WNWWWII“!WWI(IWWI}Millfllhlllllll

    

OD
NCO

    

"—1
3:8"
.cn

This is to certify that the
dissertation entitled

FACTORS ASSOCIATED WITH PATIENT ATTRITION:
FINDINGS FROM 'IWO CLINICAL TRIALS OF COGNITIVE
BEHAVIORAL INTERVENTIONS

presented by

Azfar-e-Alam Siddiqi

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Epidemiology

 

 

géflfl/flA/wgfl/L

Major Professor’s Signature

IR-il’ 100k

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

 

LIBRARY
Michigan State
University

 

 

 

on-n-n--o---n-u-o-n-o---u-o--c—.— ..--o-n-oon-n-c-a-Inc-o-.--on-o-a-.-0-0-.-.----c---o--o-o-n-0-.-o-o-u-o-o-a-n-u—

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/CIRC/DateDue.indd-p.1

 

FACTORS ASSOCIATED WITH PATIENT ATTRITION: FINDINGS FROM TWO
CLINICAL TRIALS OF COGNITIVE BEHAVIORAL INTERVENTIONS

By

Azfar-e-Alam Siddiqi

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Epidemiology

2006

ABSTRACT

FACTORS ASSOCIATED WITH PATIENT ATTRITION: FINDINGS FROM TWO
CLINICAL TRIALS OF COGNITIVE BEHAVIORAL INTERVENTIONS

By
Azfar-e-Alam Siddiqi
Randomized clinical trials and prospective observational studies are

quality epidemiological study designs for evaluating treatments and for studying
the outcomes of an exposure. Patent attrition from these studies is a potential
threat to their validity. If potential drop-outs can be identified early in a study,
then researchers can design strategies specifically targeting such patients to at
least minimize their attrition. This research was carried out on data from two
clinical trials of cognitive behavioral interventions to identify factors associated

with patient who dropped out prematurely.

In the first study 719 patients, of acute coronary syndrome (ACS),
admitted to one of the five participating mid-Michigan hospitals were recruited.
Following discharge, patients were approached by telephone at one-month
baseline (n=525), 3 months (n=440), and 8 months (n=388). Following baseline
interview patients were randomized to either a standard treatment arm or to
standard treatment plus a cognitive intervention arm. Attrition of patients,
following baseline interview (failure to complete the next scheduled interview)
was modeled using a Generalized Estimating Equation (GEE) model, on a data

set with multiple records per patient.

The second data set is a group of cancer patients undergoing
chemotherapy, at one of the six different hospitals in Michigan, Indiana and Ohio.
Two hundred and sixty three patients consented to participate. Data collection
was done at baseline (n=235), 10 weeks (n=178), 20 weeks (n=139), and 32
weeks (n: 125). Attrition following baseline was modeled in the same manner as

for the cardiac patients.

Age of the patient, smoking habits, depression and abstinence from
alcohol were found to have an association with attrition in cardiac patients.
Whereas among cancer patients, cancer of lung, time spent in trial, depression

and education were found to significantly influence attrition.

Patient attrition from longitudinal studies is mostly a function of their
mental heath, social activity and support. Non-conformation with the expected
role of the patient given the intervention under study in the trial also influences

patients’ chances of dropping out prematurely.

ACKNOWLEDGEMENTS

Author would like to thank the following for their support:

Agency for Health Services Research AHRQ: R01 H810531

Translating Research: Patient Decision Support/Coaching (Dr. Holmes-Rovner,
PI)

National Cancer Institute Grant # R01 CA79280

Family Home Care for Cancer: A Community-Based Model
In Affiliation with the Walther Cancer Institute

Indianapolis, Indiana. (Dr. Given, Pl)

TABLE OF CONTENTS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

LIST OF TABLES .............................................. .. .............................. VI
LIST OF FIGURES : VII
LIST OF ABBREVIATIONS USED .. .. VIII
CHAPTER 1 ...................................................................................................................... 1
1 INTRODUCTION -- 1
1. 1 DEVELOPMENT OF MEDICAL THERAPEUTICS ............................................................... 5
1.2 A'ITRITION IN CLINICAL TRIALS ................................................................................. 15
1.3 LITERATURE ON A’I'I‘RITION ....................................................................................... 18
CHAPTER 2 - - 22
2 METHODS 22
2.1 ETHICAL CONCERNS .................................................................................................. 22
2.2 HARP TRIAL ............................................................................................................. 22
2.2.1 Subjects and settings for HARP study ................................................................... 23
2.2.2 HARP Intervention ................................................................................................ 24
2.2.3 Study Variables and data collection ....................................................................... 25
2.3 FAMILY HOME CARE FOR CANCER TRIAL ................................................................. 29
2.3.1 Subjects and Settings for FHCC Trial ................................................................... 29
2.3.2 The nursing intervention ........................................................................................ 30
2.3.3 Study variables ....................................................................................................... 31
2.4 STATISTICAL ANALYSIS ............................................................................................. 34
CHAPTER 3 39
3 RESULTS ..... 39
3.1 HARP TRIAL ............................................................................................................. 39
3.2 FHCC TRIAL ........................................................................................................... 41
CHAPTER 4 ................................ ......................................... 44
4 DISCUSSION 44
4.1 CONCLUSION ............................................................................................................. 50
APPENDICES - - -67
REFERENCES - ........... 2.- ................... 70

 

LIST OF TABLES

Table 1: Comparison of baseline demographic characteristics of the intervention
and control groups of the HARP trial (categorical variables) .............................. 54

Table 2: Comparison of baseline demographic characteristics of the intervention
and control groups of HARP Trial (continuous variables) ................................... 55

Table 3. Factors associated with patient’s attrition based on univariate analysis
(HARP Trial) ....................................................................................................... 56

Table 4: Factors associated with post-baseline attrition in cardiac patient’s based
on the final Multivariate GEE model, (n = 525) ................................................... 58

Table 5: Comparison of baseline demographic characteristics of the intervention
and control groups in the FHCC trial (categorical variables) .............................. 59

Table 6: Comparison of baseline demographic characteristics of the intervention
and control groups in the FHCC trial (continuous variables) .............................. 61

Table 7: Factors associated with patient’s attrition based on univariate analysis
(FHCC Trial) ....................................................................................................... 62

Table 8: Factors associated with patient’s post-baseline attrition in the FHCC
trial based on the final Multivariate GEE model (n = 235). ................................. 64

vi

LIST OF FIGURES

Figure 1: Flow chart showing patient flow through various stages of cardiac

patients’ (HARP) Trial ......................................................................................... 65
Figure 2: Flow chart showing patient flow through various stages of cancer
patients’ (FHCC) study. ...................................................................................... 66

vii

ACS
AOR
ASI
CCI
CES-D
CI

EF
FHCC
GEE
HARP
IPPSR
IRB
MRC CFAS
MSU
QLMI
SD
SES
SF-36

UCRIHS

LIST OF ABBREVIATIONS USED

Acute Coronary Syndrome

Adjusted Odds ratio

Activity Status Index

Charlson Comorbidity Index

Centers for Epidemiologic Studies Depression scale
Confidence Interval

Left Ventricular Ejection Fraction (Left Ventricle)
Family Home Care for Cancer

Generalized Estimating Equation

Heart After-Hospital Recovery Planner

Institute for Public Policy and Social Research
Institutional Review Board

Medical Research Council Cognitive Function and Aging Study
Michigan State University

Quality of Life after Myocardial infarction

Standard Deviation

socioeconomic status

Short Form 36

University Committee on Research Involving Human Subjects

viii

Chapter 1
1 Introduction

In the hierarchy of Epidemiological studies clinical trials are considered the
‘gold standard’ of all study designs. When conducted carefully with close
attention to detail, with sound methodology and appropriate statistical analysis,
the findings from trials are accurate to the extent possible in an experimental
study, evaluating a new intervention against no, or standard intervention for the
problem in question. Clinical trials have a long history and have evolved over
centuries from crude ‘experiments’ of the past into the well-designed,
standardized trials of today. Despite being crude, these experiments
nevertheless laid the foundation for an epidemiologic study design that is now
considered the standard methodology for comparison of competing intervention

in the management of clinical pathologic conditions.

While the terms like ‘trial’ and ‘experiment’ were often used by the
researchers who conducted these trials, the term ‘clinical trial’ was not formally
defined until recent past. Meinert (1986) defined a clinical trial as “a planned
experiment designed to assess the efficacy of a treatment in humans by
comparing outcomes in a group of patients treated with a test treatment with
those observed in a comparable group of patients receiving a control treatment,
where patients in both groups are enrolled, treated, and followed over the same
time period” (1). Spilker (1991) defines clinical trials as “a subset of clinical
studies that evaluate investigational medicines in phase I, II and III, the clinical

Studies being a class of all scientific approaches to evaluate the effectiveness of

a treatment” (2). Piantadosi (1997) defines a clinical trial simply as “an
experiment testing medical treatment in human subjects” (3). The definition
provided by Spilker is narrow in the sense that it restricts the use of this term to
trials conducted during the stages of development of a new drug, as usually seen
in pharrnaco-epidemiologic studies often conducted by pharmaceutical
companies. Treatment or management of a disease however, often involves
modalities other than drugs that before becoming accepted as standard
treatment method need to be evaluated like any new drug. These other non-
medicinal modalities are evaluated for their efficacy in standard epidemiological
studies, namely clinical trials, just like new drugs. More recent definitions of
clinical trials that encompass drugs and also non-drug treatment modalities are
offered by the Dictionary of Epidemiology, and by National Institutes of Health
(NIH) at its official clinical trials related website, ‘clinicaltrialsgov’. The Dictionary
of Epidemiology defines clinical trials as “A research activity that involves the
administration of a test regimen to humans to evaluate its efficacy and safety...”
(4). The use of term ‘test regimen’ instead of drug allows for the application of
the term clinical trials to trials of all kinds not just drug trials. NIH and its clinical
trials website give the following definition of a clinical trial, “A clinical trial (also
clinical research) is a research study in human volunteers to answer specific
health questions”, it further states “carefully conducted clinical trials are the
fastest and safest way to find treatments that work in people and ways to
improve health. lnterventional trials determine whether experimental treatments

or new ways of using known therapies are safe and effective under controlled

environments” (5). The ‘treatment’ or ‘intervention’ under evaluation therefore
can be anything such as a vaccine, a diagnostic test, a behavioral intervention or

something as simple as education.

Depending on the main focus or objective of the trial, the NIH classifies
clinical trials into several different categories. Treatment trials test experimental
treatments, new combinations of drugs, or new approaches to surgery or
radiation therapy. Prevention trials look for better ways to prevent disease in
people who have never had the disease or to prevent a disease from returning.
These approaches may include medicines, vitamins, vaccines, minerals, or
lifestyle changes. Diagnostic trials are conducted to find better tests or
procedures for diagnosing a particular disease or condition. Screening tn'als test
the best way to detect certain diseases or health conditions. Quality of Life trials
(or Supportive Care trials) explore ways to improve comfort and the quality of life
for individuals with a chronic illness (6). What is common among all these
different kinds of trials is the ‘comparison’ of a test intervention (drug, test,
behavioral or educational intervention etc), with either no intervention or the
existing standard of treatment for the disease of interest. The purpose therefore
is to find, evaluate and later on implement new interventions that will improve

upon what exists as contemporary standard.

Clinical trials evaluating new drugs are conducted in different phases. The
trials at each phase have a different purpose and help scientists answer different
questions: In Phase I trials, researchers test an experimental drug or treatment in

a small group of healthy people usually no more than 20 to 80, for the first time to

evaluate its safety, determine a safe dosage range, and identify side effects. The
protocols for phase 1 studies are more flexible and less detailed than for the
subsequent phases. These studies also provide necessary information needed

for designing the scientifically phase II studies.

Phase II trials, are the first controlled clinical studies of the drug and
involve more subjects than phase I study but usually no more than a few hundred
(100-300 subjects). The primary objectives of phase II studies are to evaluate
the effectiveness of the drug based on clinical endpoints for a particular condition
and to determine the dosing ranges and dose/s for phase III studies. These trials
also provide information on common short-term side effects and associated risks.
Some experts further differentiate this phase into phase "A (studies to evaluate

dose) and phase 118 (studies to determine effectiveness of the drug).

In Phase III trials, the experimental study drug or treatment is given to
large groups of subjects several hundred to a few thousand to confirm its
effectiveness, monitor side effects, compare it to commonly used treatments, and
collect information that will allow the experimental drug or treatment to be used
safely. It also provides the basis for physician labeling of the drug. Phase III
trials are performed only after the preliminary evidence regarding the

effectiveness of the drug has been demonstrated.

The Phase IV trials further elucidate the incidence of adverse reactions
and determine the effect of the drug on morbidity and mortality (7-9). Clinical
trials involving non-medicinal interventions, particularly when given as an adjunct

to standard treatment are evaluated in settings similar to phase II and III trials.

The low risks (if any) posed by non-medicinal interventions such as behavioral
interventions, make it unnecessary to evaluate these interventions in phase I like

settings.

1. 1 Development of Medical Therapeutics

Since ancient times man has experimented with herbs and oils and the
like for the treatment of disease and illness. The development of ‘new’ oils,
poultices and concoctions rested upon the judgment and reckoning of the healer
and had little — if anything — to do with the comparison of one treatment against
another. The surviving records from ancient times particularly Egypt and
Babylon show that although there was an elaborate phannacopoeia the therapy
was irrational or ritualistic. Medicine in that era was closely allied to religion and
much of the treatment was aimed at the exorcism of hostile powers. This
exorcism was however, often combined with the use of poultices, purgatives and

enemas (10).

Modern medicine is said to have begun in Greece with Hippocrates the
father of modern medicine. The Hippocratic or Coan School that formed around
him was of enormous importance in separating medicine from superstition and
philosophic speculation. The studies of natural history of disease from Coan
School give rich accounts of diagnosis and prognosis, but there was less
success in therapeutics. While surgical treatment, particularly that of minor
problems was well developed; the medical therapeutics was dominated by a

priori theories. These theories however, were much less speculative than those

of the contemporary schools of thought. The Coan School provided the
foundations and placed medicine on a scientific plane based on objective
observation and critical deductive reasoning, and urged physicians to inquire into
the best treatment (11). There is however, no evidence to suggest that

experimental trials and judgment based on the results of treatment were used.

In the Middle Ages after the fall of Roman Empire, the scientific trend in
European medicine was arrested. Advances in medicine and therapeutics were
then made by Arabs for a brief period, as accounted in Zakaria Razi’s (860-932)
(better know in the west as Rhazes) book AI-Hawi (12). The first suggestions of
rules for testing drugs are found in the encyclopedic writings of Ibn-e-Sina’s (980-
1037) (better known In the west as Avicenna) ‘Qanoon Fi AI Tib’. He suggests
that in the trial of a remedy it should be given in its natural state upon
uncomplicated disease, that two opposed cases be observed, and that study be
made of the time of action and the reproducibility of the effects. He further
states, “The experimentation must be done with the human body, for testing a

drug on a lion or a horse might not prove anything about its effect on man” (13).

Arguably the oldest recorded account of an event that comes close to the
definition of a “comparative study” comes from Old Testament in the Book of

Daniel (14).

1. In the third year of the reign of Jeohiakim king of Judah came

Nebuchadnezzar king of Babylon unto Jerusalem, and besieged it...

3. And the king spoke unto Ashpenaz his chief officer, that he should bring

in certain of the children of Israel, and of the seed royal, and of the nobles...

5. And the king appointed for them a daily portion of the king’s food. And

of the wine which he drank that they should be nourished for three years...

8. But Daniel proposed in his heart that he would not defile himself with
the king’s food, nor with the wine which he drank; therefore he requested of the

chiefs of the officers that he might not defile himself...

10. And, the chief of the officers said unto Daniel: “I fear my Lord the king,
who has appointed your food and drink; for why should he see your faces sad in

comparison with the youths of your own age?”...
11. Then Daniel said to the steward...

12. Try they servants, l beseech thee, ten days; and let them give us

pulse to eat and water to drink...

13. Then let our countenances be looked upon before thee, and the

countenances of the youth that eat of the king’s food...

14. So, he hearkened unto them and tried them into in this manner, and

tried them for ten days...

15. And at the end of ten days their countenances appeared fairer and

they were latter in the flesh, than all the youths that did eat of the king’s food.

While the events in these verses fall short of qualifying as a comparative
study as it was not an intentional administration of two different diets for the
purpose of studying their outcomes, it does point towards one important aspect
of clinical trials - gaining knowledge of the effects of exposure by observing the

outcome. Another example of acquiring information through observation can be

seen in the first expedition of the East India Company to India in the 17th century.
Four ships of the company were part of this expedition and on only one of them,
that of General James Lancaster, was lemon juice provided to the crew. Of the
four ships only this ship remained almost totally free of scurvy, while the crew of
the other three ships was badly affected by the disease. The medical men of the
time did not take up this lead, but the East India Company supplied lemon juice
to all its ships in subsequent expeditions. It was not until more than a century
later that the treatment of scurvy with fruits containing vitamin C was evaluated in

a comparative trial.

The earliest documented comparative study comprising two or more
treatments of a disease was performed by James Lind (1716-1794). Lind was
appalled by the death due to scurvy of three quarters of the crew, during Admiral
(then commodore) Anson’s circumnavigation of the world from 1740-1744. Lind
planned a comparative trial of the most promising ‘cures’ of scurvy of the time

(15). Lind writes;

“On 20‘" of May, 1747, I took twelve patients in the scurvy, on board the
Salisbury at sea. Their cases were as similar as I could have them. They all in
general had putrid gums, the spots and Iassitude, with weakness of their knees.
They lay together in one place, being a proper apartment for the sick in the fore-
hold; and had one diet in common to all, viz., water-gruel sweetened with sugar
in the morning; fresh mutton-broth often times for dinner; at other times puddings,
boiled biscuit with sugar etc.; and for supper, barley and raisins, rice and

Currants, sago and wine, or the like. Two of these were ordered each a quart of

cider a day. Two others took twenty-five gutts of elixir vitriol three times a day,
upon empty stomach; using a gargle strongly acidulated with for their mouths.
Two others took spoonfuls of vinegar three times a day, upon an empty stomach;
having their gruels and their other food well acidulated with it, as also the gargle
for their mouth. The two worst patients, with the tendons in the ham rigid (a
symptom none other had), were put under a course of seawater. Of this, they
drank half a pint every day, and sometimes more or less as it operated, by way of
a gentle physio. Two others had each two oranges and one lemon given them
everyday. These they ate with greediness, at different times, upon empty
stomach. They continued but six days under this course, having consumed the
quantity that could be spared. The two remaining patients took the bigness of a
nutmeg three times a day, of an electuary recommended by a hospital surgeon,
made of garlic, mustard-seed, rad. raphan, balsam of Peru, and gum myrrh;
using common drink, barley water well acidulated with tamarinds; by a decoction
of which. With the addition of cremor tartar, they were gently purged three or four

time during the course.

The consequence was, that the most sudden and visible effects were
perceived from the use of the oranges and lemons; one of those who had taken
them, being at the end of six days fit for duty. The spots were not indeed at that
time quite off his body, nor his gums sound; but without any other medicine, than
a gargarism of elixir vitriol. He became quite healthy before we came in

Plymouth, which was on the 16th of June. The other was the best recovered of

any in his condition; and being now seemed pretty well, was appointed nurse to

the rest of the sick.”

Lind’s experiment was indeed a conscious attempt to compare several
treatments. His allusion to “..cases were as similar. .”, suggests that he was

aware of the dangers of comparison between non similar cases.

While Lind’s experiment was the trials of treatments, the earliest
documented account of a trial of preventive nature comes from the 18th century.
Eighteenth century Ottomans practiced a form of vaccination against small pox,
referred to as variolation. It involved taking material such as pus from pustules in
patients with mild disease and introducing this material into healthy individuals
through nose or skin. This practice was based on observation that patients who
survived smallpox remained immune to it for the rest of their life. An English
aristocrat, Lady Mary Montague (a smallpox victim), the wife of the English
ambassador to the Ottoman government became interested in the Turks’
variolation process and is said to be responsible for bringing variolation to
England. She and the English embassy surgeon Charles Maitland persuaded
King George I to allow a ‘trial’ of variolation on six convicts in 1721. All six
convicts survived and were released as promised (16). This trial was later
criticized as it was unclear which of the subjects had previously suffered from
disease, and the trial would have tested the safety of variolation (or inoculation)

rather than its effectiveness (17).

Another example of the use of comparative statistics is the report by

Robert Robertson in 1776. While serving as ship surgeon on Juno, of the Royal

10

Navy, Robertson ran out of the bark that was then commonly employed for the

treatment of continuous fever (malaria), and was forced to try other methods of
treatment, thus leading to an un-intentional comparative study. Robertson then
compared the case fatality percentages in the two periods (with and without

bark), and reported these rates in his report (10, 18).

Such comparative studies continued into the nineteenth century. The
comparison of statistics such as fatality percentages became more and more
refined, in parallel with the developments in statistics. The hygienic and sanitary
reform movement of the 19th century, utilized vital statistics as a justification for
proposed reforms, as they provided a quantitative measure of the community’s
health status. The work of Louis René Villermé, who compared the mortality
experience of different districts in Paris, and that of William Farr are excellent

examples (19); Wall, 1974 #288}.

One of the best examples of strength of observational epidemiology with
some degree of random assignment (though unintentional) comes from the work
of John Snow, during the London’s cholera epidemics of 1849, and 1853 (20). A
change in the water supply of London, in the period between epidemics gave
Snow the opportunity to do what he referred to as the ‘grand experiment’. Of the
three companies that supplied the city with its drinking water at that time, one
(Lambeth) moved its source of water intake to upstream Thames, from an area
that was much less polluted than the uptake areas of other companies. While
Snow’s contemporaries favored miasmatic theory, as the cause of cholera, this

Unique arrangement of water supply, enabled Snow to demonstrate that the

11

households differed considerably in terms of mortality (and incidence) of cholera
also differed by their source of drinking water. Since these households were in
the same neighborhood, even in the same street, the difference in water source
as an explanation of very different observed mortality rates made good sense.
These differences couldn’t be explained by the miasmatic theory. The
‘randomness’ with which households differed in their source of water supply, and
the ability to compare the disease pattern before and after the change in water

supply are important of modern day clinical trials.

Perhaps the most sophisticated clinical trial of the preventive type in the
nineteenth century was conducted by Ignaz Sammelweis. Sammelweis
observed that mortality in the First Division of Lying-in hospital Vienna, served by
physicians and medical students, was much higher than the Second Division,
served by midwives, he also noted that because of the stress on Ieaming
Pathology medical students and physicians were in much frequent contact with
cadavers, whereas the midwives had none. Sammelweis hypothesized that
“..cadaveric particles clinging on to the hands are not entirely removed by the
ordinary method of washing hands with soap,..”. In May 1847 he began the use
of ‘chlorina liquida’. Each student in the First Division was required to wash his
hands with this before examining a patient. This resulted in an immediate drop in
the mortality rates in the First Division to a level similar to Second Division.
Although not randomized and without a placebo, the trial did show a strong

preventive effect of using antiseptics in preventing infection (21 ).

12

Similar statistical comparisons can be found in the work of researchers in
the 19th century. In 1860 Joseph Lister used comparative statistics to evaluate
the effects of antiseptics on the mortality from amputations. Using hospital
records, he compared two years of data prior to the ‘antiseptic period’ with the
three years during the antiseptic use. Despite small numbers, which he
acknowledged, he was able to show much lower mortality following amputation in
the antiseptic period as compared to previous years (22). While statistical
comparisons mentioned above made use of a group without ‘intervention’ the use
of the term ‘control’ was probably first used by Hankin in 1890, when he used the
term ‘control-mice’ to describe untreated mice in a series of experiments of

tetanus immunization.

As the researchers continued to conduct ‘trials’ the strength of careful
observation and information gained from it became apparent to the scientific
world. This continuous evolutionary process resulted in refinement of design and
conduct as the trial moved from the primitive comparisons of the 19’h century into
the modern day trials of the 21S" century. Advancement in statistics also played a

major role in this process with more and more statistical tools applied to the later

trials.

Although researchers had alluded to the concept of randomization or
random assignment even earlier, the first clear application of the concept was
seen in agricultural experiments designed by RA Fisher, in 1926. Working at
Rothamsted Experimental Station in England, he used random allocation of

“agricultural fields to different treatments. The first formal random assignment of

13

patients in a clinical trial was done by Amberson, in his trials of Sanocrysin (a
gold compound) for the treatment of pulmonary tuberculosis. Amberson is also
credited for introducing the concept of a double blind design for the same trial.
This was achieved by giving intravenous injections of distilled water to the
controls. Diehl et al. reported the first use of a ‘saline placebo’ in their trials of

cold vaccine trials, in 1938.

While researchers perfected their study designs by introducing new
techniques, lawmakers realizing the potential benefits of a well-designed and
conducted trial, started enacting laws and regulation to ensure quality control
(and ethical acceptability) of the clinical trials. Starting with Pure Food and Drug
Act in 1906 the first regulations were laid out for regulation of clinical trials. A

brief timeline of such regulations are given in the appendix 1.

The historical account and the evolution of the regulations governing
clinical trials, as given in appendix 1 highlight the fact that clinical trials have not
only evolved over time, from primitive comparisons to sophisticated randomized
trial of today, but that their evolution is a continuous ongoing process. Being very
different from research in disciplines like Physics and Chemistry, epidemiological
research - in most cases - deals with human subjects and thus has limits on the
extent to which one can probe into the subjects and subject matter. These
restrictions imply that ‘perfection’ in study designs and methodology is virtually
unattainable and that there will always be room for improvement even in the best

conceived and conducted trials.

14

1.2 Attrition in Clinical trials

One of the limitations of clinical trials, and of all follow-up study designs
involving humans, is the subject attrition or the loss to follow-up. This loss of
participants represents a potentially large threat to validity of the trial results.
This loss can be completely random, where dropout subjects cannot be assumed
to belong to a group based on any of their characteristics, i.e. the dropout is
independent of the patients’ assigned treatment, given observed characteristics.
If this is indeed the case then patient loss of such kind will not result in drawing of
invalid conclusions (other than by chance), however the reduction in available
sample for analysis will reduce the power of the study to detect a significant
effect when there was indeed an effect of the intervention on the outcome.
Potential losses and drawback of non-significant findings due to lack of statistical
power are several and include; delay in obtaining convincing evidence of the
effect of an intervention on an outcome, as well as wasting of valuable, and
increasingly scarce, resources. This reduction in the power of the study, or
increase in the probability of the type II error, can occur despite the traditional ‘10
— 20% inflation’ of the calculated sample size, if a substantial percentage of the

sample is lost.

The non-random (or disease directed) loss on the other hand can result in
serious bias and inaccurate conclusions in addition to the loss of statistical
power. In a clinical trial, for example, if exposed patients are indeed more likely
to develop disease and at the same time, with the development of disease, more

likely to drop out of the study, the final sample may be under-representative of

15

exposed cases, and thus may result in failure to recognize the exposure’s
association with disease, or in an under-estimation of the effect of exposure on
disease at the minimum. This can work in opposite direction as well resulting in
an over-estimation of the magnitude of association between exposure and
disease. This missing data resulting from patients attrition poses particular
serious challenges to the accurate assessment of all clinical trial data, including

cancer-related QOL (23-25).

Strictly speaking refusal to participate in a trial despite eligibility cannot be
considered attrition, but this non-consent tends to impact the study results as
does attrition after randomization. Both scenarios represent subject loss and can
potentially have an impact on the validity of trial’s results. Non-consent can lead
to the ‘healthy participant effect’, in which participants report more healthy life
styles or lower risk factors than the non-participants, as underscored by several
studies (26-28). It is also generally recognized that patient attrition after
enrollment can potentially introduce bias in the study and make the results
unreliable (29). Biases originating from differential patterns of attrition from
different study groups after enrollment can result in, under or over estimation of
the magnitude of the effect of intervention or exposure, on the outcome of

interest.

Research to identify factors that distinguish study subjects who participate
in the study fill its end, from those who do not participate till the end, (dropouts)

can play a role in improving the quality of trials. Addressing this research

16

question will require an analysis comparing study completers vs. non-completers

of a trial.

Comparison of study completers with those who consent to participate but
do not complete the study may be very difficult, if not impossible. This will
depend on the point in time when the attrition occurred. If some patients dropped
out soon after consent was obtained but before any formal data collection, like
baseline or intake interview was done, the unavailability of the extensive relevant
data will make meaningful comparisons very difficult. Some information such as
that may have been collected at the time of recruitment, or that may have been
collected from patients’ medical chart records can be used to make comparisons
of study completers and those early dropouts. On the other hand when attrition
occurs at a later point in time, i.e. after patients have at least completed a
baseline or intake interview, the comparison of completers vs. non-completers of
the study is relatively easy. Such analysis can prove extremely advantageous in
revealing factors that have an association with a high probability of dropping out
of the study, and their identification can help researchers in better understanding
the extent of potential bias introduced in a given study. Additionally a before
hand knowledge of such factors when designing a study can help researchers
build checks into the study design, targeting patients at high risk of dropping out
to at least minimize their attrition. Researchers of trials and other follow-up
studies have traditionally limited their assessment of the attrition to a simple
univariate analysis of the comparison of ‘lost’ vs. ‘retained’ subjects in an attempt

. to show that statistically the completers of the study were not significantly

17

different from the subjects lost, in terms of their demographic and sometimes

other characteristics (30).

1.3 Literature on Attrition

Though the problem of patient attrition from clinical trials and its potential
impact on the study results were discussed as early as 1962, by Lasky (31), and
more recently by Leon (32), there is relatively scarce literature on
research/analysis done specifically to look for the independent effect of
participant characteristics on their probability of attrition from the study.
Additionally, the research that has been carried out has identified a somewhat
variable set of predictors of attrition, with very few factors being reported with any
consistency. Socio-demographic factors such as old age, being male, low
socioeconomic status (SES), low education, low social support and belonging to
minority groups have been found to be associated with patients’ attrition (33-36).
Physical and mental factors associated with attrition include, functional
limitations, poor self-reported health, unstable health status, cognitive impairment
and major disease (36-38). In addition, other studies have reported females,
smokers, and depressed individuals to be more likely to drop out than others (39-
41). Younger age, milder disease symptoms and fewer coexisting chronic
conditions have also been reported to be associated with patients’ attrition (42-
45). Patient depression is one factor that is perhaps most consistently reported
factor to be associated with attrition (44, 46, 47). An early work published by
Downing et al, found that younger and gainfully employed patients, those with a

lower socio-economic status and social stability were more likely to be non-

18

completers of placebo-controlled trials (48). Demographic characteristics such
as belonging to a minority group, not being married and having low education
(less than 12 years) were reported by Given et al (49). The wide variation of
characteristics reported in literature, suggests that factors influencing a person’s
chances of attrition may vary by the type of disease (if any) the participants may
have, and point to the need for further exploration of factors affecting patients’
attrition. While there can be several sources of these variations, it is intuitive that
source population would be one major cause of this variation. The nature of
disease (it any), that the patients are suffering from and/or kind of intervention
they may be undergoing may, in itself be a cause of attrition. In a study by Davis
et al, the researchers looked at the characteristics of two groups of caregivers of
elderly patients, suffering from Alzheimer’s and Parkinson’s diseases, who
dropped out of intervention trials. They showed that the nature of disease
influenced the caregivers’ attrition (50). An interesting research into factors
associated with attrition comes from Medical Research Council Cognitive
Function and Aging Study (MRC CFAS). MRC CFAS is a population-based
study of health in older population, carried out in England. The researchers in
their recent article (51 ), report that factors associated with patient attrition over a
ten year period were fairly consistent, with, patients of low socio-economic
status, lower education and older patients with poor cognitive function to be more
likely to drop out before the formal termination of study. Snow et al, studying
attrition in a clinical trial of lung cancer patients, report that patients who were

younger, unmarried and smoked cigarette were more likely to drop out. They

19

also looked if the mechanism of recruitment of study participants had an impact
on attrition or not and found that patients recruited via mass mailing, were more
likely to stay in the study till its and than those recruited through worksite or
referral methods (52). An additional source of attrition from behavioral
interventions trials is the random assignment of patient to an intervention other
than the patients’ choice. The resulting disinterest is an obvious cause of attrition
(53). The contrasting associations of age with attrition, reported in literature, may
very well be a function of the age range of participants of a given study. It is
plausible that the two ends of the age spectrum, very young and very old may be
the ones most or least vulnerable to attrition, with mid-age range representing
either a more stable or labile group in terms of continuous participation. Thus
depending on the age range of participants in any given trial, the researchers
may find, and report, older or younger patients to be more likely to dropout. All
this points to the need of more research to gain a better understanding of factors

associated with patients’ attrition.

The research reported here was conducted in two very different
populations, enrolled in two different non-drug clinical trials, one of relatively
younger, cardiac patients, and the other of cancer patients undergoing
chemotherapy. Data from both trials was analyzed separately, to identify disease
and intervention specific factors associated with patients’ attrition. Both trials
collected data via telephone interviews, at baseline (soon after recruitment) and
on at least two more occasions following baseline. This analysis was carried out

on patients who had completed at least the first (baseline) interview.

20

The objective of this research was to identify patient and disease related
factors that influence participants’ chances of dropping out from clinical trials of

cognitive behavioral interventions.

21

Chapter 2
2 Methods

The data sources for this analysis are two multi-site clinical trials, “Heart
After-Hospital Recovery Planner” (HARP), and “Family Home Care for Cancer
(FHCC) — A Community Based Model”. HARP is a clinical trial involving cardiac
patients, whereas FHCC is a cancer care intervention involving cancer patients
and their primary family caregivers. Both studies were conducted by researchers
at Michigan State University (MSU) with data collected from multiple sites and

were person-level randomized non-drug intervention trials.

2. 1 Ethical Concerns

Since the data in the original trials was collected via interviews on human
subjects, both trials were conducted after Institutional Review Board (IRB)
approval from MSU’s University Committee on Research Involving Human
Subjects (UCRIHS). Separate IRB approvals were also obtained for this

research.

2.2 HARP Trial

Patients recruited for the HARP intervention, and randomized into the
intervention arm, received a follow-up telephone coaching program during the
first three months following discharge from the hospital. The institutional review
boards of Michigan State University and each of the five participating hospitals

approved the study. HARP study was designed to target multiple risk behaviors

22

of cardiac patients, with an aim of improving overall quality of life of cardiac

patients and mortality following discharge from hospital.

2.2.1 Subjects and settings for HARP study

Patient eligibility criteria included: admission to one of five participating
study hospitals in mid-Michigan, a documented serum Troponin I level greater
than the upper limits of normal observed in each hospital, a working diagnosis of
acute coronary syndrome (ACS), and capability of being interviewed and/or
participating in telephone health behavior counseling. Exclusion criteria included:
discharge to any non-home setting, possession of any significant
mentaVcognitive impairments, lack of a home telephone, or non-English

speaking.

Between January 14, 2002 and April 13, 2003, trained nurse recruiters
approached ACS patients during their hospitalization, providing information on
study participation and attempted to obtain their consent to participate. Patients
were recruited from five hospitals located in two communities in adjacent mid-
Michigan counties. Both counties shared similar characteristics. Each had one
major city surrounded by suburbs and outlying rural and farming areas. Both
communities have diverse populations with a minority population slightly higher
than the State average (20% versus 14.5%), high unemployment rates (above
8%), and with an industriananufacturing economic base, dominated by a single
automobile manufacturing corporation as the largest employer. A mean number
(of 2.29 (SD 1.82) contacts were required with elevated Troponin patients to

determine their actual eligibility.

23

Of the 1985 eligible patients, 719 consented to participate in the study.
Shortly after hospital discharge consenting patients were called by trained survey
researchers from MSU’s Institute for Public Policy and Social Research (IPPSR),
for a 30-40 minute interview to collect the first wave (Baseline) data. IPPSR
researchers made up to three attempts to contact and interview the patients.
Patients not contacted in the three attempts or who explicitly refused interview

were not contacted any further.

Five hundred and twenty five (73%) of the 719 consenting patients were
successfully contacted and interviewed at baseline. Telephone calls were
repeated at approximately 3 months and 8 months following discharge, to collect
follow-up data. Of the 525 patients participating in baseline interview, 440 (84%)
were successfully contacted and interviewed at the first follow-up contact and
388 patients (74% of the 525 patients who completed baseline interview and
88.2% of the 440 who participated in first follow-up interview) were successfully

contacted and interviewed at the second follow-up contact (Figure 1 ).

2.2.2 HARP Intervention

Patients in the HARP Intervention arm received a six-session health
behavior change program, over a six week period, within the first three months
following discharge from hospital, delivered via telephone by a trained health
educator or “coach”. The primary health behavior goal focus of the program
included: (1) the reduction or quit smoking, (2) an increase in physical activity,
and (3) an improved healthy diet. The program adapted a relapse prevention

smoking cessation program model for use with the multiple risk behaviors of

24

cardiac patients (54). Although smoking patients were encouraged to consider
smoking cessation as a first priority, they were also encouraged to be openly
expressive about setting risk factor reduction goals they were actually prepared

to implement.

Coaching telephone sessions averaged 15 to 30 minutes. Behavior
change processes included behavioral staging, motivational interviewing, goal
setting, relapse prevention, and social support. Patients were encouraged to
identify at least one current behavior they intended to improve upon and set
weekly goal(s) in that selected behavioral area(s). Each patient and his/her
family had received a 25-page booklet and goal worksheets that provided
information related to the major health goal behaviors to consider during program
sessions. Goal setting within the intervention group is described in detail

elsewhere (55).

2.2.3 Study Variables and data collection

The outcome of interest in this analysis was patients’ attrition. For each
interview session, a patient was classified as ‘attrited’ if he/she did no complete
the next scheduled interview. Therefore patients who completed baseline and
the first follow-up interview, but not the last interview, were coded as ‘0’ (not
attrited) at the baseline interview but as ‘1’ (attrited) at the first follow-up.
Similarly patients who completed baseline interview only were coded ‘0’ at
baseline since baseline was the first and the last point in time when data could
(be available from them. Patients who completed all three interviews were coded

‘0’ (not attrited) for both, the baseline and the first follow-up interview (Appendix

25

2). In this manner were able to determine outcome at baseline and first follow-up
interview, and thus analysis included all patients who complete first follow-up

and/or baseline interview.

Of the primary interest in this analysis were the variables on patients’
demographics, severity of illness, medications and measures of quality of life.
The main measure for the severity of patient’s disease was left ventricular
ejection fraction (LVEF). Ejection fraction was obtained from the chart review
and dichotomized into two categories, LVEF > 35% and EF 3 35%. This cutoff
value has important prognostic implications in heart disease (56-58). Another
measure used was Charlson Comorbidity Index (CCI) (59). Coexisting
morbidities (not the index ACS) were counted in calculating the CCI weighted
score. In our analysis this variable was evaluated as continuous and then
categorized as a three level variable: 0 to 1 (few or no co-morbidities), 2-3

(moderate level of co-morbidity), or 3 4 (severe level of co-morbidity).

The main source of data on medication prescriptions was patient’s
medical chart. The data from medical charts was abstracted using a structured
questionnaire by trained nurse abstracters. All medicines were classified as
prescribed as before the index hospitalization (patient reporting being on a
prescription before the cardiac event), those that were prescribed during the
hospital stay and those that were prescribed at the time of discharge form
hospital. Obviously some medicines were prescribed at more than once

instance, for example, during hospital stay and also at discharge.

26

Also of interest were patients’ health behaviors of physical activity, diet,
steps taken to lose weight (if overweight) and smoking habits. Smoking status at
the time of hospitalization was established based on two sources of information:
the medical record chart audits, which contained an entry about current and past
smoking behavior, and the baseline interview. Respondents who reported that
they quitted smoking were asked how long ago that was. Comparing the time
between the baseline interview and the original hospital admission, we could
establish who quitted recently, but was still a smoker at the time of
hospitalization. In the subsequent 3-month and 8-month interviews, smoking
status (current (continuing) smoker, relapsed, previous and new quitter) was

reassessed.

Patients’ overall functional status was measured by the Activity Status
Index (ASI) (60), which provides a weighted composite score computed from
answers to questions about 12 activities of daily living of progressive intensity.
Scale scores for the ASI range from 0 to 58.2 points, with a higher composite
score indicating greater functional capacity. The ASI has been shown to be
highly correlated with oxygen uptake on treadmill exercise, and has been widely
used as an outcome in ACS (60). Prior research has demonstrated that ASI has
adequate sensitivity to show clinical changes in physical function for cardiac
patients (61, 62). The ASI was evaluated at baseline (assessed functional status

before the onset of disease), and at both the follow-up interviews.

In the interviews patients’ depression status was evaluated using the

Center for Epidemiologic Studies (CES-D) Depression scale. CES-D is a 20-item

27

self-report screening instrument designed to measure the relative frequency of
the occurrence of several depression symptoms within non-psychiatric
community populations. Each item response is scored on an ordinal scale of 0 to
3, with a possible range of the summated composite scale scores from 0 (no
depressive symptoms) to 60 (highest level of symptoms). A composite CES-D
score of 16 or greater is generally considered an indicator of at least mild
depression symptoms (63). In this analysis, the CES-D scores for all patients
were categorized into two groups. Those with CES-D scores of 16 or higher
were classified as “depressed” and those below 16 were classified as “not-
depressed”. This cutoff is consistent with that used in contemporary literature
and corresponds to 80th percentile scores for community samples and has a

95% sensitivity for diagnosing depression among low-income women (64, 65).

Additional quality of life measures such as EuroQol (66) and Quality of Life
after Myocardial infarction (QLMI) (67) were also evaluated. All measures that
can vary over time, such as ASI, CESD, QLMI, EuroOol etc. were re-evaluated at

the two follow-up interviews.

Mortality of patients in the study was obtained from a family member’s
report when the patient was called for an interview. This information was
supplemented by mortality data, which were obtained from the Michigan

department of Vital Statistics.

28

2.3 Family Home Care for Cancer Trial

2.3.1 Subjects and Settings for FHCC Trial

The family home care for caner trial focused on the care of cancer patients
at home, therefore inclusion criteria required that each otherwise eligible patients
should have a recognized family member who was the primary caregiver at home
for the patient. Dyads, patient and caregiver, instead of just patient were
recruited for the study. For participation in this study the patients had to be 21
years of age or older, be within 57 days of initiating a first course of
chemotherapy, and have a designated family caregiver, 21 years of age or older,
who also agreed to participate in the study. Once potential participants were
identified, the project was explained to the dyad, and upon agreeing to
participate, both members of the dyad signed consent forms. Following a
baseline interview, dyads were randomized to one of the two study arms; 1)
conventional care alone or 2) conventional care plus a 10-contact, 20-week
nurse-directed intervention. Subsequent interviews for both dyads in the control
and experimental arms occurred at 10 weeks (after the conclusion of the 4th
intervention contact), at about 20 weeks (after completing the intervention), and
at about 32 weeks to assess the effect of the intervention. The medical records of
all patients were reviewed at about 32 weeks after the baseline interview to verify
diagnosis and treatment information, as well as to track patients’ access of the

health care system (office visits and visits to the emergency room).

29

Of the 609 patients found eligible on screening and approached by
recruiters, 263 (43.2%) dyads (patients and their caregivers) agreed to
participate in the study and signed the consent form. Three hundred and forty six
(56.8% of 609) dyads refused participation citing a variety of reasons, common
among them were being too ill, not interested, too busy, already participating in
another study, feeling overwhelmed, or not wanting to be randomly assigned as
part of their participation and a variety of other personal reasons. In this
dissertation any patient who skipped an interview was considered to have
dropped out at that contact and from future contacts, if any. Therefore, of the
263 patients, 235 (89.4%) completed the first (baseline) interview. Of the 235
patients who completed the baseline interview, 178 (75.7%) completed the first
follow-up interview. One hundred and Thirty Nine or 78.1% of those who
completed the second follow-up and of them 125 (89.9%) completed the last
follow-up interview. For this analysis therefore, we were able to code the
outcome (attrition) at baseline and first and second follow-up contacts. Thus the
final multiple record file for analysis contained a total of 552 records, representing
235 baseline interviews, and 178 and 139, first and second follow-up contacts

respectively (Figure 2).

2.3.2 The nursing intervention

All patients randomized to the intervention arm were given a 10-contact,
20-week intervention in addition to conventional care. A nurse who specialized in
cancer care directed the intervention. The main aim of the intervention was to

assist patients and their caregivers with symptom monitoring and management,

30

education, emotional support, and coordination of services. At each visit, the
nurse intervener used a customized computer program to assess symptom
status, functional level, communication and the emotional health (anxiety &
depressive symptoms) of the patients. Assessment required the patients to rate
both the intensity of the symptom and the impact of the symptom on their quality
of life as well as appetite, normal daily activities, emotions/mood, and sleep.
Whenever a symptom had a high intensity or reached the threshold on the
impact score, the computer program automatically posted the symptom to the
patient's plan of care. Following the assessment, the nurse intervener and
patient reviewed the problems and mutually choose those problems that the
patient would work on over the upcoming weeks (next contact). Once problems
were chosen, the nurse intervener used the customized computer program to
generate interventions aimed at decreasing the intensity or impact of the
problem. Subsequent visits involved the evaluation of patients’ ability to
implement suggested interventions, as well as their ability in communicating care
needs to, and enlisting assistance in meeting care needs from, their caregivers.
A more detailed description of intervention and its effect on reducing symptom

severity can be found elsewhere (68, 69).

2.3.3 Study variables

As in the cardiac patients trial the main outcome of interest in the cancer
intervention trial was patients’ attrition, and was defined in exactly the same

manner as described above (Appendix 3). The only difference being that, since

31

4 interviews (1 baseline, three follow-ups) were conducted for this study, we were

able to determine the attrition status at three time intervals.

The covariates of interest included demographic factors such as patient’s
age, sex, race, education, marital status (social support) etc. Education was
recorded as a seven-category variable, ranging from ‘no formal education’ to
‘gradate or professional degree’. In this analysis, because some categories had
very few observations, we collapsed few categories to make education a four
category variable ‘less than high school’, ‘high school’, ‘more than high school to
completed college’ and ‘more than college’. An evaluation of this new variable’s
association with attrition in a regression model revealed that patients with ‘less
than high school’ education stood out differently from all the three other groups,
in terms of attrition, and the other three groups were not different from each
other. This variable was therefore converted to a two-level variable, consisting of
‘Iess than high school’ and ‘high school and above’ groups. Marital status
recorded as a 4 level variable, was also recoded as a two level variable ‘married’
and ‘unmarried’. Patients’ race for analytic purposes was recoded as a binary

variable (whites and minorities) as reported earlier for the cardiac patients study.

Disease related factors were also of interest such as stage of cancer and
site of cancer. Patients with cancers of seven different sites were recruited in the
original study. However, because there were relatively few patients with certain
cancers, this variable was recoded into ‘Iung cancer’ and ‘other cancers’
categories. Summary measures of variables related to patients’ symptoms, were

' also assessed for possible association with chances of patients dropping out of

32

the study. These measures included, total number of symptoms experienced at
each interview, total severity of symptoms, total limitation of daily activity caused
by symptoms, total bother felt by patient due to the symptoms. Symptom
severity was recorded on a 10-Point scale for each symptom, with ‘0’ indicting
absence of symptom (zero severity) and ‘10’ representing worst possible
severity. The total symptom severity reflects sum of the reported severities over

all individual symptoms.

The Short Form 36 (SF-36) health survey tool was used to measure health
status of patients (70). SF-36 is a multi-purpose, short-forrn health survey made
up of 36 questions. It yields an 8-Point-scale profile of functional health and well-
being scores covering both physical as well as mental health. It is a generic
measure, non-age, disease, or treatment specific. Accordingly, the SF-36 has
proven useful in surveys of general and specific populations, comparing the
relative burden of diseases, and in differentiating the health benefits produced by
a wide range of different treatments (71 -73). In this study the 10-Point physical

function sub-scale of SF-36 was used.

Other quality of life measures such as depression (CES-D) and total
optimism score were also evaluated for possible impact on attrition. Depression
was measured and used in the same manner as described earlier (binary, with
categories as <16 and 2 16). Patients’ employment status, treatment
procedures, hospitalization for chemotherapy, surgery, radiation etc. were also

evaluated.

33

2.4 Statistical Analysis

Data from the two studies were organized separately into two multiple
record files. The HARP data had either one or two records (observations) per
patient, depending upon whether that patient completed only the baseline
interview or at least baseline and the first follow-up interview. All patients
completing first follow-up interview had their outcome coded as either ‘0’ (non-
attrited) if they also completed the second follow-up interview, or as ‘1’, if first
follow-up was the last interview they completed. In the FHCC data, the outcome
variable coding process was similar to HARP data, except that maximum number

of records contributed by patients was 3 instead of 2.

Both data sets were modeled using the Generalized Estimating Equation
(GEE) method (66, 74, 75) using SAS version 9.0 (SAS software, SAS institute).
The GEE represents a class of models that are often utilized for analyzing data
that are correlated. The method can be used for both linear as well as
categorical outcomes. As in the general linear model (GLM), a function of the
mean g(u), where g is called the link function, is modeled as a linear function in
the regression parameters. For a dichotomous outcome (as in our study) the
‘Iogit’ link is commonly used. Thus the g(u) equals logit (P), where P is the .
conditional probability of Y=1 (the patient had attrition) given the values of all
covariate, define a the vector X. If there are p independent variables, this can be
expressed as; logit P(Y=1 l X) equals I30 plus the summation of the product terms

for the p independent variables times their beta coefficients. In the equation form

34

thiscan bewritten as logit PI)’ =1I x]=flo +161X1 +---+flka. This

equation is presented with acknowledgement that GEE estimates additional
correlation parameters that work in the background and are not explicitly stated
in the equation. The GEE model for dichotomous outcome data thus looks
similar to the standard logistic model, except it incorporates the correlation
between repeated outcomes from the same patient. A particular covariance
structure is assumed and employed by the GEE model for estimation of the
model parameters. In addition, the standard logistic models use likelihood
estimation function as their estimation method, but the GEE method follow quasi-

likelihood estimation method (76).

To accommodate the correlated nature of the outcome variable, the GEE
analysis of correlated data (like data presented in this analysis, with multiple
observations or records per person), requires specification of correlation
structure by the user. The correlated nature of the outcome variable can easily
be summarized by drawing a correlation matrix of observations made on an
individual patient at different points in time. A separate correlation matrix would
be needed for each stratum (person) in the study, which would require estimation
of a very large number of parameters. If, however, there are more parameters to
be estimated than the total number of observations in the data set, the model is
over-parameterized, and does not yield valid estimates. To avoid this problem,
the GEE requires that each subject have a common set of correlation
parameters. This substantially reduces the number of correlation parameters.

Although the GEE requires specification of correlation structure by the user, the

35

advantage of the GEE is that even if the correlation structure is incorrectly
specified the obtained parameter estimates are consistent. The variance of the
parameter estimate is however affected. The GEE yields two sets of variance
estimators: 1) the model-based estimators and 2) the empirical (robust)
estimators. The model-based estimators are similar to the maximum likelihood
based general linear model (GLM) variance estimators, and although the
likelihood for GEE is never formulated, the parameter estimates are consistent if
correlation structure is correctly specified. The empirical variance estimators
represent an adjustment of model-based estimators. While both approaches
make use of the specified correlation structure, the empirical estimators also take
into account the observed correlation between responses in the data. The use of
empirical estimators is advantageous as they are consistent even if the
correlation structure is incorrectly specified. This, however, does not mean that
when empirical estimators are used the correct specification of the correlation
matrix becomes redundant. The correct specification (choosing one close to real
correlation) results in gaining efficiency. This is particularly true if the number or
clusters (patients in our case) is large, meaning that incorrect specification will

result in unreliable results if the number of clusters is small (76).

In our data set we specified exchangeable correlation structure, and the
odds ratios and their confidence intervals are reported making use of empirical
(robust) estimators. The exchangeable correlation structure assumes that any
two within subject outcome responses In different time points have the same

- correlation.

36

For model building, all variables in the two data sets (if present) that were
cited in literature to be associated with patients’ attrition and all that could have a
plausible association with patient’s chances of attrition from the trials were
initially evaluated in a univariate model (one independent variable at a time) to
assess their unadjusted effect on the outcome. This served as a guide for
selecting variables for evaluation in a multivariate model. All variables having a
p-value of 0.2 or less were candidates for further evaluation; additionally any
variables that we considered to be biologically important were assessed even if
they did not meet the p-value cutoff criterion. Model building was carried out in
the manner generally recommended by standard textbooks (76, 77). The
variable found most significant in the univariate analysis was entered into the
model followed by the next most significant variable. The impact on the model,
of the addition of new variable was evaluated by change in the score test
statistic. This continued until all candidate variables were evaluated and the
impact of their addition or deletion on the model was assessed. The final model
represents a parsimonious model with variables that impact the score statistic
significantly, and where addition of further new variables does not bring a
significant change in the model. A few variables, as mentioned earlier, were kept
in the model despite their insignificant impact, because of their biological
importance. The score statistic was also used to evaluate the overall
significance of the model, evaluating all the variables simultaneously. The

purpose of multivariate modeling was to estimate the independent effect of

37

factors (variables) on patients’ chance of attrition. The details of this univariate

analysis are presented in table 3 (HARP data) and table 5 (FHCC data).

38

 

Chapter 3
3 Results

Since the analysis presented here was performed on two different
datasets that came from fairly different patient populations in terms of their

morbidity, it is presented under separate subheadings.

3. 1 HARP Trial

Participants’ progress through the study is described in figure 1.
Approximately a quarter of patients were lost between baseline interview and the
last (8 month) follow-up interview. Following the baseline interview all patients
were randomly assigned to either the intervention (n=268) or control group
(n=257). Tables 1 and 2 summarize the comparison of the intervention and
control groups, displaying comparability of the two groups. As stated earlier all
plausible variables in the data set were first evaluated in the univariate models to
identify candidates for evaluation in the multivariate model. Table 3 shows the
results of this univariate analysis for selected variables. At the univariate level,
patients’ smoking status, education (less than high school vs. High school and
more), Race (Minorities vs. whites), marital status (unmarried vs. married),
depression status (CESD 2 16 vs. < 16), alcohol consumption (consumers vs.
abstainers) and better quality of life as measured by E05, attained significance at
the level of significance of 0.05. However, all variables attaining a p-value of 0.2
or less were candidates for evaluation in the multivariate model. Following the

, univariate analysis, the process of multivariate modeling was completed and the

39

final model resulting from this exercise is described here. Because of the
longitudinal nature of the data the time variable (representing the time of
interview) was included among covariates in the model even though it did not
reach statistical significance. Additionally the group variable (that identified
patients’ group assignment as intervention or control), and the variable identifying
patients who died during the study period were also included in the final model.
Thus the effects reported from this multivariate model are adjusted for patients’
group assignment, death and the time of attrition. Death was retained in the
model to identify and estimate the effects of the factors affecting patients’ attrition

above and beyond that is caused by mortality.

Age of the patient plays an important role in attrition among the cardiac
patients. A detailed study of the effect of age over time, that included several
higher order terms, reveals a non-linear, quadratic ‘U’ shaped relationship
between patients’ age and attrition (Age coefficient, -0.2130, 95% CI of
coefficient -0.3124, -0.1136, Age Squared, coefficient, 0.0017, 95% CI of
coefficient 0.0009, 0.0025). This means that estimation of the effect of a
person’s age on his or her probability of attrition form the trial requires two and
not one coefficient. In other words, the probability of attrition of a 40 years old
patient as compared to a 30 years old patient is not same as the probability of
attrition of a 30 years old patient as compared to a 20 years old patient, despite

the fact that the difference in both cases is the same, Le. 10 years.

Patients’ self-reported smoking status was also a strong predictor of

attrition. This was a four category variable as described earlier. As compared

40

with non-smokers the smokers, recent quitters and past smokers, all were more
likely to drop out. The respective adjusted odds ratios and their 95% confidence
intervals (CI) were: smokers (AOR 2.65, 95% CI 1.27-5.51), recent quitters (AOR
2.90, 95% CI 1.41-5.98) and past quitters (AOR 2.04, 95% CI 0.94-4.41). The
past smokers although having a high odds ratio were not significantly different
from non-smokers in terms of attrition. Also smokers were not found to be
different from recent quitters in terms of attrition. Patients with a CES-D score of
16 or higher (Indicative of clinical depression) were found to be more likely to
drop out of the study than those with a CES-D score of less than16 (AOR 1.62,
95% Cl. 1.10 — 2.39). Patients who reported alcohol consumption as compared
to complete abstainers were found to be less likely to drop out from the study
(AOR 0.53, 95% CI. 0.35 — 0.81 ). In other words the adjusted odds ratio for
abstainers relative to those who reported alcohol consumption was 1.88, 95% CI
1.23 — 2.86. These results are summarized in table 4. As stated earlier, a large
number of variables were tested at univariate level for modeling purposes. The
final model reported above was based on variables that were of biological
importance and that reached statistical significance. The detailed univariate

analysis is presented in table 3.

3.2 FHCC TRIAL

Participants’ progress in FHCC study’s different stages is described in
figure 2. Almost half of the patients were lost between the baseline and the third
follow-up interviews. Following the baseline interview all patients were randomly

assigned to an intervention (n = 117) or a control (n = 118) arm. Tables 5 & 6

41

 

summarize the comparison of the intervention and control groups in terms of
various demographic and clinical characteristics, and show no significant
differences between the two arms of the trial. The approach and methodology
for analysis of patient’s attrition for the FHCC was similar to the approach
employed for the HARP trial. All plausible variables were assessed in the
univariate (one variable at a time) models and the selected candidate variables
were evaluated to develop the final multivariate model. Table 7, displays the
result of the univariate analysis of selected variables. In the univariate analysis
patients’ sex (male vs. female), site of cancer (lung vs. others), time of interview
(baseline, second, third), education, anxiety, depression (CESD 2 16 vs. < 16)
and hospitalization for chemotherapy were found to be statistically significant at
5% level. However, as in the cardiac patients’ data a cutoff of p-value s 0.2 was
used to qualify variables for further consideration and evaluation in the

multivariate model.

Table 8, summarizes the final multivariate GEE model identifying factors
associated with attrition in cancer patients. As in the previous data set, the
model described here was adjusted for mortality and patients study group
assignment. We found that lung cancer patients were more likely to drop out of
the trial than patients with other cancers (AOR 2.13, 95% CI 1.28-3.53).
Consistent with the finding in the HARP data depressed patients (CESD 2 16)
were more likely to drop out than non-depressed (AOR 1.73, 95% CI 1.03-2.91).
Education was found to be a strong predictor of attrition. Patients with less than

, high school education were more likely to drop out than those who had

42

completed at least the high school (AOR 7.53, 95% Cl 227-2497). The time
(Wave) variable identifying each data collection wave was retained in the model
because of the longitudinal nature of the study, the odds ratios for wave1 and 2

were higher as compared to wave 3 (AOR 2.19, 95% CI 1.04-4.66, and 2.87,
1.37-5.98 respectively).

43

Chapter 4
4 Discussion

Realizing that patients’ attrition from clinical trials can have its origin in
several different areas such as patients’ personal characteristics, or the disease
patients are suffering from, or the type and nature of interventions tested in the
trial, we set out to evaluate all the available information in the studies analyzed,
that fell into one of these areas. Conceptually, we focused our analysis on
evaluation of factors from the three areas mentioned, such as patients' age,
gender, education, employment status, race etc. Similarly, within each study,
disease related factors, such as severity of the disease were also evaluated. We
also looked at the patients’ compliance or non-compliance with intervention

expectations, such as quitting smoking in cardiac patients’ trial.

Despite data coming from two different clinical trials and having two very
different diseases, there were similarities in many respects. The majority of
patients were recruited from similar, if not the same, geographical areas in mid-
Michigan. The two sets of patients were very similar in their age distribution, with
mean age of cardiac patients being 59.78 (SD 12.00) years, and that of cancer
patients being 59.72 (SD 10.34) years. Caucasian whites made up the majority
of patients in both data sets (90% for FHCC, 85% for HARP). The differences
found in the factors associated with patients’ attrition in the two studies could
therefore largely be attributed to the differences in the nature of diseases patients

had or differences in the two interventions.

44

Age of the patient, among cardiac patients, plays an important and
complex role in their attrition from clinical trials, displaying a ‘U’ shaped effect
with maximum attrition at the two extremes of the age distribution. Attrition was
high in the youngest patients but declines until middle age but then climbs with
increasing age. The probability of patients’ dropping out of the study was the
lowest at age 63 and after that it started rising again. To illustrate this with an
example with two arbitrarily chosen patients of ages 53 and 73 years respectively
(10 years younger and 10 years older than the group with lowest probability of
attrition i.e. 63 years), their respective adjusted odds ratios as compared to a 63
years old patient after adjusting for the variable in the model are 1.17 and 1.20
respectively, highlighting that both the younger and older patients are more likely
to drop out than 63 years old patients. Younger age, better health, and higher
percentage of employed patients, as may be expected at the middle age, explain
why attrition may be higher among younger patients. On the other hand, higher
attrition in older patients is also not surprising, as the older patients are likely to
be unhealthier, with higher'existing co-morbidities, be more fragile and more
likely to be repeatedly hospitalized or moved into a nursing home. Our data
show a significant positive correlation between patients’ number of co-morbidities
and age (p<0.001). as well as between age and employment. Employed patients
were on average more than 11 years younger than unemployed ones (p<0.001).
Both these findings help explain higher attrition among older patients and are
consistent with the non-linear relationship of age with patient attrition. A

. significant association with patient age was not detected in FHCC data. The

45

F HCC data had a narrower age range as compared to HARP data (51 years Vs.
69 years), and the oldest patient in the FHCC trial was 87 years old, as
compared with a 100 year old patient in the HARP trial. The narrow age range
and fewer very old patients could be the reason for not finding an effect of age on
attrition. However, the fragility and other conditions associated with old age are
well recognized and for this reason patients age was retained in statistical model
despite not reaching the statistical significance. Finding a statistically significant
effect of age in HARP but not in FHCC could also be due to the difference in the
sample sizes of the two studies. FHCC trial had a smaller sample size than
HARP, and therefore an inability to find a significant effect could be due to

insufficient power to find such a difference.

Social isolation and seclusion play a role in attrition as seen, by greater
likelihood of depressed-patients to dropout of the study, in both trials. We found
that depressed patients were more likely to drop out than non-depressed (cardiac
patients AOR 1.62, 95% CI 1.10 - 2.39, and cancer patients AOR 1.73, 95% Cl
1.03 - 2.91 ). We also found that in the HARP Trial, unmarried patients were
more likely to drop out than married patients in the univariate analysis. Similarly
among cancer patients, those who scored higher on the anxiety scale were more
likely to drop out than those scoring lower on the anxiety scale. Since
depression and anxiety (78-80), and depression and marital status (81 -83) are
associated with each other, after addition of depression to the model either
variable did not add any new information and were thus removed from the final

, model. The association of social activity (or the lack of it) was further highlighted

46

by our finding that cardiac patients who reported complete alcohol abstinence
were more likely to drop out than those who considered themselves alcohol
drinkers. Since people are more likely to consume alcohol during social events,
complete abstinence could be an indicator of no or at least minimal social

participation which can explain attrition from the study (84).

Diseases, that are chronic and “incurable” tend to have a depressing
effect on those suffering from them, and if these patients also lack social support,
they tend to go into seclusion and usually diminish or cutoff contact with the rest
of the world. In contemporary literature, depression is perhaps the most
consistently reported factor associated with attrition from research studies (44,
47). Social isolation and seclusion associated with depression can explain the
higher attrition observed in depressed patients as compared to non-depressed
patients. Depressed patient with coronary artery disease also have higher
mortality than non-depressed cardiac patients (85), though mortality was lower in
the cardiac patients’ study. Studies have also shown that depressed patients are
also less likely to follow caregivers’ advice regarding the management of their co-
existing health conditions (86). In behavioral intervention trials, like the ones
reported here, requiring behavioral changes, depression may also mediate

attrition by this mechanism.

There are a few reports, in the literature, indicating a higher attrition
among smokers as compared to non-smokers (87), as also seen in the cardiac
patients study reported here. All cardiac patients, who are smokers, are advised

‘ smoking cessation by their care providers, on top of that the intervention in the

47

HARP study particularly emphasized on patients smoking habits, encouraging
smokers to quit smoking. Patients who fail to comply with their health care
provider’s advice, may feel social pressure, and choose to drop out of he study
rather than participating and remaining in the study. Similarly it has been shown
that in behavioral studies patients who fail to achieve the desired study goal tend
to drop out of the study before its completion (88). We found that patients who
continued to smoke and those who reported quitting recently (in most cases after
their cardiac event), were more likely to drop out than non-smokers. Patients
who reported quitting smoking recently, may not have completely overcome their
habit, making them more similar to smokers than non-smokers and thus were
more likely to drop out (like smokers) than non-smokers. The past smokers
(those who quitted more than 1 year prior to the interview) represent a more
health conscious group that gave up smoking on their own or on their health
providers advice before their cardiac event, and because of an initiative on their
own, rather than their adverse event, were more successful in kicking their habit
and were more similar to non-smokers, and thus not different from non-smokers

in terms of attrition.

More aggressive nature of disease (lung cancer in cancer patients’ case)
was also found to increase the likelihood of attrition. A more severe and
aggressive disease is obviously associated with higher mortality, which would
lead to higher attrition. More aggressive and severe disease is also associated
with higher morbidity, which can also lead to frequent hospital visits/admissions,

- and even admission to hospice. This would explain finding an independent effect

48

of more aggressive disease (lung cancer) over and above that caused by

mortality.

We found that among the cancer patients, education played a role in
attrition. As described earlier the effect of education was initially evaluated in
four categories, but three categories were collapsed into one, to dichotomize this
variable into, those with less than high school education, and those with high or
higher level of education. Less educated patients were more likely to drop out
than those that were better educated. This finding is consistent with similar
reports by other researchers (42, 45, 87, 89, 90). Inability to fully understand and
appreciate the value of research and participation in clinical studies, likely
because of non-exposure to research and its advantages by patients’ lower level
of education, explain why they may be more likely to drop out than better
educated patients (91 ). A similar effect of lower level of education on attrition
was also found in the univariate analysis of cardiac patients’ data (OR 1.6486,
95% CI 1.0485 — 2.5923). Smoking habits have an association with education,
with low educated people more likely to smoke than better educated, which was
also seen in our data (p = 0.04), and thus could be the reason for failure of
education to add information to our model in the presence of smoking status. We
also found that attrition of patients decreased with passage of time. This means
patients who were to drop out tended to drop out early. This was an effect
observed in both, cardiac as well as cancer patients, but reached statistical
significance only among cancer patients. Given that disease aggressiveness

~ and severity was one factor responsible for attrition of patients, it is plausible that

49

such patients, being more severely ill, would drop out earlier than later. Disease
severity (as measured by left ventricular ejection fraction) did not play as strong a
role in attrition in cardiac patients, as in cancer patients; which could explain why
the time of attrition did not reach statistical significance with cardiac patients.
Additionally the two studies differed in the number of follow-ups, and the intervals
between the follow-ups that could also contribute to significance or non-

significance.

Specification of a correlation structure in longitudinal analysis using GEE
is not always easy. In our studies where the outcome variable was of non-
recurring nature this was particularly difficult. In our studies the way the outcome
variable was defined, meant that until the interview wave where attrition occurred
outcome variable at previous time intervals were perfectly correlated. We
reviewed the correlation matrix and found our assumption of exchangeable
correlation for modeling to be reasonable. Additionally, a major advantage of
GEE is that even if the correlation structure is incorrectly specified the obtained
parameter estimates are consistent. Therefore despite the unique nature of our

outcome variable we believe our parameter estimate to be reasonably valid.

4. 1 Conclusion

Starting with the youngest cardiac patients, attrition from longitudinal
studies declines with increasing age, though in a non-linear manner. With further
increase in age the declining effect becomes less and less pronounced, and then

_ probably because of general ill health, and more co-morbidities starts climbing

50

again among the oldest patients, however in populations with diseases
associated with higher morbidity and terminal illness, such a cancer, other
psychosocial and demographic factors play a stronger role in determining
patients’ continuation (or attrition) from clinical trials. Social support and
interactions play an important role in determining patients’ attrition from research
studies. Social isolation increases and social activity and social interactions
decrease a patient’s likelihood of attrition. Depressed patients are also more
likely to drop out of the clinical trials. Presence of depression has its own effect
in addition to that it exerts through social isolation and seclusion. Other factors
such as aggressive and severe disease also make the patients drop out from the

studies.

To the best of our knowledge our study is the first of its kind to evaluate
factors associated with patients’ attrition using longitudinal models. Since such
models take into account the variation within observation (study subjects) and
also take into account the changing nature of time-varying characteristics, the
models used here are more appropriate in identifying characteristics that can

otherwise be missed by other statistical techniques.

It is not possible to ‘intervene’ to eliminate many factors that promote
attrition, but using the profiles described here, researchers carrying out clinical
trials can identify participants that are at higher risk of dropping out of the study.
With a priori idea of who is more likely to drop out during the course of the trial
and who is not the researchers should attempt to build mechanisms into their

' studies aimed specifically at high-risk patients to at least minimize their attrition.

51

 

Common sense strategies such as proper training of interview data collectors
and other study personnel who come in contact with study subjects, better
communication with subjects, expression of appreciation of patients’ participation
and time etc are recommended by experts (92). Involvement of family members
can help address the factors like depression and seclusion, which would
translate not only into improved retention of patients in the trials but will also have
a beneficial effect on the quality of life of the patients. The identification of
somewhat different set of predictors of attrition, in groups of patients with
different diseases, point to a need for studies to be carried out in patients with

other chronic conditions as well.

Clinical trials are usually designed to answer a specific research question
or a set of research questions that most often relate to evaluation of an
intervention - pharmaceutical or otherwise. Therefore the data from these trials
often lacks information on factors and characteristics that may be important from
the point of view of studying attrition. A trial conducted solely to study attrition
experience and patterns of participants, without any other research question, will
of course be illogical, or even meaningless. However, when conducting trials for
evaluation of a new treatment modality the researchers should consider
collecting, in addition to what is required for the primary purpose of the trial,
information on factors identified in contemporary literature as associated with
attrition, as well as factors that may intuitively be considered worth evaluating for

possible association with attrition.

52

Such research has the potential of further improving quality of clinical

trials, resulting in improvement of the validity of the findings from such trials.

 

53

Table 1: Comparison of baseline demographic characteristics of the intervention

and control groups of the HARP trial (categorical variables)

 

 

 

 

 

 

 

Intervention Control
Variable n=268 n=257 p-value
n(%) n(%)
Gender
Male 173 (64.6) 161 (62.6) 0.65
Female 95 (35.4) 96 (37.4)
Marital Status
Married 183 (68.3) 167 (65.2) 0.51
Unmarried 85 (31.7) 90 (34.8)
Race
Non-Hispanic white 228 (85.1) 215 (83.7)
African American 32 (11.9) 28 (10.9) 0 50
Hispanic white 5 (1.9) 7 (2.7)
American Indian 3 (1.1) 7 (2.7)
Smoking status
Never smoked 82 (30.6) 92 (35.8)
Quit > 1 year ago 90 (33.6) 89 (34.6) 0.32
Quits 1 year ago 67 (25.0) 48 (18.7)
Current smoker 29 (10.8) 28 (10.9)
Alcohol Drinking
Yes 202 (75.4) 201 (78.2) 0.47
No 66 (24.6) 56 (21.8)

 

 

 

 

54

 

Table 2: Comparison of baseline demographic characteristics of the intervention
and control groups of HARP Trial (continuous variables)

       

 

 

 

 

 

 

 

 

 

 

 

Variable 7 I f Intervention Control V ’ I p-value
Age 58.9 60.5 0.13
Years of education 13.12 13.25 0.84
Household Income (in $) 36,585 42,774 0.16
SF-36 General Health Scale 3.18 3.30 0.24
Beck’s Hopelessness score 7.75 7.93 0.11
QLMI - Symptoms subscale 5.10 5.13 0.77
QLMI - Confidence subscale 5.21 5.24 0.73
QLMI - Self-esteem subscale 5.99 5.93 0.55
EurQol 0.75 0.74 0.55
ASI 29.12 30.00 0.55
CES-D 14.00 13.09 0.32

 

 

 

 

 

55

Table 3. Factors associated with patient’s attrition based on univariate analysis
(HARP Trial)

     

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I‘Variable ' _  ‘ * ’ ’ - Odds Ratio ‘ ,   95% ci e ’
Age 0.99 0.98 - 1.01
Sex (Ref: Males)

Females 1.28 0.88 - 1.86
Smoking Status (Ref: Non-Smokers)

Current smoker 2.78 1.32 - 5.88
Recent quitter 2.58 1.21 — 5.48
Past smoker 2.14 0.97 — 4.70
Education (Ref = High School and

above)

< High School 1.66 1.06 —2.58
Race (Ref: Non-Hispanic Whites)

Minorities 1 .69 1 .05 — 2.70
Marital Status (Ref: Married)

Unmarried 1.84 1 .26 - 2.68
Employment Status (Ref: Employed)

Unemployed 1.03 0.71 - 1.49
Depression Status (Ref: CESD < 16)

CESDz 16 1.74 1.19-2.54
Alcohol consumption (Ref:

Consumers)

Abstainers 1 .78 1 .18 - 2.68
Left Ventricular Ejection Fraction (Ref:

2 35)

<35 1.14 0.74—1.76

56

Table 3 (Continued): Factors associated with patient’s attrition based on

univariate analysis (HARP Trial)

 

    

  

 

 

 

 

 

 

 

 

 

 

 

57

I _ "'gsot'corma‘m i

Variable f * " . Odds Ratio : ‘ Interval Of Odds :g
Coronary Artery Bypass Graft (CABG) 0.84 0.55 — 1.30
Angioplasty 0.86 0.29 - 2.52
Angioplasty + CABG 0.55 0.12 -— 1.51
EuroOol Score (E05) 1.69 1.05 — 2.70
QLMI Symptoms subscale 0.86 0.74 — 1.00
QLMI Self-Esteem subscale 0.88 0.76 -- 1.02
QLMI Confidence subscale 0.78 0.65 — 0.94
Activity Status Index 1.01 0.99 - 1.02
iCrSIICJj--er:(1)orbidities (Charlson’s weighted 1.08 0.93 _ 1.25

E

 

 

 

 

Table 4: Factors associated with post-baseline attrition in cardiac patient’s based
on the final Multivariate GEE model, (n = 525)

   

 

 

 

 

 

“WIPE; : . , Odds “at” Odds Ratio . ' ' (Twain), .
Age1 -o.2130’ -o.3124, b.11361 <o.o1
Age Squared‘ 0.00171 0.0009, 0.00251 <0.01
Smoking Status

Current smoker 2.65 1.27 - 5.51

Recent quitter 2.90 1.41 - 5.98 <0.01
Past smoker 2.04 0.94 - 4.41

Depression score

(CESD)

216 1.62 1.10-2.39 0.01
Drink alcohol 0.53 0.35 - 0.81 <0.01

 

 

 

 

 

 

Model adjusted for patient’s death, time of interview and patient’s group

assignment.

1Parameter estimates and their respective 95% confidence intervals,

instead of odds ratios are reported for variables age and age squared.

58

Table 5: Comparison of baseline demographic characteristics of the intervention

and control groups in the FHCC trial (categorical variables)

 

 

 

 

 

 

Intervention Control
Variable n=1 17 n=1 18 p-value
n(%) n(%)
Gender
Male 32 (27.35) 31 (26.27) 085
Female 85 (72.65) 87 (73.73)
Marital Status
Married 29 (24.79) 32 (27.35) 0.65
Unmarried 88 (75.21) 85 (72.65)
Race
Non-Hispanic white 106 (90.60) 107 (90.68) 0.98
Minorities 11 (9.40) 11 (9.32)
Education
< High School 4 (3.42) 11 (9.32)
High School 28 (23.93) 27 (22.88) 0.17
> high School to College 59 (50.43) 48 (40.68)
More than college 26 (22.22) 32 (27.12)

 

 

 

 

59

 

Table 5 (Continued): Comparison of baseline demographic characteristics of the

intervention and control groups in the FHCC trial (categorical variables)

 

 

 

 

 

Intervention Control I I 1

Variable n=1 17 n=1 18 p-value
n(%) n(%)

Cancer Site
Breast 47 (40.17) 46 (38.98)
Colon 13(11.11) 13(11.02)
Gynecological 12 (10.26) 12 (10.17)
Lung 40 (34.19) 41 (34.75) 0.98
Lymphoma 4 (3.42) 4 (3.39)
Pancreatic 1 (0.85) 1 (0.85)
Other 0 (0.0) 1 (0.85)
Lung Cancer
Yes 77 (65.81) 77 (65.25) 0.92
No 40 (34.19) 41 (34.75)
Stage of cancer
Early 40 (34.48) 37 (32.17) 0.71
Late 76 (65.52) 78 (67.83)

 

 

 

 

60

 

Table 6: Comparison of baseline demographic characteristics of the intervention

and control groups in the FHCC trial (continuous variables)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

61

 

 

Variable Intervention Contro p-value
_ Mean (SD) Mean (SD)

Age 60.50 (9.80) 58.94 (10.82) 0.25
Total Symptoms 5.42 (3.39) 5.42 (3.08) 0.99
Total Symptom Severity 28.91 (20.04) 26.86 (19.68) 0.44
Total Symptoms Limitation 23.14 (19.84) 20.52 (19.64) 0.32
Total Symptoms Bother 26.36 (21.07) 23.79 (20.93) 0.36
Total Mastery of Care 25.39 (4.84) 25.07 (4.72) 0.62
SF-36 Physical Function 63.55 (29.53) 64.21 (27.65) 0.86
SF-36 Physical Role Impact 28.92 (37.69) 29.13 (37.42) 0.97
SF-36 Emotional Role Impact 62.96 (40.54) 65.24 (40.33) 0.67
SF-36 Bodily Pain 61.97 (29.14) 64.89 (26.89) 0.43
SF-36 General Health 53.55 (24.33) 54.28 (25.60) 0.83
SF-36 Vitality 50.34 (22.93) 49.83 (24.10) 0.87
SF-36 Social Functioning 59.40 (31.09) 60.79 (31.14) 0.73
SF-36 Mental Health 73.81 (19.64) 75.17 (17.99) 0.58
Total Optimism 24.92 (3.76) 25.32 (4.41) 0.63
Total Anxiety Score 33.77 (11.60) 36.63 (13.14) 0.08
CES-D 12.19 (9.05) 13.62 (9.05) 0.24
Total Procedures Performed 1.44 (1.57) 1.42 (1.57) 0.90
Total Complications 4.08 (3.15) 3.97 (3.59) 0.80

 

 

Table 7: Factors associated with patient’s attrition based on univariate analysis
(FHCC Trial)

                 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 ' .’ 1 * g 1 , , * ' ‘ ‘. * '0' 1 . 1 ' w 5. Mam, ;
2W”? _._ w _ . e _ ,, . - w 9““: Ram 95“ 9"”‘0” . (mm)
Age 1.01 0.99 - 1.03 0.24
Sex (male) 2.10 1.29 - 3.41 0.004
Stage of Cancer 0.62 0.38 - 1.02 0.053
Lung cancer 0.46 0.30 - 0.73 0.001
Interview waves (Ref =
Wave3)
Wave1 2.79 1.49 - 5.24
0.002
Wave2 2.50 1.30 - 4.83
Education (Ref = College and
above)
< High School 4.06 1.53 - 10.76
High School 0.91 0.49 - 1.69 0014
More than high school to
College 0.72 0.41 - 1.27
Race (Ref: Non-Hispanic
Whites)
Minorities 0.82 0.37 - 1.81 0.61
Marital Status (Ref: Married)
Unmarried 0.82 0.49 — 1.35 0.42
Patient’s Mastery 0.97 0.92 - 1.02 0.22
SF-36 (Physical well-being) 0.998 0.990 - 1.005 0.62
5536 (Absence °I b°dy 1.003 0.99 - 1.01 0.46
pain)
.SF'36 (“Ck °f thSica' “"9 0.998 0.993 - 1.004 0.67
Impact)

62

 

 

Table 7 (Continued): Factors associated with patient’s attrition based on

       

univariate analysis (FHCC Trial)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

63

.. A. 95% CIforOdds p-Value
W” , F AA ;.., + . F, .; . I ,°““““‘"° ;   4. r Ratio; . __-,<tv_ne,III>_ 5:;
SF-36 (Lack of Emotional
role impact) 0.997 0.992 - 1.002 0.28
SF-36 (General health) 0.996 0.987 - 1.005 0.46
SF-36 (Vitality) 0.993 0.984 - 1.003 0.17
SF-36 (Social
Functioning) 0.997 0.991 - 1.004 0.45
SF-36 (Mental health) 0.994 0.983 - 1.01 0.35
Total Symptoms 1.06 0.99 - 1.13 0.09
Total symptom severity 1.01 0.99 - 1.02 0.082
Total limitations due to
symptoms 1.008 0.99 - 1.02 0.20
Total bother due to
symptoms 1.009 0.998 - 1.019 0.12
Anxiety 1.02 1.001 - 1.037 0.043
Depression (CESD) 1.03 1.008 - 1.056 0.011
CESD (>16)

Ref: <= 16. 1.85 1.14 - 3.01 0.02
Gainfully employed 0.83 0.51 - 1.35 0.45
Household Income 1.005 0.99 - 1.01 0.27
Complications 1.06 0.99 - 1.12 0.070
Radiation (No) 1.17 0.75 - 1.83 0.48

Surgery (No) 1.02 0.62 - 1.66 0.94
Hospitalization for chemo 1.79 1.14 - 2.81 0.01

 

 

 

Table 8: Factors associated with patient’s post-baseline attrition in the FHCC
trial based on the final Multivariate GEE model (n = 235).

 

95% Cenfidence Limits of

 

 

 

 

 

 

Variable Odds Ratio Odds ratio (Pf-g";
Age 0.99 0.98 - 1.02 0.95
Cancer Site

Lung Vs.Other 2.13 1.28 - 3.53 <0.01
Interview wave

Baseline 2.19 1.04 - 4.66

First Follow-up 2.87 1.37 - 5.98 (0'01
Depression score

(CESD)

2 16 1.73 1.03 - 2.91 0.04
Education

< Hi9“ S°h°°' VS' ”‘9“ 7.53 2.27 - 24.97 <0.01

School and more

 

 

 

 

Model adjusted for patient’s death, time of interview and patient’s group

assignment

64

 

Figure 1: Flow chart showing patient flow through various stages of cardiac
patients’ (HARP) Trial.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

65

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Consented
n=719
I Lost
I n=194
Baseline
interviewed
n=525
Intervention Random Control
n=268 assignment n=257
Lost Lost
n=43 n=42
Completed 3 M Completed
interview 09th interview
n=225 Intervrew n=215
Lost Lost
n=23 n=29
Completed 8 M Completed
interview 00‘“ interview
n=202 Intervrew n=1 86

 

 

 

Figure 2: Flow chart showing patient flow through various stages of cancer
patients’ (FHCC) study.

 

 

Consented
n=263

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I Lost
1 n=28
Completed Baseline
interview
n=235
I
& Ir
Intervention Random Control
n=1 17 assrgnment n=‘ 18
Lost , Lost
n=26 n=31
Com leted Com leted
inteIDView 1“ FSFIJOW' intepview
":91 Interview ”:87
Lost Lost
n=27 ‘—‘—‘ I"_‘ n=12
Com leted Com leted
interiew 2nd Z‘glow‘ intepview
”:64 Interview ”=75
Lost Lost
n=6 '—_I —’ n=
Completed d _ Completed
interview 3 Fuoglow interview
":58 Interview ”:67

 

 

 

66

 

 

 

 

 

 

Appendices

Appendix 1: Timeline of select important events in development of clinical trials
regulations

     

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Year i“ ' Actor reglion ' _ _ _ _ ,

1906 Pure Food and Drug Act passed

1931 Formation of US. Food and Drug Administration (FDA)

1938 US. Federal Food, Drug and Cosmetic Act
Designation of drugs as prescription or over he counter (OTC) by

1952
FDA

1966 Mandated creation of Institutional Review boards (lRBs) for funding
by the US. Public Health Service

1976 Medical Device Amendment to the US. Food, Drug Cosmetic Act

1977 Publication of General Considerations for Clinical Evaluation of
Drugs

1985 Publication of Guidelines for the Format and Content of the Clinical
and Statistical Section of an Application

1987 Treatment IND (FDA)

1992 Parallel track and accelerated approval, FDA

1997 Publication of Good Clinical Practice Consolidated Guidelines

67

Appendix 2: Format of the attrition (outcome) variable along with an example of
fixed (gender) and time-varying (CESD) covariates in the HARP dataset

    

Comment 7 Participant #' : covariates e.g. covariates 9.9.3) ’ Attrition1
' .A. . Geiger .. CESD. . ,A. w A

 

 

Completed

 

 

 

 

 

 

 

 

 

baseline only 1 Female 15 1
Completed two 2 Female 27 0
Fntervrews only 2 Female 19 1
Completed all 3| 3 Male 24 0
nterVIews 3 Male 1 5 0

 

I 1 = Attrition, 0 = No attrition until next wave

68

Appendix 3. Format of the attrition (outcome) variable along with an example of
fixed (gender) and time-varying (CESD) covariates in the FHCC dataset

   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, ’ . Fixed Time varying ‘ I“ H
Comment ,   Participant # Covariates e. g. covariates egg); . Attrition1
" ‘ ’ ‘ ‘ " ' * v“ " Gender ; _ ,QE$D_._
Completed 1 Female 15 1
baseline only
Completed 2 2 Male 24 0
Interviews only 2 Male 15

3 Female 27 0
Completed3
Interviews only 3 Female 19 0

3 Female 18 1
Completed all
'four interviews 4 Female 42 0

4 Female 31 0

4 Female 33 0
L——I——-————-.I————

1 1 = Attrition, 0 = No attrition until next wave

69

10.

11.

12.

13.

References

Meinert CL, Tonascia S. Clinical trials: design, conduct, and analysis. New
York: Oxford University Press; 1986.

Spilker B. Guide to clinical trials. New York: Raven Press; 1991.

Piantadosi S. Clinical trials: a methodologic perspective. New York: Wiley;
1997.

Last JM, International Epidemiological Association. A dictionary of
epidemiology. 4th ed. New York: Oxford University Press; 2001.

NIH. What is a clinical trial? In: Clinicaltrials.gov [lntemet], available at
http://www.clinicaltrials.gov/ct/info/whatis#whatis.

NIH. What are different types of clinical trials? In: Clinicaltrials.gov
[lntemet], available at http://www.clinicaltrials.gov/ct/info/whatis#types.

Chow S-C, Liu J-P. Design and analysis of clinical trials: concept and
methodologies. New York: Wiley; 1998.

Good Pl. A manager‘s guide to the design and conduct of clinical trials.
Hoboken, N.J.: Wiley-Liss; 2002.

Meinert CL. Clinical trials dictionary: terminology and usage
recommendations. Edition 1.0, 1st printing. ed. Baltimore, Md.: The Johns
Hopkins University; School of Hygiene and Public Health; Center for
Clinical Trials; 1996.

Bull JP. The historical development of clinical therapeutic trials. J Chronic
Dis 1959;10:218-48.

Hippocrates. Hippocrates upon air water, and situation; upon epidemical
diseases; and upon prognosticks, in acute cases especially. To this is
added (by way of comparison) Thucydides's account of the plague of
Athens. By Francis Clifton, M.D [microform]. London: printed for J. Watts;
1734.

Mettler CC, Mettler FA. History of medicine; a correlative text, arranged
according to subjects. Philadelphia [etc.]: Blakiston; 1947.

Avicenna, Gruner DC. A treatise on the Canon of medicine of Avicenna.
incorporating a translation of the first book. [New York: AMS Press; 1973.

70

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

Modern King James version of the Holy Bible. 1st ed. New York: McGraw-
Hill; 1962.

Lind J. A treatise of the Scurvy. In three parts. Containing an inquiry into
the nature, causes, and cure, of that disease. By James Lind, M.D
[microfonn]. Edinburgh: printed by Sands, Murray, and Cochran. For A.
Millar, London; 1753.

Barquet N, Domingo P. Smallpox: the triumph over the most terrible of the
ministers of death. Ann Intern Med 1997;127(8 Pt 1):635-42.

Creighton C. A history of epidemics in Britain. 2nd ed. London: Cass;
1965.

Lilienfeld AM. The Fielding H. Garrison Lecture: Ceteris paribus: the
evolution of the clinical trial. Bull Hist Med 1982;56(1):1-18.

Bourdieu J, Reynaud 8. Factory discipline, health and extemalities in the
reduction of working time in nineteenth century France. Socioecon Rev
2006;4(1):93-118.

Snow J, St. James Parish (London England). Report on the cholera
outbreak In the Parish of St. James, Westminster during the autumn of
1854 [microfomi I]. London: Churchill; 1855.

Semmelweis IF, Carter KC. The etiology, concept, and prophylaxis of
childbed fever. Madison, Wis.: University of Wisconsin Press; 1983.

Lister J. The collected papers of Joseph Baron Lister [microform.]. Oxford:
Clarendon Press; 1909.

Bernhard J, Cella DF, Coates AS, Fallowfield L, Ganz PA, Moinpour CM,
et al. Missing quality of life data in cancer clinical trials: serious problems
and challenges. Stat Med 1998;17(5-7):517-32.

Movsas B. Quality of life in oncology trials: a clinical guide. Semin Radiat
Oncol 2003; 1 3(3):235-47.

299 BC, Mok TS. Methodology for quality-of-life assessment: a critical
appraisal. Thorac Surg Clin 2004;14(3):305-15.

Klesges RC, Williamson JE, Somes GW, Talcott GW, Lando HA, Haddock

CK. A population comparison of participants and nonparticipants in a
health survey. Am J Public Health 1999;89(8):1228-31.

71

 

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

Lynch WD, Golaszewski TJ, Clearie A, Vickery DM. Characteristics of self-
selected responders to a health risk appraisal: generalizability of corporate
health assessments. Am J Public Health 1989;79(7):887-8.

Pullen E, Nutbeam D, Moore L. Demographic characteristics and health
behaviours of consenters to medical examination. Results from the Welsh
Heart Health Survey. J Epidemiol Community Health 1992;46(4):455-9.

Gordis L. Epidemiology. Philadelphia, Pa.: Saunders; 2004.

White M, Michaud G, Pachev G, Lirenman D, Kolenc A, FitzGerald JM.
Randomized trial of problem-based versus didactic seminars for
disseminating evidence-based guidelines on asthma management to
primary care physicians. J Contin Educ Health Prof 2004;24(4):237-43.

Lasky JJ. The problem of sample attrition in controlled treatment trials.
Journal of Nervous and Mental Disease 135(4) 1962:332-337.

Leon AC, Mallinckrodt CH, Chuang-Stein C, Archibald DG, Archer GE,
Chartier K. Attrition in Randomized Controlled Clinical Trials:
Methodological Issues in Psychopharmacology. Biological Psychiatry
2006;59(11):1001-1005.

Atwood JR, Haase J, Rees-McGee S, Blackwell G, Giordano L, Earnest
D, et al. Reasons related to adherence in community-based field studies.
Patient Educ Couns 1992;19(3):251-9.

Aylward GP, Hatcher RP, Stripp B, Gustafson NF, Leavitt LA. Who goes
and who stays: subject loss in a multicenter, longitudinal follow-up study. J
Dev Behav Pediatr 1985;6(1):3-8.

Norris FH. Characteristics of older nonrespondents over five waves of a
panel study. J Gerontol 1985;40(5):627-36.

Ostbye T, Hill G, Steenhuis R. Mortality in elderly Canadians with and
without dementia: a 5-year follow-up. Neurology 1999;53(3):521-6.

Wolinsky FD, Arrnbrecht ES, Wyrwich KW. Rethinking functional limitation
pathways. Gerontologist 2000;40(2):137—46.

Wilson AJ, 3rd, Webber IL. Attrition in a longitudinal study of an aged
population. Exp Aging Res 1976;2(5):367-87.

O'Leary MR, Donovan DM, Chaney EF, O'Leary DE. Interpersonal

attractiveness and clinical decision in alcoholism treatment. Am J
Psychiatry 1 979; 1 36(4B):61 8-22.

72

 

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.

Helliwell B, Aylesworth R, McDowell l, Baumgarten M, Sykes E.
Correlates of nonparticipation in the Canadian Study of Health and Aging.
Int Psychogeriatr 2001 ;13 Supp 1:49-56.

Jacomb PA, Jomi AF, Korten AE, Christensen H, Henderson AS.
Predictors of refusal to participate: a longitudinal health survey of the
elderly in Australia. BMC Public Health 2002;2z4.

Bender BG, Ellison MC, Gleason M, Murphy JR, Sundstrom DA, Szefler
SJ. Minimizing attrition in a long-term clinical trial of pediatric asthma. Ann
Allergy Asthma lmmunol. 2003;91(2):168-76.

Honas JJ, Early JL, Frederickson DD, O'Brien MS. Predictors of attrition in
a large clinic-based weight-loss program. Obes Res. 2003;11(7):888-94.

lssakidis C, Andrews G. Pretreatment attrition and dropout in an outpatient
clinic for anxiety disorders. Acta Psychiatr Scand. 2004;109(6):426-33.

Krishnan E, Murtagh K, Bruce B, Cline D, Singh G, Fries JF. Attrition bias
in rheumatoid arthritis databanks: a case study of 6346 patients in 11
databanks and 65,649 administrations of the Health Assessment
Questionnaire. J Rheumatol. 2004;31(7):1320-6.

Sugisawa H, Kishino H, Sugihara Y, Shibata H. [Characteristics of
dropouts and participants in a twelve-year longitudinal research of
Japanese elderly]. Nippon Koshu Eisei Zasshi 2000;47(4):337-49.

Zeller M, Kirk S, Claytor R, Khoury P, Grieme J, Santangelo M, et al.
Predictors of attrition from a pediatric weight management program. J
Pediatr. 2004;144(4):466-70.

Downing RWUP, Rickels K, King L, Rickels LA. Factors influencing
dosage deviation and attrition in placebo treated neurotic outpatients.
Journal of Psychiatric Research 1975;12(4):239-256.

Given CW, Given BA, Coyle BW. Prediction of patient attrition from
experimental behavioral interventions. Nurs Res 1985;34(5):293-8.

Davis LL, Weaver M, Haberrnann B. Differential Attrition in a Caregiver
Skill Training Trial. Research in Nursing & Health 2006;29(5):498-506.

Matthews FE, Chatfield M, Brayne C. An investigation of whether factors
associated with short-term attrition change or persist over ten years: data
from the Medical Research Council Cognitive Function and Ageing Study
(MRC CFAS). BMC Public Health 2006;6z185.

73

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

Snow WM, Connett JE, Sharma S, Murray RP. Predictors of attendance
and dropout at the Lung Health Study 11-year follow-up. Contemp Clin
Trials 2007;28(1):25-32.

Lewis BA, Napolitano MA, Whiteley JA, Marcus BH. The effect of
preferences for print versus telephone interventions on compliance and
attrition in a randomized controlled physical activity trial. Psychology of
Sport and Exercise 2006;7(5):453-462.

Wadland WC, Soffelmayr B, Ives K. Enhancing smoking cessation of low-
income smokers in managed care. J Fam Pract 2001;50(2):138-44.

Holtrop JS, Corser W, Jones G, Brooks G, Holmes-Rovner M, Stommel M.
Health behavior goals of cardiac patients after hospitalization. Am J Health
Behav 2006;30(4):387-99.

Effect of enalapril on survival in patients with reduced left ventricular
ejection fractions and congestive heart failure. The SOLVD Investigators.
N Engl J Med 1991;325(5):293-302.

 

Effect of enalapril on mortality and the development of heart failure in
asymptomatic patients with reduced left ventricular ejection fractions. The
SOLVD lnvestigattors. N Engl J Med 1992;327(10):685-91.

Yusuf S, Pepine CJ, Garces C, Pouleur H, Salem D, Kostis J, et al. Effect
of enalapril on myocardial infarction and unstable angina in patients with
low ejection fractions. Lancet 1992;340(8829):1173-8.

Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of
classifying prognostic comorbidity in longitudinal studies: development and
validation. J Chronic Dis 1987;40(5):373-83.

Hlatky MA, Boineau RE, Higginbotham MB, Lee KL, Mark DB, Califf RM,
et al. A brief self-administered questionnaire to determine functional
capacity (the Duke Activity Status Index). Am J Cardiol 1989;64(10):651-4.

Kaul P, Armstrong PW, Fu Y, Knight JD, Clapp-Channing NE, Sutherland
W, et al. Impact of different pattems of invasive care on quality of life
outcomes in patients with non-ST elevation acute coronary syndrome:
results from the GUSTO-llb Canada-United States substudy. Can J
Cardiol 2004;20(8):760-6.

Koch CG, Khandwala F, Cywinski JB, lshwaran H, Estafanous FG, Loop
FD, et al. Health-related quality of life after coronary artery bypass
grafting: a gender analysis using the Duke Activity Status Index. J Thorac
Cardiovasc Surg 2004;128(2):284-95.

74

63.

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

Raloff L. The CES-D scale: A self-report depression scale for research in
the general population. J. Appl. Psychol. Meas. 1977;1:385-401.

Richardson J, Barkan S, Cohen M, Back S, FitzGerald G, Feldman J, et
al. Experience and covariates of depressive symptoms among a cohort of
HIV infected women. Soc Work Health Care 2001;32(4):93-111.

Thomas JL, Jones GN, Scarinci IC, Mehan DJ, Brantley PJ. The utility of
the CES-D as a depression screening measure among low-income
women attending primary care clinics. The Center for Epidemiologic
Studies-Depression. Int J Psychiatry Med 2001;31(1):25-40.

EuroQoI--a new facility for the measurement of health-related quality of
life. The EuroQol Group. Health Policy 1990;16(3):199-208.

Hillers TK, Guyatt GH, Oldridge N, Crowe J, Willan A, Griffith L, et al.
Quality of life after myocardial infarction. J Clin Epidemiol
1994;47(1 1 ):1287-96.

Given C, Given B, Rahbar M, Jeon S, McCorkle R, Cimprich B, et al.
Effect of a cognitive behavioral intervention on reducing symptom severity
during chemotherapy. J Clin Oncol 2004;22(3):507-16.

Sherwood P, Given BA, Given CW, Champion VL, Doorenbos AZ, Azzouz
F, et al. A cognitive behavioral intervention for symptom management in
patients with advanced cancer. Oncol Nurs Forum 2005;32(6):1190-8.

Ware JE, Jr., Sherboume CD. The MOS 36-item short-form health survey
(SF-36). l. Conceptual framework and item selection. Med Care
1 992;30(6):473-83.

Johnson PA, Goldman L, Orav EJ, Garcia T, Pearson SD, Lee TH.
Comparison of the Medical Outcomes Study Short-Form 36-ltem Health
Survey in black patients and white patients with acute chest pain. Med
Care 1995;33(2):145-60.

Peek MK, Ray L, Patel K, Stoebner-May D, Ottenbacher KJ. Reliability
and validity of the SF-36 among older Mexican Americans. Gerontologist
2004;44(3):418-25.

Yost KJ, Haan MN, Levine RA, Gold EB. Comparing SF-36 scores across
three groups of women with different health profiles. Qual Life Res
2005;14(5):1251-61.

Zeger SL, Liang KY. Longitudinal data analysis for discrete and
continuous outcomes. Biometrics 1986;42(1):121-30.

75

 

75.

76.

77.

78.

79.

80.

81.

82.

83.

84.

85.

Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalized
estimating equation approach. Biometrics 1988;44(4):1049-60.

Kleinbaum DG, Klein M. Logistic regression: a self-Ieaming text. 2nd ed.
New York: Springer; 2002.

Hosmer DW, Lemeshow S. Applied logistic regression. 2nd ed. New York:
Wiley; 2000.

Mehta KM, Simonsick EM, Penninx BW, Schulz R, Rubin SM, Satterfield
S, et al. Prevalence and correlates of anxiety symptoms in well-functioning
older adults: findings from the health aging and body composition study. J
Am Geriatr Soc 2003;51(4):499-504.

Mergl R, Seidscheck l, Allgaier AK, Moller HJ, Hegerl U, Henkel V.
Depressive, anxiety, and somatoform disorders in primary care:
prevalence and recognition. Depress Anxiety 2006.

Rapaport MH. Prevalence, recognition, and treatment of comorbid
depression and anxiety. J Clin Psychiatry 2001;62 Suppl 2426-10.

Beach EK, Maloney BH, Plocica AR, Sherry SE, Weaver M, Luthringer L,
et al. The spouse: a factor in recovery after acute myocardial infarction.
Heart Lung 1992;21 (1 ):30-8.

Glass TA, De Leon CF, Bassuk SS, Berkman LF. Social engagement and
depressive symptoms in late life: longitudinal findings. J Aging Health
2006;1 8(4):604-28.

Zunzunegui MV, Alvarado BE, Del Ser T, Otero A. Social networks, social
integration, and social engagement determine cognitive decline in
community-dwelling Spanish older adults. J Gerontol B Psychol Sci Soc
Sci 2003;58(2):SQ3-S100.

Frasure-Smith N, Lesperance F, Gravel G, Masson A, Juneau M, Talajic
M, et al. Social Support, Depression, and Mortality During the First Year
After Myocardial Infarction. Circulation 2000; 1 01 (16):1919-1924.

Brummett BH, Barefoot JC, Siegler IC, Clapp-Channing NE, Lytle BL,
Bosworth HB, et al. Characteristics of Socially Isolated Patients With
Coronary Artery Disease Who Are at Elevated Risk for Mortality.
Psychosom Med 2001;63(2):267-272.

76

 

86.

87.

88.

89.

90.

91.

92.

Ziegelstein RC, Fauerbach JA, Stevens SS, Romanelli J, Richter DP,
Bush DE. Patients With Depression Are Less Likely to Follow
Recommendations to Reduce Cardiac Risk During Recovery From a
Myocardial Infarction. Arch lntem Med 2000;160(12):1818-1823.

Matthews FE, Chatfield M, Freeman C, McCracken C, Brayne C. Attrition
and bias in the MRC cognitive function and ageing study: an
epidemiological investigation. BMC Public Health. 2004;4z12.

Kaplan RM, Atkins CJ. Selective attrition causes overestimates of
treatment effects in studies of weight loss. Addict Behav 1987;12(3):297-
302.

Agosti V, Nunes E, Ocepeck-Welikson K. Patient factors related to early
attrition from an outpatient cocaine research clinic. Am J Drug Alcohol
Abuse 1996;22(1):29-39.

Young AF, Powers JR, Bell SL. Attrition in longitudinal studies: who do
you lose? Aust N Z J Public Health 2006;30(4):353-61.

Zebracki K, Drotar D, Kirchner HL, Schluchter M, Redline S, Kercsmar C,
et al. Predicting Attrition in a Pediatric Asthma Intervention Study. J.
Pediatr. Psychol. 2003;28(8):519-528.

Given BA, Keilman LJ, Collins C, Given CW. Strategies to minimize
attrition in longitudinal studies. Nurs Res 1990;39(3):184-6.

77

 

   

ltiliijjijillijiililijii