"UNIUNI“!WU”!!!I!!!WWIHIHIWIIH”1|!in.

3
THS

00.4
\J
\l-b

    

THESIS

r) \..l"“ 9*
a C \

||||llllllill\|||Will”lillllllllilillliMIMI“HHIHI

3 1293 02058 65

This is to certify that the

thesis entitled

MODEL PREDICTING SOMATIZATION STATUS OF
HIGH-UTILIZING PATIENTS IN A MANAGED CARE SYSTEM

presented by

Stacey Reimer Armatti

has been accepted towards fulfillment
of the requirements for

MS. degree in Epidemiology

 

    

Major professor

Date 4_2_;2000

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

 

 

 

 

LIBRARY

Michigan State
University

 

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

11/00 mumpss-pu

MODEL PREDICTING SOMATIZATION STATUS OF HIGH-UTILIZING
PATIENTS IN A MANAGED CARE SYSTEM

By

Stacey Reimer Armatti

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

MODEL PREDICTING SOMATIZATION STATUS OF HIGH-UTILIZING
PATIENTS IN A MANAGED CARE SYSTEM

By

Stacey Reimer Armatti

Somatizers are disproportionately high utilizers of medical services, surgical and
laboratory procedures. Somatizers have drawn the attention of Health Maintenance
Organizations (HMOs) in recent years because of their expensive health seeking
behaviors. The aim of this thesis is to develop an effective screening algorithm for
somatizing patients within an HMO setting. A logistic regression model was formulated
to predict the somatization status in high-utilizing patients using demographic and
clinical information abstracted from the patients’ medical record.

The data sample was obtained from computerized patient files from all three sites of
the Blue Care Network of Mid Michigan. In order to calculate and test the logistic model,
separate derivation and validation sets were created. Of the 883 total data subjects,
13.82% were somatizers and 86.18% were nonsomatizers. A priori discrete variables to
be used in the model included gender, the number of total visits, and percent of visits
indicative of “somatic potential”. The sensitivity and specificity in the derivation set was
96.20% and 51.47%, respectively; in the validation set, 79.07% and 44.84%. The positive
predictive value in the derivation set was 23.53%, and 19.65% in the validation set. The
negative predictive value in the derivation set was 98.87%, and 96.62% in the validation

set. The results from this research are not necessarily generalizable to other HMOs.

Copyright by
Stacey Reimer Armatti
2000

To my wonderful son.

iv

ACKNOWLEGDEMENTS

I would like to acknowledge the excellent instruction, guidance, and support I
received from the faculty and staff of the Department of Epidemiology. My advisor, Dr.
Joseph Gardiner, has been an exceptional and dedicated mentor. I would also like to
express my heartfelt gratitude to Dr. Robert Smith and Dr. Dorothy Pathak for not only
taking the time to be members of my thesis committee, but for their wisdom which has

provided me with invaluable training beyond what I learned in the classroom.

TABLE OF CONTENTS

LIST OF TABLES
LIST OF FIGURES
INTRODUCTION

CHAPTER 1
THE PROBLEM AND EPIDEMIOLOGY OF SOMATIZATION
Costs of Somatization
Etiology
Diagnosis
Bias in Somatization Studies
Prevalence
The Physician
Treatment and Management

CHAPTER 2
DATA AND METHODS
Objective
Overview
Data Collection
Data Cleanup and Counting Medical Visits
Creation of New Variables
Demographic Information
Comparability of the Data Sets
Preliminary Analysis
Derivation Data Set Univariate Analyses
Validation Data Set Univariate Analyses
Best Predictors

CHAPTER 3

THE LOGISTIC MODEL
Main Effects Logistic Regression Model
Model Fitting Steps
Constructed Model
Interpretation
ROC Curve
Choice of Cutoff Values
Appropriateness of the Model
Strengths and Limitations
Conclusion

vi

viii

13
14
15
19
21

25
25
25
25
27
29
31
32
33
34
36
37

41
41
42
44
44
46
47
48
50
52

APPENDICES
Appendix A — Special Treatment Project Chart Rating Form
References
General References

vii

56
60
65

LIST OF TABLES

Table l — Included Visits

Table 2 — Excluded Provider Types

Table 3 — Characteristics of Participants (N = 883)

Table 4 — Distribution of Somatization Status by Gender

Table 5 — Comparison of Derivation and Validation Data Sets

Table 6 — Mann-Whitney/Wilcoxon Tests and Univariate
Comparisons for Continuous Variables in the Derivation
Data Set for Somatizers vs. Nonsomatizers (N = 588)

Table 7 — Chi-Square Tests and Univariate Comparisons for
Continuous Variables in the Derivation Data Set
for Somatizers vs. Nonsomatizers (N = 588)

Table 8 — Mann-Whitney/Wilcoxon Tests and Univariate
Comparisons for Continuous Variables in the Validation
Data Set for Somatizers vs. Nonsomatizers (N = 295)

Table 9 —— Chi-Square Tests and Univariate Comparisons for
Continuous Variables in the Validation Data Set

for Somatizers vs. Nonsomatizers (N = 295)

Table 10 — Distribution of Somatization Status by TOTAL(2)
in the Derivation Data Set

Table 11 — Distribution of Somatization Status by TOTAL(2)
in the Validation Data Set

Table 12 — Model Entry and Elimination Procedures

Table 13 - Parameter and Odds Ratio Estimates Using the
Method of Maximum Likelihood

Table 14 - Adjusted Odds Ratios for Somatization Status
Table 15 - All Possible Predicted Probability and Odds

Table 16 — Prediction in the Derivation Set

viii

28

29

31

32

33

34

35

36

37

40

40

43

44

45

46

48

Table 17 — Prediction in the Validation Set

Table 18 — Model Fitting Information and Testing the
Global Null Hypothesis [3 = 0

Table 19 - Goodness-of-Fit Test
Table 20 - Residual Analysis

Table 21 - Visit Count Discrepancy

48

48

48

49

50

LIST OF FIGURES

Figure l - Distribution of Somatization Status by Gender

Figure 2 — ROC Curve

32

47

INTRODUCTION

Somatizers are disproportionately high utilizers of medical services, surgical and
laboratory procedures. Kellner (1990) estimates that one fifih of the medical budget is
spent on treating somatizers. Somatization has been identified as a mechanism to mask
and alleviate stressful emotions, and to prevent being diagnosed with a mental illness.
The somatizer usually presents with multiple symptoms in any body part or organ system,
which may imitate organic disease. The most common symptoms are pain, diffuse
muscle aches, fatigue, dizziness, shortness of breath and palpitations (Lipowski, 1988).
Typically, there exists no real underlying organic disease and the somatizer often submits
to numerous laboratory tests, possible surgery, drug treatments, and sessions with
specialists in an attempt to treat their symptoms. These may harm the patient in the
process or may help to intensify a patient’s preoccupation with their health.

Somatizers have drawn the attention of Health Maintenance Organizations
(HMOs) in recent years because of their expensive health seeking behaviors. Barsky and
Borus (1995) note an increase in health seeking for self-limited infirmities such as upper
respiratory tract infections and digestive complaints; isolated, benign symptoms such as
insomnia, motion sickness and fatigue; and physical characteristics, such as baldness and
myopia. Public health educational campaigns concerning serious diseases such as cancer
and heart disease may prompt some people to place emphasis on and seek medical
attention for benign symptoms (Barsky and Borus, 1995). Clearly, somatizers present a

problem to managed care plans.

Despite the financial pressures faced by managed care plans in treating
somatization, they present an opportunity to devise cost-effective and efficient strategies
for care of somatizers. Managed health care plans should attempt to educate primary
physicians and nurses in the identification and management somatizers. The primary
physician can be trained to effectively treat most somatizing patients. In their study,
Smith et al. (1998) concluded that residents participating in a one-month intensive
training rotation had better knowledge, attitudes, somatization management and
interviewing than residents without the training rotation. Patient satisfaction and well
being were also higher under the trained residents (Smith et al., 1998). Somatizing
patients require a supportive and therapeutic relationship with their physician (Smith,
1991). Through proper identification and treatment, physicians can avoid the many
medical, surgical and drug treatments somatizers undergo to treat organic disease where
none exists.

The goal of this thesis is to compute a mathematical model to aid physicians
within an HMO in identifying somatizers among their high-utilizing patients for cost
containment and proper medical treatment.

The data sample was obtained from computerized patient files from Health Central of
the Blue Care Network of Mid Michigan (BCNMM). BCNMM is a nonacademic, for-
profit HMO with a staff model of care. It serves over 20,000 adult primary care patients
through three local clinics. The sample consists of a random selection of patients between
20 and 55 years of age during the 1995 calendar year who had at least six distinct medical
and/or hospital visits. A resident physician reviewed medical charts for each patient in the

sample and classified the patient as possessing true organic disease or somatization. It

was determined that out of the 883 total data subjects, 122 patients were somatizers
(13.82%) and 761 were nonsomatizers (86.18%).

A logistic model for predicting somatization was constructed from approximately
two-thirds of the sample (n = 588), and validated using the remaining 295 in the sample.
The final model consists of information regarding gender, the percent of total visits
involving a varied group of diagnoses believed to be indicative of somatization, and the
total number of visits. The percent of patients classified correctly, sensitivity and
specificity in relation to probability cutoff values was computed. From this information, a
receiver operating characteristic (ROC) curve was constructed. This curve was used to
determine the best decision rule to predict the somatizing status of a given patient.
According to the best decision rule, a patient is predicted to be a nonsomatizer if their
predicted probability is less than 0.03206, or a somatizer when the predicted probability
is 0.03206 or larger. The sensitivity and specificity in the derivation set was 96.20% and
51.47%, respectively; in the validation set, 79.07% and 44.84%. The overall correct
classification in the derivation set was 66.40% and for the validation set 49.83%. The
positive predictive value in the derivation set was 23.53%, and 19.65% in the validation
set. The negative predictive value in the derivation set was 98.87%, and 96.62% in the
validation set. This model and its decision rule are not necessarily generalizable to other

HMOs but might serve as an aid to any primary physician who suspects somatization.

Chapter 1
THE PROBLEM AND EPIDEMIOLOGY OF SOMATIZATION
Costs of Somatization

The psychosomatic process currently labeled somatization has been recognized
throughout history. Most doctors are aware that many of their high-utilizers could be
somatizers, yet they lack the necessary combination of diagnostic and psychological
skills to distinguish and treat these patients. Somatization has many negative aspects for
patients and their families, friends and primary care physicians. It can lead to
inappropriate care that is both frustrating to the patient and provider because the patient
feels that his or her ailments have not been sufficiently alleviated, while physicians have
to contend with a dissatisfied patient for whom their care has not been effective.
Moreover, the cost of treatment in terms of time and resource use can be enormous.
There are great personal costs to the somatizer in terms of physical and emotional
suffering. These personal costs include time off from work, cost of medication,
deductibles and copayment charges, which are also calculable expenses. The reduction in
the somatizer’s quality of life caused by this condition cannot be quantified.

In recent years, the grth of managed and capitated plans as a means for
delivering health care has brought into focus the costs of somatization. To be
competitive, these organizations are under pressure to contain costs (Bruns, 1997).
HMOs place emphasis on health promotion and disease prevention (Barsky and Borus,
1995). High-utilizers, especially somatizers are viewed as economic liabilities for these
organizations. In a university family practice center, deGruy et a1. (1987) discovered that

somatizers had higher rates of office visits and charges incurred when compared to

matched controls. Somatizers had an average of 0.58 visits and $23.28 in charges
incurred per month and as opposed to 0.41 visits and $14.44 in high-utilizing controls.
Using the same BCNMM data that was used in this thesis, investigators Dr. John
Goddeeris and Sharon Roble calculated medical charges for each case and found that
charges were higher for somatizers than nonsomatizers. The average total charges for
1995 for all services combined were $5474 for somatizers and $4410 for nonsomatizers.
Median charges were $3454 for somatizers and $1472 for nonsomatizers. Average and
median charges for outpatient care exclusively were computed to be $3479 and $2426 for
somatizers, and only $2580 and $1019 for nonsomatizers. Office visits, diagnostic tests,
laboratory work, specialist consultation, and drug prescriptions comprise most of the
costs incurred by the patient. Effective identification and management could indeed save
thousands of dollars treating each high-utilizing patient.

Under managed care, Cantor (1996) believes that physicians may be forced to
single out somatizers and limit their access to office visits, second opinions, specialists
and procedures. Physicians would be encouraged to practice more efficiently and
effectively primarily to contain costs. They would have greater incentive to do less for
the somatizing patient, as opposed to too much (Barsky and Borus, 1995). The quality of
medical care for any patient classified as a somatizer who in fact has an organic disease,
would be greatly reduced. Somatizers would be cut off from the medical services that
they rely upon for care (Barsky and Borus, 1995). On the other hand, somatizers would
be at a reduced risk for unnecessary surgeries, medications, other iatrogenic procedures,

and fragmented care (“doctor shopping”) which is favorable (Barsky and Borus, 1995).

As managed care attempts to contain costs, it may inadvertently increase the cost
of treating somatizing patients. By restricting access to medical care, somatizers might
be compelled to invent and amplify their complaints to be allowed a visit. This may raise
an individual to a higher degree of somatization. Patients will also somatize when their
personal health expenses are affected. Many managed care providers will pay for
treatment if the patient’s diagnosis is a medical one, but if the same condition is given a
psychological diagnosis, sometimes the amount of benefits is considerably less (Bruns,
1997). Patients will choose to seek care through medical symptoms and complaints, even
if the problem is psychological.

Barsky and Borus (1995) note that most cost-containment efforts have
concentrated on limiting access to care and restricting its supply. The authors believe
that by studying somatization, the demand for care can be regulated. Various researchers
have found that the physician can effectively manage the somatizing patient once he or
she has been identified. Although guidelines are available to physicians, there is further
research to be done.

Means of identifying potential somatizers will be important in the managed care
setting, and for deploying cost-effective interventions to address the needs of these
patients. A study conducted by Smith et al (1995) demonstrated an average decrease in
1990 of $289.11 in medical costs and $51.93 in psychiatric costs for somatizers who were
assigned to an intervention. This intervention advised physicians to space visits, perform
less diagnostic tests and provide more positive diagnoses to their somatizing patients. It
appears that it would be advantageous for physicians to be educated about the

somatization process. Many traditional medical schools now include communication and

listening skills in their curricula. Medical schools based on osteopathic teaching methods
are becoming more common. Nevertheless, the additional training of primary-care
physicians to detect and manage somatizers might not always be sufficiently cost-
effective. Physicians may not have the desire or disposition to become cross-trained in
psychiatry. The predictive model presented in this thesis may aid in the identification of
somatizers among high-utilizing patients. Because this model is built on information
routinely collected by HMO’s on their patients, it may be applicable to other managed
care settings.

Etiology

The majority of patients present to their doctors for the diagnosis and treatment of
symptoms that they perceive as indicative of physical disease. Most bodily symptoms are
minor and transient, which do not require special medical care. Some symptoms are
indicative of a serious disease, usually requiring various medical treatments. Other
somatic complaints are the result of stress or psychological factors (Robbins and
Kirmayer, 1991). There are many reasons that somatizers medicalize their emotional
problems.

Many have speculated about what causes somatization. Kellner (1985) lists some
predisposing factors for somatization. First, genetics might play a role in somatization as
it does in other psychological disorders. The immediate family, both adult and
childhood, as well as personal social class and culture can influence how an individual
views somatic distress. Lipowski (1988) suggests a multifactorial causal model for
somatization and classifies the factors as predisposing, precipitating and maintaining.

Predisposing factors include genetic, developmental-learning, personality and

sociocultural elements. Life events and stressful situations may be precipitating factors;
maintaining factors would include individual predisposition and the interaction between
the patient and their family, physicians and the social system. Barsky (1981) proposes
that people are also motivated to visit doctors because of life stress, psychological
disorders, social isolation, and informational needs. Being treated for illness allows the
patient to cope with anxiety, grief, frustration and fear through postponement and
avoidance of challenges and responsibilities by adopting the sick role. Socially isolated
persons can turn to a physician for advice, interpersonal stimulation, consolation, care
and nurturing. Patients also see their physicians to gather information and education such
as etiology, treatment regimen and prognosis of their condition. Ford (1992) introduces
several factors that may promote somatization. These include: The gains of illness; social
isolation; amplification; symptoms used as communication; alexithymia; physiologic
concomitants of psychiatric disorders; cultural attitudes; religious factors; stigmatization
of psychiatric illness; economic issues; symptomatic treatment; fashionable diagnoses;
solution to a systems problem. Each individual case may exhibit none, one or many of
these factors.

Primary, secondary and tertiary gains may motivate an individual to somatize.
Psychogenic psychiatric symptoms would be primary gains. Sympathy or disability
payments would be secondary gains, and tertiary gains may apply to family members,
physicians, and attorneys (Ford, 1992). Barsky and Klerrnan (1983) note that patients
can derive both conscious and unconscious gratification from bodily symptoms and

physical suffering. Monetary gain from lawsuits, sick days and disability payments may

induce one to somatize. Increased attention from the doctor, family and friends are other
motivating factors.

Persons may visit physicians to compensate for social support. Katon et al.
(1982) describe how the somatizer within a family can use their condition for control or
manipulation over family members, avoiding responsibilities, and eliciting care and
nurturance from an external source.

Barsky and Klerman (1983) remark that a patient may simply misinterpret their
symptoms. For example, the average person may attribute nausea, fatigue or back pain to
overwork, tension or aging. The average person would not, much less even consider,
seeking medical attention. The somatizer would mistake these common symptoms for a
serious illness and visit the doctor for treatment.

Somatizers may have increased sensitivity to negative somatic and psychic
conditions (Simon and VonKorff, 1991). Differences in pain thresholds, perceptual
reactance and selective perception in somatizers can contribute to their condition.
Organic brain disease can affect how an individual perceives pain and other symptoms;
bereavement or witnessing grave illness can also act as precipitating factors for
somatization. Robbins and Kirmayer (1991) have also noted that first-hand experience
with prior physical illness or psychiatric problems influence how a patient interprets
somatic symptoms.

Numerous studies have demonstrated a link between somatic symptoms and
depression and anxiety. Kroenke et al. (1997) found that 61% of patients with a DSM-
III-R somatoforrn diagnosis had a coexisting mood disorder, and 50% had an anxiety

disorder. Somatization might serve as a defense mechanism for depression and anxiety;

depression and anxiety might manifest themselves as somatic distress. Fear, stress,
loneliness and other emotional stimuli are associated with peripheral physiological
changes. Psychophysiological responses to autonomic nervous system to stress or
voluntary nervous system responses to emotional conflict (DeVaul, 1980) indeed induce
somatic sensations. These changes may be interpreted as symptoms of illness by the
somatizer. Changes in endocrine activity at times of stress or a low pain threshold may
intensify bodily sensations.

Somatic symptoms express an “external” illness, which is more acceptable in
most societies than emotional illnesses. Certain cultures discourage open emotion, and
some do not even acknowledge emotional disorders. In most societies, psychiatric
illness carries such a negative stigma that patients will rationalize the cause of the
symptom to be of somatic rather than psychiatric origin (Ford, 1995).

Most employees in the work force will take sick leave for a somatic illness;
employees tend not to call in sick when they feel emotional stress or depression.
Insufficient reimbursement for psychiatric visits is also to blame for somatization (Ford,
1995). Many health care plans will pay for treatment if one is diagnosed as having
chronic fatigue, but considerably less if one is diagnosed with depression. This creates
pressure for somatizers to medicalize their psychological difficulties in order to obtain
treatment (Bruns, 1997). In addition, patients cannot seek psychiatric help without a
referral under most health care policies.

Somatization may serve as a defense mechanism for denial, displacement or
conflict resolution (Kellner, 1990). These emotional conflicts heighten sensitivity to

bodily sensations, increasing the concern about illness, which the patient is likely to

dwell on (Barsky, 1981). This concern will exacerbate the initial stress the somatizer had
and firrther cause somatic complaints. The majority of somatizers will not admit the
possibility that their symptoms are the result of psychological stress. Using various
attitude scales, Goldberg and Bridges (1988) observed that somatizers were more hostile
to mental illness.

Rosen et al. (1982) present a biopsychosocial model of somatization. The
individual produces a response to a stressful stimulus either somatically or
psychologically. The stimulus can arise internally or externally. Internal stimuli include
physiological and psychological stressors; external stimuli include environmental and
social stressors. The response to the stimulus is influenced by the person’s cultural
beliefs and norms, childhood experiences, psychological and personality profile, and
current environmental systems such as family dynamics, social networks, the medical
care system and the socio-political condition. The response, in turn, can alter or add to
both the stressors and the individual’s response characteristics. The authors believe that
somatization is an idiom for the expression of psychosocial stress and note that it is often
the only legitimate and culturally accepted means of seeking and receiving care from the
family, social network and medical system.

Ford (1983) also illustrates how the disease-illness process is influenced by
psychosocial stress. First, physical or psychological stress modifies the body’s defenses
to increase susceptibility to pathogenic influences. Stress can also induce emotional
responses, which in turn lead to physiologic changes. Stress may also intensify the
perception of existing physical symptoms, or cause the individual to perceive the

symptom differently than he or she would in a nonstressed state. Stress may influence an

individual to seek medical attention when experiencing a physical symptom. Lastly,
illness, or the sick role, itself can be effective in reducing stress. Goldberg and Bridges
(1988) explain that people who play the sick role can avoid responsibility and blame.
The reduction in responsibilities and blame can save the individual from anxiety and
depression. Bridges et al. (1991) feel that subacute somatization may be an adaptive
response to psychiatric illness, but it carries the risk of becoming chronic and difficult to
treat

In some instances, an organic disease may initiate somatization when it is in its
early and undiagnosed stage. Lipowski (1988) explains that somatization might be
mistaken for a disease such as multiple sclerosis or lupus in their early and undiagnosed
stages. He also explains that physicians may not acknowledge illnesses deemed
syndromes to be true diseases. Common psychosomatic syndromes include fibromyalgia,
chronic fatigue syndrome, esophageal motility disorders, nonulcer dyspepsia, irritable
bowel syndrome, urethral syndrome, and pain syndromes (Kellner, 1994). Escobar (1996)
lists other common syndromes, what he terms as functional somatic syndromes, such as
mitral valve prolapse, possible Lyme disease, reactive hypoglycemia, Gulf War
syndrome, multiple chemical sensitivity, systemic yeast infection, and others. Kellner
(1985) specifically lists several causes of somatic symptoms that cannot be detected with
routine examinations and laboratory tests. The physician may conclude that the
symptoms are not indicative of organic disease. These might originate from undetected
physical disease, central pain mechanisms, psychogenic regional pain, endogenous

depression, and chronic pain syndromes.

Diagnosis

The economic and emotional costs due to the incorrect classification of
somatization status must be weighed (Rost and Smith, 1997). The misclassification of a
truly diseased person can be lethal, and treatment undergone for somatization adds to the
cost. Unidentified somatizers will continue being treated for what is presumed organic
illness. The costs due to numerous medical visits, treatment and potential iatrogenic
disease will be high. Correct diagnosis is important.

To diagnose somatizers, the physician needs to be able to identify high-utilizers
and to rule out organic disease. Many physicians believe that additional aspects of the
patient’s medical history and personality need to be explored in order to diagnose
somatization. Rosen et a1. (1982) stress that a thorough past history should be obtained to
include both recent and childhood illnesses, drug and alcohol use, the number of
physicians consulted and the count of medical visits made in the past year. Roberts
(1994) recommends that the physician assess both somatic and psychosocial aspects of
the patient’s life. The physician should inquire specifically about any recent stresses such
as death of a spouse or parent, problems with their marriage, childbirth, job relocation or
loss, and the quality of their emotional support network of friends and family.

Patients suffering from a somatofonn disorder present to their doctor with
physical symptoms characteristic of somatic illness yet organic findings are not detected.
Once organic disease has been ruled out, psychological factors should be explored for
treatment of somatization. The fourth and most recent edition of the American
Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders (DSM-

IV, 1994) defines the following seven disorders as somatofonn disorders: somatization

l3

disorder, undifferentiated somatoforrn disorder, conversion disorder, pain disorder,
hypochondriasis, body dysmorphic disorder and somatoforrn disorder not otherwise
specified (NOS). Strict categorization is not feasible because the criteria for diagnosing
the specific disorders overlap considerably. Nearly all somatizers do not meet the narrow
DSM-IV criteria for somatization disorder. Somatizers who do not meet the DSM-IV
criteria but who are high utilizers with no detectable organic disease have been termed
“subsyndromal”. The specific diagnostic criteria for somatization disorder according to
the DSM-IV is as follows:

1) patients must have a history of many physical complaints beginning before the
age of 30 years persisting over several years and result in treatment being
sought or significant impairment in social or occupational functioning;

2) each of the following criteria must be met: A history of pain related to at least
four different sites or functions, a history of at least two gastrointestinal
symptoms other than pain, a history of at least one sexual or reproductive
symptom other than pain, a history of at least one symptom or deficit
suggesting a neurological condition not limited to pain (psuedoneurological
symptom);

3) after appropriate investigation, each of the symptoms cannot be fully
explained by a known general medical condition or the direct effects of a
substance (e. g., drug or alcohol abuse, medication); if there is a related
organic pathology, the physical complaints or impairment are in excess of
what would be expected from the history, physical examination, or laboratory
findings.

Bias in Somatization Studies

The sources of bias in somatization studies are varied and many. Bias can be

attributed to patient characteristics such as health-seeking behavior, doctor’s education

and judgment such as diagnosing certain types of syndromes as true illness or

unexplained, the investigators. The inherent bias in the investigators’ study design is also

apparent; the selection process, collection of data, diagnosis of somatizers, and data
analysis need to be carefully planned.

When primary physicians are employed to refer high-utilizing patients to a study,
a selection bias occurs. Here, somatizing patients with fewer overall visits are
underrepresented and those patients with an abundance of visits are overrepresented
(Smith et al., 1986, deGruy et al., 1987). Using medical records or a database would
correct this problem.

There is also inherent gender bias in many studies. Although it may be more
acceptable for women to express pain and emotions than men, the majority of diagnosed
somatizers tend to be women. This may be that men prefer not to present to a physician
for fear that they will be labeled “weak”. Golding et a1. (1991) have found that men are
less likely to be perceived as having multiple unexplained symptoms; it has been a cliché
for women to be “hysterical” in this case. Thus, the number of somatizing men will be
underestimated when sampling from medical records. Recall bias occurs when interviews
are used instead of medical records (Kirmayer and Robbins, 1991, Robins et al., 1984).

Classification bias can occur when patients are misdiagnosed due to incomplete
medical records or undiscovered pathological findings (Smith et al., 1986).
Inconsistencies among studies in the definition of a somatizer are problematic.
Prevalence

Studies have found that high utilizers tend to be women, the very young and old,
and those with easy access to medical services (Robins et al., 1984; deGruy et al., 1987;

Roberts, 1990; Campo and Fritsch, 1994; Kroenke et al., 1997).

15

The prevalence of somatizers in the general population is difficult to estimate and
has ranged from less than 1% to almost 50%. The figures vary across studies due to the
characteristics of the sample population and what the authors use for selection and
diagnostic criteria. Authors vary in the manner that high-utilizers are identified and
selected for further analysis. There has been little consistency between studies in the
definition of somatization and the criteria to diagnose it. Often the DSM criteria are
used, but definitions vary according to the version used, e.g. DSM-III published in 1980,
DSM-III-R published in 1987, and DSM-IV published in 1994. Various studies have
compared current diagnostic criteria and many authors have invented their own (Rief et
al., 1996; Portegijs et al., 1996; Kirmayer and Robbins, 1996;Verhaak and Tijhuis, 1994;
Tomasson et al., 1993; Bucholz et al., 1993; Katon et al., 1991; Yutzy et al., 1992;
Escobar et al., 1989, Aug 1987, July 1987; Goldberg and Bridges, 1988; Robins et al.,
1988, 1985). A problem caused by these varied definitions is that the sociodemographic
and illness behavior profiles of somatizers have been shown to deviate significantly
across the definitions (Kirmayer and Robbins, 1991).

Kroenke et al. (1997) developed criteria to diagnose undifferentiated somatoforrn
disorder, “mulitsomatoforrn disorder”, defined as having a two year history of three or
more symptoms form a list of 15 somatofonn symptoms. Using data from the Primary
Care Evaluation of Mental Disorders (PRIME-MD) study, 8.2% were diagnosed with
“mulitsomatoforrn disorder”. Based on DSM-III-R criteria, 14% had a somatofonn
disorder; 4.2% had diagnosis of somatofonn disorder NOS, hypochondriasis in 2.2% and

pain disorder in 0.8%.

16

The l-year prevalence rate of DSM-III-R somatoforrn disorders was studied in
Florence, Italy (F aravelli et al., 1997). Six hundred seventy three cases were randomly
selected and interviewed by their own, as well as four other, general practitioners to
obtain information on symptoms. Medical records were also consulted. The rates are as
follows: 0.7% somatization disorder; 0.7% body dysmorphic disorder; 4.5%
hypochondriasis; 0.6% somatofonn pain disorder; 0.3% conversion disorder; and 13.8%
undifferentiated somatoform disorder. Somatoform disorder NOS was found in 37.9% of
the patients who had no explainable organic disease.

Portegijs et a1. (1996) used the somatic symptom index (881) developed by
Escobar et al. (1989) to study the prevalence of somatization in Dutch general practice
patients. They selected patients aged 20-44 years with at least 12 visits without a somatic
diagnosis within the past three years. Of 80 cases, 45 satisfied the minimum five
complaints from the DSM-III-R list, 36 of which began before age 30, indicating a
prevalence of 45%. They also found that women had twice the risk of somatization.

Fink (1992) reports a prevalence of 19.6% of patients with 10 or more hospital
admissions based on a study conducted in The Netherlands. Records were selected from
individuals aged between 17 and 49 years who were admitted to a hospital at least 10
times between 1977-1984. Patients with a somatic disease were excluded from the initial
sample. With further exclusions such as pregnancy, substance abuse, and psychiatric
patients, final sample size consisted of 113 patients. Somatizers were defined as those
with six or more medically unexplained complaints or symptoms and comprised 56
(49.6%) of the sample of 113. Of the 56 somatizers, 84% were women and 16% were

men.

17

In their study, Kirmayer and Robbins (1991) calculated the prevalence of three
types of somatization in primary care. The sample was based on 685 patients who had
initiated a medical visit to one of two hospitals in Montreal, Canada. The overall
prevalence was 26.3%. Somatization based on the DSM-III criteria was found to be only
1% and S81 criteria was 16.6%. Hypochondriasis was measured on high scores from an
illness worry measure and was found in 7.7% of their sample.

Using data is from the Epidemiologic Catchment Area (ECA) program, Escobar
et al. (1989, Aug 1987, July 1987) discovered a prevalence of somatization disorder
using DSM-III criteria in Puerto Rico to be 0.7%. When based on 881 criteria, the
prevalence in males was 18% and for females was 20%. In Los Angeles, DSM-III
diagnosed somatization disorder was calculated to be 0.03% but using the SSI increased
to 4.4%.

Goldberg and Bridges (1988) selected patients with DSM-III criteria for
psychiatric illness from patients presenting with new illnesses from 15 family practices in
the Manchester, UK. area. They found that 5% had ‘entirely psychiatric’ illnesses, and
19% fulfilled their criteria for somatization. The patient’s diagnosis for somatization
needed to satisfy four criteria: consult for somatic symptoms; attributed symptoms to
physical disease; a DSM-III psychiatric disorder was present; finally, a research
psychiatrist felt that treatment of the affective disorder would reduce or eliminate
symptoms.

Two definitions of somatization were used in the study by deGruy et al. (1987) at
the University of South Alabama Family Practice Center. The 111 cases were derived

from patients visiting the center during a five-week period in 1986 who were interviewed

18

using DSM-III criteria to diagnose somatization disorder. Medical charts of those with
10 or more symptoms were reviewed. “Definite somatization disorder” (DSD) was based
on diagnosis of 14 symptoms for women and 12 for men among 37 symptoms, and
constituted only 5% of the sample. “Borderline somatization disorder” (BSD) was
diagnosed with 10 to 13 unexplained visits and was found in 4% of the cases.

In Arkansas, Smith et al. (1986) asked physicians to refer patients that they
believed were somatizers. Of the 51 patients referred to the investigators, 41 (80%) had a
DSM-III somatization diagnosis. It is interesting to note that when prompted to identify
somatizers within their care, these physicians diagnosed correctly (according to the DSM-
III) the majority of the time.

Using data from the ECA program, Robins et al. (1984) report somatization
disorder rates to be 0.1% in New Haven, CT, Baltimore, MD, and St. Louis, MO based
on DSM-III criteria.

The Physician

The role of the physician in the management of their somatizing patients is
complex. Physicians are primarily trained to diagnose and treat only somatic conditions
(Rosen et al., 1982). Many physicians become frustrated and angry by the somatizing
patient; it is often impossible to uncover any organic illness and consequently the course
of their treatment is unsuccessful. The physician might deem the somatizer as “difficult”
or a “crock”. The physician must be trained to recognize that the somatizing patient is
substituting somatic complaints for psychological distress.

Lipowski (1988) stresses that the doctor’s attitude, statements and actions may

either counteract or encourage somatization. Somatizers present difficult diagnostic and

19

management problems; they may or may not suffer from other psychiatric disorders.
Their office visits consume an inordinate amount of time and money for physicians,
laboratory personnel, the health insurance provider, and the patients themselves. When a
patient presents to their physician with bodily complaints, it is the physician’s duty to
correctly diagnose, provide relief and attempt to cure the patient. Somatizers are a source
of frustration and wasted time for the physician. The somatizing patient will likely be
subjected to numerous laboratory tests and diagnostic procedures, drug treatments, and
possibly exploratory surgery (Monson and Smith, 1983), yet the physician will not be
able to diagnose an organic illness.

The undiagnosed somatizing patient presents symptomatically to their physician,
and does not usually disclose psychological tensions. The somatizer believes that he or
she is truly suffering from somatic disease and may not report emotional stressors to the
physician. The physician might not suspect somatization until the cycle has become a
habit to the patient. Early detection and treatment of somatization saves time and money:
fewer workdays lost due to appointments and illness, fewer diagnostic tests, less
unnecessary medical consultations. There will be less frustration to the doctor and
resentment toward the patient when the disorder is recognized. Both the patient and
doctor will be more satisfied with the course of the treatment, and the doctor-patient
relationship itself.

Barsky (1981) notes that some patients may visit a variety of different doctors
until they encounter one that satisfies their needs. The patient may feel that a doctor does
not care, is incompetent or did not spend enough time with them. The somatizer is often

dissatisfied with their primary doctor for not diagnosing and treating their condition that

20

they will move on to another doctor and start the whole cycle over again. It is important
that the somatizer develops a strong, trusting relationship with their primary physician so
that he or she is deterred from “doctor shopping”.

Salmon and May (1995) believe that patients can in fact coerce their doctors into
the outcome they seek. These patients present symptoms to their physician and tend
exaggerate their emotional distress so that the physician feels that he or she is
incompetent, or possibly fears being charged with malpractice. Because of this effort to
treat and diagnose, the physician’s actions can be viewed as an additional stimulus to the
somatization process.

Treatment and Management

Early detection of somatization is essential to spare the patient from the numerous
physical investigations and chance for iatrogenic illness. It is important that physicians
be trained to recognize the somatizing patient under their care and for these patients to
have easy access to psychiatric services. Severe forms of somatization are best treated by
a collaboration of medical and mental health professionals (Bruns, 1997). Physicians are
able to rule out and treat physiological conditions; psychiatrists are more qualified to treat
the underlying psychological component of somatization.

The treatment process has many phases. The regimen has several steps that both
the physician and patient must comply with for effective management. First, the patient’s
physical and psychological history must be examined as well as a thorough physical
investigation to rule out organic disease. The doctor should then explain symptoms and
discuss the patient’s beliefs and lifestyle. The physician should discuss the role of

psychological factors with the patient and they should mutually agree on a treatment plan,

21

such as behavior modification or therapy. When physicians do make referrals for mental
health treatment, it is important to do so in a supportive manner. Somatizers may think
that they are being labeled as mentally ill, and experience shame and embarrassment
caused by that stigma. Consequently, they may be more reluctant in the future to discuss
their emotions with their doctor (Bruns, 1997). The physician should attempt to control
medical care, promote compliance, and avoid reinforcing the patient’s erroneous beliefs
(Mayou, 1993). Smith (1991) proposes that successful treatment depends on negotiation
and compromise between the doctor and patient. Smith (1991) further proposes that the
physician and patient establish a contract to set realistic goals, agree on diagnostic work-
ups, naming of the diagnosis and eventual treatment regimen for successful management
of the condition.

Once any organic and psychological illnesses are taken into account and treated,
the physician must establish a long-term empathetic relationship with the somatizing
patient. The physician must rule out depression, anxiety and schizophrenia (Barsky and
Klerman, 1983). This may require a psychiatric consultation. It is important for the
physician to recognize that somatizers’ lives revolve around physical distress. When new
symptoms originate, the physician should assess any signs of true organic disease without
extensive diagnostic work-ups or prescribing any medication. The physician will then
follow the patient over time.

The physician must reassure the patient that there is no serious illness and that
they understand that the patient feels hindered, concerned, preoccupied, and emotionally
and physically distressed by their symptoms. The physician should also provide the

opportunity for the patient to express any worries and answer any questions the patient

22

might have. The physician should become skilled on how to listen, talk, encourage and
ease the fears of their somatizing patients. The physician should never frighten the
patient by diagnosing a condition that a patient might interpret as life threatening
(Kellner, 1990). On the contrary, the physician should avoid telling a somatizer that
there is nothing physically wrong with him. Somatizers interpret negative diagnoses as
having been misunderstood or rejected by the doctor. Diagnoses such as chronic fatigue
or panic attack may alleviate their distress, as well as introduce them to the
psychosomatic element of their condition. These explanations may, however, reinforce
the patient’s somatizing tendencies if they feel that they have a serious illness.

The physician will primarily concentrate on alleviating the physical symptoms
together with helping the patient learn to live with the pain and distress caused by the
disorder. Smith (1985) maintains that their primary physician can successfully manage
most somatizers. He also states that the joint efforts of the primary physician and a
psychiatrist can be effective in treating difficult and severely somatizing patients. As a
precaution, regular appointments with the physician should be scheduled. This serves to
monitor the health of all patients, and more importantly any nonsomatizers who were
incorrectly misdiagnosed. Scheduling regular appointments will also prevent the
development of new symptoms in somatizers. The somatizer will already have a “ticket”
to visit the doctor, thus precluding the need to present only when symptomatic. Smith et
al. (1995) noted significantly better physical functioning in somatizing patients assigned
to an intervention that advised physicians to space visits, perform less diagnostic tests and

provide positive diagnoses. Roberts (1994) suggests variable spacing between visits; time

23

intervals between appointments should initially be short and gradually lengthen over
time.

Drugs such as antidepressants may help treat psychiatric aspects of somatization.
Relaxation and stress reduction techniques can reduce arousal levels as well as distract
the somatizer from their symptoms (Lolas, 1991). Counseling or psychotherapy may also
be needed to treat anxiety and/or depression. Meeuwesen et al. (1994) found that 49
somatizers assigned to psychiatric intervention had a decrease in depressive symptoms
compared with 57 somatizers who had none. Patients should be educated to modify the
behaviors that cause emotional tension in their lives, to use medical services judiciously,
and to manage stressful events such as loss. The patient should understand that these
stressors might manifest themselves somatically if not managed properly.

Nonsteroidal anti-inflammatory drugs may be prescribed for particular physical
symptoms. Escobar (1996) also recommends MAO inhibitors and selective serotonin
reuptake inhibitors (SSRIs) for treatment of pain. Physical therapy, acupuncture,
biofeedback and transcutaneous electrical nerve stimulation may also be recommended to

the somatizer for relief of somatic symptoms.

24

CHAPTER 2
DATA AND METHODS
Objective

The aim of this thesis is to develop an effective screening algorithm for
somatizing patients within an HMO setting. A mathematical model will be formulated to
predict the somatization status in high-utilizing patients using demographic and clinical
information abstracted from the patients’ medical record.

Overview

After routine data management and cleaning, the data set was separated into
derivation and validation sets. A randomly selected two-thirds sample was used for the
derivation set and the remainder formed the validation set. Differences between patients
in the two sets were investigated on characteristics that were regarded as important for
this study. Only the amount of copayment charge was found to significantly differ
between patients in the two data sets.

The next task was to identify variables that would be good predictors of
somatization status. The logistic model was fitted to predict the likelihood of
somatization from the derivation set. A receiver operating characteristic (ROC) curve
was graphed to find optimal cutoff points. The predictive power of the model was then
tested on the validation set. Goodness-of-fit tests and residual analysis were also
conducted.

Data Collection
Patient level data were obtained from computerized member records of all three

sites of the Blue Care Network of Mid Michigan (BCNMM) in the Lansing area. A

25

BCNMM Management Information System (MIS) Programmer/Analyst was employed to
extract the records of high medical care utilizers from computerized data tapes. High
utilizers were defined as those having six or more medical visits (65th percentile) during
the 1995 calendar year. The MIS maintains a complete record of all patient encounters by
date, type of service, primary and secondary diagnoses by ICD-9 code. Additionally,
information about the specialty of the medical provider is available.

The criteria for preliminary inclusion were continuously covered members
between the ages of 21-55 years who had at least one visit to a physician, physician’s
assistant, specialist or nurse practitioner on an outpatient basis, emergency room visit, or
a hospital admission between January 1, 1995 and December 31, 1995. This generated an
initial sample of 15,505 patients. From these records, members having less than six visits
during the 1995 calendar year were excluded, resulting in a sample of 5,423 high
utilizers.

The information base provided by BCNMM on these patients included
demographic data, itemizations of medical visits and hospital admissions, and
prescription record. A unique patient identification number tracked services provided to
each patient. Basic demographic data used in this study include gender (male/female),
relationship to subscriber (self/spouse/dependent), date of birth, type of employee
(Michigan State University (MSU)/ State of Michigan/General Motors (GM) /other),
office visit copayment amount ($0/$5/$7/$10), copayment amount ($0/$5/$7/$10). Each
medical visit records the following information: service date, provider type/specialty code
(e.g., cardiology, family practice), type of service code (e.g., medical visit, surgery), place

of service code (e.g., doctor’s office, inpatient hospital), and primary diagnosis and

26

secondary diagnoses identified by ICD-9-CM codes. Service dates were used to compute
the number of visits a patient made during the 1995 calendar year. Hospital admission
and prescription data were not used in this present study.

The study was supported by a grant from the Institute for Managed Care at
Michigan State University (Robert C. Smith, MD, Principal Investigator).

Data Cleanup and Counting Medical Visits

We randomly selected a sample of 1,000 patients for inclusion into the study from
the 5,423 member records provided by BCNMM. A first-year resident physician was
trained by the Principal Investigator to review and rate the medical charts of these
selected patients to determine the presence of organic disease, somatization or if data was
inadequate to determine status, the latter rarely occurring. The chart rating form can be
viewed in the appendix of this thesis. The internal reliability and validity of the
determination of outcome, either somatization, nonsomatization or incomplete was
assessed by periodic rating of charts by both the resident and the principal investigator.
The resident physician trained to review and rate the charts was provided with specific
written instructions and the level of agreement with the principal investigator varied
between 97% and 100%.

Upon chart review of these 1,000 patients, 27 were excluded because they were
BCNMM employees, 66 were excluded because of pregnancy, substance abuse or
psychiatric diagnoses and 23 were excluded due to incomplete data. One duplicate record
was eliminated from the data set. These 117 exclusions yielded a final sample size of

883; true organic disease was found in 761 patients and somatization in 122 patients.

27

The first step in counting the number of medical visits was to identify and count

only the visits where the patient presented due to a physical complaint. Patients

undergoing treatment for substance abuse or psychiatric therapy were excluded because

of the comorbidity associated with somatization, and the need for a different regimen for

treating somatization. The following procedure refines the definition of a medical visit so

that encounters or claims such as laboratory, radiology, pharmacy, allergy shots, dialysis,

and physical therapy, are not counted as a visit. To warrant its count, a visit must be

comprised of one of the following eight combinations of type and place of service in

Table 1; however, medical visits with the particular provider types listed in Table 2 were

not counted as a visit.

Table 1 - Included Visits

 

 

 

Type of Service Place of Service
1 Medical Visit Doctor’s Office
2 Consultation Doctor’s Office
3 Medical Visit Emergency Room
4 Surgery Emergency Room
5 Surgery Ambulatory Surgical Center
6 Consultation Emergency Room
7 Outpatient Services Emergency Room
8 Other Medical Services Emergency Room

 

28

Table 2 - Excluded Provider Types

 

 

Provider Type Provider Type
Allergy Pathology
Ambulance Pediatric Surgery
Anesthesiology Pediatric
Audiology/Hearing Aids Pediatric Allergy
Chiropractic Pediatric Cardiology

Dental Endodontics
Dental General
Dental Oral Surgery
Dental Orthodontics
Dental Periodontics
Dental Surgery - Med.
Geriatrics

Home Health Care
Laboratory

Maternal Fetal Medicine
Neonatology

Nuclear Medicine
Nursing Home

Oral Surgery

 

Pediatric Endocrinology
Pediatric Hematology/Onocology
Pediatric Neurology

Pediatric Opthalmology

Pediatric Pulmonology

Pharmacy

Physical Therapy

Podiatry

Prosthodontics

Psychiatric/ Substance Abuse Facility
Purchase/Rental Order
Radiology

Terminations

 

Creation of New Variables

After these exclusions, the encounter file for the 883 patients contained 20,807

service records. A program was written using the statistical sofiware SAS to count the

number of visits for each patient in the data set. For each patient there exists a record for

each encounter or other service based on the ICD-9—CM code and the date associated with

it. The number of distinct service dates was counted individually for all primary

diagnoses consisting of:

l. Digestive System diagnoses (ICD-9-CM codes 520 through less than 580) denoted by

“Number of GI Visits”

2. Symptoms, Signs, and Ill-defined Conditions diagnoses (ICD-9—CM codes 780
through less than 800) denoted by “Number of MISC Visits”

29

3.

4.

Musculoskeletal and Connective Tissue diagnoses (ICD-9-CM codes 710 through less
than 740) denoted by “Number of MS Visits”

Nervous System and Sense Organs diagnoses (ICD-9-CM codes 320 through less
than 390) denoted by “Number of NS Visits”

All other ICD-9-CM codes are denoted by “Number of OTHER Visits”. The
supplemental coding system using E- prefixed ICD-9-CM billing codes was not
included in the count of Number of OTHER Visits. Diagnoses prefixed by the letter V
were included. Specific diagnoses included:

Infectious and Parasitic Diseases (ICD-9-CM codes 001 through less than 140);
Neoplasms (ICD-9-CM codes 140 through less than 240);

Endocrine, Nutritional, Metabolic, and Immunity Disorders (ICD-9-CM codes 240
through less than 280);

Blood and Blood Forming Organs (ICD-9-CM codes 280 through less than 290);
Mental Disorders (ICD-9-CM codes 290 through less than 320);

Circulatory System (ICD-9-CM codes 390 through less than 460);

Respiratory System (ICD-9-CM codes 460 through less than 520);

Genitourinary System (ICD-9-CM codes 580 through less than 630);

Skin and Subcutaneous Tissue (ICD-9-CM codes 680 through less than 710);
Congenital Anomalies (ICD-9-CM codes 740 through less than 760);

Injury and Poisoning (ICD-9-CM codes 800 through less than 1000).

Based on the five variables that were computed above, the following additional

variables were calculated:

1.

The total number of visits, “Number of TOTAL Visits”, was computed by combining
the above five subtotals for each patient.

The variables “Percent GI Visits”, “Percent MISC Visits”, “Percent MS Visits”,
“Percent NS Visits”, and “Percent OTHER Visits” were created by taking the
appropriate nrunber of visits in each category and dividing by the number of total
visits, “Number of TOTAL Visits”, for each patient.

Grouping the above variables created the following categorical variables:

. The variable “SOMATIC POTENTIAL” is a dichotomous variable that indicates if

the sum of the variables Percent GI Visits, Percent MISC Visits, Percent MS Visits
and Percent NS Visits exceeds Percent OTHER Visits.

SOMATIC POTENTIAL = “Yes” if

(Percent GI Visits + Percent MISC Visits + Percent MS Visits +Percent NS Visits)
> 50% of the Number of TOTAL Visits.

30

SOMATIC POTENTIAL = “No” if

(Percent GI Visits + Percent MISC Visits + Percent MS Visits +Percent NS Visits)
5 50% of the Number of TOTAL Visits.

Therefore, a patient with visits indicative of “somatic potential” would have more
than 50% of his or her visits labeled as Digestive System diagnoses; Symptoms,
Signs, and Ill-defined Conditions diagnoses; Musculoskeletal and Connective Tissue
diagnoses; or Nervous System and Sense Organs diagnoses.

2. “TOTAL” categorizes the Number of TOTAL Visits as:
1. 0 to 7 total visits
2. 8 to 11 total visits
3. 12 to 17 total visits
4. 18 or more total visits
Although the initial sampling procedure selected patients with six or more visits, a
small number of patients in our final data set had fewer than six visits. This was a
result of further exclusions and missing data.

Demographic Information
Table 3 describes the demographics of the 883 patients in our study.

Table 3 — Characteristics of Participants (N = 883)

 

 

Characteristic Subgroup Total N (%) Somatizer Nonsomatizer
N (%) N ("/o)
Age Group, 20—29 118 (13.4) 16 (13.1) 102 (13.4)
years 30—39 277 (31.4) 30 (24.6) 247 (32.5)
40—49 376 (42.6) 59 (48.4) 317 (41.7)
50+ 112 (12.7) 17 (13.9) 95 (12.5)
Gender Male 283 (32.0) 21 (17.2) 262 (34.4)
Female 600 (68.0) 101 (82.8) 499 (65.6)
Employer State of MI 208 (23.6) 31 (25.4) 177 (23.3)
GM 246 (27.9) 27 (22.1) 219 (28.8)
Other'r 429 (48.6) 64 (52.5) 365 (48)
Copay,$ 0 487 (55.2) 67 (54.9) 420 (55.2)
5 275 (31.1) 35 (28.7) 240 (31.5)
7 or 10 121 (13.7) 20 (16.4) 101 (13.3)
Relationship to Self 554 (62.7) 71 (58.2) 483 (63.5)
Subscriber Spouse 296 (33.5) 48 (39.3) 248 (32.6)
Dependent 33 (3.7) 3 (2.5) 30 (3.9)

 

 

I BCNMM employees were excluded from sample
The distribution of somatization status by gender is given below in Table 4 and

igure 1. There are twice as many females in the somatizer group.

31

Table 4 - Distribution of Somatization Status by Gender

 

 

Somatization Male Female
Status % %
Nonsomatizer 92.6 83.2
Somatizer 7.4 1 6.8

 

 

Status by Gender

I Nonsomatizer i l
I Somatizer

  
   

100
80

‘# ,
1

l

l

I Percent

} 20

1 Male 4 ‘ Feale ‘
l

L,

Gender l
Figure 1 - 0138111151 of Soniaazaiién Status by Gender d If I —'V I
Comparability of the Data Sets

In order to derive and test a logistic model to predict somatization status, separate
derivation and validation sets were created. The derivation data set was composed of a
randomly selected two-thirds of the total 883 patients (588 patients) and the validation set
the remaining one-third (295 patients). It is important to assess comparability of the two
data sets to construct an adequate model and to test its reliability with another similar data
set.

For continuous variables, the nonparametric Mann-Whitney/Wilcoxon Test was
used to test for any differences between the cases selected for the derivation and the
validation data sets. Despite the large sample size, the Mann-Whitney/Wilcoxon Test was

preferred because the underlying distributions were extremely skewed and did not

32

approximate a normal distribution. For categorical variables, we used chi-square tests.

The validation data set had slightly higher Number OTHER Visits but the more

meaningful variable Percent OTHER Visits did not significantly differ between the sets.

The results from these analyses are shown in Table 5. A statistical difference was

detected between the data sets in office visit copayment amounts. Within the derivation

set and validation sets there was no association between copayment amount and

somatization status (p = .810 in derivation set, p = .464 in validation set).

Table 5 — Comparison of Derivation and Validation Data Sets

 

 

Variable Probability of
Significant Difference

Number of GI Visits .246
Number of MISC Visits .592
Number of MS Visits .603
Number of NS Visits .268
Number of OTHER Visits .055
Percent GI Visits .336
Percent MISC Visits .450
Percent MS Visits .469
Percent NS Visits .301
Percent OTHER Visits .436
Number of TOTAL Visits .405
Age in Years .104
Somatization Status .643
Relationship to Subscriber .589
Employee Group .653
Copayment Amount .01 7*
Gender .191
Age Group .653
TOTAL .333
SOMATIC POTENTIAL .096

 

 

Preliminary Analysis

The derivation data set was used to assess differences between somatizers and

nonsomatizers on the above variables. The best predictors were hypothesized a priori

33

and labeled as having “somatic potential”, which can be found in a later section of this

thesis titled “Best Predictors”. This preliminary analysis was an aid to the selection of

candidate variables that will be entered in the logistic regression model.

Derivation Data Set Univariate Analyses

The Mann-Whitney/Wilcoxon Test was used to detect any differences between

somatizers and nonsomatizers for continuous variables. The results in Table 6 indicate

that Percent GI Visits, Percent MISC Visits, Percent MS Visits, and Percent OTHER

Visits differ between the groups in the derivation data set.

Table 6 - Mann-Whitney/Wilcoxon Tests and Univariate Comparisons for Continuous
Variables in the Derivation Data Set for Somatizers vs. Nonsomatizers (N = 588)

 

 

 

 

 

 

 

Variable Somatizers Nonsomatizers Probability
Median Interquartile Median Interquartile . (3f
Range Range Significant
Dtflerence

Percent G1 0.00 0.00 — 10.00 0.00 0.00 — 0.00 .004*

Visits

Percent MISC 8.33 0.00 — 25.00 0.00 0.00 — 14.29 .001 *

Visits

Percent MS 34.62 11.11 — 57.14 0.00 0.00 - 16.67 <.0001*

Visits

Percent NS 0.00 0.00 — 10.00 0.00 0.00 -— 10.82 .267

Visits

Percent OTHER 36.00 22.22 — 50.00 71.43 50.00 — 87.08 <.0001*

Visits

AGE 40.84 35.54 — 46.84 40.74 33.75 - 46.83 .830

 

 

Chi-square tests were performed to detect any statistically significant differences

between somatizers and nonsomatizers in categorical variables. As indicated in Table 7,

somatization status differs between GENDER, TOTAL and SOMATIC POTENTIAL.

There are significantly fewer male somatizers than female somatizers. There are fewer

somatizers with less than eight visits and more somatizers with twelve or more visits.

34

Somatizers are more likely than nonsomatizers to have visits indicative of having somatic

potential than would be expected by chance alone. There is no reason to believe that

somatization status differs among any other categorical variables.

Table 7 —- Chi-Square Tests and Univariate Comparisons for Discrete Variables in the

Derivation Data Set for Somatizers vs. Nonsomatizers (N = 588)

 

 

 

 

Variable Somatizer Nonsomatizer Probability of

N (%) N (%) Significant
Diflerence

Age Group, years .932

20—29 12 (15.2) 70 (13.8)

30—39 23 (29.1) 166 (32.6)

40—49 34 (43.0) 208 (40.9)

50+ 10 (12.7) 65 (12.8)

Gender .001*

Male 14 (17.7) 183 (36.0)

Female 65 (82.3) 326 (64.0)

Employer .414

State of MI 22 (27.8) 122 (24.0)

GM 17 (21.5) 145 (28.5)

OtherI 40 (50.6) 242 (47.5)

Copay,$ .810

0 46 (58.2) 287 (56.4)

5 20 (25.3) 146 (28.7)

7or10 13 (16.5) 76 (14.9)

Mimi) to Subscriber .550

Self 46 (58.2) 316 (62.1)

Spouse 31 (39.2) 172 (33.8)

Dependent 2 (2.5) 21 (4.1)

TOTAL <.0001*

0-7Visits 4 (5.1) 202 (39.7)

8 — 11 Visits 16 (20.3) 188 (36.9)

12 - 17 Visits 30 (38.0) 82 (16.1)

18+ Visits 29 (36.7) 37 (7.3)

SOMATIC POTENTIAL <.0001*

Yes 58 (73.4) 102 (20.0)

No 21 (26.6) 407 (80.0)

 

 

I Does not include BCNMM employees.

35

 

Validation Data Set Univariate Analyses
The results in Table 8 indicate that Percent MISC Visits, Percent MS Visits,
Percent OTHER Visits and AGE differ between the groups in the validation data set.

Table 8 - Mann-Whitney/Wilcoxon Tests and Univariate Comparisons for Continuous
Variables in the Validation Data Set for Somatizers vs. Nonsomatizers (N = 295)

 

 

 

 

 

 

 

Variable Somatizers Nonsomatizers Probability
Median Interquartile Median Interquartile . .Of
Range Range Significant
Difference
Percent GI 0.00 0.00 — 9.09 0.00 0.00 — 4.71 .073
Visits
Percent MISC 12.50 0.00 — 27.78 0.00 0.00 — 14.29 .001*
Visits
Percent MS 15.38 0.00 — 33.33 0.00 0.00 — 17.80 .001*
Visits
Percent NS 0.00 0.00 — 11.11 0.00 0.00 — 11.11 .436
Visits
Percent OTHER 50.00 40.91 — 58.33 70.29 50.74 — 85.71 .000*
Visits
AGE 45.96 39.25 — 49.76 41.67 34.96 — 47.02 .007*

 

 

Chi-square tests were performed to detect any statistically significant differences
in the following categorical variables between somatizers and nonsomatizers. As
indicated in Table 9, somatization status differs between GENDER, TOTAL and
SOMATIC POTENTIAL.

There are significantly fewer male somatizers than female somatizers. There are
fewer somatizers with less than eight visits and more somatizers with at least twelve
visits. These results are similar in the derivation data set. The distribution of SOMATIC
POTENTIAL among somatizers and nonsomatizers is statistically significant yet does not

agree with the associations in the derivation data set.

36

Table 9 — Chi-Square Tests and Univariate Comparisons for Discrete Variables in the

Validation Data Set for Somatizers vs. Nonsomatizers (N =295)

 

 

 

 

 

Variable Somatizer Nonsomatizer Probability of

N (%) N (%) Significant
Diflkrence

Age Group, years .122

20—29 4 (9.3) 32 (12.7)

30—39 7 (16.3) 81 (32.1)

40—49 25 (58.1) 109 (43.3)

50+ 7 (16.3) 30 (11.9)

921421 .044*

Male 7 (16.3) 79 (31.3)

Female 36 (83.7) 173 (68.7)

Employer .652

State of MI 9 (20.9) 55 (21.8)

GM 10 (23.3) 74 (29.4)

OtherI 24 (55.8) 123 (48.8)

Copay, S .464

0 21 (48.8) 133 (52.8)

5 15 (34.9) 94 (37.3)

7or 10 7 (16.3) 25 (9.9)

Relationihip to Subscriber .457

Self 25 (58.1) 167 (66.3)

Spouse 17 (39.5) 76 (30.2)

Dependent l (2.3) 9 (3.6)

TOTAL <.0001*

0—7Visits 7 (16.3) 88 (34.9)

8 —— 11 Visits 11 (25.6) 94 (37.3)

12 - 17 Visits 14 (32.6) 55 (21.8)

18+ Visits 11 (25.6) 15 (6.0)

SOMATIC POTENTIAL .009*

Yes 16 (37.2) 49 (19.4)

No 27 (62.8) 203 (80.6)

 

 

I Does not include BCNMM employees.

Best Predictors

The benefits of effective screening of somatizing patients have been presented.

This thesis will offer an algorithm to aid physicians in the BCNMM ascertain if a high-

utilizing patient is a somatizer. A logistic regression model was used to calculate the

probability of somatization. The hypothesized best predictors were system—based
symptoms, based on the ICD-9—CM classifications, that had “somatic potential”. The
symptoms identified as having “somatic potential” were diagnoses from the digestive
system; musculoskeletal and connective tissue; nervous system and sense organs; and
symptoms, signs, and ill-defined conditions based on ICD-9-CM codes. The choice of
predictor variables was primarily based on a pilot study conducted at St. Lawrence
Family Health Center in Lansing, Michigan by the principal investigator Robert C.
Smith, MD, of Michigan State University. In this study, it was found that patients
having eight or more visits with “somatic potential” were indicative of chronic
somatization. The predictor variables were also selected from a review of the literature
(Saxe et al., 1994; Escobar et al., 1987, 1989; Smith et al., 1986); preliminary analysis of
the current data; and the clinical experience and expertise of Robert C. Smith, MD, the
principal investigator of this study. Next, an ROC curve was graphed using the model’s
predicted probabilities. The ROC curve was used to aid in the selection of optimal cutoff
points in determining somatization status. The derivation data set, consisting of two-
thirds of the sample provided by BCNMM, was used to build the logistic model and
calculate the ROC Curve. The model was validated using the validation data set,
contained the remaining third of the BCNMM data.

Three types of variables were considered for entry into the logistic model:
demographic variables; variables possessing “somatic potential”; and the total number of
visits. Examples of demographic variables include gender and age. In the St. Lawrence
pilot study, medical visits for gastrointestinal (GI), musculoskeletal (MS), nervous

system (NS) and ill-defined conditions (MISC), were found predictive of somatization, or

38

that they possessed “somatic potential”. Findings from the pilot study also revealed that
25% of patients with eight or more medical visits were somatizers. When only
diagnoses with “somatic potential” were counted, 36% of patients having eight or more
visits were somatizers.

In this thesis, the dichotomous variable SOMATIC POTENTIAL equals “Yes”

when the percent of visits with “somatic potential” exceeds 50% of the Number of I

TOTAL Visits. Otherwise, it is “No” when the percent of visits with “somatic potential”

is less than 50% of the Number of TOTAL Visits. In the derivation data set, 73.4% of

 

somatizers and 80.0% of nonsomatizers were classified correctly using SOMATIC L:
POTENTIAL as previously shown in Table 7. The positive predictive value is 36.3%
and negative predictive value is 95.1%. In the validation data set, 37.2% of somatizers
and 80.6% of nonsomatizers were classified correctly using SOMATIC POTENTIAL as
was exhibited in Table 9. The positive predictive value here is 24.6% and negative
predictive value is 88.3%.

Simple inspection of the total number of visits relating to somatization status
results in another interesting relationship. To simplify the relationship, reduce the
variable denoting the total number of visits in grouped form, “Number of TOTAL
Visits”, as displayed in Table 7 and Table 9, to a dichotomous variable called
“TOTAL(2)”. If TOTAL(2) is dichotomized (no significant difference between
derivation and validation data sets, p = .403) at eight visits, prediction of somatization by
eight or more visits and nonsomatization by less than eight visits results in high
sensitivity and negative prediction value. In the derivation data set, 94.9% of somatizers

and 39.7% of nonsomatizers were classified correctly using TOTAL(2) as shown in Table

39

10. The positive predictive value here is 19.6% and negative predictive value is 98.1%.
In the validation data set, 83.7% of somatizers and 34.9% of nonsomatizers were
classified correctly using TOTAL(2) as shown in Table 11. The positive predictive value
here is 18.0% and negative predictive value is 92.6%.

Table 10 - Distribution of Somatization Status by TOTAL(2) in the Derivation Data Set

 

 

T 0TAL(2) Somatizer Nonsomatizer I?
< 8 Visits 4 202
2 8 Visits 75 307
Total 79 509

 

 

 

Table 11 - Distribution of Somatization Status by TOTAL(2) in the Validation Data Set

 

 

TOTAL(2) Somatizer Nonsomatizer
< 8 Visits 7 88
2 8 Visits 36 164
Total 43 252

 

 

Contingency table analyses with the dichotomous variables SOMATIC
POTENTIAL and TOTAL(2) variables are simple ways to tentatively predict
somatization status of a patient if the clinician knows the total number of visits or the
relative percents of the various classifications of visits as given above. These variables

have high negative predictive values.

40

CHAPTER 3
THE LOGISTIC MODEL
Main Effects Logistic Regression Model
The logistic model is given by the following equation:
logit(p) = ln (p/l -p) = or + [31x1 + [32x2 + + 13ka
= ln (odds of being a somatizer)

where 0 < p < 1, x1, x2, xk are either continuous or categorical predictor (independent)
variables; [31, [32, ..., Bk are weights, called regression parameters, estimated by the
method of maximum likelihood; and or is a constant called the intercept parameter.

The probability of somatization given by the logistic regression model is of the
form:

p = probability = P(somatizer l x1, x2, ..., xk) = 1 / (1 + e "a + Bl“; + [’22 + + 3151))
=e (or + 131.1] + 112.1; ”CW (1 + e (a + 1351+ 1152+ +13k-rk))

The following dichotomous variables were selected for entry into the main effects
model because of their association with somatization status in the derivation data set:
GENDER: The variable GENDER is coded l for female and 0 for male genders. Males
have been designated as the referent group for GENDER.

TOTAL(2): This variable is coded 1 if the Number of TOTAL Visits is greater than or
equal to eight and coded 0 if the Number of TOTAL Visits is less than eight.

SOMATIC POTENTIAL: SOMATIC POTENTIAL = 1 (“Yes”) if the sum of
Percent GI Visits, Percent MISC Visits, Percent MS Visits, and Percent NS Visits
exceeds 50% of the Number of TOTAL Visits. SOMATIC POTENTIAL = 0 (“No”) if
the sum of Percent GI Visits, Percent MISC Visits, Percent MS Visits, and Percent NS
Visits is less than or equal to 50% of the Number of TOTAL Visits.

41

 

Model Fitting Steps

Many steps were involved in developing a model to fit these data. First the
variables hypothesized to differentiate between somatizers and nonsomatizers were
entered: GENDER, TOTAL(2) and SOMATIC POTENTIAL. Next, other potential
predictive factors were entered: Age group, Employer, Copay and Relationship to
Subscriber. None were significant nor were they determined to be confounders. Using a
hierarchical model, all interactions were eliminated because of non-significant p-values.

Steps are given in Table 12 below. SPSS software was used in these analyses.

 

42

Table 12 — Model Entry and Elimination Procedures

 

Logistic Model Stages

 

Hypothesized mlain effects are entered into model

SEX (p = .0014)

TOTAL(2) (p < .0001)

SOMATIC POTENTIAL (p < .0001)

All main effects are significant and will be retained in model.

Main effects entered using forced entg; other covariates tested for remaining in model
using stepwise backward procedure

Relationship to Subscriber (p = .8369)

Copay (p = .8217)

Age Group (p = .7431)

Employer (p = .2446)

These variables were removed from the model. The above main effect model using
GENDER, TOTAL(2) and SOMATIC POTENTIAL will be maintained.
Confounder Assessment

Compare odds ratios of main effects with and without each separate covariate (see
above) to assess confounding. Odds ratios are not significantly altered between models.
Beta Coefficients for Main Effects Model:

SEX 1.1262

TOTAL(2) 2.6476

SOMATIC POTENTIAL 2.6263

Beta Coefficients with Age Group Added to Main Effects Model:
SEX 1.0983

TOTAL(2) 2.6659

SOMATIC POTENTIAL 2.6833

Beta Coefficients with Employer Added to Main Effects Model:
SEX 1.1063

TOTAL(2) 2.6858

SOMATIC POTENTIAL 2.6638

Beta Coefficients with CopayAdded to Main Effects Model:
SEX 1.1629

TOTAL(2) 2.6368

SOMATIC POTENTIAL 2.6177

Beta Coefficients with Relation Added to Main Effects Model:
SEX 1.1081

TOTAL(2) 2.6417

SOMATIC POTENTIAL 2.6326

Interaction Assessment Using Hierarchical Model
SEX*TOTAL(2)*SOMATIC POTENTIAL eliminated (p= .9504)
SEX*TOTAL(2) (p = .8850)

TOTAL(2) * SOMATIC POTENTIAL (p = .6817)
SEX*SOMATIC POTENTIAL (p = .5136)

No significant interaction effects between main effects.

 

43

 

Constructed Model
The final model and information pertaining to its fit are given below.
Interactions and potential confounders are not included in the final model, as they were of
no importance. All independent variables are indicator variables.
The final model is given by the following equation where GENDER, TOTAL(2)
and SOMATIC POTENTIAL are previously detailed:
1n (odds) = -6.0552 + ( 1.1262 x GENDER) + (2.6476 x TOTAL(2))
(2.6263 x SOMATIC POTENTIAL)
The probability of being a somatizer can be calculated by the following formula:
P = odds/(1+odds), where odds can be derived by exponentiating the ln(odds) above.
Interpretation
The probability of being a somatizer can be computed from the logistic regression
model. The parameters to be used in the model are given in Table 13. The decision rule,
or cutoff point, will be decided upon according to the desired sensitivity and specificity.
Contingency table analysis and the ROC curve aid in this determination.

Table 13 - Parameter and Odds Ratio Estimates Using the Method of Maximum
Likelihood

 

 

 

Variable Parameter Chi- Probability Odds Lower Upper
Estimate Square Ratio 95% 95%
(d.f.) Cl. C].
CONSTANT -6.0552 10.2234 < 0.0001 *
(or) (1 d-f)
SEX 1.1262 23.9777 0.0014* 3.0838 1.5463 6.1501
(131) (1dr)
TOTAL(2) 2.6476 76.4793 <0.0001* 14.1207 4.8934 40.7472
(132) (14.0
SOMATIC 2.6263 88.8915 <0.0001* 13.8219 7.6727 24.8994
POTENTIAL (1 d.f)
(133)

 

 

 

 

 

 

 

*All independent variables are significant.

44

 

 

Odds ratios and their confidence intervals for the three variables in the model are
shown in Table 14. An independent variable’s regression parameter can be
exponentiated to obtain the odds of being a somatizer. It is the odds of being a somatizer
for that independent variable while controlling for all other independent variables.

Table 14 — Adjusted Odds Ratios for Somatization Status

 

 

 

 

 

 

Variable Odds Ratio 95% Confidence
(Referent) Interval

SEX 3.0838 1.5463 - 6.1501
(MALE)
TOTAL(2) 14.1207 4.8934 - 40.7472
(<8 Visits)
SOMATIC POTENTIAL 13.8219 7.6727 - 24.8994
(GI+MISC+MS+NS s 50%)

 

For example, the odds of somatization for a female compared to a male is 3.0838 when
all other variables are held fixed. The odds of somatization for an individual with eight
or more visits versus an individual with less than eight visits is 14.1207. With all other
variables held fixed, the odds of somatization for an individual with more than 50% of
total visits having “somatic potential”, i.e. GI+MISC+MS+NS > 50%, versus an
individual with 50% or less is 13.8219.

Various combinations of independent variables can produce up to eight different
predicted probabilities. For an individual in the referent group (male with less than eight
visits and 50% or less visits with “somatic potential”) odds for somatization is exp(-

6.0552) = .0023. Odds for all predictors are listed in Table 15.

45

 

Table 15 — All Possible Predicted Probability and Odds

 

 

 

 

SEX TOTAL(2) SOMA TIC POTENTIAL PREDIC TED ODDS
PROBABILITY

Female 8+ Visits GI+MISC+MS+NS > 50% .58537 1.4118
Female 8+ Visits GI+MISC+MS+NS S 50% .09268 .1021
Female <8 Visits GI+MISC+MS+NS > 50% .09089 .1000
Female <8 Visits GI+MISC+MS+NS S 50% .00718 .0072
Male 8+ Visits GI+MISC+MS+NS > 50% .31404 .4578
Male 8+ Visits GI+MISC+MS+NS S 50% .03206 .0331
Male <8 Visits GI+MISC+MS+NS > 50% .03140 .0324
Male <8 Visits GI+MISC+MS+NS _<_ 50% .00234 .0023
ROC Curve

The ROC curve is a plot of the false positive rate (1 — specificity) by the true
positive rate, (sensitivity) for various probability cutoffs. The SAS procedure PROC
LOGISTIC calculates the sensitivity and specificity for these probabilities using 2x2
contingency tables based on each cutoff. The ROC curves were then constructed using
Microsoft Excel.

The area under the ROC curve as well as the height above the diagonal, or chance
line, are important attributes of the curve. (Hanley and MCNeil, 1982) Perfect prediction
by the model would yield an area exactly equal to one; prediction based on chance would
have an area equal to 0.5. The ROC curve for the final model has area equal to .8640 in
the derivation data set and .6651 in the validation data set. The greater the height above
the chance line indicates better discrimination between the somatizer and nonsomatizer
groups. The ROC curve in Figure 2 indicates that the optimal probability cutoff is when
p > .03206 to predict somatization since this value is furthest from the chance line. In the

data set, this cutoff yields 96.2% sensitivity and 51.47% specificity. In the validation

46

 

data set, this cutoff has 79.07% sensitivity and 44.84% specificity. These values are

shown in Table 16 and Table 17.

 

 

ROC Curve

100 ,4, -_ _.. . , -
l Derivation , . _ : Z 7 t
90 ' Data Set f L { 1

Validation
Data Set
p = .03206

 

 

 

 

 

 
    
 

  
  

 

 

 

 

 

 

   

 

 

 

 

 

 

 

 

. a .
l '5 1
'8 p l 1
l a ; ‘ i Y l _
l ° 3 9 i ? 1 g
I a "" . , - , i - 1
‘ '- ' l * 1
l 30 T k} f l . T» - l 44.472; ‘
i
i 0 1

. i i g . .
J ——.— L ——.7 r—s.___.,_.__.—.L-.__ -—--—$— ..__ - -.J
, .

0 1O 20 30 40 50 60 70 80 90 100

False Positive

 

“ FiguTe2— ROC—Curve

Choice of Cutoff Values

The desired sensitivity and specificity can be modified as needed. Depending on
the cost of making a true or false prediction, the choice of cutoff points can be altered.
The cutoff value p = .03206 is the best choice when high sensitivity is needed; it is
approximately 96% sensitive and 51% specific in the derivation set. The positive

predictive value is 23.53% and the negative predictive value is 98.87%. This decision

47

 

rule has 79% sensitivity and 45% specificity in the validation set, with positive predictive

value 19.65% and negative predictive value 96.62%.

Table 16 — Prediction in the Derivation Set

 

 

 

 

 

 

 

 

 

 

 

 

Probability Prediction Diagnosed Diagnosed
Nonsomatizers Somatizers

P < .03206 Nonsomatizer 262 (51.47%) 3 (3.8%)

P 2 .03206 Somatizer 247 (48.53%) 76 (96.20%)

Table 17 — Prediction in the Validation Set

Probability Prediction Diagnosed Diagnosed
Nonsomatizers Somatizers

P < .03206 Nonsomatizer 113 (44.84%) 9 (20.93%)

P 2 .03206 Somatizer 139 (55.16%) 34 (79.07%)

 

 

 

 

 

 

 

Appropriateness of the Model

Table 18 indicates that the null hypothesis that the independent variables in the
model are not significant is rejected. The variables used to calculate the logistic
regression parameters are significant. Pseudo R-Squared measures were computed: Cox
& Snell's = .220 and Nagelkerke's = .404.

Table 18 - Model Fitting Information and Testing the Global Null Hypothesis [3 = 0

 

 

Test Intercept Intercept and C hi-Square Probability

Statistic Only C ovariates (d. f.)

-2 Log L 317.589 225.152 146.437 < 0.0001
(3 d.f.)

 

 

Hosmer and Lemeshow’s goodness-of-fit test confirms that the data fits the model

very well.

Table 19 — Goodness-of-Fit Test

 

 

Test Statistic C hi-Square Probability
(d-f)
Hosmer and Lemeshow 1.6404 0.9496
(6 d.f.)

 

 

48

We should examine potential outliers in the data. Table 20 exhibits 26 cases

from the derivation data set with large residual values. Residual analysis indicates that

the following might be considered outliers because the difference between their true

somatization status and their expected probabilities was large. These cases are close in

value to the small value used for the probability cutoff. Only one somatizer who had a

large residual value was misclassified. No outliers were eliminated from the data; all

cases were used to construct the logistic regression model.

Table 20 - Residual Analysis

 

 

 

Status GENDER TOTAL(2) SOMA TIC POTENTIAL Model ’s Studentized
Predicted Residual
Probability
Somatizer" male <8 Visits GI+MISC+MS+NS > 50% 0.0314 5.5537—
Somatizer male 8+ Visits GI+MISC+MS+NS S 50% 0032" 5-4946
Somatizer male 8+ Visits GI+MISC+MS+NS g 50% 0.0321 5.4946
Somatizer female <8 Visits GI+MISC+MS+NS > 50% 0.0909 3.1626
Somatizer female <8 Visits GI+MISC+MS+NS > 50% 0.0909 3.1626
Somatizer female <8 Visits GI+MISC+MS+NS > 50% 0.0909 3.1626
Somatizer female 8+ Visits GI+MISC+MS+NS S 50% 00927 31289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 00927 3.1239
Somatizer female 8+ Visits GI+MISC+MS+NS _<_ 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 00927 31289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS S 50% 00927 31239
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 00927 31289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 00927 3-1289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS _<_ 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 00927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS s 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS 5 50% 0.0927 3.1289
Somatizer female 8+ Visits GI+MISC+MS+NS g 50% 0.0927 3.1289

 

*Misclassified case

49

Strengths and Limitations

The strengths of this study outweigh any shortcomings. One discrepancy was
between the number of visits computed in our analysis and those counted by BCNMM.
The algorithm used by BCNMM was to extract only those patients with six or more
visits, the lower cutoff value for high-utilization set by the principal investigator.
However, when applying our exclusion criteria to the data set, the total number of visits
was diminished in many cases, which lead to 3.4% (30 cases) of our sample to have less
than six visits. All these were nonsomatizers. Excluding these cases from the data did not
significantly alter the results of this thesis.

Examination of the data indicated that 3.4% (30 cases) of our sample had less
than six visits and that these were all nonsomatizers. Excluding these cases from the data
did not significantly alter the results of this thesis.

Table 21 - Visit Count Discrepancy

 

 

Number of Frequency Percent of 883 cases
Visits %

3 3 0.34

4 l 0 l . 1 3

5 17 1.93

Total 30 3.4

 

 

Another limitation of the study is that without direct patient contact, the
investigators were not able to diagnose a specific somatoforrn disorder had to exclude
organic disease using just chart review.

The extraction of cases through BCNMM computerized database eliminates any
selection, measurement, and participation bias. Classification bias might exist in this

study if the chart rater misdiagnosed the somatization status of any patients. If this were

50

the case, the discriminatory capability of the model would be reduced. The more cases
that are classified correctly will enhance the model’s ability to predict somatization
status.

Potential interactions were carefully considered. Effect modification and
confounding would have been adjusted for in the logistic regression model had they
existed.

A model including continuous variables utilizes much more information from the

collected data compared to categorical, and especially dichotomous, variables. There is

‘w'.’ :Lrl' In.

only one fi-coefficient to estimate for each continuous variable; for categorical variables,

 

one must estimate B-coefficients for one less than the total number of categories. For
example, if there are five categories for variable X then four B-weights will need to be
estimated. The use of arbitrary cutoff points as in categorical variables is not an issue
when using continuous variables. The ROC curve is constructed with several more
points giving a greater choice of potential cutoff points. The continuous model is not
sensitive to empty cells, which can be problematic in categorical models.

A model consisting of three dichotomous variables results in the estimation of one
B-coefficient for each variable in the model, thus a total of eight possible predicted
probabilities for somatization. The benefits of a model containing only categorical
variables include an uncomplicated calculation of predicted probabilities as well as a
model that makes intuitive sense.

The derivation and validation data sets do not significantly differ on any of the
variables entered into the final model. The sample sizes were large in both data sets. It is

expected that the results will vary when the model is tested with a different data set. The

51

differences in sensitivity, specificity and predictive value are due to slight differences
between the data sets. Overall, the validation step in the analysis showed that the model
was capable of good discrimination and predictive value.

The model’s sensitivity, specificity, positive and negative predictive values
indicate that the model is reliable and can be used as an aid for diagnosis of somatization.
It should be noted that changes in the distribution of somatization status within different
populations and differences in a physician’s skill at diagnosing symptoms would alter the
relative prevalence of somatizers. The positive predictive value can be compared from
the following formula (Schulzer, 1994):

Positive predictive value = Pr (somatizer | predicted somatizer using decision rule)
= (sensitivity x Pr(somatizer))/

((sensitivity x Pr(somatizer)) + (false positive rate x Pr(nonsomatizer))).
Changes in prevalence of somatization would affect the predictive value of the logistic
regression model as well as the selection of optimal cutoff values.

A shortcoming of the logistic regression model is that it predicts only slightly
better than the simple variables SOMATIC POTENTIAL and TOTAL(2) in contingency
tables. If the percentage of GI, MISC, MS and NS visits exceed the number of OTHER
visits (SOMATIC POTENTIAL), or if the patient has eight or more total number of visits
(TOTAL VISITS), suggest that somatization is very likely. This is an extremely
uncomplicated and reliable way to identify somatizers among high-utilizers.

Conclusion
The resulting model can serve as a guideline for physicians in the BCNMM to

identify their high-utilizing somatizing patients. It is important that the physician

52

 

continue to trust their own clinical experience, judgement, and familiarity with the
individual patient rather than rely solely upon the results of the model. The methods
described here can be used to develop models for other HMOs. This specific model may
be applicable to other HMOs, but it is recommended that each individual HMO develop
its own model. Differences in the patient demographics and prevalence of somatization,
will influence the logistic regression model. Consequently, the variables and their
weights will change and the predicted probabilities will vary accordingly. Cutoff values
will have to be modified for the distribution of predicted probabilities among somatizers

and nonsomatizers for the desired levels of sensitivity and specificity.

53

 

APPENDICES

54

 

 

APPENDIX A

55

r?

 

 

APPENDIX A

SPECIAL TREATMENT PROJECT CHART RATING FORM

DATE:

PATIENT ID #: DOB:

 

RATER:

PERIOD OF TIME RATED

 

# OF VISITS RATED # OF PHONE
CONTACTS

 

 

COMPLAINTS/SYMPTOMS DIAGNOSES
(combine symptoms that occur together as one complaint)

1.

 

 

 

2.

 

 

 

 

 

 

 

 

DOCUMENTATION TO DETERMINE PERSENCE/ABSENCE OF ORGANIC
DISEASE (for each complaint/symptom)

 

 

 

 

 

56

OVER
CONCLUSION (For each complaint, assign one of the following letters:)

. ORGANIC DISEASE (assign Charleston weight to each applicable organic disease)

. SOMATIZATION

. DATA INADEQUATE TO DETERMINE CLASSIFICATION (no more than 10% of
all patients rated)

 

 

 

 

 

 

 

 

 

. A WEIGHT B C
. A WEIGHT B C
. A WEIGHT B C

A WEIGHT B C

 

 

 

TOTAL

 

57

 

 

 

WEIGHTED INDEX OF COMORBIDITY
ASSIGNED WEIGHTS FOR DISEASES

l

b)

CONDITIONS

Myocardial infarct
Congestive heart failure
Peripheral vascular disease
Cerebrovascular disease
Dementia

Chronic pulmonary disease
Connective tissue disease
Ulcer disease

Mild liver disease
Diabetes

Hemiplegia

Moderate or severe renal disease
Diabetes with end organ damage
Any tumor

Leukemia

Lymphoma

Moderate or severe liver disease
Metastatic solid tumor
AIDS

 

 

 

Assigned weights for each condition that a patient has. The total equals the score.
Example: chronic puhnonary (1) and lymphoma (2) = total score (3).

58

 

REFERENCES

59

 

 

REFERENCES

American Psychiatric Association, Diagnostic and Statistical Manual of Mental
Disorders, Fourth Edition. Washington DC: American Psychiatric Association, 1994

Barsky, AJ, “Hidden Reasons Some Patients Visit Doctors”. Ann Intern Med, 1981;
94(1): 492-98

Barsky, AJ, Klerman, GL, “Overview: Hypochondriasis, Bodily Complaints, and Somatic
Styles”. Am J Psychiatfl, March 1983; 140: 273-83

Barsky, AJ, Borus, JF, “Somatization and Medicalization in the Era of Managed Care”.
JAMA, 1995; 274(24): 1931-4

Bridges, K, Goldberg, D, Evans, B, et al., “Determinants of Somatization in Primary
Care”. Psychol Med, 1991 May; 21(2): 473-83

Bruns, D. The Problem of Somatization. [Online] Available http://www.healthpsych.com,
11/26/97.

Bucholz, KK, Dinwiddie, SH, Reich, T, et al., “Comparison of Screening Proposals for
Somatization Disorder Empirical Analyses”. Compr Psychiatry, 1993 Jan-Feb;
34(1): 59-64

Campo, JV, Fritsch, SL, “Somatization in Children and Adolescents”. J Am Acad Child
Adolesc Psychiatry, 1994 Nov-Dec; 33(9): 1223-35

Cantor, C, flantom Illness: Shattering the Myth of Hypochondria. New York, NY:
Houghton Mifflin Company, 1996

deGruy, F, Columbia, L, Dickinson, P, “Somatization Disorder in a Family Practice”.
J Fam Pract, 1987; 25(1): 45-51

DeVaul, RA, “Hysterical Symptoms” In: Hall, CW (editor), Psychiatric Presentations of
Medical Illness. Som_atopsychic Disorders. Chicago, IL: Spectrum Publications, 1980

Dickson, LR, Hays, LR, Kaplan, C, et al., “Psychological Profile if Somatizing Patients
Attending the Integrative Clinic”. Int J Psychiatry Med, 1992; 22(2): 141-53

Escobar, JI, Golding, JM, Hough, RL, et al., “Somatization in the Community:

Relationship to Disability and use of Services”. Am J Public Health, July 1987;
77(7): 837-40

60

 

Escobar, JI, Burnam, MA, Karno, M, et al., “Somatization in the Community”. Arch Gen
Psychiap'y, Aug 1987; 44: 713-8

Escobar, JI, Rubio-Stipec, M, Canino, G, et al., “Somatic Symptom Index (SS1): A New

and Abridged Somatic Construct. Prevalence and Epidemiological Correlates in Two
Large Community Samples”. J Nerv Ment Dis, 1989; 17(3): 140-6

Escobar, JI, “Overview of Somatization Diagnosis, Epidemiology, and Management”.
Psychopharmacol Bull, 1996; 32(4): 589-96

Faravelli C, Salvatori S, Galassi F, et al., “Epidemiology of Somatoform Disorders: A
Community Survey in Florence”. Soc Psychiam Psychiatr Epidemiol, 1997 Jan;
32(1): 24-9

- 1I_Jm. .h'. —

Fink, P, “The Use of Hospitalizations by Persistent Somatizing Patients”. Psychol Med,
1992 Feb; 22(1): 173-80 I

 

Fink, P, “Somatization-Beyond Symptom count”. J Psychosom Res, 1996; 40(1): 7-10

Ford, CV, The Somatizing Disorders. New York, NY: Elsvier Science Publishing Co.,
1983

Ford, CV, “Illness as a Lifestyle. The Role of Somatization in Medical Practice”. Spine,
1992 Oct; 17(10 Suppl): 8338-43

Ford, CV, “Dimensions of Somatization and Hypochondriasis”. Neurol Clin, 1995 May;
13(2): 241-53

Goldberg, DP, Bridges, K, “Somatic Presentations of Psychiatric Illness in Primary Care
Setting”. J Psychosom Res, 1988; 32(2): 137-44

Golding, JM, Smith, GR Jr., Kashner, TM, “Does Somatization Disorder Occur in
Men?”. Arch Gen Psychiatry, 1991 Mar; 48: 231-5

Hanley, JA, McNeil, BJ, “The Meaning and Use of the Area under a Receiver Operator
Characteristic (ROC) Curve”. Radiology, 1982 Apr; 143(1): 29-36

Katon, W, Kleinman, A, Rosen, G, “Depression and Somatization: A Review, Part I”.
Am J Med, Jan 1982; 72: 127-35

Katon, W, Kleinman, A, Rosen, G, “Depression and Somatization: A Review, Part 11”.
Am J Med, Jan 1982; 72: 241-7

Katon, W, Lin, B, Von Korff, M, et al., “Somatization: A Spectrum of Severity”, Am J
Psychiatg, 1991 Jan; 148(1): 34-40

61

Kellner, R, “Functional Somatic Symptoms and Hypochondriasis”. Arch Gen Psychiatry,
Aug 1985; 42: 821-33

Kellner, R, “Psychosomatic Syndromes, Somatization and Somatoform Disorders”.
Psychother Psychom, 1994; 61 (1 -2): 4-24

Kellner, R, “Somatization. Theories and Research”. J Nerv Ment Dis, 1990 Mar; 178(3):
150-60

Kirmayer, LJ, Robbins, JM, “Three Forms of Somatization in Primary Care: Prevalence,
Co-occurrence and Sociodemographic Characteristics”. J Nerv Ment Dis, 1991 Nov;
179(11): 647-55

Kirmayer, LJ, Robbins, JM, “Patients Who Somatize in Primary Care: A Longitudinal
Study of Cognitive and Social Characteristics”. Psychol Med, 1996 Sept; 26 (5): 937-51

Kroenke, K, Spitzer, RL, deGruy, FV, et al., “Multisomatoform Disorder. An Alternative
to Undifferentiated Somatoform Disorder for the Somatizing Patient in Primary Care”.

Arch Gen Psychiatry. Apr 1997; 54: 352-8

Lipowski, ZJ, “Somatization: The Concept and Its Clinical Application”. Am J
Psychiatpy, Nov 1988; 145: 1358-68

Lolas, F, “Attention, Meaning and Somatization: A Psychophysiological View”.
Psychopathology. 1991; 24(3): 147—50

Mayou, R, “Somatization”. Psychother Psychom, 1993; 59(2): 69-83

Meeuwesen, L, Huyse, FJ, Meiland, FJM, et al., “Psychiatric Consultations in Medical
Outpatients With Abdominal Pain: Patient and Physician Effects”. Int J Psychgtry Med.
1994; 24(4): 339-56

Monson, RA, Smith, GR Jr., “Current Concepts in Primary Care. Somatization Disorder
in Primary Care”. N Engl J Med, June 1983; 308(24): 1464-5

Pathak, D, “Multiple Logistic Regression Analysis”. Class Notes HM826 Research
Methods in Epidemiology: Michigan State University, Fall Semester 1996

Portegijs, PJM, van der Horst, FG, Proot, IM, et al., “Somatization in Frequent Attenders
of General Practice”. Soc Psychiatry Psychiatkr Epidemiol, 1996 Jan; 31(1): 29-37

Rief, W, Heuser, J, Mayrhuber, E, et al., “The Classification of Multiple Somatoform
Disorders”. J Nerv Ment Dis. 1999 Nov; 184(11): 680-7

62

 

 

Robbins, JM, Kirmayer, LJ, “Attributions of Common Somatic Symptoms”. Psychol
Med, 1991 Nov; 21(4): 1029-45

Roberts, SJ, “Psychosocial Assessment of the High Utilizer of Occupational Health
Services”. AAHONJ, 1990 Feb; 38(2): 62-6

Roberts, SJ, “Somatization in Primary Care. The Common Presentation of Psychosocial
Problems Through Physical Complaints”. Nurse Pract, 1994 May; 19(5): 47, 50-6

Robins, LN, Helzer, JE, Weissman, MM, et al., “Lifetime Prevalence of Specific
Psychiatric Disorders in Three Sites”. Arch Gen Psychiatry, Oct 1984; 41: 949-58

Robins, LN, Helzer, JE, Orvaschel, H, et al., “The Diagnostic Interview Schedule”. In:
Eaton, WW, Kessler, LG (editors), Epidemiological Field Methods in Psychiatry. The "
NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, 1985

 

Robins, LN, Wing, J, Ulrich Wittchen, H, et al., “The Composite International Diagnostic
Interview. An Epidemiologic Instrument Suitable for Use in Conjunction With Different
Diagnostic Systems and in Different Cultures”. Arch Gen Psychiatry.

Dec 1988; 45:1069-77

Rosen, G, Kleinman, A, Katon, W, “Somatization in Family Practice: A Biopsychosocial
Approach”. J Fam Pract, 1982; 14(3): 493-502

Rost, KM, Smith, GR Jr., “Improving the Effectiveness of Routine Care for
Somatization”. J Psychosom Res, 1997; 43(5): 463-5

Salmon, P, May, CR, “Patients’ Influence on Doctors’ Behavior: A Case Study of Patient
Strategies in Somatization”. Int J Psychiatry Med, 1995; 25(4): 319-29

Schulzer, M, “Diagnostic Tests: A Statistical Review”. Muscle & Nerve, 1994 July;
17(7): 815-9

Simon, GE, VonKorff, M, “Somatization and Psychiatric Disorder in the NIMH
Epidemiologic Catchment Area Study”. Am J Psychiatry, 1991 Nov; 148(11): 1494-500

Smith, GR, Monson, RA, Ray, DC, “Patients With Multiple Unexplained Symptoms”.
Arch Intern Med, 1986 Jan; 146: 69-72

Smith, GR Jr., Post, K, Kashner, TM, “A Trial of the Effect of a Standardized
Psychiatric Consultation on Health Outcomes and Costs in Somatizing Patients”. Arch

Gen Psychiatry, 1995 Mar; 52(3): 238-43

Smith, RC, “A Clinical Approach to the Somatizing Patient”. J F am Pract, 1985;
21(4): 294-301

63

Tomasson, K, Kent, D, Coryell, W, “Comparison of Four Diagnositc Systems for the
Diagnosis of Somatization Disorder”. Acta Psychiatr Scand, 1993 Nov; 88(5): 311-5

Verhaak, PFM, Tijhuis, MAR, “The Somatizing Patient in General Practice”. Ir_rt_J
Psychiatry Med, 1994; 24(2): 157-77

Yutzy, SH, Pribor, EF, Cloninger, CR, et al., “Reconsidering the Criteria for
Somatization Disorder”. Hosp Communig Psychiatm, 1992 Nov; 43(11): 1075-6, 1149

 

GENERAL REFERENCES

American Psychiatric Association, Diagpostic and Statistical Mapual of Mental
Disorders, Third Edition. Washington DC: American Psychiatric Association, 1980

American Psychiatric Association, Diagnostic and Statistical Manual of Mental
Disorders, Third Edition - Revised. Washington DC: American Psychiatric Association,
1987

Berkow, R (editor), The Merck Manual (16’h edition). Rahway, NJ: Merck Research
Labs, 1992

Cody, RB, Smith, JK, Applied Statistics and the SAS Programming Language(4th
edition). Upper Saddle River, NJ: Prentice-Hall, 1997

D’Agostino, RB, “Two Group Logistic Regression Analysis”. Class Notes MA684
Advanced Statistics: Boston University, Fall Semester 1991

de Blay, F, Zana, H, Offner, M, et al., “Receiver Operating Characteristic Analysis: A
Useful Method for a Comparison of the Clinical Relevance of Two in vitro IgE Tests”.
J Allergy Clin Immunol, 1993 Aug; 92(2): 255-63

Detrano, R, “Accuracy Curves: An Alternative Graphical Representation of Probability
Data”. J Clin Epidemiol, 1989; 42(10): 983-6

Dolin, L, “Somatization and Medicalization (Letter)”. JAMA, May 1996; 275(18): 1398

Eliastam, M, “Somatization and Medicalization (Letter)”. JAMA, May 1996;
275(18): 1398-9

Erdreich, LS, Lee, ET, “Use of Relative Operating Characteristic Analysis in
Epidemiology”. Am J Epidemiol, 1981; 114(5): 649-62

Fabian, RH, Langsjoen, HA, Kent, TA, “Somatization and Medicalization (Letter)”.
JAMA, May 1996; 275(18): 1398

Fedorkow, DM, Sand, PK, Retzky, SS, et al., “The Cotton Swab Test. Receiver-
Operating Characteristic Curves”. J Reprod Med, 1995 Jan; 40(1): 42-6

Guess, HA, Resseguie, LJ, Melton, LJ, et al., “Factors Predictive of Seizures Among
Intensive Care Unit Patients with Gram-Negative Infections”. Epilepsia, 1990 Sept-Oct;
31(5): 567-73

65

 

Hanley, JA, McNeil, BJ, “A Method of Comparing Areas under Receiver Operator
Characteristic Curves Derived from the Same Cases”. Radiology, 1983 Sept;
148(3): 839-43

Huguet, J, Castifieiras, MJ, Fuentes-Arderiu, X, “Diagnostic Accuracy Evaluation Using
ROC Curve Analysis”. Scand J Clin Lab Invest, 1993; 53:693-9

Jensen, AL, Poulsen, J SD, “Evaluation of Diagnostic Tests Using Relative Operating

Characteristic (ROC) Curves and the Differential Positive Rate. An Example Using the
Total Serum Bile Acid Concentration and the Alanine Aminotransferase Activity in the
Diagnosis of Canine Hepatobiliary Diseases”. J Vet Med, 1992; 39(9): 656-68 1'

Katz, D, Foxman, B, “How Well Do Prediction Equations Predict? Using Receiver
Operating Characteristic Curves and Accuracy Curves to Compare Validity and
Generalizability”. Epidemiology, 1993; 4(4): 319-26

 

Kleinbaum, DG, Kupper, LL, Morgenstem, H, Epidemiologic Research. New York, NY:
Van Nostrand Reinhold, 1982

Kleinbaum, DG, Kupper, LL, Muller, KE, Applied Regression Analysis and Other
Multivariable Methods. Boston, MA: PWS-KENT Publishing Company, 1988

Lee, W-C, Hsiao, CK, “Alternative Summary Indices for the Receiver Operating
Characteristic Curve”. Epidemiology, 1996; 7(6): 605-11

Leibovici, L, Alpert, G, Laor, A, et al. , “A Clinical Model For Diagnosis of Urinary Tract
Infection in Young Women”. Arch Intern Med, 1989 Sept; 149: 2048-50

Leibovici, L, Greenshtain, S, Cohen, O, et al., “Bacteremia in Febrile Patients”. Arch
Intern Med, 1991 Sept; 151: 1801-6

Linnet, K, Brandt, E, “Assessing Diagnostic Tests Once and Optimal Cutoff Point Has
Been Selected”. Clin Chem, 1986; 32(7): 1341-6

Loeb, S (editor), Professional Guide to Diseafis (4th edition). Springhouse, PA:
Springhouse Corp, 1992

Plebani, M, Borghesan, F, Basso, D, et al., “Receiver-operating Characteristic (ROC)
Curves: A Fundamental Tool for Improving the Clinical Usefulness of in vitro IgE
Tests”. Allergy, 1996 June; 51(6): 407-11

Rodin, GM, “Somatization: A Perspective from Self Psychology”. J Am Acad
Psychoanal, 1991 Fall; 19(3): 367-84

66

Rosner, B, Fundamentals of Biostatistics (4th edition). Belmont, CA: Wadsworth
Publishing Company, 1995

Saxe, GN, Chinman, G, Berkowitz, R, et al., “Somatization in Patients with Dissociative
Disorders”. Am J Psychiatry, 1994; 151(9): 1329-34

Schuchter, L; Schultz, DJ, Synnestvedt, M, et al. “A Prognostic Model for Predicting
lO-Year Survival in Patients with Primary Melanoma”. Ann Intern Med, Sept 1996;
125: 369-75

Schultz, EK, “Multivariate Receiver-Operating Characteristic Curve Analysis: Prostate
Cancer Screening as an Example”. Clin Chem, 1995; 41(8B): 1248-55

Selvin, S, Statistical An_alysis of Epidemiologc Data (2"d edition). New York, NY:
Oxford University Press, 1996

Sensky, T, “Somatization: Syndromes or Processes?”. Psychother Psychom, 1994;
61(1-2): 1-3

Simel, DL, Samsa, GP, Matchar, DB, “Likelihood Ratios with Confidence: Sample Size
Estimation for Diagnostic Test Studies”. J Clin Epidemiol, 1991; 44(8): 763-70

Smith, PJ, Thompson, TJ, Engelgau, MM, et al., “A Generalized Linear Model for
Analyzing Receiver Operating Characteristic Curves”. Stat Med, 1996; 15: 323-33

Starlanyl, D, Copeland, ME, F ibromyalgizy & Chronic Myofascial Pain Syndrome. A
Survivors Manual. Oakland, CA: New Harbinger Publications, 1996

Swets, JA, “ROC Analysis Applied to the Evaluation of Medical Imaging Techniques”.
Invest Radiol, Mar-Apr 1979; 14(2): 109-21

Swets, JA, “Measuring the Accuracy of Diagnostic Systems”. Science, 1988 June 3;
240: 1285-93

Tomasson, K, Kent, D, Coryell, W, “Somatization and Conversion Disorders:
Comorbidity and Demographics at Presentation”. Acta Psychiatr Scand, 1991 Sept;
84(3): 288-93

von Scheele, C, Nordgren, L, Kempi, V, et al., “A Study of So-Called Hypochondriasis”.

Psychother Psychosom; 1990; 54(1): 50-6

Walker, LS, Garber, J, Greene, J W, “Somatization Symptoms in Pediatric Abdominal
Pain Patients: Relation to Chronicity of Abdominal Pain and Parent Somatization”.
J Abnorrn Child Psychol, 1991 Aug; 19(4); 379-94

67

 

 

Wasson, JH, Sox, HC, Neff, RK, et al., “Clinical Prediction Rules. Applications and
Methodological Standards”. N Engl J Med, Sept 1985; 313(13): 793-9

Wickramasekera, IE, “Somatization. Concepts, Data, and Predictions from the High Risk
Model of Threat Perception”. J Nerv Ment Dis, 1995 Jan; 183(1): 15-23

Wright, AF, “A Study of the Presentation of Somatic Symptoms in General Practice by
Patients with Psychiatric Disturbance”. Br J Gen Pract, 1990 Nov; 40(340): 459-63

68

 

"11111111111111.1111111’