study, however, and is often not defined at all.

This study was designed to help sort out these problems. It includes all
patients who met eligibility requirements for the triple test, whether they had a
biopsy or follow-up. It does not select for patients in whom malignancy was
suspected, and includes a minimum follow-up period of 12 months as part of
the eligibility requirements. It attempts to characterize the population in detail
and to clearly define the criteria used to classify CBE findings. Although the
patient population is derived form a university-based Breast Center, the large
number of self-referred patients and the volume of referrals for anxious women
decreases the pro-test likelihood of disease, or prevalence, to a level that is
more likely to be reflective of average surgical practices throughout the country
rather than a tertiary care center.

MATERIALS AND METHODS

uantitative Scorin S stem: Prelimina Work:

 

In academic year 1993-94, a pilot study in the Comprehensive Breast Health
Clinic (CBHC) at Michigan State University was conducted to investigate a
method to quantify the principles of triple diagnosis. The goal of the study was
to design a quantitative model for each of the methods used for breast mass
evaluation in the CBHC with the purpose of arriving at a summation score. The

intent was for the score to be used to predict for benign vs. malignant disease

65

and to be used prospectively to triage patients between open biopsy and follow-
up. A numerical score was assigned to each of 4 separate assessments of
breast masses, which included

(1) clinical impression of the breast mass on CBE

(2) mammography results

(3) F NAB results and

(4) assessment of mass texture when aspirated.
For clinical impression on CBE categorization, three numerical scores between
1 and 5, 1 for benign, 3 for indeterminate, and 5 for suspicious for malignancy
were assigned. Components of the clinical impression used in the
categorization included the risk factor profile, the history, the CBE results, the
mammogram assessment by the surgeon, and the internal texture of the mass

when fine needle aspiration was performed. For the CBE results, categories

similar to those of Di Pietro et al were collected 91.

The BI-RADS lexicon system of the American College of Radiology was
used to categorize the mammography findings. This system includes the 5
categories found below.

(1) Category 1 — normal;

(2) Category 2 —benign finding;
(3) Category 3 — indeterminate;
(4) Category 4 - suspicious;

(5) Category 5 — malignant.

66

 

For categorization of fine needle aspiration biopsy cytology, the literature was
reviewed and the most commonly observed categories were assigned
numerical scores between 0 and 5 consistent with their positive predictive
value. Scores assigned were as follows:

(1) 0 — no epithelial cells; adipose tissue;

(2) 1 — benign cells;

(3) 2 — slight atypia, consistent with benign disease;

(4) 3 — many atypical cells;

(5) 4 — suspicious;

(6) 5 — malignant.
For textures of aspirates, three numerical scores between 1 and 5 were
assigned, based on descriptions by Roberts 119 and Kline 120.

(1) 1 - soft or rubbery;

(2) 3 - soft or rubbery but gritty;

(3) 5 - hard and gritty.
To conduct the preliminary study, the cases seen in 1992 by a single surgeon
whose patients had been evaluated only by her for breast masses and for
whom a complete data set was available were analyzed. Numerical scores
were assigned for each of the 4 evaluation components as discussed above
and a summation score obtained. One hundred thirty seven masses in 103
patients were evaluated; all had either open biopsy or a minimum of one-year
follow-up. A one year time was chosen based on the natural history of the

growth of palpable breast masses that represent carcinoma and on a literature

67

review of screening mammography standards 136, 151. Sensitivity, specificity,
negative predictive value (NPV), and positive predictive value (PPV) were
individually calculated for each of the 4 categories, for different score
assignments in each of the categories, and for a summation score once optimal
numerical assignments had been determined. Possible scores ranged between
2 and 20. A cutoff score of 7 or less essentially ruled out breast cancer (N =
124), and a score of greater than 7 had a sensitivity of 100% and a specificity of
98%. The results of the preliminary study can be found in Table 6. All
malignancies were diagnosed within 4 weeks of presentation. Based on these
results, a prospective study was begun, which is the topic of the remainder of
the discussion.

Study Design:

Data Collection

To develop a mathematical model to predict for breast cancer using the
principles of triple diagnosis, a prospective database was constructed. Two
forms were developed to collect demographic, epidemiologic, historical, and
clinical data. The epidemiologic portion of the database was completed by the
patient at the time of the initial evaluation in the CBHC and pertinent information
updated at each visit. Epidemiologic variables collected for analysis included
age, height, weight, body mass index, personal history of breast cancer,
reproductive factors (age at menarche, parity, age at first delivery, number of
deliveries, menopausal status, age at menopause), exogenous hormone use

(oral contraceptives, estrogen replacement therapy, progestin replacement

68

therapy, including duration of use), family history of breast cancer (first and
second degree relatives, including age at diagnosis and unilateral or bilateral
involvement), history of childhood or early adulthood radiation therapy, and
tobacco and alcohol use, including duration and frequency. The follow-up
database contained clinical data appropriate to each patient visit and allowed for
ongoing prospective data to be collected from each of the clinicians practicing in
the CBHC. Clinical variables collected included mass characteristics (size,
shape, texture, mobility, consistency, mammogram results, results of fine needle
aspiration biopsy (internal consistency and cytology results), and clinical
disposition of the patient. Additional variables were collected for interest but
were not a major focus of the study. These included ultrasound results and
mammography density in the area of the mass. When the forms were not
complete, data was extracted from the medical record using clinic notes and
letters to referring physicians. This occurred for at least one variable in
approximately 15% of the cases. Variables related to the diagnostic evaluation
which are routinely included in assessment by triple diagnosis were defined
according to criteria detailed in the following sections.

Components of Clinical Impression:

The first evaluation component of triple diagnosis involves the clinical

impression of the physician examining the patient and represents the

overall probability of cancer based on a variety of factors assessed during

the clinical encounter. These include assessment of the risk factor

profile, CBE results, mammogram results, and assessment of the internal

69

consistency of the mass when the needle is placed during fine needle
aspiration. Clinical impression therefore represents an outcome rather
than a true explanatory variable. Because we were interested in
analyzing each of the components of the clinical impression separately, a
decision was made to substitute the results of CBE for the overall clinical
impression commonly used in triple diagnosis.

Clinical Breast Examination (CBE):

Mass size was assigned to 1 of 4 categories: (1) less than or equal to 1
cm; (2) 1.1 — 2.0 cm; (3) 2.1 — 5.0 cm; (4) greater than 5 cm. Mass shape
was assigned 3 categories: (1) vague thickening, (2) round, oblong or
lobulated mass, or (3) irregular shape. Mobility was assigned 2
categories: (1) mobile or (2) fixed to skin or chest wall. External texture
was assigned 2 categories: (1) smooth or (2) irregular. Consistency was
assigned 3 categories: (1) soft, (2) hard or (3) neither.

Mammography:

The surgeon’s evaluation of the mammogram in the area of the palpable
abnormality was recorded according to the BI-RADSTM classification
system as detailed above. Although much literature has been written
about mammogram results in the evaluation of a breast mass, very little
has been written about the timing of the mammogram with respect to
other aspects of the mass evaluation. There is some literature that points
out that mammography must be done before FNAB to avoid false-positive

results from possible hematomas precipitated by the aspiration. If the

70

aspiration is done before mammography, a waiting period of 2 weeks to
allow for hematoma resolution is suggested 153» 159. All of the
mammograms in this study were done using these guidelines.

Few authors have addressed the timing of mammography evaluation
in relation to the appearance of a breast mass. Langmuir et al used a
period of 6 months on either side of the time of the FNAB in their study,
as did Steinberg et al 95, 101. Although admittedly arbitrary, a decision
was made to include patients in this study whose mammograms had
been done within 4 months before the first presentation of the breast
mass or within 6 months afterward.

Fine Needle Aspiration — Internal Consistency and Cflology;

Internal mass consistency refers to the perception of the clinician
regarding the way that the mass felt when performing an FNAB. It was
categorized as follows using descriptions by Roberts 119 and Kline 120:
(1) very soft, consistent with adipose tissue; (2) rubbery or soft, not
consistent with adipose tissue; (3) either hard and non-gritty, or soft or
rubbery but gritty; (4) both hard and gritty.

Fine needle aspiration biopsy was performed using a modification of
the technique originally introduced by Martin in 1930 160. A 22 gauge
needle attached to a 10 cc syringe held by an lnradTM syringe holder was
used. The skin was entered once, 3-5 cos of vacuum applied, and
aspiration performed with several small strokes in several directions.

Vacuum was released and 4 smears were immediately prepared, 2 for

71

Papanicolaou and 2 for Giemsa staining. Two of the slides were
smeared and immediately fixed in 95% alcohol and 2 were smeared and
air — dried. If there was not enough specimen to prepare 4 slides, the
procedure was repeated. The needle and syringe were rinsed in
cytopathology fixative and the specimens sent to the pathology
department. If needed, the pathologist could retrieve cells by performing
a cytospin retrieval on this specimen. Fine needle aspiration cytology
was classified using 7 categories: (1) adipose tissue; (2) benign cells;
(3) few atypical cells consistent with benign disease; (4) many atypical
cells, indeterminate; (5) suspicious for malignancy; (6) malignant; and (7)
insufficient. These categories are consistent with those most recently
published following a Consensus Conference at the National Institutes of
Health in 1996 123.

Follow-up:

For patients who did not undergo open biopsy after the initial evaluation,
a follow-up appointment was routinely scheduled before the patient left
the clinic. Routinely, this occurred at 3 months in women who had been
examined at the optimum phase of their menstrual cycle. For those who
were not at the optimal phase in whom the phase of the cycle was not
clear, 6 week follow-up was scheduled. Rarely, 6 month follow—up was
scheduled; typically this would be in cases of a clinically diagnosed
fibroadenoma. The third and fourth follow-up visits were usually

scheduled at 3 month intervals from the previous visit.

72

Study Setting:

The study was conducted at the Comprehensive Breast Health Clinic (CBHC) at
Michigan State University. The CBHC represents a single-site referral center
for both physician and self-referred patients attended by 5 surgeons and 1
nurse practitioner. Of the new patients, approximately 85% of the patients are
physician-referred and 15% are self-referred. Approximately 35% of the clinic
population consists of follow-up high-risk patients, patients with a previous
history of breast cancer, or women with active problems. The vast majority of
the clinic population is drawn from a 30—mile radius around Lansing, Michigan.
Period for Data Collection:

Data collection forms were completed prospectively by the 5 surgeons and 1
nurse practitioner with clinical practices in the CBHC beginning in November
1994 and extending through the end of February of 1998. No data was
abstracted from the charts until approval was granted by the University
Committee on Research Involving Human Subjects. This approval was initially
granted on 3/4/96 and renewed for one year’s time on 2/20/97.

Selection of Study Population:

Patients who were potentially eligible for the study included all patients
presenting to the CBHC who had billing codes for fine needle aspiration biopsy
or open biopsy for a mass during the times of the study. The billing office used a
computer-generated system of identification to provide the list of potentially
eligible women. Codes for 823 FNAB or open biopsy procedures were reviewed

to determine patient selection for the study. The medical records were then

73

examined for each patient. Patients were included in the study if they were
female, 30 years of age or older, and met all of the following eligibility
requirements:
0 Physical examination results recorded by the examining clinician
o Mammogram within 4 months before or 6 months after presentation of the
mass
0 Fine needle aspiration cytology of adequate cellularity on first or repeat
aspiration; or, if hypocellular, consistent with the clinical assessment that
adipose tissue was aspirated and represented the mass in question
0 Hypocellular findings that were repeated with adequate cellularity for
evaluation
c Mass disposition that met one of the following criteria:
0 Open biopsy
0 Mass disappearance within 12 months of follow-up
0 A minimum of 12 months of follow-up by the evaluating clinician with no
change in the size or the consistency of the mass
Abstraction of Data from the Medical Record and Database Creation:
Data was collected from the medical records and transferred manually onto a
data abstraction form. To assure accuracy of data abstraction, the results of the
first 20 charts abstracted by the data abstractor were compared with those
abstracted by the PI of the project . A goal of achieving a kappa coefficient of

agreement of .9 or better between the abstractor and the Pl was set prior to the

study according to the suggested study design criteria of Fleiss 161. The Kappa

74

scores for inter-observer agreement between the data abstractor and the Pl
were .98 using the abstraction forms created. Data was entered into a database
created in Lotus Smart Suite Approach software and imported into the SAS"
System for analysis.

Definitions:

The mathematical model constructed from the data is termed BREAST AID
(Breast Cancer Bisk Evaluation and §coring System to Aid in the Biagnosis of
Mammary Masses). To construct it, multivariate logistic regression analysis
was performed using SAS® System software version 6.12. The data was
divided into predictor (epidemiologic and clinical) variables and outcome
variables.

Epidemiolggic predictor variables included age, height, and weight; body
mass index (bmi expressed as weight/ height2 ); personal history of breast
cancer; reproductive factors (including age at menarche, parity, age at first
delivery, number of deliveries, menopausal status, age at menopause);
exogenous hormone use (oral contraceptives, estrogen replacement therapy,
progesterone replacement therapy), including duration of use; family history of
breast cancer (first and second degree relatives) including age at diagnosis and
unilateral or bilateral involvement; history of radiation treatment in childhood or
young adulthood; and tobacco and alcohol use, including duration and
frequency of exposure.

Clinical predictor variables included characteristics of the breast mass

(size, shape, mobility, external texture, consistency); results of mammography

75

(negative, benign, indeterminate, suspicious, malignant); and fine needle
aspiration results (internal consistency of mass — soft , rubbery, hard, gritty; and
cytological evaluation — adipose tissue, mild atypia, many atypical cells,
suspicious, malignant).

Outcome variables included either histologic results at open biopsy or
clinical disposition after a minimum of 12 month follow-up. The gold standard
for BREAST AID evaluation was the presence or absence of cancer defined as
ductal carcinoma in-situ, invasive ductal carcinoma, or invasive lobular
carcinoma.

Data Analysis and Statistics:

Staged Application of Bayes’ Theorem:

Application of Bayes’ theorem provides a method to use the pre-test probability
of disease and modify it by applying the results of one or more diagnostic tests
(in the form of a likelihood ratio) to arrive at a post-test probability of disease.

The rationale for this approach is described in detail by Weinstein and Fineberg

162 and by Sackett et al 163 and is briefly discussed in the subsequent section
entitled “Staged Approach to Model Building”. We were interested in modeling
the diagnostic prediction for breast cancer in a way as similar as possible to the
thinking process that a clinician uses in evaluating a breast mass. To
accomplish this, a 3-step modeling process was used. Stage 1 represented the
“pre-test” predicted probability for breast cancer based on epidemiologic and
historical data gathered during the initial clinical encounter with the patient.

Stage 2 represented the probability of breast cancer based on results of CBE

76

and mammography in combination with the epidemiology and historical
assessment from Stage 1. Stage 3 represented the predicted probability for
breast cancer based on the FNAB findings in combination with the Stage 2
model results.

Multiple Lpgistic Regression MogaLipg

To begin construction of the mathematical model, initial screening of each
predictor variable was done using bivariate logistic regression analysis with
breast cancer as the target disorder. Odds Ratios (OR) and 95% confidence
intervals (Cl) were calculated to assess the magnitude of the association
between each predictor variable and the outcome variable. Statistical
significance was evaluated using the likelihood ratio chi-square (LRCS) statistic.
Variables determined to be important m1 (defined as age and menopausal
status for this analysis) or those with a p value 0.30 or less were selected for
inclusion in the multivariate logistic regression model building phase. These
covariates were entered into the multiple regression model and tested for

significance using backwards elimination according to the method discussed by

Kleinbaum 164. Possible tvvo-way interactions were assessed using similar
methodology. The final model was the most parsimonious model that was as
accurate in prediction for the probability of breast cancer as was possible.

Sta ed roach ta Mogal Bagging;

Stage 1 analyzed the epidemiologic variables in order to define the pre-test
odds or likelihood of disease. The results of the multivariate analysis of these

variables, referred to as the pre-test probability of disease, represented the

77

probability of breast cancer that a clinician might estimate after evaluation of
only the clinical history and risk factor information. Application of Bayes’
theorem requires the conversion of the pre-test probability of disease to the pre-
test odds of disease using the formula

Pre-test odds of disease = P(D) I 1 — P(D)
where P(D) = probabilityof disease.

In Stage 2 the clinical variables from the CBE and mammography results
were analyzed in a multivariate logistic regression model. The output of this
model represented the probability of breast cancer based on the clinical
variables of CBE and mammography alone. In order to apply Bayes’ theorem
the logistic model was used to generate a likelihood ratio which was used in
combination with the pre-test odds from Stage 1 to generate a post-test odds
using the following equation:

Pre-test odds X Likelihood Ratio (LR) = Post-test odds
(from Stage 1) (from Stage 2)

The LR expresses the odds that a given level of a diagnostic test result would

be expected in a patient with, as opposed to without, the target disorder 153,
and is a common measure of the discriminating ability of a diagnostic test.

In this case, the LR from the logistic model represents the likelihood of
cancer based on the combination of CBE and mammography results. The post-
test odds represents the odds of disease for the combination of epidemiologic
(Stage 1) and CBE and mammography (Stage 2) data combined.

Stage 3 repeated the steps of Stage 2 by generating a logistic regression

model for the FNAB data. A second likelihood ratio (LR2) representing the

78

odds of breast cancer based on the F NAB variables alone was generated, and
then multiplied by the post-test odds from stage 2 to obtain the final odds for
cancer. This final post-test odds was then converted to the final post-test or
predicted probability of cancer. A summary of these calculations for the 3
stages is presented below.

S_ta9e_1

Epidemiologic variables —> MODEL 1 (Pre-test probability) —+
pre-test probability I 1 — pre-test probability = pre-test odds

S_tage_2
0 Clinical variables —> MODEL2 —> likelihood ratio1 (LR1)

. Pre-test odds X LR1 = Post-test odds

S_tage_§
o FNAB variables —-) MODEL 3 a likelihood ratio 2 (LR2)

o Post-test odds X LR2 = Final odds
0 Final odds I 1 + final odds = Final Probability of Cancer

Final Adjustment of the Model based on Disease Prevalence:

The relationship between the probability of disease generated in the multiple
. . . . . . . . 165
logistic regresswn model and the likelihood ratio (LR) has a logistic form
and the regression coefficients of the logistic model can be used to generate an
. . . 165
accurate LR according to the followmg equation

LR = exp (a0 + B1X1 4' 32X2 +---+kak)

79

When the prevalence of disease is not exactly 50%, an adjustment must be
made based on the prevalence of disease in the population under study in order
to generate an accurate LR. This is achieved by adding log P(D—) I P(D) to the

intercept term of the logistic model according to the methods described by

Albert 165 and Reeves 166.

Adjusted intercept = ao + log P (D-)I P (D)
\Miere P(D-) = the prevalence of non-diseased individuals and P(D) = the
prevalence of diseased patients in the population under study. The LR
equation derived from the logistic model therefore has the following form under

these circumstances:

LR = exp (a0 + log P (D-) I P (D) + Z kaxk) or

LR = P (D-) l P (D) exp (a0 + 2 kaxk)
This represented the final step in our analysis to determine the probability of
cancer for each patient based on her unique covariate pattern.
Modfel Goodpess - of - Fit_:
Each logistic model was evaluated using the Hosmer-Lemeshow goodness-of-
fit chi square statistic. The latter statistic is a means to assess how well the
model fit the data from which it was generated. It divides the data into groups
and then calculates expected numbers of cases of disease based on the

predicted probabilities from the model, and compares these numbers with

observed values 167. This validation procedure was applied to all of the logistic

models generated. A non-statistically significant result indicates that there was

80

insufficient evidence that the null hypothesis of adequate goodness-of-fit be
rejected.
BREAST AID Evaluation:
Once constructed, the test performance of BREAST AID was evaluated using
two techniques: (1) measurement of test characteristics and (2) construction of
an ROC curve. Test characteristics are defined as the sensitivity, specificity,
and negative and positive predictive value of a test based on a 2 X 2 table —
see Figure 1 for an example of a 2 X 2 table and Table 7 for definitions of test
characteristics. An ROC curve is a graphic representation of the test results,
with sensitivity depicted on the “y” axis and 1 - specificity depicted on the “x”
axis of the graph. This plot is used to assess various cutoff points for the final
probability of cancer that best predicts outcome (see Table 7 and Figure 2 for
definitions).

RESULTS
Population Studied:
There were 279 female patients with a breast mass who were at least 30 years
of age seen in the CBHC during the period of the study included in the study.
They had 336 occurring synchronously (simultaneously) or metachronously (at
different points in time) masses that met all of the eligibility requirements.
There were 616 potentially eligible patients. Reasons for non-inclusion in the
study and estimates of the number of lesions by reason include: (1)
mammogram not done within the defined period for mammography (N = 36);

(2) open biopsy, but no fine needle aspiration done (N = 39); (3) charts not

81

 

found on at least 3 attempts (N = 27); (4) FNAB hypocellular and not consistent
with adipose tissue, not repeated or repeated with persistent inadequacy (N =
157); and (5) FNAB hypocellular on visit 1; patient sent for open biopsy (N =
52). Some of the patients had more than one reason for non-inclusion. In
these cases, hypocellular results were tracked in preference to other reasons
for non-inclusion. Estimations of reasons for non-inclusion were necessary
because at the beginning of the study, specific reasons for exclusion were not
tracked on 300 charts, although numbers of ineligible patients were. To
account for this, 50% (N = 150) of the ineligible charts were reexamined and
reasons for ineligibility listed. Data for the entire portion of the population of 300
patients was then estimated based on these results. These results were
compared with the charts for whom detailed reasons for ineligibility were
tracked in the latter half of the study (N = 316). The percentages of reasons for
ineligibility between the 300 original charts and the 316 charts from later in the
study demonstrated agreement of reasons for non-inclusion of within 10% of

one another.

Population Characteristics and Results of Bivariate Laglstic Ragression

Biopsy and Follovv-up Rates

The usual practice in the CBHC involves open biopsy for discordant triple
diagnosis results (at least one of the three tests with an other-than—benign
result) or for patient preference. Using these criteria, 135 of 336 lesions
underwent open biopsy. These included 86 (63.7%) biopsies within 1 month of

visit one, 35 (25.9%) biopsies within 4 months of visit 1, 12 (8.9%) biopsies

82

within 7 months of visit 1, and 2 (1 .5%) biopsies within 10 months of visit 1.
Thirty-two malignancies were diagnosed after visit 1 and 4 after visit 2 using this
approach. One of these four lesions was in a postmenopausal patient who
initially declined biopsy at visit 1, even though the surgeon felt it necessary; she
consented after the FNAB cytology indicated malignancy. The other 3
malignancies were diagnosed in premenopausal women. Two had a
concordant triple diagnosis for benign disease after visit 1 and one had many
atypical cells diagnosed on FNAB. All of these were diagnosed at visit 2 with
persistent masses that had become more prominent at visit 2. All cancers were
diagnosed within 3 months of presentation. Visit 2 follow-up periods ranged
from a period of 6 weeks to 6 months. Six-week follow-ups were usually
scheduled for pre-menopausal patients for purposes of examination at a more
optimum phase of the menstrual cycle and six month follow-up for patients
diagnosed with clinical fibroadenomas. Othenrvise, 3 month follow-up was
scheduled. Table 8 lists the histopathological results for the 135 biopsied
lesions and Table 9 compares them with histopathological frequencies from
other published series. Thirty-six lesions were diagnosed with cancer; 27 with
invasive ductal carcinoma, 5 with invasive lobular carcinoma, and 4 with ductal
carcinoma-in-situ. This represents a malignant: benign ratio of open biopsy of
1:275 (i.e., 36:99). Ten high-risk lesions were identified, 2 with lobular-
carcinoma-in-situ and 8 with atypical ductal hyperplasia. Of the benign lesions,
60 represented fibrocystic changes, 21 were fibroadenomas, and the remaining

8 represented miscellaneous benign histopathological findings.

83

Descriptive and Bivariate L_pgistic Regression Resalg

The prevalence of cancer and bivariate logistic regression results for each of
the categorical epidemiologic predictor variables are listed in Table 10. The
same data for the categorical clinical predictor variables are listed in Table 11.
Table 12 shows the same results for predictor variables specified on a
continuous scale.

Epidemiologic Variables

All of the data concerning the categorical epidemiologic variables can be found
in Table 10. Table 12 includes data on the continuous epidemiologic and
clinical variables.

Aga

Age ranged between 25 and 84 years, with a mean of 47.6 years. The mean
age of the benign lesions was 46.8 years and for the malignant lesions, 53.9
years. Seventeen of 221 lesions were malignant in women 49 years of age or
less (7.7% malignancy rate) and 19 of 115 were malignant in women 50 years
of age and older (16.5% malignancy rate). The comparison between the two
was significant with a p value of .02 (data not shown). When patients were
grouped into those less than 40 (reference group), 40-49, 50-59, and greater
than 60, the risk of cancer increased sequentially with values of 1.0, 2.1, 3.7,
and 4.9, respectively. Associated 95% confidence intervals were 0.67-9.5, 1.1-

17.3, and 1.4-23.2, respectively. This result was significant at the p = .05 level.

84

My Mass Index (BMI)

Degree of obesity was measured by the Quetlet, or Body Mass Index (BMI).
This was defined by converting height in inches to height in meters and weight
in pounds to weight in kilograms. The formula weight! height2 was applied to
calculate the BMI. The mean BMI for those women with cancer was 26.1
kglmz, compared to 24.6 kglm2 in the group without cancer. For purposes of
data analysis, the BMI was grouped into those above and below the mean of
24.73 kg/m2 (range 16.7-58.5 kg/mz), with an associated odds ratio of 2.09
(95% CI = 0.75-5.34) for those above the mean. This was significant at the p =
.04 level.

Personal History of Breast Cancer

Only nine patients indicated a history of previous breast cancer, and one of
these was diagnosed with a new primary cancer in this study. There was no
significant difference In the frequency of a personal history of breast cancer
between the two groups (OR 1.04, 95% CI = 0.06-5.93, p = .97).

Age at Menarche

The average age of onset of menarche for women with breast cancer was 12.5
years compared with those without malignancy whose average age of onset
was 12.9 years. Age at menarche was categorized into those less than 12
years and 12 years or more. Bivariate analysis showed that women whose age
at menarche was less than 12 years of age were at higher risk of breast cancer
(OR = 1.64, 95% CI = 0.69-3.60. The p value of .25 was not significant, but met

the criteria for inclusion in the construction of the multivariate model.

85

Menogusal Status

Women were classified as premenopausal if they had had any menstrual
activity within the previous 12 months and were not on hormone replacement
therapy, or if they had a history of hysterectomy alone or a hysterectomy and
unilateral oophorectomy and were less than or equal to 50 years of age.
Women were classified as postmenopausal if they had had no menstrual
activity for 12 months or if they were on hormone replacement therapy, if they
had a history of bilateral oophorectomy, or if they had a history of hysterectomy
alone or hysterectomy and unilateral oophorectomy and were greater than 50
years of age. Fifty years of age was chosen as the cut-off age for menopause
because 90% of the population that knew their age at the onset of menopause

had reached menopausal status by that time. This approach is the same as

that used in the Nurses Health Study 163. Breast cancer was diagnosed in
6.7% of the premenopausal patients and 18.8% of the postmenopausal patients
who presented with a breast mass. Menopausal status was strongly associated
with the risk of breast cancer (OR = 3.2, 95% CI = 1.60-6.60), with a p value of
less than .01.

Age at Menopause

Although 112 women were classified as postmenopausal, only 97 knew the
exact age at which they reached menopause. The average age at menopause
for the 97 women in whom it could either be determined or estimated was 46.4
years. The average age at menopause was 46.4 years for the group without

cancer and 46.5 years for the malignant group. Sixty-nine of the women

86

underwent natural menopause and 91% of these women had menopause at or
before the age of 50. Sixty-four women had a surgical procedure that resulted
in cessation of menses, 30 had a bilateral oophorectomy while 34 had at least
one ovary left intact Thirty-one of these procedures were done before the age
of 49 and 34 were done at age 50 or over. Women were categorized into those
who had reached menopause at or after age 50 and compared to those who
had reached menopause before age 50. The OR of 1.37 (CI = 0.44-4.17) was
not significant and did not reach the criteria for inclusion in the multivariate
model (p = .58).

Reproductive Factors - Pragnang Status, Number of Pregnancies, Number of
Beliveries, and Age of First Delivegy

 

The frequency of nulliparity in our population was 12.5%. Only 11% of the
women with cancer were nulliparous. Comparison of this group to a group who
had delivered at least one child yielded an OR of 0.86 (95% CI = 0.25-2.32),
which was not significant and did not reach the criteria for inclusion in the
multivariate model (p = .78). The average number of pregnancies did not differ
between the two groups (n = 2.8). When the number of pregnancies were
categorized into one, two, and three or more pregnancies, there was again no
significant relationship with cancer risk. Similar results were obtained for
number of deliveries. The average number of deliveries for the benign group
was 2.3 and for the malignant group, 2.5. When comparing groups who had
had 2 or 3 or more deliveries with those who had only one, there was no

significant relationship with cancer risk. The average age of first delivery for the

87

benign group was 24.7 years and for the malignant group, 24.2 years. Age of
first delivery comparisons into groups of ages <19, 20-24, 25—29, and >29 also
yielded OR that were not significant when tested by bivariate analysis; the p
value of 0.78 did not meet the criteria for testing in the bivariate model.

Oral Contraceptive Use

In this study, 78.3% of the patients were ever users of oral contraceptives, but
only 8 (0.4%) of these patients were current users. Ever as compared with
never use of oral contraceptives was highly significantly associated with cancer
risk with a p value of <01. However, the magnitude of association was the
reverse that would have been predicted based on past literature (OR for ever
vs. never use 0.29, 95% CI = 0.14-0.61). This finding was hypothesized to
represent confounding by age due to the cohort effect, and was strengthened
by the finding that duration of use had no effect on outcome. The mean
duration of oral contraceptive use in the cancer patients was 5.3 years,
compared with 5.7 years in the benign group. Bivariate OR estimates for 3
months - <5 years of use vs. 5 — 10 years vs. >10 years were 1.0, 0.89, and
0.69, respectively, with non-significant confidence intervals (p = .82).

anpgan Replacementfia

In this study, 82.1% of the postmenopausal patients were pastor current users
of estrogen replacement therapy (94 current and 20 past users). Comparison
of pastor current use with never use demonstrated respective OR of 0.46 and
0.67 and respective 95% CI of 0.06-2.55 and 0.22-2.32. These associations

were not statistically significant (p = .69). The average duration of estrogen

88

replacement use was 6.7 years for the benign group and 10.6 years for the
malignant group. Duration of estrogen replacement use demonstrated an OR
of 2.1 (95% CI = 0.71-7.12) when comparing <5 years of use vs. 5 or more
years (p = .14). This variable met the criteria for testing in the multivariate
model. Because of this, ERT use (never, past, current) was included in the
model as well.

Progestin Raplacement [Jae

In this study, 58% of the postmenopausal patients were past or current users of
progestin replacement therapy (47 current and 18 past users). Comparison of
past vs. current use with never use demonstrated respective OR of 1.62 and
0.74 and 95% CI of 0.45-5.65 and 0.24-2.18. These values were not
statistically significant (p = .51), and progestin use was not included in the
multivariate model. The average duration of progestin replacement use was
low, 3.3 years for the benign group and 5.5 years for the malignant group.
Duration of progestin replacement use demonstrated an OR of 1.5 (95% CI =
0.40-5.48) when comparing <5 years of use vs. 5 or more years (p = .53). This
was not significant and this variable was not tested further.

Family Histogy of Breast Cancer, Age at Diagnosis, and Laterality of Disease

In the overall population, the frequency of a family history of breast cancer in a
first and/or second-degree relative was forty-five percent and was 21.1% for at
least one affected first-degree relative. The magnitude of the association was
1.42 (95% CI = 0.40-5.48) for any family history, 1.76 (95% CI = 0.79-3.71) for a

family history in a first-degree relative, 1.45 (95% CI = 0.41-4.07) for family

89

history in a sister, and 2.28 (95% CI = 0.95-5.08) for family history in the
mother. The associated p values were less than .3 only for family history in a
first-degree relative or history in the mother. Because these variables could not
be simultaneously used for testing in the multivariate model because one is a
subset of the other, the strongest magnitude of association was chosen and
history in the mother was tested (p = .06). Forty-five women knew the age of
the mother’s diagnosis. The mean age of diagnosis in the benign group whose
mother had breast cancer was 55, compared to 53 for the group with cancer.
Comparison of those diagnosed at greater than age 50 years with those
diagnosed at 50 years of age or less demonstrated an OR of 1.96 with a p
value of .39. Forty-six women knew if their mothers were diagnosed with
unilateral or bilateral disease. Seven had bilateral breast carcinoma.
Comparison of this group with those of unilateral disease demonstrated an OR
of 4.13 (95% CI = 0.69-24.00). Because of small numbers of patients, this was
not statistically significant but did reach the criteria for inclusion in the
multivariate model (p = .12).

History of Radiation msure
There was a history of radiation exposure in childhood or early adulthood in

 

4.4% of the population. Cancer was diagnosed in 9.8% of the population
without the exposure and 28.6% of the pOpuIation with the exposure (OR 3.67,

95% CI 0.96-11.72), p = .06).

90

Alcohol and Tobacco Bmosure

Current alcohol use was compared with no use, demonstrating an OR of .64
(95% CI = 0.32-1.32) and a non-significant p value of .22. This did meet the
criteria for testing in the multivariate model, although the magnitude of the
association was the reverse of that expected based on the majority of the
epidemiologic literature. Alcohol dosage and frequency of use (number of
drinks/week) were not significant in the bivariate model, nor was tobacco
exposure when tested for ever vs. never use or numbers of pack years of
cigarette smoking. Neither were tested further.

Clinical Variables

All of the data concerning the categorical clinical variables can be found in
Table 11. Table 12 contains data concerning the continuous clinical and
epidemiologic variables.

Mass Size

Mass size was documented for 330 of 336 lesions (98.2%). The mean mass
size was 1.6 cm; 1.5 cm for the benign group and 2.4 cm for the malignant
group. Mass size was categorized into <= 1 cm (N = 129); 1.1-2.0 cm (N =
150), 2.1-5.0 cm (N = 46), and > 5 cm (N = 5). The frequency of cancer for the
respective categories were 4.7%, 10.7%, 23.9%, and 60%. The latter three
categories were compared with those 1 cm in size or less using bivariate
logistic regression. Corresponding OR were 2.45, 6.44, and 30.75,
respectively, and respective 95% Cl were 0.97-7.0, 229-199, 436-2712.

These values yielded a highly significant p value of <01.

91

Mass Shape

Mass shape was available for 317 lesions, determined by either written
description or a picture representation from the chart. Shape was categorized
as a vague thickening (N = 147), a round or lobulated mass (N = 156), or as an
irregular mass (N = 14). Cancer frequencies for the respective categories were
6.1%, 13.5%, and 42.9%. Using round or lobulated masses as the referent
category, odd ratios of .42 (95% CI = 0.18-.92) for vague thickenings and 4.82
(95% Cl 1.46-15.30) for irregular masses were obtained. These were
associated with a significant p value of <0.01.

Mass Mobility

Masses were classified as either fixed (if they were fixed to the chest wall or to
skin), or mobile. Three-hundred thirty-five of 336 lesions were mobile. The one
lesion that was fixed represented cancer, as expected. Because of the highly
skewed distribution, this variable was not analyzed further.

mes Consistency

Data for this variable was available for only 29% of the lesions. Its distribution
was also highly skewed. None of the lesions categorized as soft represented
cancer, whereas all of the lesions that were categorized as hard were
malignant. However, of the 97 lesions with data available, 84.5% were neither
hard nor soft on palpation. For these reasons, the variable was not analyzed

further.

92

Mass External Texture

Data was available for the impression of external texture for 318 lesions. Of the
41 lesions classified as irregular (12.9%), 17 (41.5%) were in the cancer group,
whereas only 19 of the 277 smooth lesions had cancer (6.9%). A highly
significant association between breast cancer and external texture was found
(OR = 9.62, 95% CI = 4.42-21.08, p = <.01).

Mammggram Results

Mammogram results were available for all 336 lesions. The distribution of the
five categories of mammogram readings were as follows: Category 1: Negative
- 56.0%; Category 2: Benign - 32.1%; Category 3: Indeterminate - 7.4%;
Category 4: Suspicious - 2.4% and Category 5: Malignant - 2.1%. The
frequency of cancer in each of these categories was: Category 1: Negative —
1.6%; Category 2: Benign - 7.4%; Category 3: Indeterminate — 44%; Category
4: Suspicious — 87.5% and Category 5: Malignant - 100%. For purposes of
further analysis, Category 4 :Suspicious and Category 5: Malignant readings
were combined. The three remaining categories were compared with those
classified as Category 1 using bivariate logistic regression. Corresponding OR
were 4.93 (95% CI = 139-2290), 48.45 (95% CI = 13.42 — 233.64), and 863.33
(95% CI = 12249-99900) respectively, yielding a highly significant p value of
<01.

Internal Consistency an F NAB

lntemal consistency refers to the perception of the consistency of the mass to

the examiner during F NAB. Internal consistency results were available for 336

93

lesions at visit 1. The distribution of the four original categories of lntemal
consistency were as follows: 1 - soft, consistent with adipose tissue — 14.3% (N
= 48); 2 - soft (not consistent with adipose tissue) or rubbery — 70.0% (N = 235);
3 - hard (not gritty), or soft or rubbery with some grittiness - 14% (N = 47); 4 -
hard and gritty - 1.8% (N = 6). All of the 48 lesions in category 1 were benign,
whereas all 6 in category 4 were malignant. Soft (not consistent with adipose
tissue) or rubbery lesions were found to represent cancer 6% of the time, and
for the category of hard (not gritty) or soft or rubbery with some grittiness, 34%
of the time. For purposes of further analysis, categories 1 and 2 were
combined, used as the referent category, and compared with the combination of
categories 3 and 4. The corresponding OR was 13.64 (95% CI = 624-2998).
This result was highly significant, with a p value of <01.
fll.AB_CvtologitRes_ufl§

FNAB cytology results were available for 317 lesions at visit 1. Nineteen cases
had FNAB readings that were hypocellular and in which the internal consistency
was not consistent with adipose tissue. Repeat aspirations were done for these
lesions at visit 2 but no further malignant cases were diagnosed. For purposes
of analysis, the visit 1 cases were examined. The distribution of the six
categories of FNAB cytology readings were as follows: 1 — adipose tissue -
11.4% (N = 36); 2 - benign epithelial cells - 63.1% (N = 200); 3 - few atypical
cells — 11.4% (N = 36); 4 - many atypical cells — 5.0% (N = 16); 5 - suspicious
for malignancy - 4.1% (N = 13); and 6 - malignant - 5.0% (N = 16). Cancer

was associated with each of these categories with the following frequencies:

94

adipose tissue - 0%; benign epithelial cells - 3.5%; few atypical cells - 8.3%;
many atypical cells - 18.8%; suspicious for malignancy — 53.9%; malignant —
100%. For purposes of further analysis, the first and second categories were
combined, as were the fifth and sixth categories. Using the adipose
tissue/benign cases as the referent category, the remaining categories were
tested for association with cancer using bivariate logistic regression.
Corresponding OR were 2.97 (95% CI = 062-1129) for the category read as
few atypical cells, 7.55 (95% CI = 1.50-30.85) for many atypical cells, and
125.41 (95% CI = 41 79-44475) for suspicious/ malignant readings. This was
highly significant, with a p value of <01.

Results of Multivariate Logistic Ragression Analysis

The results of multivariate testing are presented according to the 3 stages used
in model building. There were no significant two-way interactions encountered
in the modeling process.

$999.1

Because menopausal status was so strongly predictive of malignancy in the
bivariate model, we considered premenopausal and postmenopausal women
separately in the analysis of the epidemiologic variables.

Premenopausal Women

The epidemiologic variables that passed the initial criteria of a p value of 0.3 or
less using bivariate logistic regression for the premenopausal women included
age, weight, body mass index, age at menarche, oral contraceptive use, family

history of breast cancer in the mother, history of radiation exposure, and current

95

alcohol use. After multivariate analysis of these epidemiologic variables related
to the 223 lesions and 15 cancers in premenopausal women, none were found
to be significant in predicting for breast cancer.

The goal of Stage 1 was to establish the pre-test probability of disease,
which, in this case, represents the closest estimate of the likelihood of breast
cancer prior to the application of a diagnostic test. Because the pre-test
probability of breast cancer for the premenopausal women in the study could
not be predicted from epidemiologic variables pafi, the best estimate of it was
through the use of disease prevalence. Because disease prevalence is age-
dependent, we used age groups <40 years (N = 65), 40—44 years (N = 74), and
>44 years (N = 84) to calculate prevalence in our study population and establish
the pre-test probability of breast cancer. The associated probabilities
(prevalence) by age group were 4.6%, 6.8%, and 8.3%, respectively.
Postmenopausal Women
For the postmenopausal women, the same variables were tested as in the
premenopausal women, with the addition of estrogen replacement therapy use
and duration for the postmenopausal group. Multivariate analysis of the
epidemiologic variables in the 112 lesions and 21 cancers in this group
demonstrated two to be significant: family history in the mother and a history of
radiation exposure. The final model obtained for the postmenopausal women is

summarized below

96

Table 13: ngistic Ragression Model of the Risk of Breast Cancer for 112
Postmenopausal Women based on Epidemiologic Variables

 

Wald

Variable B s_e_l§ QB 95% CI p Value
Intercept -2.44 2.00
Age (years) 0.01 0.03 1.01 .95, 1.08 0.77
History in Mother

No (referent) - - 1.00 - -

Yes 1.51 0.69 4.52 1.15, 17.85 0.03
History of Radiation

No (referent) - - 1.00 - -

Yes 2.23 0.99 9.27 1.34, 79.19 0.02

The associated pre-test probability of malignancy for each covariate pattern,
based on the presence or absence of these factors, ranged between 11% and
66%.

Results of Hosmer-Lemeshow Goodness-of-Fit Chi Sguare: Stage 1

The GOFCS was generated for the 103 postmenopausal women. The post-test
probabilities were ranked in ascending order and then grouped into quartiles.
The groups were not equal because many patients shared the same pre—test
probability result, especially in the lower half of the probability rankings. The
number of predicted and observed cases of breast cancer for each quartile of

predicted probability or risk is summarized below.

Table 14
Observed and Expected Freguencies within Quartiles of Risk: Stage 1*
Cancer = 1 Cancer = 0
Group Total Observed Expected Observed Expected
1 41 4 4.6 37 36.4
2 30 5 4.9 25 25.1
3 13 3 2.4 10 10.6
4 19 9 9.0 10 10.0

 

0 Legend on following page

97

* Postmenopausal women only

# Quartiles were created by ranking the pre—test probabilities and then dividing
them into sequential groups

+ The expected frequencies were calculated by summing the pre—test
probabilities for all women within each grouping

There is good agreement between the number of observed and predicted cases

of cancer for each quartile. The GOFCS was 0.26 (df = 2, p = 0.88). The

model therefore fit the data well and appears to provide an accurate estimate of

the probability of breast cancer for postmenopausal women.

From a diagnostic perspective, the results from stage 1 make it clear that
epidemiologic data alone is not able to differentiate those women with a breast
mass that might be safely triaged to follcvv-up, as evidenced by the distribution
of cancer cases across all quartiles of predicted probability.

Final Step in Stage 1: Evaluation of Pre-test Probabilities:

Data from Stage 1 modeling was merged for the premenopausal and
postmenopausal women and used to set the pre—test (or prior) probability of
disease for application to the Stage 2 model. The final pre-test probabilities for
breast cancer ranged between 4.6% and 66% based on Stage 1 results, with an
average of 11.0%. The mean pro-test probability was 20.2% for the cancer
patients and 9.9% for the patients without cancer. However, the range of pre-
test probabilities was similar between the two groups; 4.6% — 52.2% for the
non-malignant group and 4.6% - 66.4% for the cancer patients.

fleas-2

Stage 2 involved analysis of the clinical variables obtained from the CBE and

mammogram that passed the initial screening criteria. Age was included in the

98

analysis regardless of the statistical significance as an ‘a priori’ confounder.
Specific candidate variables tested included mass size, mass shape, mass
mobility, mass external texture, and mammogram results. Because a Category
5 (malignant) mammogram was associated with malignancy 100% of the time, it
was necessary to group Category 4 (suspicious) and Category 5 mammograms
into one category in the multivariate analysis.

Multivariate analysis of the clinical variables was performed on 330 lesions
that had all data available. The 6 missing cases lacked a specific mass size on
CBE. After stepwise backwards deletion of non-significant variables (p>0.05)
only age, mass size and mammogram results remained in the final model. The

final model obtained for the clinical variables is summarized below

Table 15: Lpgistic Ragression Model of the Risk of Breast Cancer in 330
Breast Masses Based on CBE and Mammography Results

 

Wald
Variable B M O_R 95% CI p Value
Intercept -7.91 1 .61
Age 0.06 0.03 1.07 1.02, 1.12 0.01
Size (centimeters)
5 1.0 (referent) - - 1.00 - -
1.1-2.0 0.70 0.65 2.01 0.59, 7.92 0.28
2.1-5.0 1.30 0.79 3.68 0.78, 18.10 0.10
> 5 2.75 1.34 15.68 0.94, 227.30 0.04
Mam Findings
Normal (referent) - - 1.00 - -
Benign 1.49 0.71 4.43 1.20, 21.11 0.03
lndeterm 4.08 0.75 59.39 15.20, 310.85 <0.01
Susp/Malig 6.00 1.22 402.40 52.96, 999.00 <0.01

The associated post-test probability of malignancy for each covariate pattern,
based on the presence or absence of these factors, ranged between 0.28% and

98.07%.

99

Results of Hosmer-Lemeshow Goodness-of—Fit Chi Sguare: Stage 2

The GOFCS was generated for Stage 2 in a similar manner to that described
for the Stage 1 results. In this case, however, the post-test probabilities were
ranked in ascending order and then grouped into nine groups of approximately
equal size according to their post-test probability rankings. These groups were
chosen by the SAS program as the optimal number for GOFCS testing. The
number of predicted and observed cases of breast cancer for each of the nine
groups of risk is summarized below.

Table 16

Observed and Expected Freguencies within Groups of Risk: Stage 2

Cancer = 1 Cancer = 0
Group # Total Observed Expected Observed Expected

1 42 0 0.17 42 41.8
2 33 0 0.28 33 32.7
3 33 0 0.44 33 32.6
4 36 1 0.57 35 35.4
5 32 1 0.83 31 31.2
6 43 2 1.56 41 41.4
7 40 4 2.49 36 37.5
8 33 3 4.67 30 28.3
9 38 25 24.99 13 13.0

 

# Groups were created by ranking the post-test probabilities and then dividing
them into sequential groups of approximately equal size

+ The expected frequencies were calculated by summing the post-test
probabilities for all women within each grouping

There is good agreement between the number of observed and predicted cases

of cancer for each group. The GOFCS was 3.07 (df = 7, p = 0.88). The model

for Stage 2 fit the data well and appears to provide an accurate estimate of the

probability of breast cancer for this data set.

100

Merging of Data and Modification of F ina_l Post-Test Probability and Likelihood

Ratio for Stage 2
The output generated by the models for during Stage 1 and Stage 2 was

merged so that both epidemiologic data and data related to mass size and
mammogram results could be evaluated. This merge, as outlined in the
methods section, was done because we were interested in the ability of the
model to predict for breast cancer in women with a breast mass evaluated only
by history, physical exam, and mammogram, as is done in many primary care
and surgical offices. After the data was merged, the likelihood ratio generated
from merged dataset was adjusted based on the proportion of cases and
controls in the dataset as described in the Methods section. The mean
adjusted likelihood ratio for the merged dataset was 2.93, with a range from
.001 to 374.88. The mean adjusted post-test probability was 10.66%, with a
range between 0.10% and 99.71%.

Test Characteristics for Merged Dataset Based on Stages 1 and 2:

Various cut-points for triage to biopsy were explored by calculating sensitivity,

 

specificity, and positive and negative predictive values using the combination of
Stage 1 and Stage 2 data. The best test sensitivity (97.2%) was obtained using
a post-test probability cut-point of 1%. However, the corresponding specificity
was only 37% and 214 of 320 lesions would have been triaged to biopsy. One
cancer would have been missed. The most efficient cut-point in test
performance using the criterion of the maximum average of sensitivity and

specificity combined was a post-test probability of 4%. Using this value, the

101

sensitivity was 86.1%, the specificity was 74.7%, the positive predictive value
was 30.4%, and the negative predictive value was 97.7%. Of the 320 lesions,
102 would have been triaged to biopsy for a malignant: benign ratio of 121.83.
However, this level of efficiency came at the price of 5 of 36 missed cancers

(1 3.9%).

SL892;

Stage 3 involved analysis of the data obtained from FNAB; internal consistency
and cytology. Age was again included in the analysis as an ‘a priori’
confounder. Both FNAB variables passed the initial screening criteria and were
entered into the multivariate model. The F NAB variable of lntemal consistency,
originally consisted of 4 levels (1 - adipose tissue, 2 - soft or rubbery, 3 - gritty
with associated soft or rubbery findings, 4 - hard and gritty). Two of these
categories needed combining for entry into the multivariate model. The adipose
tissue category represented a 100% benign distribution and the hard and gritty
category was 100% malignant. Because of this, any hard finding, or any gritty
finding, (whether associated with soft, rubbery, or hard findings), was combined
into a single category and was compared with another category that included
adipose tissue, soft, or rubbery findings. In addition, the six categories of FNAB
cytology were combined into four for the same reasons. The new combined
categories were: 1 - adipose tissue or benign cytology; 2 - few atypical cells; 3 -
many atypical cells and 4 - suspicious or malignant cytology. Data was
available for all 336 lesions for Stage 3 multivariate analysis. Both internal

consistency and cytology remained as predictors of malignancy in the final

102

multivariate model. The final model obtained for the F NAB data is summarized

below

Table 17: Lpgistic Ragression Model of the Risk for Breast Cancer in 336
Breast Masses Evaluated with FNAB

 

 

Wald

Variable B _s_e_l3 .O_R 95% CI p Value
Intercept -5.53 1 .22
Age 0.03 0.02 1.03 0.99, 1.09 0.15
lntemal consistency

Adipose/soft/rubbery

(referent) - - 1 .00 - -

Hard and/or gritty 1.64 0.54 5.15 1.75, 14.95 <0.01
FNAB Cfiolpgy
Benign (referent) - - 1.00 - -
Few atypical 1.01 0.77 2.73 0.51, 11.39 0.19
Many atypical 2.40 0.79 11.08 2.05, 49.90 <0.01
Suslealig 4.22 , 0.63 68.26 20.89, 255.97 <0.01

The associated post-test probability of malignancy for each covariate pattern,
based on the presence or absence of these, ranged between 0.57% and
95.97%, with a mean of 10.71%.

Resplts of Hosmer-Lemeshow Goaness—of—Fit Chi Sguare: Stage 3

The GOFCS was generated for Stage 3 in a similar manner to that described
for the Stage 1 and 2 results. In this case, however, the post-test probabilities
were ranked in ascending order and then grouped into eight groups according
to their post-test probability rankings. The groups were not equal because
many patients shared the same covariate pattern and thus post-test probability
result. These groups were chosen by the SAS program as the optimal number
for GOFCS testing. The number of predicted and observed cases of breast

cancer for each of the eight groups of risk is summarized below.

103

Table 18

Observed and Exgted Freguencies within Groups of Risk: Stage 3

Cancer = 1 Cancer = 0
Group # Total Observed Exgcted Observed Exfled

1 48 0 0.27 48 47.73
2 30 0 0.43 30 29.57
3 107 2 1.73 105 105.27
4 12 0 0.29 12 1 1.71
5 51 2 2.1 1 49 48.89
6 37 2 2.74 35 34.26
7 34 15 13.01 19 20.99
8 17 15 15.41 2 1.59

 

# Groups were created by ranking the post-test probabilities and then dividing
them into sequential groups

+ The expected frequencies were calculated by summing the post-test
probabilities for all women within each grouping

There is good agreement between the number of observed and predicted cases

of cancer for each group. The GOFCS was 1.88 (df = 6, p = 0.93). The model

for Stage 3 fit this data well, and appears to provide an accurate estimate of the

probability of breast cancer for this data set.

Merging of Data and Generation offipal Post-Test Probabilities

The model generated during Stage 3 was used in sequence with the combined

data generated at the end of Stage 2 so that the contribution of FNAB results to

the combined model could be evaluated. Three hundred and twenty lesions

had all data necessary for calculations for all 3 models. The likelihood ratio

generated from the final dataset was adjusted based on the proportion of cases

and controls in the dataset. The mean adjusted likelihood ratio for the final

dataset was 169.61, with a range from .00004 to 2,113.24. The mean final

post-test probability was 10.12%, ranging from 0.004% to 99.99%.

104

Test Characteristics for BREAST AL): Fina_l Merged Dataset:

Various cut-points of the final post-test probabilities were explored by

calculating sensitivity (Se), specificity (Sp), positive predictive value (PPV) and

negative predictive value (NPV) for each cut-point. A classification table

demonstrating the values follows:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 19
‘ Test Characteristics for BREAST AID by Final Probability of Cancer
2 % Final # Biopsies % Missed
Probability Se (%) Sp (%) PPV (%) NPV (%) Needed Cancers
95 55.6 100 100 94.7 20 44.4
90 55.6 100 100 94.7 20 44.4
80 58.3 99.6 95.5 95.0 21 41.7
70 58.3 99.2 91.3 94.9 23 41.7
60 61.1 99.2 91.7 95.2 24 38.9
50 66.7 99.2 92.3 95.9 26 33.3
40 77.8 98.6 87.5 97.2 32 22.2
30 80.6 98.6 87.9 97.6 33 19.4
20 86.1 97.2 79.5 98.2 39 13.9
10 86.1 95.1 68.9 98.2 45 13.9
5 94.4 89.8 54.0 99.2 63 5.6
4 94.4 88.4 50.7 99.2 67 5.6
3 97.2 85.6 46.1 99.6 76 2.8
2 97.2 80.3 38.5 99.6 91 2.8
1 97.2 74.3 32.4 99.5 108 2.8
.5 97.2 62.0 26.3 99.4 133 2.8
.4 97.2 56.3 22.0 99.4 159 2.8
.3 97.2 50.3 19.9 99.3 176 2.8
.2 97.2 37.0 16.4 99.1 214 2.8
.1 100 26.4 14.7 100 245 0

 

These cut-off points were plotted and an ROC curve constructed. An ROC

curve is simply a graphic means of plotting sensitivity against 1 - specificity. It

is used to assess the ability of a test to distinguish between, in this case, cancer

and benign disease. Figure 2 illustrates an ROC curve for a perfect test (solid

line) and a test with no discriminating ability (dotted line). Figure 3 illustrates

105

 

the ROC curve constructed for BREAST AID. In general, the point on an ROC
curve that is closest to the upper left hand comer of the graph represents the
most desirable cut point for a test (assuming that the “costs” of false negative
and false positive results are equal) . The most efficient cut-point in terms of
maximizing sensitivity and specificity in test performance for BREAST AID was
3%. Using this value, the sensitivity was 97.2%, specificity was 85.6%, positive
predictive value was 46.1%, and negative predictive value was 99.6%. Of the
320 lesions, 76 would have been triaged to biopsy and 35 cancers would have
been detected for a malignant: benign biopsy ratio of 1:1.17. One malignancy
was missed with this approach. This patient was also missed on the initial visit
using the triple diagnosis approach. She presented as a 43 year-old
premenopausal patient with a 1.2 cm vague thickening, a normal category 1
mammogram, a rubbery lntemal consistency on FNAB, and a benign fine
needle aspiration cytology result. She was diagnosed on 6-week follow-up at
the best phase of her menstrual cycle, when her mass became more prominent.
The cancer rate for the 244 lesions triaged into follow-up was 0.41% and for
the 76 lesions triaged into biopsy, 46.1%. Of the cancer cases, sixteen lesions
had malignant F NAB cytology results and all were malignant on histopathology.
Seven had Category 5 mammograms and all of these were malignant as well.
Suspicious FNAB readings were found in 13 cases and seven of these were
malignant at biopsy. Suspicious mammograms (Category 4) were found in 8

cases, 7 which were malignant.

106

Reatllts of Individual Diagnostic Tests Compared with ResJults of BREAST AID:
CBE findings were not evaluated independent of mammography findings in this
study, and calculation of test characteristics for CBE is therefore not possible.
However, this study is one of the first in the past two decades to characterize
the clinical characteristics of a series of breast masses. This is relevant
because, as discussed above, the clinical findings suggestive of malignancy on
CBE are decreasing in prevalence, especially since the widespread use of
mammography. Twenty-two of 36 cancers were 2 cm or less in size, and 6
were 1 cm or less in size. Only 6 cancers presented with an irregular shape on
palpation; the remaining 30 represented smooth or lobulated lesions or vague
thickenings. Only 1 of 36 cancers was fixed; the rest were mobile. Seventeen
malignancies were assessed to be irregular on palpation of external texture, the
remaining 19 were smooth. These findings highlight the low sensitivity of CBE
for diagnosing malignancy. Although abnormal physical examination signs may
increase the likelihood of cancer, lack of physical signs of malignancy by no
means assures a benign diagnosis.

Unlike CBE, it is possible from this study to report the test characteristics for
mammography. Eleven of 36 cancers (30.6%) were classified as Categories 1
or 2 on mammography by the same surgeons who had palpated the mass
within minutes of the mammogram reading. The possibility of false negative
results due to interpretive errors seems remotely low under these conditions. It
was unusual for the surgeon to disagree with the mammographic findings of the

radiologist, and when it did occur, the reading was upgraded by the surgeon

107

based on the CBE findings. Fifteen mammograms were classified as either

Category 4 or Category 5 and 14 (93.3%) of these were malignant. These

fourteen results represented 38.8% of the 36 malignant lesions. The other 11

cancers were classified as Category 3 mammographically (30.6% of malignant

cases). The following table demonstrates these points and compares them with

results from BREAST AID.

   

Table 20: Sensitivi S cifici

and Predictive Values far Mammoggghy

Alone with Comparison to Test Characteristics of BREAST AID*

 

 

 

 

 

Se Sp PPV NPV # Biopsies % Missed
Test (%) (%) (%) (%) Needed Cancers

Mammogram

Category 3, 4 or 5 69.4 95.0 62.5 96.3 40 30.6
Mammogram

Category 4 or 5 38.9 99.7 93.3 93.2 15 61.1
BREAST AID

Probabiliy a 3% 97.2 85.6 46.1 99.6 76 2.8

 

 

 

 

 

 

 

* Se = sensitivity, Sp = specificity, PPV = positive predictive value and

NPV = negative predictive value.

The findings of FNAB were not much better. Twenty of 36 malignancies

were soft or rubbery on internal consistency, for a sensitivity of 44.4%. Ten of

36 malignancies had FNAB readings of benign epithelial cells (7) or a few

scattered atypical cells (3) for a sensitivity of 72.2%. The following table

demonstrates the FNAB cytology results and compares them with results from

BREAST AID.

108

 

Table 21

SensitivityI Specificity, and Predictive Values for Fine Needle Asplration
Cytology Alone with Comparison to Test Characteristics of BREAST AID“

 

 

 

 

 

 

 

 

 

 

 

 

Se Sp PPV NPV # Biopsies % Missed
Test (%) (%) (%) (%) Needed Cancers
Many atypical,
suspicious or malignant
cells on FNAB 72.2 93.7 57.8 96.6 45 27.8
Suspicious or malignant
cells on FNAB 63.9 98.0 79.3 95.8 29 36.1
BREAST AID
Probability a 3% 97.2 85.6 46.1 99.6 76 2.8

 

* Se = sensitivity, Sp = specificity, PPV = positive predictive value,
NPV = negative predictive value, and FNAB = fine needle aspiration biopsy.

One could argue that all cases of either suspicious or malignant
mammogram or FNAB cytology results would result in biopsy. One or both of
these results occurred in 26 of 36 cancers in this study. The ten remaining
malignancies had indeterminate or lower findings on both mammography and
FNAB. The challenge for the surgeon is to decide how many lesions assessed
to be equivocal or less in nature should be biopsied. BREAST AID was capable
of sorting out this dilemma in 9 of 10 cases, requiring only an additional 40
biopsies to make the distinction. Table 22 in the appendix summarizes the
findings of these ten cases.

DISCUSSION
When evaluating a breast mass, the most important question facing the
physician and the patient differs from the usual one to be answered in the
evaluation of a physical sign. Rather than the question “Mat is causing this
problem?”, the question is, instead, “Does this mass represent cancer?” When

the challenge of the clinical evaluation is to rule out cancer, the clinician is likely

109

 

to tolerate many false positive results in a diagnostic test, favoring high
sensitivity (and high negative predictive value) at the expense of lower
specificity (and lower positive predictive value). The consequence of this
approach in the case of a breast mass, in which the diagnostic procedure is
mass removal, is that a large number of biopsies for benign disease are
performed in the process of ruling out breast cancer. Although many consider
open biopsy to be a minor procedure with few consequences as compared with
a missed diagnosis of breast cancer, women pay a significant emotional toll
associated with the anxiety of the procedure itself. In addition, there is
emotional distress related to the necessary delay between the clinical
evaluation and the biopsy, and the biopsy and the final pathology result, a
process that can take several days or even weeks. In addition, although skin
scars usually resolve over time, some women experience breast disfigurement
as a consequence of open biopsy, and subsequent screening by CBE and
mammography can be more difficult after open biopsy because of breast tissue
scarring. Finally, the economic costs of liberal use of tissue biopsy to rule out
breast cancer are enormous — approximately $4 billion per year in the United
States alone. This has resulted in tensions between health economists and
surgeons, with the former group quoting unnecessary biopsy rates and the
latter group citing the expectation of diagnostic perfection on the part of patients
and the legal community.

The ability of BREAST AID to assist in the management of breast masses by

triaging malignant lesions to biopsy and benign lesions to follow-up is based on

110

its ability to incorporate several complementary pieces of diagnostic information
to arrive at one final probability. Although scoring systems have been
attempted applying the principles of triple diagnosis in the past, arriving at a
single, and thus definitive, quantitative score has been difficult. This is due in
part to the lack of definition of the clinical impression component of the triple
test, which depends on clinical expertise with undefined results of CBE, and on
results of several diagnostic tests which result in an overall gestalt Impression
of benign vs. malignant disease that is difficult to characterize. BREAST AID
removes this undefined characteristic of triple diagnosis while incorporating all
of its fundamental principles. It adds to triple diagnosis as well, by incorporating
important epidemiologic information into the interpretation of the test results. By
arriving at a quantitative result that predicts accurately for malignancy, it helps
remove the diagnostic uncertainty that a clinician faces when evaluating
equivocal findings.

Potential Stugy Biases: Patient Selection and Pppulation Characteristics
The question of the representativeness of the population in a study to the world
of patients who present to their doctors with a symptom is an important one,
especially when the study is aimed at developing a mathematical model to
predict for a target disorder. How representative the population under study is
in this report, then, becomes critical when judging the applicability of the results.
To answer this question, it becomes necessary to consider the criteria for

selection, and the causes of non-inclusion once selected.

111

Selection Criteria
The selection criteria for entry into this study required the performance of
certain diagnostic tests within a defined time period. By definition, then, the
study design resulted in the elimination of patients who undenNent immediate
biopsy without mammography and/or adequate FNAB results. It is unlikely,
except in a very young women, that an open biopsy would be done without
having obtained a diagnostic mammogram. It is much more likely that an open
biopsy might be done without performing an FNAB. This is most likely to occur
when the surgeon is certain of the need for biopsy or the patient Is adverse to
F NAB. It is possible under these circumstances that more malignancies would
be present in the group that did not undergo FNAB than in the group that did. if
this is the case, then the malignancy rate would be underrepresented in our
study. The effect that this might have on the mathematical model would not be
substantial however, because the model would certainly triage patients with
clear indications of malignancy on either mammography or FNAB into biopsy.
Of greater concern is the group of patients who underwent FNAB in their
diagnostic work-up but whose results were hypocellular and not repeated. This
group was not eligible for this study according to our selection criteria, but data
was collected on these patients. The fact that patients with hypocellular FNAB
results have a higher rate of benign disease than an overall group of patients
who undergoes evaluation with FNAB has already been discussed. To
substantiate this, and to document the outcome of our group with hypocellular

FNAB results (31.8% of all subjects who underwent F NAB), a separate analysis

112

will be conducted later. The other 21.6% of patients who were not included in
the study will not be analyzed further. These patients include 27 in whom the
charts could not be located, and these are likely to be missing at random. The
36 patients who did not have a mammogram evaluation within the time period
specified by the study probably do differ from the overall population studied.
They are more likely to present with a clinically benign finding than the overall
group. In contrast, the 39 patients who had an open biopsy without a fine
needle aspiration biopsy are likely to have a higher rate of malignancy than the
overall group. This is especially true in the practice at the CBHC, where it is
routine to evaluate patients with F NAB unless he patient prefers to defer the
procedure of unless the clinician judges that the FNAB result would be unlikely
to change management options.

There is also the possibility of referral bias to consider. One should question
whether patients with a breast mass referred to a breast clinic are
representative of patients with breast masses overall, and if patients referred to
the CBHC specifically are representative. If any bias exists, one would expect
an overrepresentation of the target disorder. This study documents a
prevalence for breast cancer of 10.7%. Whether this prevalence is reflective of
the prevalence of cancer in women with breast masses will of course be
dependent on the age of the population being studied and the stages of disease
in the sampled population. Most of the surgical literature quotes much higher
malignancy rates (Tables 2-5), but in many of these studies, the rates are given

only for patients who underwent open biopsy. This suggests either referral or

113

selection bias possibilities in these studies, and a resulting overrepresentation
of malignancy in some of the surgical literature. A study of 2000 FNAB
aspirations by Horgan et al from the University Hospital Breast Unit in Galaway,
Ireland specifically states that they used consecutive cases of breast masses
seen between 1982 and 1989 in their study and that they did not attempt to
select patients for biopsy based on clinical findings. Their malignancy rate of
13.2% is close to our own 130. Selzer’s report of 1,000 consecutive patients

referred to his surgical clinic with a breast mass documented an 8.2%

malignancy rate in his population 5. In Barton’s retrospective study of 2400
patients in a primary care setting who were followed over time to measure

frequency of breast complaints, 157 patients presented with or were discovered

to have a breast mass on CBE; of these, 10.7% were malignant 169. It seems
likely that the malignancy rate in our study is representative of the women with
breast masses overall and is not an underrepresentation of the frequency of
malignancy in women with a breast mass.

Another reason that may account for the prevalence of malignancy in our
surgical clinic is the recent increase in mass screening rates with CBE and
mammography. The result of this trend is the identification of more breast
masses than would otherwise have been identified. Although the goal of
screening is to identify cancer at an earlier stage than would be the case
without screening, the inevitable consequence of this effort is the identification

of higher rates of what eventually prove to be benign findings.

114

Population Characteristics

The mean age of 47.6 years in our study population is similar to others that

report age in women with breast masses, the range of which varies between 45
years and 54 years 7. 89’ 9°: 93: 94: 96. 97’ 101. The exception is the study by

Grobler et al, whose average age was 62 years 75. The mean age of 46.8
years for benign lesions in this study is essentially the same as a large multi-

institutional study of Ciatto et al, who reported a mean age of biopsy for benign

lesions of 46.7 years for 754 collected cases 13.

The mean BMI in this study was 24.73 kg/mz, which compares with a mean
value of 24.3 kg/m2 in the Adventist Health Cohort Study that studied exposure
variables related to cancer and heart disease in 20, 341 women. Some
question may arise with the comparison of BMI in our population with the
Adventist Health Cohort Study because the latter population is largely
vegetarian. However, in a population-based case control study by Stanford et

al of women 50-64 years of age, 45.6% of the cases and 50.2% of the controls

had a BMI below 23.7 170. Thus, the BMI in our population seems
representative of women in general.

The average age of onset of menarche of 12.8 in this study compares with

the same value in the Adventist Health Cohort Study 171. In Rockhill’s study of
population attributable fraction estimations for breast cancer risk factors using
the data from the Carolina Breast Cancer Study conducted between 1993 and
1996, 19.7% of the 958 Caucasian women had an age at menarche of less than

12 years. This compares with 18.3% of our population. These findings support

115

the representativeness of our population. Further support is found in the 1987
National Health Interview Cancer Epidemiology Supplement, which reported
that 18.3% of women less than 50 years of age from the Midwest have an age

at menarche of less than 12 years and that 14% of women 50 and over from the

Midwest had such a history 172.

The frequency of nulliparity in our population was 12.5%. This compares
with a frequency of 14.6% in Rockhill’s study 48 and 10.8% in the study by

Stanford 170. Age distribution could account for the variations in frequency of
nulliparity in these two studies, since the former included women 20-74 while
the latter was in women 50-64. The frequency of nulliparity in our population
appears representative, confirmed also by the frequency of nulliparity in women

between the ages of 50-79 living in the Midwest to be 13.5% in the 1987

National Health Interview Cancer Epidemiology Supplement 172 This study,
which is a supplement study to the larger National Health Interview Study
(NHIS), found a prevalence of nulliparity of 29.6% in the Midwest for women 20-
49 years of age.

The average age of first delivery of 24.7 years in our study is higher than in
the Cancer and Steroid Hormone Study of women 20-55 years of age, where

the mean age of delivery was found to be 21.8 years for control subjects in

seven of eight SEER regions 173. The distribution of age at first birth is also
not representative of predicted frequencies. Our study had frequencies of
16.0%, 33.2%, 33.9%, and 16.8% for age of first delivery of <20, 20-24, 25-29,

and >29 years, respectively. This represents a lower frequency of early age of

116

delivery and a higher frequency of late age at delivery compared to other

studies 170. 172. It is possible that these distribution of factors in our
population reflects the fact that our population is at higher risk for breast cancer,
by virtue of the fact that they are presenting with a breast symptom.
Alternatively, these findings could be explained by the relatively young age
group that we are studying, and the trend of delay in childbirth for this group of
women.

The frequency of number of deliveries in this study was 2.4, 2.3 for the
benign group and 2.5 for the group with cancer. This compares to the mean

number of deliveries in the seven SEER regions other than the San Francisco

Bay area of 2.6, and 2.1 for the San Francisco Bay area 173, and indicates
population representativeness.
The average age of menopause of 46.4 years compares with 44.4 years in

the seven SEER regions other than the San Francisco Bay area, and 44.9

years for the San Francisco Bay area 173, and also indicates population
representativeness.
The rate of ever use of oral contraceptives in our population was 78.3%.

This compares with a frequency of 70.7% in the Adventist Health Cohort Study

171 but only 41.3% in the Nurses’ Health Study 174. (The latter finding is
probably explained by the age of the cohort studied and the time of initiation of
the study.) The unexpected protective association of OCP use with breast

cancer found in this study has already been discussed.

117

The rate of ever use of postmenopausal estrogens in our population was

82.1%. Of these, 68.8% were current users. This is higher than ever-use of
post-menopausal estrogens of 52.4% in the Adventist Health Cohort Study 171,
41.4% in the Nurses’ Health Study 174, and 34.1% in the Midwest as reported

in the 1987 National Health Interview Cancer Epidemiology Supplement 172.
This may reflect a practice pattern specific to the Lansing area, or the current
prevalence of HRT use in the population in the US as a whole, but the
possibility that hormone replacement therapy may be associated with an
increased risk of breast masses must also be considered.

This study found a history of breast cancer in first-degree relatives to be

21.1% and in mothers only, 14.1%. This compares with a family history in first-

degree relatives in Rockhill’s study of 22.0% 48. However, these are higher
numbers than those usually quoted. For instance, the Nurses’ Health Study

found an age-standardized distribution of family history in first-degree relatives

in the Midwest to be only 8.2% 175, and the Adventist Health Study found a
breast cancer frequency in first-degree relatives of only 4.2%. It is possible that
women with a family history are more vigilant about breast self-examination
and/or have greater medical-seeking behavior than women without a family
history. This would lead to an increased frequency of breast lump detection in
the former group.

The population characteristics in this study are similar, but not equivalent, to
those of others. Just how the differences might affect the results of the

multivariate model is open to conjecture at this point. It is very likely that

118

because the number of women with the target disorder is low in this study (36
cases of breast cancer), that additional risk factors may contribute to the model

once the numbers of patients is increased in a larger study.

SUMMARY AND CONCLUSIONS

Assuming that BREAST AID is a mathematical representation of the
decision-making process facing a clinician, it is not surprising that the surgical
approach to the evaluation of breast masses has traditionally required open
biopsy for a definitive diagnosis. In this study, the final predicted probabilities
for a breast mass based on the final steps in Stage 3 and representing
malignancy ranged between .004% and 99.99% for the population of breast
masses overall, and between 3% and 99.99% for the lesions triaged into
biopsy. Despite the low predictive probability for cancer of 3% from a raw value
standpoint, 46.1% (35/76) of the lesions whose final probability was at or above
this level had malignancy. It is unrealistic to think that clinical decision-making
is capable of making differentiations this small. Herein lies the potential
application of this model; it combines multiple pieces of routine diagnostic
information, much of it inter-related, into one final prediction that is capable of
efficiently triaging patients into open biopsy or follow-up. In this regard, it
satisfies both the economic and medical-legal concerns so commonly
associated with the symptom of a breast mass.

BREAST AID is a unique prediction model for breast cancer that has the
potential to improve the current accuracy of diagnosis while minimizing the rate

of open biopsy. I believe it does this without compromising the timely diagnosis

119

of breast cancer. This latter caveat is dependent on close follow-up of non-
biopsied lesions for a minimum of 3 months and for at least one year if the
mass persists. If the size of the mass increases or the consistency of the mass
becomes more pronounced on follow-up, open biopsy is indicated. Prior to the
widespread application of BREAST AID, a demonstration study to validate and
test the generalizability of the model will be necessary. This will require a large-
scale mum-institutional study incorporating large numbers of patients with both
malignant and benign disease. If the predictive ability of the model is
confirmed, BREAST AID has the potential to conserve health care dollars while
simultaneously offering both patients and physicians the assurance that clinical

follow-up is a safe alternative to open biopsy.

120

APPENDIX

121

TABLE 1 Probability of developing invasive breast cancer in

 

SEER" areas, women, all races, 1987-88

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, AGE: w j * f i1 ’ .1 w ‘ Cumulative probability (%)“
20-24 .005
25-29 .04
30-34 .16
35-39 .46
40-44 1.1
45-49 2.0
50-54 3.0
55-59 4.2
60-64 5.8
65—69 7-4
70-74 8.9
75-79 10.4
80-84 1 1.5
85-90 12.1
90-94 12.5
95+ 12.6

 

 

 

 

* SEER = Surveillance, Epidemiology and End Result program of
the National Cancer Institute

** Cumulative probability of developing invasive breast cancer from

birth

122

 

come 88: co :20 .2 9.33 9

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8 8 8 8 8 :98 so: a 8 8.8 6: e8 82 8.85:?
8 R 8 8 8 828 so: we 02 ~55 8 8 82 95>
8 2 E 8 t. 8v .88 ca 62 New 8 82 .58 8>
8 R 8 2 8 :98 6e we 62 8% so: 8m #82 .9888
8 vs 8 8 2 scam so: em 3 88 6a 32 22 .8882
8 R no 8 8 528 so: 2 8 Es 8 8m 82 8:65.
8 .8 8 8 8 8238 aé em 8 6.5 8.2. 8: :82 .___cecs_
B 2 8 8 2 826. so: 88 8 8v .88 at 82 23228..
8 E 8 8 S 88 so: we 2: 52m 6: N8 22 .583.
8 8 8 8 as 155 a be 02 as; 8 48 a 82 82.58.
8 8 2 8 8 88 so: he 62 :96. so: 88 82 68885:
8 8 8 8 8 8v .28 we 62 828 so: «8 82 .8255...
8 8 8 8 K aev 8 he 2: 88 8 Ba 82 839.0
8 2 8 8 8 2.5 8 we 2: 8.8 .8 com .282 .= :98
8 R 8 8 8 ace 8 we 62 828 .o: 82 .282 ._ :98
8 8 8 8 8 8v :8 .N .8 aé 9 8 B2 .2855
8 E E 8 t. 8v :8 .N .8 are 2. v8 82 dead 5
8 on k 8 8 i we 2: t 88 82 .8880
8 E 8 8 8 8.5 8 ca 82 68 9. a: 82 8:8

at 35:25 a...

2mm :55 a: 9am 8.885 2 88> co£=<

>mz >9... comm meow 2.32 on.» onus. .>>o=on_ >835 om<

 

 

 

 

 

 

 

 

 

 

 

 

885.92... 85 So: 3.858820 88... - cocoa—2 888.6 83898.". 53> .3803 c. Eaxm .88.»??— N 2nd...

123

«mm 758.. 88638 = :98 6.3 78an 88638 _ :98

958:: 82:88 8 88 8283288 8:388: .8955

8:288. E222: 8:8 :mEon 508 38852982 38 8E3...

8:88:88 9 28 8.8.8: 2.8862282 38 .025...

8880: E85888 8898_:8 :0 Eu v :85 88888.: 60:88.8: :8 6 88:88.2

>828 So: 828588 8:88:88 8322.0 2825.0

E88898_:8 b8___x8 Lo :ozogcc :28 o: 85:9: N 555, 8888858: .8835 So: 8:o_88_ =4.
88“ 888 58288 ._. >8 358:8 88888 88:88 :_ 38 860.92: .6 88:83.2.

You 888 ”88888 .3 858:8 8% v6 88:83:...

<<

++

 

124

New 7 Pom? 02038 = :95 ”mum Tan? 08638 _ :95 .2
8:80:80 0. 2:0 8:28: 2.850.985 38 0:5... +
85:08 N 55:5 8888858.. 088803 “0: 8:288. __< s

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8 8 8 2 8 :88 .o: 8 8 :88 5: 88 82 8.85:?
8 K 8 8 e8 :88 so: 8:: :2 a8 «8 8: 82 98>
8 8 8 8 t. .1818: s: :2 8: 8 82 .80 :o>
8 8 E 8 8 :88 .o: 8: o2 8% so: 8 82 8:88:88
8 R 8 8 8 :98 so: 455 2 8 28 9. F8 82 8:65.
k 8: 8 2 8: :98 so: as; 8 8 88 8 82 82 .___o:8.2
8 8 8 8 2 :98 so: 8.“ 8 8v s8 «8 82 8:88..
8 8 s. 8 s. :88 so: 8:: o2 :88 6: N8 82 .8388.
8 E 2 8 a: 1:8 8 8:: 2: Es 8 e8 82 8:89.
8 8 8 8 8 :98 .o: 8:: :2 :88 so: 8: 82 88:85.8...
8 8 8 8 8 :86. so: 8: o2 8% so: «8 82 88:85.8...
.8 8 8e 8 :8 8.5 8 c: 82 88 8 82 82 .5390
- - - 8 2: :88 so: a: 82 :88 so: 8: 82 80.868
8 8 8 8 8 3.5 8 8: o2 82816: as. .282 .= :98
8 8 8 8: 8 Es 8 8: o2 :88 so: 88 .282 ._ :98
8 E 2 2 8: 8v s8 .8 .8 38 t. 88 82 .2855
8 8 k 8 t. 8v s8 .8 .8 8.5 8 88 82 685 _o
8 E 8 8 3. 88 8 8: we" 6.5 8 8 82 .588
e.
>az >2. 8:8 8:88 at so: .83 85:95 98: 88>. z 88> .553.
9.8.2 on? 888.2 cargo—.0". >305 8m<

 

 

 

 

 

 

 

 

 

 

 

 

8.38:3... 85 50...— 8o=8_:888._8:o «88... 1 888885. .8885 83828:. 5.3 :8803 c. 33.608585. 8 8.98...

125

88.85:...8 8.8: 88.6 5:0 88888.: 2.0m 2

888 T .83 88.838 .. 580 .83 7832 88.838 . :95 5.
2:0 >0:8:m_.8:. 5.2. 2:838: .0“. <

8:80:80 0: >.:0 8.885 2.8052088... 88.8 5:5... +
8:“:2: N 55.2. 8888868.. 88.8805 .0: 8:0.88. ._< ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8 :2 :2 8 8 82878: 2:5 2 28 8:5 8 88 82 8:5...
8 8 8 8 8 828 .o: 8 8 :85 8 82 82 .___8:8..._
8 88 8 8 2 :8... .o: 8.8 8 8v .8: 88 82 22:88..
8 8 8 8 8 :82: .o: 8.: :2 828 .o: 88 82 .888.
8 8 8 8 .8 +25 8 8.: :2 3:5 8 88 82 8:832.
8 8 8 8 2 828 .o: 8.: :2 :82: .o: 8.2 82 :8qu
8 8 8 8 8 8:8 .o: 8.: :2 8:8 .o: 88 82 .8858:
8 8 8 8 8 :38 .o: 8.: :2 828 .o: 8 8 82 888:8...
8: ..8 8 8 2.. :25 8 8.: :2 :25 8: 82 82 8.8.0
8 8 8 88 8.. :28 .o: 8.: :2 :88 .o: 282 82 .28.“.
8 8 8: 8 88 828 .o: 8.: :2 828 .o: 82 82 8.88...
8 8 8: 8 8 es. 8 8.: :2 828 .o: 88 .282 .__ :93
8 8 8 8 8 €55 8 8.: :2 828 .o: 88 .282 ._ :98
8 8 8 8 8 8v 088 .8 .8 2:5 8.. 88 82 .9855
8 8 8 8 8 8 v2.8 .8 .8 :25 8 :8 82 .28.: _o
8 8 8 88 8 828 .o: 8 8 828 .o: 882 82 .28.:
8 8 8 8 2 828 .o: 2 8 :82: .o: 82 8:2 .880
88 8: 8 8 8 :25 8 8.: :2 :25 8 8: 82 8:8
8 8 8 88 .8. +8 2 8.8 .o: 8 22 .882 .8328:
.8
>n.2 >8. 08am 8:88 8:8”. .80. 88.8 85:02:. 3.. 888m. .888... z 88> ...0£:<
9.85. 888.2 8.3.32.0“. >8ao.m 8m<

 

 

 

 

 

 

 

 

 

 

 

 

9:56..me 0r: SOC flOBmtquNhNLO «ah.

I 88888.2 .885 8.0828... £..s :8EO>> :. 30.9.6 :0_:8...n.8< 8.0882 8:. n.

V 838...

126

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

09:38 “O: 9.9:??? mums—00.09.: 556:: 0>Emoa a ma 690328 3:50. _moo>_36m in
2:0 mucozma Uflwaofi how 95 mozmtmuomhmco “mm“ 620:6

”scamjaon .9399 52.8 m_ $855.3 5 89:5: .3533 uflwaoéflfi 30:8 2 .5383: “mg“. g.
8 mm 5 mm x. 82m 6: m2 2: 52m 6: ~va 52 .093
mm 8 8 E. 9. :95 6: m2 o2 5% 6c owfi .82 18:3
3 S 8 8 3 52m 6: NF :96 “o: 5% 6.. wow 82 .9862,
8 8 8 E 5 52m 6: m mm :95 6: F2 82 .5628?
:w 3 8 E 2 :95 “o: «E 2: ES 2. New E2 5855
8 2 8 3. 5m 52m 6: «E 2: 82m 6: now 82 9&5
8 08 08 8 5 52m .8 «E 2: Es mm 9. 32 95>
mm E mm mm mm :93 6: E: 2: :96 6: SN #82 .9352m
mm 8 8 8 3 :95 6: E: 02 :95 8: $2 82 .0925
mm 8 8 8 E 5 mm :95 6c 88 6: mm :95 “o: o E 88 Sim

mm mm 5 mm 3 52m 6: vm mm :25 “8 93 122 $7.55.:
8 8 mm mm NV 5% 8: Sc 2: 52m 8: we 52 £3
3 8 mm 5 gas 2 :96 “o: 8% 6: X” :96 6: “mo 82 .Essoaa
mm mm 2 3 AS. :95 6: 3c 2: 52m 6: 8m 32 .__mz_o

on
>n_2 >mm ovum 95$ 23. 3.8 on? Ange—.95 ex; 8mm. 3.62: 2 am; ...o£:<
9:22 anus. Q:.>>o=o..._ >305 om<

 

 

 

 

 

 

 

 

 

 

 

 

6.233 e asap

127

.0E050EE0E 05000: 5.2, 0005032 .0 35:86.5 .00:0:._.:0:0 .6 0:0.00. .0:0_0.00:0n:w. in.

0032.2 50: 0.:0500 mo PR 0%
:0.:0n 2.00....00508808 0:0 >._00_:..0 0:02, 0:000. ..< .0

>303 :000 :0 00... 0005 3 =0 00 b00200 5:0:m__0E 05 0E. 0>on0 .0 m 3 00.000 .8 5500000005

5:000:00: 0.0: 0.0083: 05 5:05:05 :00 0>000 0:0 0 0:0 00000.0 5.50 .6 0.90.0000 00 v ..0 0 .6 000000 :0
00000 :000 :00 $09 .8 0:_0> 050.0000 05000: 0:0 05.000 0:0 5.0.0.0000 5.5.050 0 .00: 0.000 «005 0.0.; ++

3.00:00 05 2:0 0.30.5 2.00.0220... 00.0 006:... +
05:08 N 55.3 0000050: 00.0003 00: 0:0.00. ._< ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

- - - 3 02 E5 8 0.: 02 525 00 8 52 .20.?
02 v0 E 02 v0 5% 0o: 0.: 8. E5 00 0.. 82 90>
2: 00 3 02 #3 820 .8 5.5 5 £0 e5 0.. #00 «02 .2058
2: E S 8. 00 520 8: 55 S S .55 2‘ 60 $88 0522

- - - 8 0v 5% 0o: .0 00 65 om 8: 82 .___m:0_2
8 E 00 8 :V 520.2 0.: 2: =90 .8 08 002 03:20.
02 E S 2: % +455 3 02 02 E5 v0 v00 32 55.500.
8 00 8 00 33 520 6: NF 3 520 .8 $00 008 0:39.503
2: .0 E o? 5 5%. .8 0.: 09 820 6: 000 $2 $20.50:
02 00 00 2: 00 820 0o: 00: 02 520 6: 30 v02 52052..
02 8 00 2: 8 6.5 8 0.: 09 9,5 00 v2 82 00320
8 E 00 00 on ouv $8 .0 .00 25 0V :0 $2 .3905
8 00 8 00 0V omv $3 .0 .00 ~95 9. .00 002 .9005

- - - mm 2: 820.8 0.: 02 00v $02 00 .8058
02 00 00 02 S. 6.5 on 0.: 2: 6.5 0.. 00 89 0250

3.0 .0555 3...
>02 >00 0000 0:00 300 E005 00.0 0: 050m .0003 00> 0053.0
9.0.2 000.2 {50:00 >000.m 0m< z

 

 

 

 

 

 

 

 

 

 

 

 

0:30:00... 05 805 0050.005000030 .00... I 00000.... «000.5 0300.00 5.3 0080.5 :. 0.00: 0.0 0.0... .. m 0.00...

128

 

 

.5

_:fi JmflGF.y

 

ON

 

my

 

or

 

t

 

mp

 

mp

 

3.

 

2‘

 

NF

 

:

 

or

 

 

 

CDIDONOOOflNMOPI-O

w¢OwoooFOwoocoR
1-5

 

5‘

E.

 

_‘v

;

 

on

on

 

NF

NF

 

o

OOOOOOFOFOOONNNOPFO_2-

 

 

o

 

NMV’IOCDNQO

 

 

 

Hmmoom Mafi

 

 

 

>35 $55.35 Q< 5<mmm so: 958 3 938m c _com 9, £29.22

w 035...

129

Table 7

Definition of Test Characteristics Used in BREAST AID

 

 

 

 

 

 

 

 

 

 

 

 

 

CHARACTER SYMBOL
TERM DEFINITION FORMULA* FORMULA*
Sensitivity Proportion of patients TP a
with breast cancer who TP + FN a + c
have a positive test
Specificity Proportion of patients TN d
without breast cancer TN + FP b + d
who have a negative
test
Positive Proportion of patients TP a
predictive with a positive test who TP + FP a + b
value (PPV) have breast cancer
Negative Proportion of patients TN d
predictive With a negative test who TN + FN c + d
alue (NPV) do not have breast
v cancer
Likelihood Odds that a given test
Ratio result would be a
expected in a patient Sensitivity a + c
who has breast cancer 1 - Specificity b
compared to one who D + d
does not
An acronym for Receiver or Response Operating
ROC Curve Characteristic, which is a graph representing a plot of the

 

sensitivity and 1 -— specificity rates for various cut-off points
for a diagnostic test. A perfect diagnostic test has a
sensitivity of 1.00 and a specificity of 0; the “best" cutoff
point represents the point on an ROC curve that is closest
to the upper left-hand comer of the graph (Figure 1).

 

Refers to 2 X 2 table shown in Table 10

130

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

v o.” .050
N 3 «Eon:
N 3 £8.82 a".
.N «.2 «Eocoumosm
cm 1.3. 2.285.".
8 er... 25.3.. 5.5%
w Qm 23.90%: .3334.
N m.., 3.2.. mEOEofio 5.390..
2 3. Ex :9:
1 9n 3.9:. mEoEPEo .3030
m En mEoEQfio .2390. 0232..
um 98 mEoEan _Sosu o>.wm>:.
8 9mm $025

2. H_ . .. , 33%... s2 2._ , . .33m .3058

 

 

 

32 E535»... ucm vamp EonEo>oz 525.03 ovumsfi>m
«$me 550.5 can E0... «no.3... m2 .0 3.33m. >moo.m a 22m...

131

00.20.30... 3:0 0.990 505.; .0 5.? 009.050 02.0.0505 00.03.05 .

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

.5 020.00 20.2mm, ,, ,,.. 1
,2 2 >20... zmmo 2.0 0530.... 50.05....

 

 

 

.s. «325. .

o; o._. _ 0.. 0.3 mém - - 0N. me adv ow. .Emn. c0>
v... Om 0.0.. For .05.. fiov v.0 0... 00¢ om. ..__0t0.>.
- - - ode 0.8 m... - IN 0.0 0.3. v0. .20.:me
0.0 - - 0N. mdm 0.0 - 0.0 0.0 0.5 mm. .050... .0
Wm - EN v.3 Bum - - - wd v.9. om. .._0_.3m
0.0 me m. 0.3 TS. 0.0 we o.m Em 0.8 00. £0.50
.050 0E2... 0.00.002 080:000 0.3.6 0.0.0.... 20.. man. .0050 00:00
.0“. 65¢ -05.". -..0n.>I 32390.. .8030
30.054. 0>.00>:_ 02002..
.3. v.21 I0.I

 

 

20:54. .

 

 

030500.53... horn. :23 0:300”. 30.90.... >035 E0050 .0 :00..00E00 m 0.00...

132

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.18. F med .0 F. t... F . FF 2...?
9F CF. o... «E «5&3
.3. F «Fe .8. 0.8 aw... .805 $0.... .600
26.8. N... F .F F. 3: E. 85:. 0%
80-3. .8. F a. v. F F o. 8:8. 3.0.3
8.0%. 8. a. a.» .2: 8.0:. 3-8
o. F a. «d S 8.0:. «muv
we. n 5.0 an. m.oF «8 29......
3.0.0... Ed a. 03 an 2258. SFA
3 .9... 3 8m 28:2. 2:5
2. F NF.~ an. 0.8 8.” 2.0.25
«.8... F a... a. «.2 3 8R
n..F-F.F hm 8 F. 3: 5 9.8.. $5...
0.3.... F.~ .3. F6. 3F 28.. 212.
3 a. m... «2 e8. mnuv
.8. n 3... an. ....oF 8n om<
.0 $33 1.0.3.3.. 2.3. «.80 30> .. .0 20:00 .5 035:... 25.8. m._m<.m<>
0.020.:— 10 0:0”. 00:00 .0 09:52
8:00:80 085......

 

 

003 30350". 0:0 <03 .0nE0>oz :00250: 00.03_0>m 30.00.. .0005 on» :. m0.n0...0> 8.0.005
00.30300 0.020.800.0m .0. 0.300: :0.00050m. 000.00.. 0.0...0>.m 0:0 00:00 .0 00:0_0>0..n. 0F 0.00h

133

00: 30.05. .:0E000_00. 0:0E.0£0m0 .0 0.00.. on A 0:0 >E0.00.0:000
_0.0.0__:: 4+ >Eo.00.0.0>_.=>:.0.00.0:000 _0.0.0__£0:.:OE NF 55? 000:0... 02 n .003000:0E.0on_ n3.
0.00.. on «V 000 0:0 .2.-5.00.2300 _0.0.0__:: 4+ >E0.00.0.0>£00E «_- cEgB 000:05. n _003000:0E0.n_ n ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0-00. 00. .0. 0.0 0.. .052
3 .00. 0.3 000 .20
0... F 00. .00. 0.2 000 00.0.0 0000000...
5.0.3. .0.F E 0.2 00 0.00. 00%
0F .0. 0.2 00 0.00. 00 uv
00. F 00. .0: 0.0F .0 00:03:05. .0 00<
- 00.0-00. F 00.0 :0. 0.0 F 0 F F 00 003000000000.
3 .2. .0 000 0 000008052.
-F0.v F 00.2 .00. 0.2 000 00.0.0 00000000:
00.0-00. 00F .0. 0.: F0 28. «Fv
0F .00. 0.0 0.0 0.00. «F "A
00. F 00. F .00. 0.2 000 0:20:02 .0 00<
00.0-00. 00. F AF. F. FF 0 00>
3 .00. ..0F .00 oz
.0. F F0. .00. ..0F 000 .8000 .0020
20.0.: _0:00.00_
_0 .000: 1.0.0004 2.00 00.00 «20> 0 .0 20:00 _00 00050:. 0:202 m-_m<_m<>
0_0>.0.:_ mo 2.01 .00:00 0.. .0952
000000000 08500.:

 

 

00.08. 2 0.00.

134

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.3. I 0 F. 00. .00. 0.. 000 .00..:0..0>m
3 .0.. 0. F0 0. .052
00:
..0.v F 00.2 .00. ..0. 000 900800000 .05
0.....0... 00.F .0. 0.0. 00 0.00.. 00K
0F.0-F... ...F .0. 0.0 F0 000.. 00-00
0. F .0. 0.0 00 0.00.. 00-00
00.0- .0. 00.F .0. 0..: 0.. 0.00.. 0.nv
0.. 0 00.F .00. 0.2 000 0.03.00 .0... .o 00...
0 F0 - 00. 00.F .0F. 00F F0. 0R
0F.0 - 0... 00.. .0: 0.0. 0F. 0
0.. .0. 0.0 0.. F
00. 0 0.. .00. 0.: 000 00002.00... 00.002
00.0-F... 0F.. .0: 0. FF 00. 02
2.0.00. 00. .0. 0.0 F0 0
0.. .0. 0.0. .0 F
F0. 0 0F. .00. 0.0. 000 00.000000...
.0 39552
.0 0000.. 100.00.. 0000. 0000 00.0., a .0 20:00 .00 000.02.. 0.5.8. m..m<.m<>
0_0>.0.:_ «O 2.01 .00:00 0.. .3532
000000000 085.00....

 

 

.0008. 2 0.00.

135

3:0 000.03 .00:000:0E.000 0. 00.00.00”. ...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.0.0. 0.0 E 0.0. .0 .00:00
00.0-00. 00.F .0. 0.0 0F .00.
0.. .0. 0.2 .0 .052
.0. 0 00.. ..0. 0.2 0.. 0.0058000”.
$3000.00 00:
0F..-F.. ..0 ..F. 0.2 00 0.00.. 0 "A
3 .0. 0.0 00 0.00.. 0 v
0 .. F 0... .0 F. 0.0. 0 FF 205000.000.
:000..0m
€039.30
00.0-00. .00 .0.. 0.8 .. .00:00
00.0-00. 00.0 .0. 0.0. 0. .00.
3 .0. 0.00 00 .052
.0. 0 .0. ..0. 0.2 0F. 0000.80.00:
:0mob0m .0 00...
.00-00. 00. .0. 0.0 .0 0.00.. 0FA
00.0-00. 00. .0. 0.. 08 0.00.. 2 I 0
3 .0: 0.0 0 FF 0.00.. 0v
00. 0 00. .00. 0.. 000 002.080.0000
3.0 £059.30
.0 0.00: r 6.000.. 2.00. 0000 00.0., 0. .0 0.03.00 0.0 .0005... 000.00. m..m<.0.<>
0339:. mo 03...”. .00:00 0\0 .3532
8:00.000 085.00....

 

 

 

 

00:8. 0. 0.00.

136

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

:0 F00. 00.F .0. ..00 00 0.00. 00uv
0F .0. 0.2 00 0.00. 000
00. F 0.. .0. 0..F 00 0.000005 F0
.0505. 00 090
00.0-00. 00.0 .0. 0.2 00 00>
3 ..0. 0.0 000 oz
00. F 00.0 .00. 0. FF 000 .0505. -
30.0.: 2.80“.
F..0.0.. 0..F A F F. 0.2 00 00>
3 .00. ..0 .00 oz
0 F. F 0.0 30. 0. FF 000 00.000. .2 I
3000.: 2.5.0".
00.0-F.. 00.F .0: 0.0F .2 00>
3 ..F. 0.0 0.F oz
00. F 00. .00. 0.: 000 .00:00 0000.0.
0.20.... 2.0.0“.
00.0-00. 00.F .0. F.0F F0 0 "A
3 .0. 0. FF 00 0 v
0:0—0000.000.
00. F 00. .FF. F.0F 00 0000020
9.056030
.0 0.00... I 6.000.. 0000. 0.0.00 00.9 0 F0 0.00.00 0.0 0.00505 0.3.00. m-.m<.m<>
m_m>.__0uc_ KC 0351 .00—.50 0\0 008532
02.0.0050 080.00....

 

 

 

 

 

 

 

 

 

.0008. 2 0.00.

137

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0-00. 0F.F .0: 0.0F .2 00>
3 .00. 0.2 0 F0 oz
00. F 00. .00. 0.: 000 02.0.00
00500 N
I 020.: $00“.
0.0-00. 00. .0. F. FF 0F 0.00. 00 uv
0F 0. 0.00 0 0000. 00A
00. F 0.. .0. 0.0F 00 0.80005 .0
00.0.0 00F .o 00<
.0.:0. 00.F .0. 0.0F .0 00>
0. F .00. ..0F 000 oz
00. F 00. .00. 0. F F 000 00.0.0
- 020.... 2.0.0“.
00.00-00. 2.0 .0. 0.00 . 0.0.0.5
0F .0. 0.0F 00 0.0.0.0:
0 F. F 00.0 .0. 0.0F 00 00080.. - 000000
0o 3.0.0.0..
.0 A.000: : 60.00.. 2000 0000 00.0., 0 .0 000000 .00 000050. 00200. m..m<.m<>
0.9.8:. , mo , 0.001 .00:00 0.. 03:52
00000008 0 , . . 080000....

 

 

.0008. 2 0.00.

138

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00. F-0F. 00. a. 0.0 00 0.200.000 0.-F
3 a 0.2 00F xEFv
00. F 00. :00 0.0 000 0000000.".
.282
0.0-0 F. 00. A00 0.0F 0F 0A
0... F-00. 00. E 0.2 00 0v FA
00. F-00. 00. A F F0 0.0 00F an 0A
3 a: 0.2 02 0
00. 0 00. F 88 F. FF 000 000.5 uFo 0000.02
00. F-00. 00. :00 0.0 000 00>
3 a: 0.0F 02 oz
00. F 00. F an 0.2 000 .282 .00:00
00. F F00. 00.0 A 0 0 0.00 0F 00>
3 800 0.0 000 oz
00. F 00.0 A000 0.2 20 00.02.00
£03065”. x..—
.o .000: 160.00.. 2.00 0000 00.0., 0 .0 0.0000 .00 0000.000 00200. 0.5.0.5;
wiznuc— MO Enum— uoocao 3 39:32
80.000000 080.000....

 

 

.0008. S 0.000

139

 

 

 

 

 

 

 

 

 

 

 

 

 

0080.090 2.8000000 u .
00.0. F0. 00. F .0. 0.0F F0 000
000.00. 00.F .0 F. F. FF 02 00v 00
0F .0: 0.0 02 0
00. 0 00. .00. F.0F 000 0000.090
0000.. 0.0.00.
8.0.00. 00.F .0: F... 00F .0>m
0F .0: 0.0 00F 0.02
00. F 00. .00. 0.2 000 000 800000
«0003:0050
.0 0.00: 1.0.0.00... 0.000 0000 00.0.3. .0 0.0000 .00 ..00050. 00200. m-.m<.m<>
0.02.2:— . 10 0.001 .00:00 .x. .00—002
,, ”00000008 _ . , . . . 080000....

 

 

 

 

 

 

 

.0008. 8 0.000

140

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0900; 00.000 .F. 02 F 00.0".
.00. 0.2 000 00022
.00. F 00.0 .00. 0.3 000 0.0.02... 000....
-00.0F-00.F 00.0 .0. 0.00 3 00.0000.
00.0.0 3 00.0 .0. F0 00F 00.000.00.00 0:00>
0F :0. 0.0F 00F 020.003.0080.
.F0.v 0 .0.: .00. 0. FF 0F0 00000 000s.
0. F00-00.0 00.00 .0. 0.00 0 .00 00
0.0F-00.0 00.0 .FF. 0.00 00 0.0 0.0-F0
0.0-00. 00.0 .0: 0.2 00F 0.0 0.0-F.F
3 .0. 0.0 00F 0.0 0. Fuv
..F0.v 0 00.2 .00. 0.2 000 00.0 000.2
.0 .000: 1.0.0.000 2.00. 0000 0:.0> 0 00 0.00.00 .00 0000.00. 00200. m-.m<.¢<>
«3?.qu KC 0361 00050 0x. ban—:32
00000080 080000....
08.. . 0:000". 0:0 002 000.0902 0002500 0000:_m>m 000.00.. 00005 can 0. 0050...;

0000.020. _00_..00900 .0256 .50 00.000”. 0.200050”. 030.00.. 000...0>_m 0:0 .00:00 00 00020500. FF 0.00...

141

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.00-00.0 00.2 .00. 0. F0 00 00.00.2000 0.0...
3 .00 0.0 000 0000000000
.00—003.0000
.F0.v F 00.00 .00. 0.3 000 .000000.
.00.000-00.00F 00.000 .0 F. 0.00 0F 0000000.).
3:20.096
00.000-00.2 00.00 .FF. 0.00 00 0.00.0..00000.
00.00-00.F 00.0 .0. 0.0 02 00.000
0. F .0. 0. F 00 F 0.00002
..F0.v 0 00.02 .00. 0.8 000 0.0.8050...
.00.F0-00.0 00.0 .0 F. 0. F0 F0 00.0000.
3 .0: 0.0 000 0.80.0
.F0.v F 00.00 .00. 0. FF 20 0.00.00 00.000
8.00.00 00.000 .0. 02 0 0.0:
8.00.0 00.000 .0. 0.: 00 0000.02
3 .0. 0.0 0 com
.F0.v 0 00.00 00F. 0.0F 00 8000.00
.3 00.020.00.00
.0 .0000 100.00.. 2.00 0000 00.0., 0 00 0.00.00 .00 0000.00. 00200. 0.005005
«.0330: m0 0.03. .00:00 .0. .00—=02
00000008 080000....

 

 

.0008. FF 0.000

142

 

 

 

 

 

 

 

 

acmoEcgw >=wowflumuw H 0
00.000050 F0.0~F .00. 0.00 00 00000005.
.msofifimsw
.00.00-00.F 00.0 .0. 0.0F 0F 0.00.2 0.00.2
0~.FF-~0.0 00.0 .0. 0.0 00 0.0.90 2,0”.
3 E 0.0 000 00030309000.
.F0.v 0 00.00 .00. 0.: =0 000.98 00000.02
0.0002 00.".
_o .000: 1.0.0000 2.00. 0000 00.0.. 0 .0 0000.00 .00 000.000. 000.00. m..m<.m<>
, 0.05.3... «O 0.000. .00:00 .x. 03.032
00000008 080000....

 

 

 

 

 

 

 

 

 

.0008. FF 0.00.0

143

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_ _ _ _ _ . . . 0.....
00.F-00. 00.0 00. F 00. .00. 0.00 .000. 0.00 0.00 0:02.00 0F00<
00.F-00. _ 0F.F _ 00. _ F _ 00. _.00. 0.0 _ .000. 0.0 _ 0.0 _ 00002.00 000000.02
_ _ H _ 4 _ ¥ _ 00.000—5000.
00.F-00. 00.0 00. F 3. .00. 0.0 :8. 0.0 0.0 .0 0000.02
_ . _ v A _ _ . 0.....
0F.F-00. 00.F 00. F F0. .0: 0.00 .00. 0.00 0.00 00000000.... 00<
00.3.0. _ 00. _ 00. _ F_ 00.F _.00. 0.0F _.000. 0.0F _ 0.0F _.0000...0000000s. 00<
.0F.F-00.F _ 00.F _.F0.v_ F. 00.1 _.00. 0.00 _ .000. 0.00 _ 0.00 _ .0000...00<
.0000000. 0. 000.0500.
FF.F-00. 00.F 0F. F 00.0 .00. 2.0 .000. 0.00 0.00 0. 20.03.0000.
000.2 ..Uom
00.F-00.F _ F0.F _ 00. _ F _ 00.0 _ .00. 0.00: .000. F.00F_ 0.00F _ .00000&.0fl0§
0F.F-00. 00.F 0... F 0F. .00. 0.00 .000. 0.00 0.00 000000290:
.0 0.03.5 €00.02... €00E30.
0.003-003 00.00. 00.0> 00 .00 0.00000 0000000 000.000. 0.00..0>
0.030.... 0300 0 0.00”. .00:00 09:00 :09:
000000000 00 009.000.... .0000. .0002 _.0005

 

 

 

 

 

 

 

 

 

 

0000 E03000". 0:0 009 000.0052 c0050.. 0203.05 000.00.. «000.5 000 :. «0.0.0.0.; 020.00....—
_00_:__o 0:0 0.00—0.0.02.5 0:030:00”. 08 8.300”. 00.000500. 0.00.00... 20.0035 0:0 0000.00 00 00:205....—

uuw 030%

144

00.00. 00 .00—=32 u z

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

~.00.0.00.F 00.F .F0.v F 0.0.0: .00. 0.0 .000. 3 .2 0.0.0.0008.
mmo :o 006 000.2
.0000... 0.00.505

0F.F-00. 00.F 00. F 00. .00. 0.00 .00. 0.00 0.00 .0 00.0.0
0F 0. 00.0. .o 00<
$0.00... 0.090905

00. F-F0. 00. 00. F 00. .0. 0.00 .00. F.00 0.00 .0 00.00....
5 000m 00 090

.0000...

000800200”.

F0.F.00. 0F.F 2. F 00.0 .FF. 0.0 .0... 0.0 0.0 0000020
00... 00.00.50

.0000...

0:0—000.0.me

FF.F-00. 00.F 00. F 00.0 .0: 0.2 .00. 0.0 0.0 0002.00.
on: 02.0.05

.0000...

00.F-00. 00. 00. F 0F. .00. 0.0 .000. 0.0 0.0 09000000000
_000 00: 02.0.50

0.0306 tan—0:5 ..00E::.
.0 00000000.. 0000. 00.0., .0 .00 0.00000 0.00000 000.000. 0.00.00>
0.020.... 0.000 0. 00.01 .00:00 09:00 0.00:.
000000000 00 080.00.... .000.). .0005. ..0.0>0

 

 

 

 

 

 

 

 

 

 

.0008. 0F 0.000

145

0000.0 .0. 00000.00. 00 0.0 A .0 0.30090 02 F0500

 

 

 

 

 

 

 

 

 

 

 

 

0005.200 mczo> 8 0802.00 0. 0.30098 00.00.00. 003000.00. n 00.0.: 00.00.00”. .0....
.00:00 .0005 Fo >800... .0520... u >.o.0.... 00.0.0.2 00
60300000000000 u 0000,5000. 000300000000... u 00000.20. 1,
0.00> 0. 0m< 0
0.000 n 002008 m<20 00 0.00.
01.00:.00 m>00 .50. 0.00 09 000 0. F 00 00> 000000.000 F 0
$300.00 030 .50. .800 09000 0. F 00 00> 000000000 F 0
01000 F .3 .0000: >.000:. 09000 o.o 00 00> 000000000 mm
«1330.0 .0....00 0.0... 09000 0. F 00 00> 000000000 um
£0000. Fv 0>00 >00. >.0003. 0.00.0..00000. 0F o: 00 00000.0... 3
10.0.00. FN .00:00 .0000... 0.00.0..00000. N. F O0 00 00000.0... 00
10300.0 .00:00 0000:. 09000 0. F 00 00> 000000.000 mm
000$ F F . 0 09000 >.000:. 0.00.8.00000. oN 00 00 00000.0... 9.
000.00de a>00 >00. >.000:. 09000 c. F 00 00 3000.0... mv
£6 F .o 09000 E0003. .00:00 N. F 00 00 00000.0... NV
>00.00090... m<z... 00 0.800.008 00. 0 3 ..
0.4. >mo.3>o >00000.0000 0. 005 >300... 320.: 030.0 0m<
...m<mmm m<z.... .00.0.0. 0.0525005. 00.00.. 00.00.03. .00.0F0.2 .00:000002

 

 

 

 

 

 

 

 

 

 

 

 

D.< hw<mmm 0 0000.695 .00:00 F0
000000 .0005 00 00000 00 .F 0. 920 000 0 0. 00.0.0.2 0o 0 0.00.... 000.. .o 200.00.30.00. .0 0:300”.

«N 0.00...

146

 

 

 

z _u + n o + a
u + o u o
Azb A2“:
@3392 03:. 025me 3.x...
a + a n w
E”: Eb

 

o>_:mon_ 025".

 

w>=_mon_ 2:...

 

 

mmoz<0 OZ

mmoz<0

.. hwmh

+ hwmh.

147

m>~50 00¢ hmmh. hmm0>> \ hwmh. hwwm "N MMDOE

 

b65026 - _
a ad ad 9.: N6 9
. _ _ _ _ c
i I H6
\ I N...
\ I Md

 

 

 

:5 3.83 .. .I. I \\ I #6

 

.39 .momlT \\ I m.©

 

 

\\\ l 9.:
\I\ ”.6

\\ fi a...
l\ O H

\
\
I

l‘.

Q

 

 

Mumsuas

 

 

148

m$.50 00m. o_<._.m<mmm "n $50.".

 

Eumuunmé
83 82 8o... 8:. 83 8.8
— _ _ \F p sane
-82
m
8.85
n
3 M
82m
82

 

-8":

149

10.

11.

12.

13.

REFERENCES

Greenlee RT, Murray T, Bolden S, Wingo PA. Cancer Statistics, 2000.
CA Cancer J Clin 2000; 50:5—33.

Leis HP, Jr. Clinical diagnosis of breast cancer. J Reprod Med 1975;14:
231 -40.

Donegan WL. Evaluation of a palpable breast mass. N Engl J Med 1992;
327(13):937—42.

Goodno JA, Jr. Breast mncer detection: experience in a suburban
community. Am J Obstet Gynecol 1990; 162(6):1393-6.

Seltzer MH. The significance of breast complaints as correlated with age
and breast cancer. Am Surg 1992; 58(7):413-7.

Seltzer MH. Preoperative prediction of open breast biopsy results.
Cancer

1997; 79(9): 1 822—7.

Martelli G, Pilotti S, Coopmans de Yoldi G, et al. Diagnostic efficacy of
physical examination, mammography, fine needle aspiration cytology
(triple-test) in solid breast lumps:an analysis of 1708 consecutive cases.
Tumori 1990; 76(5):476 -9.

Eddy DM. Screening for breast cancer. Ann Intern Med 1989;
1 1 1(5):389—99.

Eddy DM. Clinical decision making: from theory to practice. Cost-
effectiveness analysis. Is it up to the task? Jama 1992; 267(24):3342—8.

Schmidt JG. The epidemiology of mass breast cancer screening—a plea
for a valid measure of benefit. J Clin Epidemiol 1990; 43(3):215—25.

Egan RL. Breast biopsy priority: cancer versus benign preoperative
masses. Cancer 1975; 35(3):612-7.

Physician's Insurance Association of America, Breast cancer study, New
York, 1995.

Ciatto S, Del Turco MR, Marrazzo A, et al. Time trends of benign/

150

 

 

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

malignant breast biopsy ratios a multicenter Italian study. Tumon' 1996;
82(4):325-8.

Lay SF, Crump JM, Frykberg ER, Goedde TA, Copeland EMd. Breast
biopsy. Changing patterns during a five-year period. Am Surg 1990;
56(2):?9-85.

Bondy ML, Vogel VG, Halabi S, Lustbader ED. Identification of women at
increased risk for breast cancer in a population-based screening
program. Cancer Epidemiol Biomarkers Prev 1992; 1(2):143-7.

Vogel VG, Graves DS, Vernon SW, Lord JA, V\finn RJ, Peters GN.
Mammographic screening of women with increased risk of breast cancer.
Cancer 1990; 66(7): 1 61 3—20.

Newell GR, Vogel VG. Personal risk factors. What do they mean?
Cancer 1988; 62(8 Suppl): 1 695-701 .

American Cancer Society, Breast cancer facts and figures, 1999.
American Cancer Society, 1999.

Feuer EJ, Wun LM, Boring CC, Flanders WD, Timmel MJ, Tong T. The
lifetime risk of developing breast cancer. J Natl Cancer Inst 1993;
85(11):892—7.

Key TJA, Pike MC. The Role of Oestrogens and Progestagens in the
Epidemiology and Prevention of Breast Cancer. Eur J Cancer Clin Oncol
1988; 24:29-43.

Pike MC, Krailo MD, Henderson BE, T CJ, Hoel DG. "Hormonal" risk
factors, "breast tissue age" and the age-incidence of breast cancer.
Nature 1983; 303:767-770.

Trichopoulos D, MacMahon B, Cole P. Menopause and breast cancer
risk. J Natl Cancer Inst 1968; 41 2315-329.

Russo J, Russo IH. The etiopathogenesis of breast cancer prevention.
Cancer Letter 1995; 90:81 -89.

MacMahan B, Cole P, Lin TM, al. 9. Age atfirst birth and breast cancer
risk. Bulletin of the World Health Organization 1970; 43:209-212.

Janerich D, Hoff M. Evidence for a crossover effect in breast wncer risk
factors. Am J Epidemiol 1982; 116:737-742.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

de Waard F. Premenopausal and postmenopausal breast cancer: One
disease or two? J Natl Cancer Inst 1979; 63:549-552.

Huang Z, Hankinson SE, Colditz GA, et al. Dual effects of weight and
weight gain on breast cancer risk. Jama 1997; 278(17):1407-11.

Breast cancer and hormonal contraceptives: collaborative reanalysis of
individual data on 53 297 women with breast cancer and 100 239 women
without breast cancer from 54 epidemiological studies. Collaborative
Group on Hormonal Factors in Breast Cancer. Lancet 1996;
347(9017):1713-27.

Breast cancer and hormone replacement therapy. Combined reanalysis
of data from 51 epidemiologic studies involving 52,705 women with
breast cancer and 108,411 women without breast cancer. Lancet 1997;
350:1947—1059.

Persson l, Adami H-O, Bergkvist L, et al. Risk of endometrial cancer
after treatment with oestrogens alone or in conjunction with
progestogens: Results of a prospective study. Br Med J (Clin Res) 1989;
298:147—1 51.

Stanford JL, Thomas DB. Exogenous progestins and breast cancer.
Epidemiologic Reviews 1993; 15298-107.

Robbins GF, Berg JW. Bilateral primary breast cancers: a prospective
clinicopathological study. Cancer 1964; 17:1501-1527.

Weber BL, Garber JE. Familial breast cancer. In: Harris JR, Lippman
ME, Morrow M, Hellman S, eds. Diseases of the Breast. Philadelphia:
Lippincott—Raven, 1996; 168-185.

Pharoah PD, Day NE, Duffy S, Easton DF, Ponder BA. Family history
and the risk of breast cancer: a systematic review and meta-analysis. Int
J Cancer 1997; 71 (5):800—9.

Biesbecker BB, Brody L. Genetic susceptibility testing for breast and
ovarian cancer: A progress report. JAMWA 1997; 5222-2 7.

Dawson DA, Thompson GB. Breast cancer risk factors and screening:
United States, 1987. Vital Health Stat 10 1990(172):iii-iv, 1-60.

Brinton LA, Hoover R, Fraumeni JF, Jr. Interaction of familial and

hormonal risk factors for breast cancer. J Natl Cancer Inst 1982;
69(4):817-22.

152

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

Colditz GA, \Afillett WC, Hunter DJ, et al. Family history, age, and risk of
breast cancer. Prospective data from the Nurses' Health Study. Jama
1993; 270(3):338—43.

Boice JD, Jr., Monson RR. Breast cancer in women after repeated
fluoroscopic examinations of the chest. J Natl Cancer Inst 1977;
59(3):823—32.

Boice JD, Jr., Preston D, Davis FG, Monson RR. Frequent chest X—ray
fluoroscopy and breast cancer incidence among tuberculosis patients in
Massachusetts. Radiat Res 1991; 125(2):214-22.

Shore RE, Hildreth N, Woodard E, Dvoretsky P, Hempelmann L,
Pasternack B. Breast cancer among women given X—ray therapy for
acute postpartum mastitis. J Natl Cancer Inst 1986; 77(3):689—96.

Shimizu Y, Mabuchi K, Preston DL, Shigematsu l. Mortality study of
atomic-bomb survivors: implications for assessment of radiation
accidents. World Health Stat Q 1996; 49(1):35-9.

Shimizu Y, Kato H, Schull WJ. Risk of cancer among atomic bomb
survivors. J Radiat Res (Tokyo) 1991; 32 Suppl 2:54—63.

Smith-Warner SA, Spiegelman D, Yaun SS, et al. Alcohol and breast
cancer in women: a pooled analysis of cohort studies [see comments].
Jama 1998; 279(7):535—40.

Tseng M, Weinberg CR, Umbach DM, Longnecker MP. Calculation of
population attributable risk for alcohol and breast cancer (United States).
Cancer Causes Control 1999; 10(2):119-23.

Seidman H, Stellman SD, Mushinski MH. A different perspective on
breast cancer risk factors: some implications of the nonattributable risk.
CA Cancer J Clin 1982; 32(5):301-13.

Madigan MP, Ziegler RG, Benichou J, Byrne C, Hoover RN. Proportion
of breast cancer cases in the United States explained by well—established
risk factors. J Natl Cancer Inst 1995; 87(22):1681—5.

Rockhill B, Weinberg CR, Newman B. Population attributable fraction
estimation for established breast cancer risk factors: considering the
issues of high prevalence and unmodifiability [see comments]. Am J
Epidemiol 1998; 147(9):826-33.

Goodson WHd. Clinical breast examination. West J Med 1996;
164(4):355—8.

153

50.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

Morrow M. Identification of the woman at risk for breast cancer: problem
solved? Recent Results Cancer Res 1999; 151 :85—95.

Anderson DE. Breast cancer in families. Cancer 1977; 40(4 Suppl):
1855—60.

Anderson DE, Badzioch MD. Risk of familial breast cancer. Cancer 1985;
56(2):383—7.

Ottman R, Pike MC, King MC, Casagrande JT, Henderson BE. Familial
breast cancer in a population-based series. Am J Epidemiol 1986;
123(1):15-21.

Claus EB, Risch N, Thompson WD. Autosomal dominant inheritance of
early-onset breast cancer. Implications for risk prediction. Cancer 1994;
73(3):643—51 .

Gail MH, Benichou J. Validation studies on a model for breast cancer
risk. J Natl Cancer Inst 1994; 86(8):573—5.

Bondy ML, Lustbader ED, Halabi S, Ross E, Vogel VG. Validation of a
breast cancer risk assessment model in women with a positive family
history. J Natl Cancer Inst 1994; 86(8):620-5.

Costantino JP, Gail MH, Pee D, et al. Validation studies for models
projecting the risk of invasive and total breast cancer incidence. J Natl
Cancer Inst 1999; 91 (18):1541-8.

Devitt JE. Value of the history in the office diagnosis of breast cancer.
Can Med Assoc J 1977; 116(10):1127—8.

Morrow M, Bland K, Foster R, Breast cancer surgical practice guidelines.
Society of Surgical Oncology, 1997.

The palpable breast lump: information and recommendations to assist
decision-making when a breast lump is detected. The Steering
Committee on Clinical Practice Guidelines for the Care and Treatment of
Breast Cancer. Canadian Association of Radiation Oncologists [see
comments]. Cmaj1998; 158 Suppl 3:33-8.

Dawson C, Armstrong MW, Michaels J, Faber RG. Breast disease and
the general surgeon. II. Effect of audit on the referral of patients with
breast problems. Ann R Coll Surg Engl 1993; 75(2):83—6.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

Barton MB, Harris R, Fletcher SW. Does this patient have breast cancer?
The screening clinical breast examination: should it be done? HOW?
Jama 1999; 282(13):1270—80.

Newton P, Hannay DR, Laver R. The presentation and management of
female breast symptoms in general practice in Sheffield. Fam Pract
1999; 16(4):360—5.

Nichols S, Waters WE, Wheeler MJ. Management of female breast
disease by Southampton general practitioners. Br Med J 1980;
281 (6253): 1450-3.

Roberts MM, Elton RA, Robinson SE, French K. Consultations for breast
disease in general practice and hospital referral patterns. Br J Surg
1987; 74(11):1020—2.

Mikulin T, Blarney R. The problems referred to a breast clinic. The
Practioner 1988; 232:606-610.

Ellis H, Cox PJ. Breast problems in 1,000 consecutive referrals to
surgical out-patients. Postgrad Med J 1984; 60(708):653-6.

Barclay M, Carter D, Horobin JM, Preece PE, Wood RA. Patterns of
presentation of breast disease over ten years in a specialised clinic.
Health Bull (Edinb) 1991; 49(4):229-36.

Kleinfeld G. Diagnostic value of age in breast cancer. Am Surg 1972;
38(5):257-8.

Byrd BF. Close-up. Standard breast examination. CA CancerJ Clin
1974; 24(5):290-3.

Holleb Al. The technique of breast examination. CA Cancer J Clin 1966;
16(1):7-9.

McGinnis LS. The importance of clinical breast examination. Cancer
1989; 64(12 Suppl):2657—60.

Baines CJ, Miller AB, Bassett AA. Physical examination. Its role as a
single screening modality in the Canadian National Breast Screening
Study. Cancer 1989; 63(9): 1816-22.

Zajicek J, Franzen S, Jakobsson P, Rubio C, Unsgaard B. Aspiration

biopsy of mammary tumors in diagnosis and research—a critical review of
2,200 cases. Acta Cytol 1967; 11(3):169—75.

155

75.

76.

77.

78.

79.

80.

81.

82.

83.

84.

85.

Kreuzer G, Boquoi E. Aspiration biopsy cytology, mammography and
clinical exploration: A modern setup in the diagnosis of tumors of the
breast. Acta Cytol 1976; 20:319-323.

Grobler SP, du Toit RS, Brink C, Divall PD, Middlecote BD, Nel CJ. Pre-
operative evaluation of palpable breast tumours. S Afr J Surg 1990;
28(4):128-32.

Wertheimer MD, Costanza ME, Dodson TF, D'Orsi C, Pastides H, Zapka
JG. Increasing the effort toward breast cancer detection. Jama 1986;
255(10):1311-5.

Knight JA, Libstug AR, Moravan V, Boyd NF. An assessment of the
influence of clinical breast examination reports on the interpretation of
mammograms in a breast screening program. Ontario Breast Screening
Program Radiologists Research Group. Breast Cancer Res Treat 1998;
48(1):65-71.

Cady B, Stone MD, Schuler JG, Thakur R, Wanner MA, Lavin PT. The
new era in breast cancer. Invasion, size, and nodal involvement
dramatically decreasing as a result of mammographic screening. Arch
Surg 1996; 131(3):301-8.

Venet L, Strax P, Venet W, Shapiro S. Adequacies and inadequacies of
breast examinations by physicians in mass screening. Cancer 1971;
28(6): 1 546-51 .

Mushlin Al. Diagnostic tests in breast wncer. Cliniwl strategies based
on diagnostic probabilities. Ann Intern Med 1985; 103(1):79-85.

Rimsten A, Stenkvist B, Johanson H, Lindgren A. The diagnostic
accuracy of palpation and fine-needle biopsy and an evaluation of their
combined use in the diagnosis of breast lesions: report on a prospective
study in 1244 women with symptoms. Ann Surg 1975; 182(1):1-8.

Ciatto S, Rosselli del Turco M, Catarzi S, et al. Causes of breast cancer
misdiagnosis at physical examination. Neoplasma 1991; 38(5):523-31.

Boyd NF, Sutherland HJ, Fish EB, et al. Prospective evaluation of
physical examination of the breast. Am J Surg 1981; 142:591-594.

Boerner S, Sneige N. Specimen adequacy and false-negative diagnosis
rate in fine-needle aspirates of palpable breast masses [see comments].
Cancer 1998; 84(6):344—8.

 

 

 

 

86.

87.

88.

89.

90.

91.

92.

93.

94.

95.

Meden H, Neues KP, Roben—Kampken S, Kuhn W. A clinical,
mammographic, sonographic and histologic evaluation of breast cancer.
Int J Gynaecol Obstet 1995; 48(2): 1 93—9.

Hermansen C, Skovgaard Poulsen H, Jensen J, et al. Palpable breast
tumours: "triple diagnosis" and operative strategy. Results of a
prospective study. Acta Chir Scand 1984; 150(8):625-8.

Feig SA, Schwartz GF, Nerlinger R, Edeiken J. Prognostic factors of
breast neoplasms detected on screening by mammography and physical
examination. Radiology 1979; 133(3 Pt 1):577-82.

Di Pietro S, Fariselli G, Bandieramonte G, et al. Diagnostic efficacy of the
clinical-radiologiwl—cytological triad in solid breast lumps: results of a
second prospective study on 631 patients. EurJ Surg Oncol 1987;
13(4):335-40.

Butler JA, Vargas HI, Worthen N, Wilson SE. Accuracy of combined
clinical-mammographic-cytologic diagnosis of dominant breast masses.
Arch Surg 1990; 125:893-896.

Di Pietro S, Fariselli G, Bandieramonte G, et al. Systematic use of the
clinical-mammographic-cytologic triplet for the early diagnosis of
mammary carcinoma. Tumori 1985; 71 (2):179-85.

Hermansen C, Skovgaard Poulsen H, Jensen J, et al. Diagnostic
reliability of combined physical examination, mammography, and fine-
needle puncture ("triple-test") in breast tumors. A prospective study.
Cancer 1987; 60(8):1866-71 .

Kaufman Z, Shpitz B, Shapiro M, Rona R, Lew S, Dinbar A. Triple
approach in the diagnosis of dominant breast masses: combined
physical examination, mammography, and fine-needle aspiration. J Surg
Oncol 1 994; 56:254-257.

Morris A, Pommier R, Schmidt W, Shih R, Alexander P, Vetto J. Accurate
evaluation of palpable breast masses by the triple test score. Arch Surg
1998; 133:930—934.

Steinberg J, Trudeau M, Ryder D, et al. Combined fine-needle aspiration,
physical examination and mammography in the diagnosis of palpable
breast masses: their relation to outcome for women with primary breast
cancer. Can J Surg 1996; 39:302—31 1.

157

96.

97.

98.

99.

100.

101.

102.

103.

104.

105.

106.

van Dam PA, Van Goethem ML, Kersschot E, et al. Palpable solid breast
masses: retrospective single— and multimodality evaluation of 201
lesions. Radiology 1988; 166(2):435-9.

Vetto J, Pommier R, Schmidt W, et al. Use of the "triple test" for palpable
breast lesions yields high diagnostic accuracy and cost savings. Am J
Surg 1995; 169(5):519-22.

VIfinchester DP, Sener S, lmmerman S, Blum M. A systematic approach
to the evaluation and management of breast masses. Cancer 1983;
51(12 Suppl):2535—9.

Cardona G, Cataliotti L, Ciatto S, Rosselli Del Turco M. Reasons for
failure of physical examination in breast cancer detection (analysis of
232 false-negative cases). Tumori 1983; 69(6):531-7.

Dixon JM, Anderson TJ, Lamb J, Nixon SJ, Forrest AP. Fine needle
aspiration cytology, in relationships to clinical examination and
mammography in the diagnosis of a solid breast mass. Br J Surg 1984;
71(8):593—6.

Langmuir VK, Cramer SF, Hood ME. Fine needle aspiration cytology in
the management of palpable benign and malignant breast disease.
Correlation with clinical and mammographic findings. Acta Cytol 1989;
33(1):93—8.

Osuch JR. Abnormalities on Physical Examination. In: Harris JR,
Lippman ME, Morrow M, Hellman S, eds. Diseases of the Breast.
Philadelphia: Lippincott - Raven, 1996; 110-114.

Devitt JE. Benign breast disease in the postmenopausal woman. World J
Surg 1989; 13(6):731-5.

Bassett LW, Hendrick RE, L BT, et al., Quality Determinants of
Mammography. Clinical Practice Guidelines. Agency for Health Care
Policy and Research, Public Health Service, US. Department of Health
and Human Services, 1994.

Linver MN, Osuch JR, Brenner RJ, Smith RA. The mammography audit:
a primer for the mammography quality standards act (MQSA). AJR Am J
Roentgenol 1995; 165(1):19—25.

Coveney EC, Geraghty JG, O'Laoide R, Hourihane JB, O'Higgins NJ.

Reasons underlying negative mammography in patients with palpable
breast cancer. Clin Radiol 1994; 49(2):123-5.

158

107.

108.

109.

110.

111.

112.

113.

114.

115.

116.

117.

118.

Walker OJ, 0 LA. The misuse of mammography in the management of
breast cancer. Med J Aust 1986; 145:185.

Edeiken S. Mammography and palpable cancers of the breast. Cancer
1988; 61:263.

Chew SB, Hughes M, Kennedy C, Gillett D, Carmalt H.
Mammographically negative breast cancer at the Strathfield Breast
Centre. Aust N Z J Surg 1996; 66(3): 1 347.

Dawson AE, Mulford DK, Taylor AS, Logan-Young W. Breast carcinoma
detection in women age 35 years and younger: mammography and
diagnosis by fine—needle aspiration cytology. Cancer 1998; 84(3):163-8.

Burns PE, Grace MG, Lees AW, May C. False negative mammograms
causing delay in breast cancer diagnosis. J Can Assoc Radiol 1979;
30(2):74—6.

Langlands AO, Tiver KW. Significance of a negative mammogram in
patients with a palpable breast tumour. Med J Aust1982; 1(1):30—1.

Walker QJ, Gebski V, Langlands A0. The misuse of mammography in
the management of breast cancer revisited [see comments]. Med J Aust
1989; 151(9):509—12.

Lister D, Evans AJ, Burrell HC, et al. The accuracy of breast ultrasound
in the evaluation of cliniwlly benign discrete, symptomatic breast lumps.
Clin Radiol 1998; 53(7):490—2.

Reinikainen HT, Rissanen TJ, Piippo UK, Paivansalo MJ. Contribution of
ultrasonography and fine-needle aspiration cytology to the differential
diagnosis of palpable solid breast lesions. Acta Radiol 1999; 40(4):383—9.

Breast Imaging Reporting and Data System (Bl-RADS). American
College of Radiology (ACR), Reston, VA, 1993.

D'Orsi CJ, Getty DJ, Swets JA, Pickett RM, Seltzer SE, McNeil BJ.
Reading and decision aids for improved accuracy and standardization of
mammographic diagnosis. Radiology 1992; 184(3):619-22.

Sterns EE. Relation between clinical and mammographic diagnosis of

breast problems and the cancer/biopsy rate. Can J Surg 1996;
39(2): 1 28—32.

159

119.

120.

121.

122.

123.

124.

125.

126.

127.

128.

129.

130.

Roberts JC, Rainsbury RM. 'Tactile sensation': a new clinical sign during
fine needle aspiration of breast lumps. Ann R Coll Surg Engl 1994;
76(2):1 36—8.

Kline TS, Kline IK. Guides to Clinical Aspiration Biopsy - Breast. New
York: lgaku-Shoin, 1989.

Somers RG, Sandler GL, Kaplan MJ, Najjar D, Anderson AV, Cohen MH.
Palpable abnormalities of the breast not requiring excisional biopsy. Surg
Gynecol Obstet 1992; 175(4):325-8.

Zarbo RJ, Hovvanitz PJ, Bachner P. lnterinstitutional comparison of
performance in breast fine-needle aspiration cytology. A Q-probe quality
indicator study. Arch Pathol Lab Med 1991; 115(8):743-50.

The uniform approach to breast fine-needle aspiration biopsy. National
Cancer Institute Fine-Needle Aspiration of Breast Workshop
Subcommittees. Diagn Cytopathol 1997; 16(4):295-311.

Ciatto S, Bonardi R, Cariaggi MP. Performance of fine-needle aspiration
cytology of the breast- multicenter study of 23,063 aspirates in ten Italian
laboratories. Tumori 1995; 81 (1):13—7.

Barrows GH, Anderson TJ, Lamb JL, Dixon JM. Fine-needle aspiration of
breast cancer. Relationship of clinical factors to cytology results in 689
primary malignancies. Cancer 1986; 58(7):1493-8.

Cariaggi MP, Bulgaresi P, Confortini M, et al. Analysis of the causes of
false negative cytology reports on breast cancer fine needle aspirates.
Cytopathology 1995; 6(3):156-61 .

Park IA, Ham EK. Fine needle aspiration cytology of palpable breast
lesions. Histologic subtype in false negative cases. Acta Cytol 1997;
41(4):1131-8.

Lazda EJ, Kocjan G, Sams VR, Wotherspoon AC, Taylor |. Fine needle
aspiration (F NA) cytology of the breast: the influence of unsatisfactory
samples on patient management. Cytopathology1996; 7(4):262-7.

Watson DP, McGuire M, Nicholson F, Given HF. Aspiration cytology and
its relevance to the diagnosis of solid tumors of the breast. Surg Gynecol
Obstet 1987; 165(5):435—41 .

Horgan PG, Waldron D, Mooney E, O'Brien D, McGuire M, Given HF.
The role of aspiration cytologic examination in the diagnosis of
carcinoma of the breast. Surg GynecolObstet1991; 172(4):290-2.

160

131.

132.

133.

134.

135.

136.

137.

138.

139.

140.

141.

142.

Frable WJ. Needle aspiration of the breast. Cancer 1984; 53(3
Suppl):671-6.

Casey 'I'l', Rodgers WH, BaXter JW, Sawyers JL, Reynolds VH, Page
DL. Stratified diagnostic approach to fine needle aspiration of the breast.
Am J Surg 1992; 163(3):305-1 1.

Abele JS, Miller TR, Goodson WHd. Hunt TK, Hohn DC. Fine-needle
aspiration of palpable breast masses. A program for staged
implementation. Arch Surg 1983; 118(7):859-63.

Abele JS, Wagner LT, Miller TR. Fine-needle aspiration of the breast:
cell counts as an illusion of adequacy. A clinical cytopathologist's point of
view . Cancer 1998; 84(6):319-23.

Abele J, Stanley MW, Rollins SD, Miller TR What constitutes an
adequate smear in fine-needle aspiration cytology of the breast? [letter;
comment]. Cancer 1998; 84(1):57-61.

Wolberg WH, Tanner MA, Loh W-Y. Fine-needle aspiration for breast
mass diagnosis. Arch Surg 1989; 124:814—818.

Layfield LJ, Mooney EE, Glasgow B, Hirschowitz S, Coogan A What
constitutes an adequate smear in fine-needle aspiration cytology of the
breast? Cancer 1997; 81 (1):16—21.

Stanley MW, Abele J, Kline T, Silverman JF, Skoog L. What constitutes
adequate sampling of palpable breast lesions that appear benign by
clinical and mammographic criteria? Diagn Cytopathol1995; 13(5):
473-82.

Layfield LJ, Glasgow BJ, Cramer H. Fine needle aspiration in the
management of breast masses. Pathol Anna 1989; 24:23.

Moriarty AT. Fine-needle biopsy of breast: when is enough, enough?
Diagn Cytopathol 1995; 13(5):373-4.

Sneige N. Should specimen adequacy be determined by the opinion of
the aspirator or by the cells on the slides? Cancer 1997; 81 (1)23-5.

Salami N, Hirschowitz SL, Nieberg RK, Apple SK. Triple test approach to

inadequate fine needle aspiration biopsies of palpable breast lesions.
Acta Cytol1999; 43(3):339-43.

161

143.

144.

145.

146.

147.

148.

149.

150.

151.

152.

153.

154.

Zajdela A, Ghossein NA, Pilleron JP, Ennuyer A. The value of aspiration
cytology in the diagnosis of breast cancer: experience at the Foundation
Curie. Cancer 1975; 35(2):499—506.

Troxel DB, Sabella JD. Problem areas in pathology practice. Uncovered
by a review of malpractice claims Am J Surg Pathol 1994; 18(8):821-31.

Dixon JM, Dobie V, Lamb J, Walsh JS, Chetty U. Assessment of the
acceptability of conservative management of fibroadenoma of the breast.
Br J Surg 1996; 83(2):264-5.

Linsk J, Kreuzer G, Zajicek J. Cytologic diagnosis of mammary tumors
from aspiration biopsy smears. ll. Studies on 210 fibroadenomas and
210 cases of benign dysplasia. Acta Cytol 1972; 16(2):130-8.

Benoit JL, Kara R, McGregor SE, Duggan MA Fibroadenoma of the
breast: diagnostic pitfalls of fine-needle aspiration. Diagn Cytopathol
1 992; 8(6):643-7.

Stanley MIN, Tani EM, Skoog L. Fine-needle aspiration of fibroadenomas
of the breast with atypia: a spectrum including cases that cytologically
mimic carcinoma. Diagn Cytopathol 1990; 6(6):375-82.

Maygarden SJ, McCall JB, Frable WJ. Fine needle aspiration of breast
lesions in women aged 30 and under. Acta Cytol1991; 35(6):687-94.

Smallwood J, Khong Y, Boyd A, et al. Assessment of a scoring scheme
for the preoperative diagnosis of breast lumps. Ann R Coll Surg Engl
1 984; 66(4):267-9.

Logan-Young VWV, Hoffman NY, Janus JA Fine-needle aspiration
cytology in the detection of breast cancer in non-suspicious lesions.
Radiology 1992; 184:49-53.

Layfield LJ. Can fine-needle aspiration replace open biopsy in the
diagnosis of palpable breast lesions? Am J Clin Pathol 1992; 98(2):
1 45—7.

Yelland A, Graham MD, Trott PA, at al. Diagnosing breast carcinoma in
young women 8m] 1991; 302(6777):618-20.

Matsuda M, Nagumo S, Koyama H, Terasawa T. Fine needle aspiration
cytology of "minimal" breast cancer. Acta Cytol 1986; 30(5):501-4.

162

155.

156.

157.

158.

159.

160.

161.

162.

163.

164.

165.

166.

167.

168.

Brown LA, Coghill SB, Povvis SA. Audit of diagnostic accuracy of FNA
cytology specimens taken by the histopathologist in a symptomatic
breast clinic. Cytopathology 1991; 2(1):1-6.

Giard RW, Hermans J. Interpretation of diagnostic cytology with
likelihood ratios. Arch Pathol Lab Med 1990; 114(8):852—4.

Giard RW, Hermans J. The value of aspiration cytologic examination of
the breast. A statistical review of the medical literature. Cancer 1992;
69(8):2104—10.

Horobin JM, Matthew BM, Preece PE, Thompson AJ. Effects of fine
needle aspiration on subsequent mammograms. Br J Surg 1992; 79:
52-4.

Sickles EA, Klein DL, Goodson WH, Hunt TK. Mammography after
needle aspiration of palpable breast masses. Am J Surg 1983; 145:395.

Martin H, Ellis E. Biopsy by needle puncture and aspiration. Ann Surg
1930; 92:169-181.

Fleiss J. The design and analysis of clinical experiments. New York:
John Wiley and Sons, 1966.

Weinstein M, Feinberg H. Clinical Decision Analysis. Philadelphia: W.B.
Saunders, 1980.

Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology: A
Basic Science for Clinical Medicine. Boston: Little, Brown and Company,
1 991 .

Kleinbaum DG. Logistic Regression. New York: Springer-Verlag, 1994.

Albert A. On the use and computation of likelihood ratios in clinical
chemistry. Clin Chem 1982; 28:1 1 13—1119.

Reeves M, Curtis C, Salman T, et al. Validation of logistic regression
models used in the assessment of prognosis and the need for surgery in
equine colic patients. Preventive Veterinary Medicine 1992; 13:155-172.

Hosmer D, Lemeshow 8. Applied logistic regression. New York: VIfiley
lnterscience, 1989.

Colditz GA, Sankinson SE, Hunter DJ, VIfiIlatt W, et al. The use of

estrogens and progestins and the risk of breast cancer in
postmenopausal women. N Engl J Med 1995; 332: 1 589-1 593.

163

169.

170.

171.

172.

173.

174.

175.

Barton MB, Elmore JG, Fletcher SW. Breast symptoms among women
enrolled in a health maintenance organization: frequency, evaluation,
and outcome. Ann Intern Med 1999; 130(8):651-7.

Stanford JL, Weiss NS, Voigt LF, Daling JR, Habel LA, Rossing MA.
Combined estrogen and progestin hormone replacement therapy in

relation to risk of breast cancer in middle-aged women. Jama 1995;
274(2): 1 3742.

Fraser GE, Shavlik D. Risk factors, lifetime risk, and age at onset of
breast cancer. Ann Epidemiol 1997; 7(6):375-82.

Sturgeon SR, Schairer C, Gail M, McAdams M, Brinton LA, Hoover RN.
Geographic variation in mortality from breast cancer among white
women in the United States. J Natl Cancer Inst 1995; 87(24):1846-53.

Robbins A, Brescianini S, Kelsey JL. Regional differences in known risk
factors and the higher incidence of breast cancer in San Francisco. J
Natl Cancer Inst 1997; 89:960-965.

Colditz GA, Rosner BA, Speizer FE. Risk factors for breast cancer
according to family history of breast cancer. For the Nurses' Health Study
Research Group. J Natl Cancer Inst 1996; 88(6):365-71.

Laden F, Spiegelman D, Neas LM, et al. Geographic variation in breast
cancer incidence rates in a cohort of US. women. J Natl Cancer Inst
1997; 89(18):1373-8.

164

 

 

 

 

 

 

I ‘4

I

 

I,
III]

I
I

STQTE UNIV

I II I I
31293020

 

2‘ II
M

 

 

nICHIan
I
I
I

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

7 I
JHMWHJ I: L. y .
z :2. 5 1,: A .. 3:!
E . x . .
Mr. L. 1 =7. .15 . .-