University of Southern California Dissertations and Theses

The construct validity of the Health Utilities Index in patients with chronic respiratory disease in a managed care population

(USC Thesis Other)

The construct validity of the Health Utilities Index in patients with chronic respiratory disease in a managed care population

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INFORMATION TO U SERS
This manuscript has been reproduced from the microfilm master. UM I films
the text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the
copy subm itted. Broken or indistinct print, colored or poor quality illustrations
and photographs, print bleedthrough, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send UM I a complete manuscript
and there are missing pages, these w ill be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and continuing
from left to right in equal sections with small overlaps.
ProQuest Information and Learning
300 North Zeeb Road. Ann Arbor, M l 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
THE CONSTRUCT VALIDITY OF THE HEALTH UTILITIES INDEX IN
PATIENTS WITH CHRONIC RESPIRATORY DISEASE IN A MANAGED CARE
POPULATION
by
Patrick William Sullivan
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment o f the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)
December 2001
Copyright 2001 Patrick William Sullivan
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3073853
Copyright 2001 by
Sullivan, Patrick William
All rights reserved.
___ < g )
UMI
UMI Microform 3073853
Copyright 2003 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SOUTHERN CALIFORNIA
The Graduate School
University Park
LOS ANGELES, CALIFORNIA 90089 1695
This dissertation, written b y
Under the direction o f hJs.. Dissertation
Committee, and approved b y all its members,
has been presented to and accepted b y The
Graduate School, in partial fulfillment o f
requirements for the degree o f
DOCTOR OF PHILOSOPHY
r D t i
Desui o f Graduate Studies
Date August 6 , 2002
D/SSERTA TION COMMITTEE
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Dedication
This research is dedicated to Beatrice. Pauline. Florie Ann. Faustine. Liam and
Domitille. whose tremendous support, patience and understanding made this work
possible.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table of Contents
Section Title Page
Dedication ii
List of Tables v
1.0 Introduction
1.1 Purpose
1.2 Organization o f the Dissertation
I
2.0 Background 3
2.1 Respiratory Disease
2.2 The Need for Comprehensive Assessment o f Health Interventions
in Respiratory Disease: QALY
2.3 Utility Measures
2.4 Construct Validity
2.5 Measures o f Severity of Illness
2.6 The Health Utilities Index
2.7 Reliability and Validity o f the HUI
2.8 Severity o f Illness in Pharmacy Claims Databases
2.9 Asthma-Specific Severity of Illness
2.10 Validity o f the SF-36 in Chronic Respiratory Disease
3.0 Methodology 32
3.1 Study Objective
3.2 Population Assessed/Data Specifications
3.3 Variables Included in the Primary Data Set
3.4 Study Variables and Measures
3.5 Exclusion Criteria
3.6 Data Dropout
3.7 Response Rates
3.8 Study Hypotheses
3.9 Econometric and Statistical Models
4.0 Results 59
4.1 Descriptive Statistics
4.2 Validity o f the Clinical Model of Severity o f Illness
4.3 Validity o f the Chronic Respiratory Disease Scale (CRD Scale)
4.4 Construct Validity of the Health Utilities Index
4.5 Data Dropout
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ty i i / i
4.6 Response Rates
5.0 Conclusion 91
.1 Summary
.2 Contribution to the Field
.3 Limitations of the Current Research
.4 Future Research
6.0 Bibliography 98
7.0 Appendix 103
7.1 Chronic Disease Comorbidity Using the American Hospital Formulary Service
(AHFS) Medication Coding System
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Tables
Table Page
Table 2.1 Chronic Disease Score Algorithm 25
Table 2.2 Asthma Specific Severity Scale 29
Table 3.2 Chronic Respiratory Disease coding using Rx’ s
(AHFS coding) 38
Table 3.3 Severity o f Illness Clinical Algorithm for Chronic
Respiratory Disease Based on 1992 NAEPP Guidelines: Three
Categories 39
Table 3.4 Severity o f Illness Clinical Algorithm for Chronic
Respiratory Disease Based on 1992 NAEPP Guidelines: Two
Categories 40
Table 3.5 Chronic Respiratory Disease Scale (CRD scale) 41
Table 3.6 Data Description 45
Table 4.1 Sociodemographics for CRD Sample. USC/Kaiser Sample
and the California and U.S. Population 62
Table 4.2 Comparison o f Health Status (SF-36) of CRD Sample
with USC/Kaiser Sample and the U.S. Population at Endpoint (b) 63
Table 4.3 Health Utility (HUI2 and HU13) and Health Status
(SF-36 t-scores) for USC/Kaiser Sample and CRD Sample by
Severity at Endpoint (ti) - Unadjusted Means 64
Table 4.4 Estimated Effect o f Clinical Severity on HRQoL (SF-36
Scores) Using a Random Effects or Fixed Effects Model Formulation
(n=289, t=3) 70
Table 4.5 Estimated Effect o f Clinical Severity on Logged Total Cost
Using a Random Effects Model Formulation (n=289. t=3) 70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.6 Chronic Respiratory Disease Scale (CRD scale) and
Distribution o f Frequencies (n=289 at t=2)
Table 4.7 Estimated Effect of a Scale of Chronic Respiratory Disease
on HRQoL (SF-36) Using a Random Effects or Fixed Effects Model
Formulation (n=289. t=3)
Table 4.8 Estimated Effect of a Scale of Clinical Severity on Logged
Total Cost Using a Random Effects Model Formulation (n=289. t=3)
Table 4.8a Expected Value of Total Cost by Severity Using Smearing
Estimates from the Random Effects Linear Regression of Logged
Total Cost (n=289. t=3)
Table 4.9 Estimated Effect of Clinical Severity on Utility (HUI2 and
HUD) Using OLS at t=2 (n=609)
Table 4.10 Estimated Effect of a Scale of Clinical Severity (CRD
Scale) on Utility (HUI2 and HUD) Using OLS at t=2
Table 4.11 Estimated Effect of Clinical Severity on LAS Using a
Random Effects Model Formulation (n=289. t=3)
Table 4.12 Estimated Effect of Clinical Severity on LAS Using a
Random Effects Model Formulation (n=289. t=3)
Table 4.13 Estimated Effect of Clinical Severity on LAS Using a
Random Effects Model Formulation (n=289, t=3)
Table 4.14 Adjusted Means Comparing Patients with Severity
Classification at All Three Periods and Those with Fewer than
Three Periods
Table 4.15 Selected Results of Cross Sectional Regressions by Time
Including Dropouts Using OLS
Table 4.16 Complete Responses for SF-36 and HUD in General
Population Compared to CRD Population
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
1.0 Introduction
1.1 Purpose
The purpose of this dissertation is to examine the construct validity of the Health
Utilities Index (HUI) in Chronic Respiratory Disease (CRD) using a pharmacy
claims database. This will be accomplished by using established methods to
compare the HUI Mark II and Mark III (HUI2 and HU13) with external measures of
severity o f illness to determine whether the relationship is consistent with a priori
hypotheses. In addition, the secondary aim of the research is to develop and validate
a pharmacy-based measure o f severity o f respiratory disease. Two models o f severity
will be created from clinical guidelines and previous research in respiratory disease.
These two models will be validated using standard methods to compare their
association with total cost and health-related quality of life (HRQoL) as measured by
the SF-36.
1.2 Organization of the Dissertation
The next chapter will outline the motivation and background o f the proposed
research, as well as providing a literature review o f relevant subjects. Section 2.1
will discuss the burden of respiratory disease and the need for comprehensive
assessment methods. Section 2.2 will delineate the type of comprehensive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
assessment methods available with focus on the QALY. Section 2.3 will discuss the
methods available for utility elicitation. The theoretical process of construct
validation will be discussed in Section 2.4 followed by a discussion o f the rationale
behind the measures of severity used in this research. Sections 2.6-2.10 provide a
discussion o f the background and literature o f the HUI. generic and disease-specific
pharmacy-based risk assessment methods and the validity of the SF-36 in respiratory
disease.
Chapter three gives an outline of the various methodologies involved in the
research such as data exclusion criteria, an explanation o f the primary source of data,
a description o f the variables and measures used, study hypotheses and econometric
and statistical models and methods. Chapter four presents the results o f the analyses
and Chapter five provides a conclusion, including a summary of research findings,
limitations of the current research, contribution to the field and directions for future
research. The appendix includes the Chronic Disease Score coding system.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.0 Background
2.1 Respiratory Disease
Respiratory disease represents a significant burden to our society. Asthma alone
affects 14 million to 15 million persons in the U.S.. including an estimated 4.8
million children. It is the cause of 100 million days of restricted activity. 470.000
hospitalizations and 5.000 deaths annually (NAEPP. 1997). Among children, asthma
is associated with 200.000 hospitalizations and 13 million physician visits per year
(Taylor and Newacheck 1992). The total economic burden of asthma represents
$12.7 billion in direct and indirect costs in the U.S. alone (Weiss and Sullivan 2001).
Internationally, the estimated annual average societal cost o f asthma ranges from
$326 to $1315 per afflicted person (in 1991 U.S. dollars) (Global Initiative for
Asthma: National Institutes of Health 1995).
In addition, the literature demonstrates that more severe asthma is associated with
higher cost. In one study based on the National Medical Expenditure Survey.
Malone et al. found that 20% o f asthma patients were responsible for 80% o f all
direct medical costs (Malone. Lawson, and Smith 2000). In Spain. Serra-Batlles et
al. also found severity' o f asthma to be associated with significantly higher cost:
Expenditures for moderate patients were twice as high as expenditures for mild
asthma patients and likewise for severe versus moderate patients (Serra-Batlles et al.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1998). Perhaps an additional reflection of severity, higher rates of pharmacy
utilization are also associated with greater cost (Malone. Lawson, and Smith 2000).
In particular, higher rates of utilization of short-acting beta-agonists have been shown
to be associated with higher cost. Annual charges for asthma-related treatment were
3.0 times higher for high utilizers of short-acting beta-agonists than the average
patient with asthma ($1,347 versus $447) (Stempel et al. 1996).
As well as differences across levels o f severity, there appear to be gender
differences in the healthcare utilization associated with asthma. In the US. studies
show that women are more likely to utilize healthcare resources for asthma than men
are. In one study, the female-to-male ratio of asthma-related hospital admission was
nearly 3:1 for adults aged 20 to 50 years. After the age o f 50 years, females were
admitted for asthma at a rate o f approximately 2.5:1 compared with their age-
equivalent male counterparts (Skobeloff et al. 1992). This trend seems apparent
outside o f the U.S. as well. In Europe, asthma is more common in adult females than
males (Sunyer et al. 1997): and in Denmark, women have a higher risk of being
admitted to the hospital for asthma than men (Prescott. Lange, and Vestbo 1997).
Respiratory disease is clearly a tremendous burden to society in terms of
morbidity , mortality and expense, with a disproportionate weight falling on more
severe patients and women. Further investments in the research and development o f
novel interventions to alleviate this burden are much needed. However, any new
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
treatment will face close scrutiny as an additional expense in an environment of
ubiquitous strapped healthcare budgets.
2.2 The Need for Comprehensive Assessment of Health Interventions in
Respiratory Disease: QALY
Concern about the increasing cost of medical treatments within the confines of
limited societal resources has lead to increasing discernment requiring proof of value
for each dollar spent on health care. More comprehensive assessment methods can
help guide decision-makers to make rational choices in allocating scarce healthcare
resources. For example, expenditures on pharmaceuticals have increased
dramatically in recent years. As a result, pharmacy budgets have come under
pressure to control escalating costs. Often this results in the categorical reduction in
the pharmacy budget of a particular institution or health plan. However, the proper
utilization o f pharmaceuticals may result in very significant reductions in overall
health system costs, and equally important improvements in patient outcomes, both
clinical and humanistic (such as quality of life). Thus a more inclusive analysis of
the impact of increasing utilization o f select pharmaceuticals may substantiate an
increase rather than a decrease in the pharmacy budget.
Documenting these reductions in system-wide costs and/or improvements in
clinical and humanistic outcomes requires a comprehensive approach. One o f the
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
most compelling methods o f assessing the value o f healthcare interventions in a
comprehensive fashion is cost-effectiveness analysis. Cost-effectiveness analysis is
an analytical tool in which the costs o f an intervention are compared to the effects, or
health outcomes, to provide a ratio of the overall impact of an intervention (Gold
1996).
In addition to the importance of a complete identification o f the costs associated
with a given intervention, it is imperative that measures of effectiveness analyze the
impact of a given intervention in a comprehensive manner. As well as being
clinically meaningful, measures of effectiveness should incorporate outcomes that
are important to patients in order to more appropriately guide an optimal allocation
of resources. In particular, the evaluation of outcomes for patients with respiratory
disease should include measures of the impact of treatment alternatives on the health-
related quality of life (HRQoL) of patients. There are many instruments available to
measure HRQoL in respiratory disease, including both disease-specific measures like
the Asthma Quality o f Life Questionnaire (AQLQ) and generic measures like the SF-
36. However. HRQoL instruments are of limited value in conducting cost-
effectiveness and other economic analyses because of the inability to correlate
multidimensional measures of HRQoL with preferences and treatment cost in a
meaningful way.
Health Status Classification Systems (HSCS) and utility scales are a superior
means of collecting effectiveness information for use in economic analyses because
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
they incorporate HRQoL and provide a single measure of utility that is interval
scaled and preference based. The advantage of the preference-based measures of
utility, like the HUI. is the single score that captures the morbidity associated with a
particular disease; Researchers can then combine this measure o f morbidity with
mortality data in a summary effectiveness measure, such as the Quality Adjusted Life
Year (QALY). The QALY is a more comprehensive assessment o f patient health
outcomes than other measures of effectiveness because it incorporates health-related
quality of life with measures of mortality (as opposed to measures that only
incorporate mortality, such as the life year gained); One year of life gained as a result
o f chemotherapy when a patient is miserable, for example, may not be equivalent to
one year of life gained when a patient is in perfect health.
2.3 Utility Measures
The cornerstone of the QALY is the measurement of utility. There are many
available methods for the measurement of utility. These methods can be separated
into two categories: direct preference measures and multi-attribute health status
classification systems (MAHSCS). Examples of direct preference measures include
the standard gamble (SG). time trade-off (TTO) and visual analogue scale (VAS).
The VAS. also known as the linear analogue scale, consists of a straight line
anchored by the worst possible health state at one end and the on the other. The
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respondent manifests preferences by marking the desirability of the different health
states in rank order on the line. The VAS offers ease o f use. transparency and,
consequently, high rates of completion (Froberg and Kane 1989). Although the VAS
has been used to elicit preferences for health states quite extensively, it is not
theoretically compatible with the axioms o f Von Neumann Morgenstem (VNM)
expected utility theory because o f the lack o f choice alternatives and uncertainty
(Von Neumann and Morgenstem 1947).
The Time trade-off (TTO) method was developed by Torrance et al. as an
alternative to the standard gamble for ease o f administration (Torrance. Thomas, and
Sackett 1972). In the TTO approach, respondents are given a choice between two
certain alternatives and asked how much time in a state o f full health (x) would be
equivalent to more time in a less healthy state (t>x). The less healthy state is the
state for which the utility is being measured. The resulting ratio (x/t) is the utility the
person ascribes to that health state (Torrance 1986). Although the TTO does include
an alternative choice, it fails to incorporate the inherent uncertainty that is part o f
normal decision-making regarding future events. As a result, the TTO method, like
the VAS. violates the axioms of VNM expected utility theory.
The third method of direct utility elicitation is the standard gamble (SG). The SG
approach was first developed by Von Neumann and Morgenstem in accordance with
the principles of expected utility theory (Von Neumann and Morgenstem 1947). It
offers the respondent a choice between a certain outcome and a gamble. The certain
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
outcome is typically the health state o f interest compared to a gamble that will result
in full health with a probability o f p. and death with a probability of (1-p) (Torrance
1986). The respondents choose the probability, p, which would make them
indifferent between the certain alternative (the health state o f interest) and the
gamble. The resulting value o f p is the utility for the health state of interest.
Although burdensome to administer, the SG has foundations in VNM expected
utility theory and is thus the theoretically preferred method o f utility elicitation for
economic and decision analysis (Torrance et al. 1996).
The Health Utility Index (HUI). Quality o f Well-Being Scale (QWB) and EuroQol
(EQ-5D) are examples of multi-attribute health status classification systems
(MAHSCS) with empirically based utility functions capable o f measuring patient
preferences for numerous health states. As a result. MAHSCS provide a
comprehensive yet compact elicitation of health status (Feeny et al. 1999). O f these,
only the HUI uses the SG approach to elicit utilities for the respective health states
(Torrance. Boyle, and Horwood 1982). Hence, for reasons discussed previously, the
HUI is the only MAHSCS consistent with the theoretical axioms delineated under
VNM expected utility theory. While the direct preference measures may be
appropriate for specific applications, like the assessment o f health states for patients
in a particular study, the MAHSCS have a much broader appeal and are not restricted
to comparisons within specific populations. Because of the common framework of
HRQoL assessment based on the HSCS and the use of community preferences,
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
MAHSCS like the HUI facilitate comparisons of health interventions and health
status within and between populations. This enables their use in specific
applications, like assessing health status at points in time or changes over time in
clinical trials, as well as broader applications like comparing the cost-utility o f very
different societal interventions across different diseases and populations. For
example, the MAHSCS utilities provide a common framework capable o f comparing
very different interventions in very different populations such as the cost-utility of
seat belts for the entire nation versus anti-hypertensives for survivors o f myocardial
infarction.
In summary, there are many methods for eliciting preferences. As discussed, the
HUI is the only MAHSCS consistent with the axioms of VNM expected utility
theory and. therefore, appropriate for economic and decision analysis, such as
computing QALYs for cost-effectiveness analyses. (The background and
development of the HUI will be discussed in greater detail in Section 2.6 below.)
2.4 Construct Validity
The focus of the discussion thus far has been on the importance o f using a
comprehensive measure o f effectiveness when assessing the health outcomes
associated with respiratory disease. As discussed, the measurement o f utilities is the
basis of this type o f assessment in the QALY. The measurement o f QALYs is
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
important for comparing the cost-effectiveness o f different health interventions in a
comprehensive maimer and therefore providing guidance on the allocation of scarce
resources. The HUI is a theoretically sound MAHSCS capable of eliciting
preferences for numerous health states, from which QALYs may be calculated.
While the HUI is a theoretically sound method o f utility elicitation, it is crucial to
ensure that it is a valid measure of utility in respiratory disease. If the HUI is not
accurately measuring the morbidity associated with respiratory disease, biased
QALYs may result in dramatically underestimating the real value of health
interventions that treat respiratory disease. This has significant policy ramifications:
policy makers may base decisions on inaccurate information regarding the cost-
effectiveness of particular health interventions for asthma or other respiratory
diseases, causing a misal location of resources to undeserving interventions or
treatments away from more cost-effective interventions.
Therefore, it is imperative to assess the validity of the HUI as a measure of utility
in respiratory disease. Validity is the extent to which an instrument measures what it
is intended to measure (Staquet. Hays, and Fayers 1998). This is analogous to
throwing darts at a dartboard: the dart thrower aims (intends) to hit the center of the
target each time: validity would be represented by how close, on average, the darts
come to the center of the target. In contrast, reliability would be measured by how
consistently the darts hit the target of aim (McDowell and Newell 1996). Typically,
comparing the measure o f interest to a gold standard provides an assessment of
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
validity known as criterion validity. For example, comparing a new screening
method for detecting tuberculosis to the usual standard of care, the chest x-ray.
would provide insight on the criterion validity o f the new test. This process may
involve only one or two studies.
However, in the case of HRQoL or MAHSCS no gold standard measure exists
(McDowell and Newell 1996). Consequently, the simple assessment of criterion
validity is not possible. In this case the assessment of validity requires an iterative
process o f comparing the measure of interest to multiple indicators, known as
construct validation. The process of construct validation involves hypothesizing how
a measure should behave compared to other measures and confirming or
disconfirming these a priori hypotheses (Staquet. Hays, and Fayers 1998).
Specifically, it involves comparing the construct to be measured, in the present
research a MAHSCS. to external criteria with a clearly defined theoretical
relationship between the measure and the external criteria (hypotheses) (McDowell
and Newell 1996).
A thorough construct validation process o f a particular instrument would involve
three stages: first, specifying the domain o f the variables or delineating the
construct(s) to be measured; second, establishing the internal structure of the
observed variables; and third, verifying the theoretical relationships between the
instrument or construct(s) and external criteria (McHomey, Ware, and Raczek 1993;
Nunnally 1978). The third step is an ongoing process and reflects the iterative nature
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f construct validation: assessing the correlation with various measures o f interest to
further define and delimit the measurement capabilities of the instrument. For the
purposes o f examining the construct validity o f the HUI. the first two steps have been
documented extensively. The focus of the present research is on the last o f these
three steps where additional contribution is most needed. Although the construct
validity o f the HUI as a generic measure has been assessed extensively, its
measurement capabilities in respiratory disease have not been clearly established.
(The construct validity of the HUI as a generic and disease specific measure will be
discussed more thoroughly in the next section).
One specific method consistent with step three outlined above for assessing the
construct validity of HRQoL and utility involves comparing the instrument to
external measures of disease severity (Stewart and Ware 1992). This method has
been used extensively in the literature to validate many different instruments (for
examples see McDowell and Newell. 1996). In addition to direct measures of
severity, indirect measures that reflect severity have been used, such as utilization or
total health care cost. The present discussion will elucidate the application of this
method o f construct validation using three well-accepted examples.
In an analysis of the construct validity o f the physical and mental health constructs
o f the MOS SF-36. the authors used mutually exclusive groups differing in severity
o f medical and psychiatric conditions as the external criteria of comparison
(McHomey. Ware, and Raczek 1993). To distinguish patient groups differing by
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
severity, the authors used disease-specific severity scales. For example, to define
severity o f psychiatric conditions, an eight-item depression symptom scale was used
in addition to psychiatric diagnostic criteria. They assessed the validity of the
physical health construct by comparing it to groups differing in severity of chronic
medical conditions related to physical health. Similarly, they examined the validity
of the mental health construct by comparing it to groups differing in severity of
psychiatric disorders. The authors hypothesized that more severe psychiatric patients
w ould have lower HRQoL scores on the mental component o f the SF-36 than milder
patients and. likew ise, that more severe patients (in terms o f physical function-related
chronic disease) would have lower scores on the physical component o f the SF-36.
The results confirmed the a priori hypotheses and provided evidence o f the construct
validity o f the mental health and physical health components o f the SF-36.
Two other examples using both direct measures of severity of illness and indirect
measures like total cost in the construct validation process within respiratory disease
are described in detail in Section 2.9. Briefly. Revicki et al. examined the theoretical
hypothesis that utility is lower for more severe asthmatic patients in an attempt to
assess the construct validity of the Asthma Symptom Utility Index (ASUI) (Revicki
et al. 1998). The authors used two asthma-specific severity scales and the HUI2 as
external criteria to test this hypothesis and provide insight to the construct validity o f
the ASUI. In addition. Eisner et al. applied the same method to assess the construct
validity of an asthma-specific severity scale (Eisner et al. 1998). The authors used
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
four measures previously shown to correlate with asthma status as external criteria to
test the theoretical hypotheses that more severe disease is associated with lower
levels o f HRQoL and utility and higher levels o f utilization. The General Health and
Physical Function domains of the SF-36. the Asthma Quality o f Life Questionnaire
(AQLQ). subject-perceived asthma severity and measures o f utilization were used as
external criteria to test the hypotheses and examine the construct validity of the
asthma-specific scale. In addition, the discussion of the literature supporting the
validity o f the Chronic Disease Score in Section 2.8 provides several examples of
using total cost as the external criteria by which to examine the construct validity of
an instrument.
2.5 Measures of Severity of Illness
The present research will apply the method delineated in Section 2.4 above to
assess the construct validity o f the HUI in respiratory disease. Specifically, the
theoretical hypothesis to be tested is that lower levels of utility are associated with
more severe respiratory disease. Comparing the HUI to external criteria of severity
of illness will examine the nature o f this relationship, and shed additional light on the
measurement capabilities o f the HUI. Thus, in order to assess the construct validity
of the HUI in CRD. we must first determine an appropriate measure of severity of
illness.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, in searching for suitable measures of severity in CRD, health
researchers are faced with resource constraints. Health researchers in many cases do
not have access to an unlimited number o f standard measures of severity due to
limited resources and lack of available data. The ideal criteria of severity in
respiratory disease may be comprised of several measures: clinical markers of
disease severity (such as FEV1 percent predicted): a disease-specific measure of
HRQoL (such as the Asthma Quality of Life Questionnaire (AQLQ)); a generic
measure o f HRQoL (such as the SF-36); measures of total cost or utilization; and
symptom history and utilization-based measures (such as the Asthma Disease
Severity Scale or Eisner's severity o f asthma scale). In practice, however, health
researchers often do not have such an unlimited breadth of options and must choose
one or tw o measures based on limited resources and limited availability o f clinical
data.
Given these resource constraints and lack o f available data, it is beneficial for the
field to develop and validate new methods for measuring severity in CRD that can be
used commonly by health researchers. Health researchers typically have access to
pharmacy utilization data and. consequently, pharmacy-based risk assessment
methods (such as the Chronic Disease Score) have recently burgeoned in interest and
application (Fishman and Shay 1999). Several pharmacy-based risk assessment
measures have been shown to be reliable and valid measures o f severity o f illness
(these measures are described in greater detail in Section 2.8). The further
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
development and validation of pharmacy-based risk assessment methods for
respiratory disease would significantly contribute to the field and would provide a
valid measure o f severity of illness for the purposes o f examining the construct
validity of the HUI.
Therefore the current research will develop and validate two retrospective data-
based measures o f severity in CRD. After separately validating these two measures
of severity, they will be used as the external criteria by which to assess the construct
validity of the HUI. The HUI will be compared to the two data-based measures of
severity of illness in accordance with the methodology for assessing construct
validity described above. Similar to previous research discussed above, the a priori
hypothesis is that high levels o f severity are associated with low levels o f utility. The
extent to which this relationship holds true in the current research will shed light on
the construct validity of the HUI in respiratory disease.
2.6 The Health Utilities Index
The Health Utilities Index is widely used to assess utility values for cost-
effectiveness analysis. The HUI is a comprehensive Health Status Classification
System (HSCS) complemented by a multi-attribute utility function (MAUF) capable
o f defining 24.000 (HUD) or 972.000 (HUD) unique health states and corresponding
preferences. The HUD system consists o f six attributes: sensory, mobility, emotion,
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
cognition, self-care and pain (Feeny et al. 1995). Each o f these attributes contains 4
or 5 levels o f function ranging from inability to function to normal capacity. For
example, the attribute cognition ranges from level 6 ("Unable to learn and
remember") to level 1 ("Leams and remembers normally for age”). The more recent
HUI3 system is comprised of eight attributes: vision, hearing, speech, ambulation,
dexterity, emotion, cognition and pain (Feeny et al. 1995). Each of these attributes
contains 5 or 6 levels of function ranging from inability to function to normal
capacity. For example, the attribute vision ranges from level 6 ("Unable to see at
all") to level 1 ("Able to see well enough to read ordinary newsprint and recognize a
friend on the other side of the street, without glasses or contact lenses"). In the HUI
system, function is defined by capacity' as opposed to performance or preference for
performance.
As discussed in the introduction, one o f the advantages of the HUI system in
economic evaluation is its ability' to capture morbidity in a single measure, which
allows researchers to combine measures of morbidity' with mortality. As a result of
the multi-attribute utility function, a global utility score can be derived and applied to
survival data to determine quality-adjusted life years (QALY). Based upon the levels
of function in each of the attributes, the HUI can be used to determine a community-
based utility score for a specific health state (Torrance et al. 1996).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.7 Reliability and Validity of the HUI
Previous research has provided evidence o f the reliability and validity o f the HUI.
Boyle et al. examined the test-retest reliability o f the HUI3 in the Canadian General
Social Survey using telephone interviews (Boyle et al. 1995). Reliability varied by
attribute with substantial kappa estimates (0.61-0.80) for vision, ambulation and
emotion: moderate estimates (0.41-0.60) for hearing, cognition and pain: and low
estimates for speech (0.137) and dexterity (0.347). However, the low kappa
coefficients for speech and dexterity may have been due to the small number of
subjects experiencing deficits in those attributes. The authors also found substantial
test-retest reliability for the global HUI3 score (intra-class correlation coefficient of
0.77).
In addition. Gemke et al. have demonstrated the inter-rater reliability o f the
attributes of the HUG in a population o f children admitted to intensive care (Gemke
and Bonsel 1996). The reliability of the global HUB score has been shown among
stroke patients and their caregivers in the assessment of patient health status and
HRQL (Mathias et al. 1997). In this research. Mathias et al. found moderate to high
levels o f inter-rater reliability. Raina et al. found high test-retest reliability o f the
HUB system in several Canadian population health surveys (Raina et al. 1999).
Evidence of the reliability o f the scoring formula has been demonstrated for the HUI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Mark 2 (HUI2). the predecessor to the HUI3. in a sample o f the general population
and in a sample of parents o f childhood cancer patients (Feeny D 1996).
In addition to having substantial evidence supporting its reliability, previous
research has also documented the construct validity o f the three versions of the HUI.
Gold et al. found evidence to support the predictive validity of the HUI in the
National Health and Examination Survey I Epidemiologic Follow-up Study (NHEFS)
(Gold. Franks, and Erickson 1996). Results indicated that a version o f the HUI.
constructed from the NHEFS. predicted future hospitalizations, mortality and self-
rated health.
There is evidence of the construct validity o f the HUD between particular disease
states. Grootendorst et al. provide evidence that supports the construct validity o f the
HUD for the measurement of health-related quality o f life (HRQOL) and attribute-
specific morbidity in the Ontario Health Survey, a population health survey
(Grootendorst. Feeny. and Furlong 2000). They found that HUD measures of global
utility were lower for subjects with stroke, arthritis and both conditions compared to
those with neither condition. In addition, attribute-specific utility scores were
consistent with clinical expectations: Stroke subjects had lower attribute-specific
utility scores for speech and cognition while arthritis subjects had slightly higher
scores for pain.
Although there is a significant body o f literature supporting the validity of the
three versions o f the HUI. studies examining the construct validity o f the HUI in
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respiratory disease have had mixed results and are limited to the HUI and HUI2.
Juniper et al. compared the validity of four measures of health-related quality of life
(HRQL) in pediatric asthma patients: the Pediatric Asthma Quality of Life
Questionnaire (PAQLQ). the Health Utilities Index (HUI). the Feeling Thermometer
and the Standard Gamble. The authors found that the PAQLQ performed the best in
discriminative and evaluative measurement properties, followed closely by the
Feeling Thermometer. Measurement properties in the Standard Gamble were
weaker. In particular, they documented that, although the HUI was reliable in
pediatric asthma patients, it was unable to measure asthma-related severity of disease
(Juniper et al. 1997).
In contrast. Revicki et al. found the HUI2 to be a valid measure of utility in adult
asthmatics. As mentioned briefly in the introduction, they analyzed the correlation
between severity o f illness and HUI2 scores in 161 adults with asthma (59% female,
mean age 35 +/- 11 years) (Revicki et al. 1998). The mean HUI2 score was 0.84.
ranging from 0.17 to 1.01; the median was 0.90 with a mode of 1 (n=31). The
distribution was peaked with a kurtosis o f 3.28. The HUI2 showed a consistent
correlation with two multi-dimensional severity-rating schemes, the Physician
Severity Rating Scale (PSRS) and the Asthma Disease Severity Scale (ADSS). The
PSRS is a physician global assessment o f the severity of asthma on a scale of 1
(mild) to 6 (severe) based on pulmonary function tests and medical and symptom
history information. Patients were classified into one o f four groups: mild;
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mild/moderate: moderate and moderate/severe. The ADSS is a composite of
resource use. spirometry and symptoms with a final score ranging from 0 to 7.
Patients were categorized into the same four groups as for the PSRS. The HUI2
discriminated between asthma severity groups under both schemes: PSRS (F 3.151 =
5.91. p<0.001. R squared = 0.11) and ADSS (F 3.153 = 6.28. p<0.001. R squared =
0.11). However, the HUI2 was not able to differentiate four distinct severity
classifications in post hoc analyses: mild from either mild/moderate or moderate or
mild/moderate from moderate. In the same study, the construct validity o f the HUI2
was supported by its correlation with the Asthma Symptom Utility Index (ASUI). an
asthma-specific measure of utility. The Pearson correlation coefficient between the
ASUI and the HUD was statistically significant in this population o f adult asthmatics
(r = 0.36. p<0.001).
In conclusion, the literature regarding the construct validity of the HUI in CRD
has produced ambiguous results. Although both the HUD and HUD appear to be
valid and reliable measures of generic utility, their validity in respiratory disease is
not clear.
2.8 Severity of Illness in Pharmacy Claims Databases
The most widely used measure o f generic severity of illness in retrospective
claims data analysis is the Chronic Disease Score (CDS). The CDS is a scale that
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
measures chronic disease status as a function of prescription claims aggregated over
1-year periods (Von Korff. Wagner, and Saunders 1992). In designing the CDS. a
multidisciplinary group o f physicians, pharmacists, epidemiologists and health
services researchers defined consensus scoring rules based on the following
principles: the CDS should increase with the number of chronic diseases under
treatment, but not with the number o f times a particular class o f medication was
used: the CDS should increase with the complexity of the regimen used to treat a
given chronic disease: potentially life-threatening or progressive diseases should
receive a higher score than stable and benign diseases; and medication regimens
contributing to the score should target diseases, not symptoms. Based on these
principles, the group used a consensus judgment process to identify weights assigned
to different drugs. Multiple medications from the same class were only counted as
one prescription. The resulting scoring rules are listed in Table 2.1.
A variety o f analyses were conducted to demonstrate the stability and validity of
the CDS. The CDS was correlated with primary care physician ratings o f the severity
o f physical disease (well. mild, moderate, severe) in a pilot sample o f high utilizers,
including the top 10% o f total ambulatory visits for the patient's age-sex group in
1985 (n = 219. Pearson correlation, r = 0.57. p < 0.05). In a second random sample
o f 722 patients in the panel of one physician. CDS was also correlated with physician
rating of severity o f illness (r = 0.46. p< 0.05). In a probability sample o f HMO
enrollees, the correlation of CDS with self-rated health status was significant and
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
slightly higher than the correlation between physician visits and self-rated health
status (n = 1016. Pearson correlation, r = 0.23. p < 0.05 vs. r = 0.19). CDS was also
found to be correlated with functional disability in a sample o f 2247 patients aged 65
or older (chi squared = 180.9. d f - 12. p< 0.001). In addition. CDS was strongly
associated with subsequent probability of death and probability of being hospitalized.
The reliability, construct and predictive validity o f the CDS were replicated in a
subsequent study of automated pharmacy data in the Northwest Region o f Kaiser
Permanente (Johnson. Hombrook. and Nichols 1994). Johnson et al. demonstrated
the test-retest reliability o f the CDS over time, both quarterly and yearly. They also
documented the construct validity o f the CDS by comparing it to the SF-36 and the
BSI-8 depression screener. The predictive validity was replicated by comparing the
CDS to subsequent health care visits and hospitalizations. The results were similar
to those obtained by Von Korff et al. In addition, the CDS was a highly significant
predictor o f next year's total medical care expense per person, after adjusting for age
and gender in a health risk-assessment model: In a split-sample analysis, the CDS
model explained 2.4% of the variation in expense in the estimation sample and the
error in the mean predicted expense was less than 3% in the second sub sample.
In conclusion, the CDS is a valid and reliable measure for generic chronic disease.
The methods used in developing and validating the CDS as a pharmacy-based
measure may be useful in the creation and validation of a respiratory-specific
measure o f severity.
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.1 Chronic Disease Score Algorithm
Qifwijf disease Mediation chutes) « « g niles
Heart disease (1) Anti-coagulaats, hemostatics
(2) Caidiae agents. ACE inhibiton
(3) Diuieac loop
One dam * 3
Two dasses-4
Three dasaes - S
Respiratory dinar (1) Isoproterenol
(2) Beta-adrenergic, misc.
(3) Xanthine products
(4) Respiratory products indudmg
bronchodOaton and mucoiytics
but exdudmg cromolyn
(5) Epinephrine
One dam * 2
Two or more dasaes- 3
AsthflUt rheumatism Glucocorticoids Score- 3
Rheumatoid arthritis Gold salts Score* 3
Cancer Anuneopiasba Score *3
Parkinson's L-Oope Score- 3
Hypertension (1) Andhypcrtsnsivcs
(except ACE inhibiton)
or calcium channel Mockers
(2) Beta blockers.
Diuretics
I f d a u ( t) - 2
If dam (2) 4 not (1) - 1
Diabetes Insulin
Oval hypoglycemics
Score * 2
Epilepsy Anticonvulsants Score * 2
Asthma, rhinitis Cromolyn Score * 2
Acne (1) Andacae tretinoin
(2) Topical macrolides
fitte r dam with
2 ♦ prescriptions file d * I
Uken
fim tiM B iw
Score-1
Glaucoma Ophthalmic miotics Score- 1
Gout, hypsniricsmia Uric acid agents S co re-1
Hi|h cholesterol Aatilipcmics Score-1
Migraines Ergot derivatives S co re-1
Tuberculosis Antitubercular agents S co re-1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.9 Asthma-Specific Severity of Illness
The CDS has been shown to be a reliable and valid measure o f generic severity of
chronic disease. The current research will focus on the development of a method of
measuring disease-specific severity o f illness in chronic respiratory disease based on
pharmacy utilization data. However, there is limited research on the measurement of
severity o f illness in chronic respiratory disease. Most research has examined the
development o f asthma-specific measures o f severity o f illness, but not specifically
based on pharmacy claims data. Typically, measures o f severity o f illness in
respiratory disease rely on pulmonary function and other clinical markers unavailable
in pharmacy claims data. For example, as mentioned previously. Revicki et al. used
two multi-dimensional rating scales to demonstrate the construct validity of the
Asthma Symptom Utility Index (ASUI) and. indirectly, the HUI: the Physician
Severity Rating Scale (PSRS) and the Asthma Disease Severity Scale (ADSS)
(Revicki et al. 1998). The Physician Severity Rating Scale (PSRS) has been shown
to be a valid and reliable instrument for measuring severity o f illness in asthma
(Diemer FB 1997). Based on the asthma medical history instrument, the PSRS is a
physician global assessment o f severity o f illness in asthma on a scale from 1 (mild)
to 6 (severe). The PSRS includes measures o f pulmonary function, medical history
and symptoms. The Asthma Disease Severity Scale (ADSS) is based on the Asthma
Control Scale (Juniper et al. 1993). The ADSS includes measures of pulmonary
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
function (FEV1 percent predicted < or = 70%). symptoms (chronic cough or chronic
phlegm, chronic wheeze, chronic breathlessness and chronic nighttime symptoms)
and resource use (ER visits > or = 1 in the last year, hospitalizations > or = 1 in the
last year). Each item adds one point to the composite score, which ranges from 0 to
7.
The ADSS and PSRS are valid and reliable measures of severity of illness in
asthma. However, data on symptoms and pulmonary function are often not available
in retrospective claims analyses. Hence, it is imperative to find a measure of severity
of illness based upon pharmacy claims data and measures of resource use. Although
designed as a survey questionnaire. Eisner et al. developed and validated an asthma-
specific measure of severity o f illness that does not include measures of pulmonary
function (Eisner et al. 1998). Their scale is based on the "stepwise" treatment
approach outlined in the National Asthma Education and Prevention Program
(NAEPP) guidelines (National Asthma Education and Prevention Program (National
Heart Lung and Blood Institute) 1997). The authors use clinical judgment based on
the effect on asthma severity to determine the weights assigned to prescription drug
class and dose, recent asthma symptoms and whether the patient had ever been
hospitalized or intubated. The scale has 13 items and a score range from 0 to 28. The
authors used clinical judgment to devise the components rather than formal factor
analysis. The resulting 13 items are designed to measure four principle components
of asthma severity: frequency o f current asthma symptoms, use o f systemic
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
corticosteroids, use o f other medications and history of hospitalizations and
intubations. The questionnaire is reproduced in Table 2.2.
In an analysis of the construct validity' o f the scale, the authors demonstrate a
strong association between the asthma-specific severity o f illness scale and three
measures correlated with asthma status in patients treated by family practitioners: the
Asthma Quality o f Life Questionnaire (AQLQ): the General Health and Physical
Functioning domains of the SF-36: and subject-reported perceived asthma severity.
The scale was also associated with subject-reported emergency department visits for
asthma, urgent physician visits for asthma and restricted activity days in the last year.
In addition, the authors have previously demonstrated a strong correlation with work
disability', the asthma quality' of life instrument, health care utilization and pulmonary
function as measured by FEV1 percent predicted in patients treated by allergists and
pulmonologists (Blanc et al. 1996).
In conclusion, many instruments exist to measure severity o f illness in respiratory
disease. However, all depend on measures not available in pharmacy-based
utilization data. Eisner's severity o f asthma scale appears to be a useful and valid
measure of severity o f illness that could be amenable to use in retrospective
pharmacy claims analyses.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.2 Asthma Specific Severity Scale
S e v e rity -o f-A s th m a S co re
Ite m s* S c o re X <%>
S y m p to m s
1. T ro u b lin g s y m p to m s , last 2 w k
X o n c O 3 (2)
M in im a l 1 3 3 (22)
O c c a s io n a l 2 5 5 (37)
M o s t d a y s o r n ig h ts 3 3 8 (25)
E v ery d a y o r n ig h t 4 21 (14)
S y ste m ic c o rtic o ste ro id s*
2. E v e r u s e d 2 9 4 (63)
3. U s e d in p a s t y e a r 2 5 3 (35)
4. U s e d S 3 m o n th s of lost i yi 3 2 2 (15)
O th e r m e d ic a tio n u s e , last 2 w k
5- (5-A g o n is t by m c tc rc d -d o se
in h a le r
N o n e o 24 (lb )
U se < 2 p u ffs p e r d a y 1 18 (12)
U se 2 2 p u ffs p e r day 2 108 (72)
b. C.'orticoN teroid by' m e te ie d
c lo se in h a le r
N o n e u 70 (47)
U se < 2 0 p u ffs p e r d ay 1 77 (51)
U se 2 2 0 p u ffs p e r d a y 2 3 (2)
7 A n y c ro m o lv n bv m e te re d - 1 2 2 (15)
d o s e in h a le r
8 A n y ip ra tro p iu m b ro m id e by 1 13 (9)
m e te r e d - d o s e in h a le r
9. A n y th e o p h y llin e o r o th e r oral J 5 2 (35)
3 - a g o n is t
10 A n y a n tih is ta m in e . ‘ 1 79 (53)
d e c o n g e s ta n t, o r n a s a l sp ray
11. A n y h o m e n e b u liz e r u s e 1 3 0 (20)
H o s p ita liz a tio n s /in tu b a tio n s
12. E v e r h o s p ita liz e d for a s th m a 3 51 (34)
13. E ve r in tu b a te d fo r a s th m a 5 7 ( 5 )
*A t o ta l o f 13 in d iv id u a l s c a le ite m s a r e n u m b e r e d 1
to 13. M a x im u m p o s s ib le s c o re — 28. M e d ia n s c o r e * * 8,
2 5 — 7 5 th i n t e r q u a r t i l e r a n g e 5 — 13. C r o n b a c h 's a lp h a =
0 .7 5 .
r N o t m u t u a l l y e x c lu s iv e c a t e g o r i e s . S u b j e c t s c a n
r e c e iv e b e t w e e n O a n d 7 p o i n t s fo r s y s te m ic c o r t i c o
s te r o id u s e .
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.10 Validity of the SF-36 in Chronic Respiratory Disease
The current research will develop two models o f severity o f illness in CRD. In
order to demonstrate the construct validity o f these models, it is necessary to
compare them to an independent, previously validated measure of health status as
outlined in Section 2.4. One such measure is the Medical Outcome Study SF-36.
The SF-36 is a health status instrument designed to measure eight domains: physical
function: role limitation due to physical health problems; bodily pain; social
functioning; general mental health (psychological distress and psychological well
being); role limitations due to emotional problems; vitality (energy/fatigue): and
general health perceptions (Ware and Sherboume 1992). In addition, there are two
constructs created from the eight health concepts to calculate a mental health and a
physical health summary score (Ware and New England Medical Center Health
Institute 1994).
Although the SF-36 has been well validated as a measure o f general health status
(McDowell and Newell 1996). there are a limited number o f studies examining the
validity of the SF-36 in respiratory disease. Eisner et al., as previously discussed,
examined the relationship between severity of illness and health status by comparing
a severity scale with the General Health and Physical Function domains of the SF-36.
In this analysis, the SF-36. in particular the General Health and Physical Function
domains were highly correlated with severity o f illness in respiratory disease.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bousquet et al. examined the validity of the French version of the SF-36 in
patients with asthma (Bousquet et al. 1994). The authors found all domains of the
SF-36 to be highly correlated with severity o f illness in asthma. In addition, they
demonstrated the internal consistency of the SF-36 as a measure of HRQoL in adult
asthmatics. Other studies have used the SF-36 as a measure o f health status in
respiratory disease, but not in the context of validation. For example, McDermott et
al. used the SF-36 to measure HRQoL. in addition to other outcomes, to compare
emergency diagnostic and treatment unit (EDTU) care and inpatient care in the
management of acute asthma (McDermott et al. 1997). They found the PF. RE. SF,
MH and VT domains to be significantly better for EDTU vs. inpatient care.
However, this does not necessarily support the validity o f the SF-36 (or certain
domains o f the SF-36) in CRD.
In conclusion, the SF-36 is a valid measure o f general HRQoL. and evidence
supports its validity as a measure o f health status in respiratory disease. In particular,
it appears from the available literature that the most sensitive SF-36 domains to
respiratory-related morbidity are Physical Function and General Health.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.0 Methodology
3.1 Study Objective
The objective of this research is to examine the relationship between severity of
illness and utility for patients with chronic respiratory disease (CRD). To
accomplish this, both a clinical algorithm and a scale of severity o f illness in CRD
will be developed and validated using the construct validation process delineated in
Section 2.4. The CRD scale will be adapted from Eisner's scale to conform to the
pharmacy-based data set. To validate both the clinical algorithm and the CRD scale,
each model of severity of illness will be compared to two external criteria previously
validated in respiratory disease. First, both models will be compared to the SF-36 to
test the a priori hypothesis that high levels o f severity of illness are associated with
low levels of HRQoL. Second, each model of severity of illness will be compared to
total cost to assess the a priori hypothesis that higher levels of severity are associated
with greater total cost. After the clinical model and the CRD scale have been
validated as measures of CRD-related severity of illness, both will be compared to
the HUI (HUI2 and HUI3 scores) to assess the a priori hypothesis that high levels of
severity are associated with lower utility scores.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.2 Population Assessed/Data Specifications
Data come from the Kaiser Permanente/USC Patient Consultation Study. The
purpose o f this study was to compare patient outcomes from different models of
pharmacist consultations. A more comprehensive outline o f the study design has
been published previously (McCombs et al. 1995). Briefly, enrollees completed
surveys including sociodemographic characteristics, the SF-36 and overall health
assessed by a visual analogue scale at baseline (calendar year 1992). Follow-up
surveys were completed after the first year (April 1. 1993 to March 30. 1994) and in
the final 11 months (April 1. 1994 to February 28. 1995). The Health Utilities Index
was administered during the final follow-up survey. The data set included both a
random assignment set and a geographic area set. In the geographic area set
(n=4.600). enrollees in six counties in Southern California were assigned to
treatment group based on the location o f their service providers. In the random
assignment group (n=6.000). members of Kaiser Permanente were assigned
randomly to one of three Pharmaceutical Care models: the Kaiser Permanente model,
State model or control group.
In addition, all patients were stratified into four groups based on their use of
prescription medications. Three mutually exclusive groups o f high-risk patients were
specified to correspond to populations targeted for pharmacy consultation under the
Kaiser model: those taking five or more medications, those taking pre-specified
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
target medications, or both. A sample o f "normal" patients who filled at least one
prescription during the prior year of the survey but did not meet any of the high-risk
criteria was also selected.
Patients enrolled in the study completed surveys annually for three years (1992 to
1995). Seventy-five percent of the random assignment subjects (n=4.500) and sixty-
three percent of the geographic sample (n=2.606) responded to surveys at all three
periods. Data imputation methods were used to impute SF-36 domain scores where
possible. Patients with two or more missing domains were deleted. The resulting
data set contained 6.921 patients with complete responses.
3.3 Variables Included In The Primary Data Set
Sociodemographic: Variables include age. gender, ethnicity (Asian. Black. Latino.
.American Indian and Caucasian), marital status (married, single, divorced and
widowed), education (6 levels), employment (employed, unemployed, retired,
disabled and working in the home or school) and income (8 levels). These variables
were included in the baseline survey.
Health Status: The SF-36 was used to measure generic HRQoL at baseline and both
follow-up surveys. Survey responses from the eight SF-36 domains and the two
physical and mental health component scales were converted to normalized t-scores
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(z scores multiplied by 10, plus 50). The normalized t-scores result in a standard
mean o f 50 and a standard deviation of 10. Imputed values were created for patients
who failed to complete the survey questions required to calculate any one o f the eight
domains of the SF-36 as recommended by Ware et al (Ware et al. 1993). The
imputed values were based on statistical models that regressed each o f the domains
on the remaining seven domains and other patient characteristics. A random
component w as also added to the imputed value to reflect the variation in the study
population for the missing data. Patients with two or more missing values for SF-36
health domains were dropped from the database.
Utility: The HUI Mark-II and III (HUI2 and HUD) were used to measure health
utility during the third survey.
Value: A global visual analog scale was used to assess value at baseline and both
follow-up surveys. The global VAS consisted of a single question. "How would you
rate your overall health?" The subject was asked to place an X on a 100mm line
between 0 (worse than death) and 100 (excellent health). The score was then read in
millimeters with a range o f 0 to 100.
Chronic Disease: The original version of the Chronic Disease Score was computed
from pharmacy data for all three periods.
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Pharmacy Data: Detailed pharmacy information including prescription dose and date
was collected from Kaiser’s automated database.
Utilization Data: Number o f hospital, emergency room care, urgent care and
outpatient visits were collected for the duration o f the study.
3.4 Study Variables and Measures
Chronic Respiratory Disease (CRD)
The following method will be used to identify the presence of CRD for each year:
Any prescription for the treatment of asthma. COPD. chronic bronchitis, or
emphysema from the list in table 3.2 (a proxy for chronic respiratory disease in the
absence of a diagnosis code).
Severity of Illness: Clinical Algorithm
The clinical algorithm o f severity o f illness in CRD will be based on the 1991
clinical practice guidelines from the National Asthma Education and Prevention
Program of the National Institute of Health (National Heart Lung and Blood Institute.
National Asthma Education Program. Expert Panel on the Management o f Asthma
1991). According to the guidelines, patients should be categorized as mild, moderate
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
or severe with correlated prescription drug use (table 3.3). The 1992 guidelines were
used rather than the more recent revised guidelines o f 1997 to more accurately reflect
treatment patterns at the time of data collection (1992-1995). As discussed in section
2.9. previous research has failed to differentiate four distinct severity groups. In the
event that three categories are not supported by the severity of illness model, the
guidelines have been interpreted to create two. rather than three, mutually exclusive
categories for clinical severity of illness: mild and severe (table 3.4).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.2. Chronic Respiratory Disease coding using Rx's (AHFS coding)
Diagnosis | AHFS Class Drug (generic name)
1. Chronic Respiratory Disease:
Sympathomimetic agents 12:12:02 Albuterol sulfate
Ephedrine sulfate
Terbutaline sulfate
Sympathomimetic agents 12:12:04 Albuterol
Bitolterol Mesylate
Ipratropium Bromide
Isoetharine HCL
Isoetharine Mesylate
Pirbuterol
Salmeterol Xinafoate
Adrenals: Inhaled
corticosteroids
68:04 Beclomethasone Dipropionate
Flunisolide
Metaproterenol Sulfate
Triamcinolone Acetonide
Respiratory smooth muscle
relaxants
86:16 Aminophylline
Oxtriphylline
Theophylline
Theophylline Anhydrase
Theophylline/EPH
Theophylline-GG
Theophylline - Eph
Unclassified therapeutic
agents
92:00 Cromolyn Sodium
Nedocromil Sodium
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.3. Severity of Illness Clinical Algorithm for Chronic Respiratory
Disease Based on 1992 NAEPP Guidelines: Three Categories________
Mild Moderate Severe
Inhaled beta agonist
AND/OR Cromolyn
Inhaled beta agonist AND
(Inhaled corticosteroids
OR
Cromolyn)
Inhaled beta agonist AND
(Inhaled corticosteroids OR
Cromolyn) AND
Oral corticosteroids burst ( lOmg 4xday
for 1 week than 1 xday for 1 week WITH
OR WITHOUT
Theophylline (sustained release)
AND/OR
Oral beta agonist WITH
Extra beta agonist from metered dose
inhaler or nebulizer
WITH or WITHOUT any other agent
Any ONE of the
following:
Beta agonists:
Albuterol (< 1
can/mo.)
Isoetharine HCL
Isoetharine Mesylate
Pirbuterol
Bitolterol
Cromolyn Sodium
Inhaled beta agonist AND
(Inhaled corticosteroids
OR
Cromolyn) AND
Theophylline (sustained
release) AND/OR
Oral beta agonist
Uncontrolled beta agonist with any other
agent
Inhaled beta agonist AND
Theophylline (sustained
release)
Ipratropium with any other agent
Short course of oral
corticosteroids followed by
inhaled corticosteroids
>=1 CRD - related Hospital admit within
previous 12 months + moderate or severe
classification
Theophylline >=1 urgent care visit within previous 12
months + moderate or severe
classification
Saimeterol >= 1 E.R. visit within previous 12 months
+ moderate or severe classification
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.4. Severity of Illness Clinical Algorithm for Chronic Respiratory
Disease Based on 199" , NAEPP Guidelines: Two Categories
Mild Severe
Inhaled beta agonist
AND/OR Cromolyn
Inhaled beta agonist AND
(Inhaled corticosteroids OR
Cromolyn) AND
Oral corticosteroids burst ( lOmg 4xday for 1 week then 1 xday
for 1 week WITH OR WITHOUT
Theophylline (sustained release)
AND/OR
Oral beta agonist WITH
Extra beta agonist from metered dose inhaler or nebulizer
WITH or WITHOUT any other agent
Any ONE of the following:
Beta agonists:
Albuterol (< I can/mo.)
Isoetharine HCL
Isoetharine Mesylate
Pirbuterol
Bitolterol
Cromolyn sodium
Theophylline
Salmeterol
Uncontrolled beta agonist with any other agent
Inhaled beta agonist AND
(Inhaled corticosteroids OR
Cromolyn)
Ipratropium with any other agent
Inhaled beta agonist AND
Theophylline (sustained
release)
>=1 CRD - related Hospital admit within previous 12 months *
mild or severe classification
Inhaled beta agonist AND
(Inhaled corticosteroids OR
Cromolyn) AND
Theophylline (sustained
release) AND/OR
Oral beta agonist
>=1 urgent care visit within previous 12 months -r mild or
severe classification
>=l E.R. visit within previous 12 months + mild or severe
classification
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Severity of Illness: Chronic Respiratory Disease Scale (CRD Scale)
The newly created CRD scale will be used as a second measure o f severity of
illness in respiratory disease. The CRD scale is based upon Eisner's 18-item scale
and has been adapted to pharmacy claims data. The revised CRD scale contains 10
items with a maximum possible score o f 17. The CRD scale is reproduced in table
3.5. The methodology for establishing the validity of the CRD scale will be
discussed in Section 3.8 below.
Table 3.5 Chronic Respiratory' Disease Scale (CRD scale)
Item Points
Systemic corticosteroids:
1. Used at all during past 3 years
2
2. Used at all in the past year
2
3. Used more than 2 months in the last year
2
Other medications used in the last year:
4. Inhaled beta agonists 1
5. Uncontrolled inhaled beta agonist (>2 canisters per
month)
2
6. Inhaled corticosteroids 1
7. Cromolyn 1
8. Ipratropium 1
9. Long-acting beta agonists, oral beta agonists or
Theophylline
1
10. Hospitalization in last 3 years (for respiratory disease) 4
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Total Cost
Total cost (inpatient, outpatient and prescription drug cost) will be used as a
dependent variable to examine the validity o f both the CRD scale and the clinical
algorithm of severity of illness. Outpatient cost information is not available in the
data set. Total number of outpatient visits is captured, however, and will be used in
conjunction with external cost information to determine total outpatient cost. The
methods used to compute inpatient reimbursement and drug cost have been published
previously (Mccombs et al. 1997). Estimates o f outpatient costs come from
McCombs et al. (They use $70 as an average cost o f an outpatient visit, which is the
cost used by Kaiser Permanente management for resource allocation decisions).
Non-Respiratory Chronic Disease Score (NRCDS)
As mentioned previously, the chronic disease score is a stable, reliable and valid
means o f controlling for chronic disease severity. However, because the CDS has a
respiratory portion that contributes to the overall score, using it as a control for
chronic disease severity in the analyses of this research would result in severe
multicolinearity. As a result, the respiratory portion o f the original version of the
CDS will be omitted. The NRCDS will be calculated based on Von KorfFs original
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
algorithm excluding respiratory medications. The NRCDS will be used to control for
non-respiratory chronic disease severity.
3.5 Exclusion Criteria
For the current analysis, patients in the Kaiser/USC database were identified
based upon their utilization of one or more prescriptions for the treatment o f asthma,
COPD. chronic bronchitis, or emphysema (listed in table 3.2). Any medication listed
in table 3.2 was sufficient to include patients in the data set. Note the exception of
systemic corticosteroids. Systemic corticosteroid utilization independent of other
CRD-related drugs was not sufficient to include patients in the analysis because it
may not be specific to respiratory disease. CRD patients use systemic corticosteroids
as an adjunct to other medications; therefore it was not considered a primary means
of inclusion. Patients had to have taken one o f the listed prescriptions for at least 30
days to be included. The resulting data set included 3.116 patients who had taken at
least one o f the listed drugs for at least 30 days. However, these inclusion criteria
were not sufficiently specific to CRD patients. Therefore, the data set was further
restricted to include only those patients who could be categorized as having mild or
severe CRD at any one of the three periods according to the clinical algorithm
delineated in table 3.4. The resulting CRD sample contained 2.188 unique patients
with severity information for at least one period; 573 CRD patients have severity
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
information for all three periods, but may not have complete responses for selected
variables o f interest. There were 1.134 CRD patients in the baseline year, 1,319 in
year one and 1,045 in year two. However, the number of patients with complete
utility and health status responses varies by measure and time (table 3.6). The
number of patients in period 0 with complete SF-36 responses was 1.111: there were
723 with complete SF-36 responses in period 1 and 648 with complete SF-36
responses in period 2. At the endpoint (t2). there were 647 patients with complete
HUI2 responses and 679 with complete HUB responses. For the CRD sample used
in the cross sectional analysis o f utility, there were 610 patients with complete
responses on the HUB and all covariates and 638 patients with complete responses
on the HUB and all covariates. For the CRD sample used in the panel data analysis,
there were 289 patients with complete responses on all variables o f interest over all
three periods.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.6. Data Description
USC/KP 6.921 Complete responses
USC/KP 13.682 Hospital and survey data
CRD Rx 3.116 Table 3.2 criteria
CRD (table 3.4) 2.188 Table 3.4 criteria
CRD (table 3.4 by time) t=0: n= 1.332 t=l: n=1.499 t=2: n= 1.212
CRD Sample SF-36 and
covariates complete
responses by time
t=0 : n = l.l 11 t=l: n=723 t=2: n=648
CRD Sample HUI2 (t=2) 647 HUD complete responses
CRD Sample for cross
sectional regression (t=2)
610 HUD and covariates complete
responses
CRD Sample HUD (t=2) 679 HUD complete responses
CRD Sample for cross
sectional regression (t=2)
638 HUD and covariates complete
responses
CRD Sample for panel data
analysis (t=0,1.2)
289 Complete responses on all
variables over all periods
3.6 Data Dropout
In the panel data analyses that follow, patients who are not classified as mild or
severe (according to the table 3.3 guidelines) for all three periods were excluded.
Patients are categorized as mild or severe based upon prescription drug utilization. If
respiratory symptoms improve, it is likely that the patient will not continue to need
the same level of prescription drug utilization. It is possible, in particular with
asthmatics that form the majority of this population, that mild episodes are followed
by no drug utilization. As a result, the patient will not utilize the same prescription
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
drugs in the next period and thus may not be categorized as mild or severe. The
patients whose symptoms deteriorate will not dropout because their utilization of
prescription drugs will increase. Therefore, the hypothesized reason for certain
patients not being categorized for all 3 periods is that they improve, become too mild
to be classified as mild by the table 3.3 guidelines, and are thereby excluded from the
panel data analysis. On the contrary, it is possible that certain patients remain severe
(or progress in severity) but become recalcitrant to following their prescription drug
treatment. These patients would also dropout due to a lack of prescription drug
utilization, but they would be much more severe and costly.
In order to assess the nature o f the bias from excluding patients without
classification for all 3 periods, the analysis will determine whether dropout patients
are milder than those with severity classification at all 3 periods. The extent to
which dropout introduces bias in the CRD sample will be assessed by three methods:
First, patients who are not categorized as mild or severe for all three periods
(dropouts) w ill be compared to those who have been categorized as mild or severe
for all three periods by analyzing the differences in unadjusted mean logged total
cost. % o f patients categorized as mild. CRD scale. CDS. HUI2. LAS. SF-36 score
(all eight domains), age and gender. A t-test will be used to determine if the
difference is statistically significant.
Second, adjusted means will be compared to determine if the difference is
statistically significant when controlling for NRCDS, age, gender, income and
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
education. Ordinary Least Squares will be used to regress three different dependent
variables (log total cost, and the Physical Function (PF) and General Health (GH)
domains o f the SF-36) on DROPOUT (a dichotomous variable: DROPOUT = 1 if
patients are classified for less than 3 periods and DROPOUT = 0 if patients are
classified for all 3 periods) and the aforementioned control variables.
Third, the panel data analyses (discussed in detail in Section 3.9) will be
replicated for each cross section with dropout patients included. The cross sectional
results including dropout patients will be compared to the panel data results
excluding dropout patients. This comparison will provide evidence o f the effect of
any bias introduced due to excluding dropout patients in the panel data analyses.
3.7 Response Rates
In addition to excluding patients who were not categorized as mild or severe for
all three periods, patients with incomplete responses on any o f the model variables
are excluded from the panel data analysis. Response rates for the CRD sample are
compared to those for the larger Kaiser/USC sample in order to determine if
systematic differences exist. Response rates for the main variables o f interest (SF-
36. HUI2 and HUB scores) are also compared to determine systematic differences in
response.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.8 Study Hypotheses
The two primary- research questions will be addressed by the following three
hypotheses:
Construct validity of the HUI2:
Hi: An increase in severity of illness is associated with a decrease in utility for
patients with CRD.
Multivariate cross sectional (t=2) regression analysis will be used to assess the
relationship between utility (as measured by the HUI2 and HUI3) and severity of
illness (as measured by the clinical algorithm and CRD scale) controlling for age,
gender, income, education, and non-respiratory chronic disease severity.
Validity of the Severity of Illness Model:
Hi An increase in severity of illness is associated with a decrease in HRQoL for
patients with CRD.
Multivariate random effects panel data analysis will be used to assess the
relationship between HRQoL (as measured by the SF-36) and severity o f illness (as
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
measured by the clinical algorithm and the CRD scale) controlling for age. gender,
income, education, and non-respiratory chronic disease severity.
H3: An increase in severity of illness is associated with an increase in total cost
for patients with CRD.
Multivariate random effects panel data analysis will be used to assess the
relationship between total cost (outpatient, inpatient and prescription) and severity of
illness (as measured by the clinical algorithm and the CRD scale) controlling for age.
gender, income, education, and non-respiratory chronic disease severity.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.9 Econometric and Statistical Models
Econometric Methods
There are both cross sectional data structures and panel data structures in this
research. Where possible, the available panel data structure will be exploited using a
random-effects or fixed-effects model formulation. The advantage of the fixed or
random-effects formulation is the ability to control for unobserved individual
characteristics to avoid specification bias and improve efficiency o f the estimates
(Hsiao 1986). In the panel data models discussed in this chapter, the random-effects
model formulation is preferred because the nature of the inference drawn from the
sample appeals to a broader population. However. Mundlak criticized the random-
effects model formulation, stating that the intercept and the independent variables
cannot be independent (Mundlak 1978). In response to Mundlak’s criticism o f the
random-effects model formulation, the Hausman specification test will be used to
test for independence of the intercept and the independent variables. Hsiao shows
that Mundlak's model formulation to address the lack of independence is equivalent
to that of the fixed-effects model (Hsiao 1986). Therefore, if the Hausman
specification test shows that they are independent, the random-effects model
formulation will be used. If not. a fixed-effects formulation will be used to reduce
the risk o f specification bias.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The method o f choice for cross sectional data structures is ordinary least squares
(OLS) because it is the best, linear, unbiased estimator (BLUE).
The significance level for all analyses will be reported individually. Where
multiple statistical testing is conducted, the significance level will be corrected to
adjust for the number o f tests.
Construct Validity of the Clinical Model of Severity of Illness
Health-Related Quality of Life (HRQoL)
To analyze the relationship between HRQoL (SF-36) and severity of illness
(clinical model) the following panel data model is proposed:
Yk = fj + Oj + ) 6 i t +S'Zi t + u it
where t = 0. 1.2
a, is the individual-specific effect for individual i across time
Yu is the SF-36 score for CRD-related domains (a separate equation will be estimated
for each domain)
p is the effect o f mild disease severity
S„ =1 if patient i is severe at time t. 0 otherwise
Z„ is a (lxk) vector of K exogenous covariates. (Z ,„ ,..., Z k« ) = Z 'u (including
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
age, gender, income, education, and non-respiratory chronic disease severity)
uu is the stochastic error term
Total Cost
To analyze the relationship between total cost (inpatient, outpatient and
pharmacy) and severity of illness (clinical model) the following panel data model is
proposed:
=M + a i + }S u + S , Z u + u it
where t = 0.1.2; i = 1,2 N; N = the number of individuals with CRD
a, is the individual-specific effect for individual i across time
Ylt is the total cost
p is the effect o f mild disease severity
Stl =1 if patient i is severe at time t. 0 otherwise
Z n is a (lxk) vector o f K exogenous covariates, (Zi«- •••• Z ku) = Z'u (including
age, gender, income, education, and non-respiratory chronic disease severity)
un is the stochastic error term.
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Construct Validity of the CRD Scale
Health-Related Quality of Life (HRQoL)
To analyze the relationship between HRQoL (SF-36) and severity o f illness (CRD
scale) the following panel data model is proposed:
Yu = v + a i +)5il+SZit+uit
where t = 0. 1.2
or, is the individual-specific effect for individual i across time
is the SF-36 score (a separate equation will be estimated for each domain)
p is the pooled intercept
S„ is the CRD scale for individual i at time t
Z„ is a ( I xk) vector o f K exogenous covariates. (Zu, ..... Z k ,i)= Z'u (including
age. gender, income, education, and non-respiratory chronic disease severity)
ult is the stochastic error term
Total Cost
To analyze the relationship between total cost (inpatient, outpatient and
pharmacy) and severity of illness (CRD scale) the following panel data model is
proposed:
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ya = fj + a, + 0M„ + ]S„ + 8'Zit + u„
where t = 0.1.2: i = 1.2 N: N = the number o f individuals with CRD
a , is the individual-specific effect for individual i across time
Ylt is the total cost
ji is the effect of mild disease severity
S„ is the CRD scale for individual i at time t
Z„ is a (1 xk) vector of K exogenous covariates. ( Zu, Z k,i ) = Z'u (including
age. gender, income, education, and non-respiratory chronic disease severity)
u„ is the stochastic error term
Construct Validity of the Health Utilities Index Mark II and III (HUI2 and
HUB)
Clinical Model of Severity of Illness
To analyze the relationship between utility (HUI2 and HUD) and severity of
illness (clinical model), the following cross-sectional model is proposed for t = 2 :
Yj = a + ySt +8Zi+ « ,
Yj is the HUD, HUD or LAS score for individual i
a is the intercept and coefficient for mild severity o f illness
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S, =1 if patient i is severe. 0 otherwise
Z , is a vector o f covariates (including age, gender, income, education, and non-
respiratory chronic disease severity)
u, is the stochastic error term
Chronic Respiratory Disease Scale
To analyze the relationship between utility (HU12. HUI3 and LAS) and severity of
illness (CRD scale), the following cross-sectional model is proposed for t = 2:
Yj = a + ySj + S'Z, + w,
Yj is the HUI2. HUD or LAS score for individual i
a is the intercept and coefficient for mild severity of illness
S, is the CRD scale score for individual i at time t
Z , is a vector o f covariates (including age. gender, income, education, and non-
respiratory chronic disease severity)
u, is the stochastic error term
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The Visual Analog Scale
Clinical Model of Severity of Illness
To analyze the relationship betw een value (V AS) and severity of illness (clinical
model) the following panel data model is proposed:
Yu =M + a i +ySil+S'Zil+uil
where t = 0. 1.2
a, is the individual-specific effect for individual i across time
is the VAS score
p. is the pooled intercept
S„ = 1 if individual i is severe at time t. 0 otherwise
Z „ is a(Ix k ) vector of K . exogenous covariates. ( Zu, Z ku ) = Z'u (including
age. gender, income, education, and non-respiratory chronic disease severity)
utl is the stochastic error term
Chronic Respiratory Disease Scale
To analyze the relationship between value (VAS) and severity of illness (CRD
scale) the following panel data model is proposed:
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Yu = fj + ctj + }S U + S'Zit+ u it
where t = 0. 1.2
or, is the individual-specific effect for individual i across time
Ya is the VAS score
H is the pooled intercept
S u is the CRD scale for individual i at time t
Z „ is a (lxk) vector ofK exogenous covariates. ( Zut Z ku ) = Z'u (including
age. gender, income, education, and non-respiratory chronic disease severity)
u„ is the stochastic error term
Dropout
In order to assess the impact of excluding dropouts in the aforementioned panel
data analyses, three separate cross sectional regressions will be estimated including
dropout patients. The results o f these three regressions will be compared to the panel
data analyses to determine the extent of bias introduced by excluding dropouts in the
four panel data analyses.
The following four cross sectional models are proposed for each period (t = 0, 1
and 2):
Yj = a + y&i + S'Z, + u,
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Yj is the SF-36 domain score or logged total cost for individual i
a is the intercept
St = 1 if patient i is severe. 0 otherwise: or the CRD scale score for individual
Z , is a vector of covariates (including age. gender, income, education, and non
respiratory chronic disease severity)
ul is the stochastic error term
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.0 Results
4.! Descriptive Statistics
Demographics
The study sample for the current analysis (CRD sample) is approximately the
same age as the Kaiser/USC sample (median 50 years) but slightly older than the
U.S. population (U.S. Bureau o f The Census. 1995). The CRD sample contains a
much greater proportion of females than the Kaiser/USC sample or the nation
(68.2%)(table 4.1). This is not unexpected given the trends outlined in section 2.1.
The proportion of Blacks. Asians and Latinos is higher (18.4%. 5.2% and 12.0%)
than the nation: as is the proportion o f college graduates (29%). In summary, there
are no unexpected differences between the USC/Kaiser Permanente Patient
Consultation Study or the U.S. population and the current CRD sample.
Health Utility and Health Status
The CRD sample has lower mean raw scores for all eight domains o f the SF-36
than the USC/Kaiser Permanente sample and the U.S. population (table 4.2). This
reflects the fact that the CRD sample is a select group with chronic disease. In
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
addition, the USC/Kaiser Permanente sample was selected based on at least one
prescription fill during the demonstration period. Hence. SF-36 scores for this
sample are slightly lower than the nation.
Measures of utility (HUI2 and HUD) and all eight domains o f health status (SF-
36) are consistently lower for the CRD sample compared to the USC/Kaiser
Permanente sample (table 4.3). again reflecting the chronic disease status o f the CRD
sample. Furthermore, the particular domains o f the SF-36 with the largest
decrements relative to the USC/Kaiser Permanente sample reflect the respiratory
morbidity o f the CRD sample: The General Health and the Physical Function
domains were identified in Section 2.10 as the most valid measures of respiratory
morbidity. Domains related to Mental Health are the least sensitive to respiratory
morbidity. The decrements are most substantial for General Health and Physical
Function, while Mental Health and Role Emotional are the least substantial (table
4.3). SF-36 scores in table 4.3 are expressed in t-score metric, using the U.S.
population scores as the normalizing means. As discussed in Section 3.3. the result
is domain scores that cluster around fifty and standard deviations o f approximately
ten points.
The mean utility scores (HUD and HUD) and health status (SF-36 t-scores) for
varying degrees of severity' (as categorized by the clinical algorithm in Section 3.4,
table 3.4) are also shown in table 4.3. Mean utility scores are lower for severe
patients than for mild patients: more severe patients have consistently lower utility
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
scores for both the HUI2 and HUD. However, only HU12 scores are statistically
significantly different for severe patients versus mild at any level of significance less
than or equal to 0.05. HUI2 scores are statistically significantly different by severity
after adjustment for multiple tests (0.05/10 = 0/005). The question remains as to
whether these differences are clinically significant. Grootendorst, Feenyet al. have
suggested that a 0.03 point difference in HUD scores is significant because it
corresponds to differences of one level within an attribute score (Grootendorst.
Feeny. and Furlong 2000). This may or may not be the same as a clinically important
difference, but is the only previously published metric o f significance. By this
definition, the differences in table 4.3 appear to be significant (>=0.03).
In addition, all eight domains of the SF-36 are lower for greater severity of
respiratory disease. Again, the more sensitive domains of the SF-36 show a more
substantial absolute difference between mild and severe patients. The Physical
Function. Role Physical and General Health domains show the most substantial
absolute differences, while Mental Health shows the least. Furthermore, Mental
Health is the only domain for which there is no statistically significant difference
between severe and mild patients at any level of significance. Role Emotional and
Social Function were also not statistically significantly different after adjusting for
multiple tests, but were at the higher significance level o f 0.05. Severe patients were
statistically significantly different than mild for all other domains after adjusting for
multiple tests.
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ware et al. have suggested that a 5-point difference in SF-36 scores is clinically
significant (Ware et al. 1993). By this definition, all o f the differences are clinically
significant (>=5 points) except for General Health, which approaches clinical
significance.
However, the absolute, clinical and statistical differences that are reported in table
4.3 are unadjusted: A more thorough analysis following the descriptive statistics
section will illuminate the significance o f these differences.
Table 4.1. Sociodemographics for CRD Sample, USC/Kaiser Sample and the
California and U.S. Population _________________ _________________
CRD
Sample
(N=2,188)
USC/Kaiser
QoL Sample
(N=6.921)
US
Population*
Median age
(years)
50.0 50.0 35.2
Female (%) 68.2 63.9 51.2
Ethnicity (%)
Black 18.4 21.7 12.7
Asian 5.2 7.0 3.8
Latino 12.0 14.1 11.2'
White 64.4 56.5 82.5
Education (%)
College grad 29.0 19.8 23.9
* USA Statistics in Brief. 1999. http://www.census.gov/statab/
Persons of Latino or Hispanic origin in the US population may be of any race. Hence, the summation
of the proportions exceeds one hundred percent.
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.2. Comparison of Health Status (SF-36) of CRD Sample with
USC/Kaiser Sample and he U.S. Population at Endpoint fa)
CRD
Sample
Mean Raw Score
(N = U I2 )
USC/K.P
Consultation Study'
Mean Raw Score
(N=6.921)
US General
Population*
Mean Raw Score
(N=2,471)
SF-36 Domains
General Health
Perceptions (GH)
58.6 68.1 72.2
Role limitation due to
Physical problem (RP)
58.4 70.7 81.2
Bodily Pain (BP) 61.5 66.3 75.5
Mental Health (MH) 71.2 72.9 74.8
Physical Functioning (PF) 68.5 77.6 84.5
Vitality (VT) 52.6 57.4 61.1
Role limitation due to
Emotional problem (RE)
73.5 77.8 81.3
Social Functioning (SF) 70.9 78.4 83.6
‘ (McCombs et al. 1995)
* (Ware and New England Medical Center Health Institute 1994)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 43 . Health Utility (HUI2 and HUI3) and Health Status (SF-361-
scores) for USC/Kaiser Sample and CRD Sample by Severity at
Endpoint fo) - Unadjusted Means ________________________ ______
USC/KP
Sample
Mean (Std.
Dev.)
(N=6,921)
CRD Sample
Total
Mean (Std.
Dev.)
(N=I,2I2)
CRD Sample
Mild
Mean (Std.
Dev.)
(N=617)
CRD Sample
Severe
Mean (Std.
Dev.)
(N=595)
Health Utility Index
(HUI2)
0.*0(0.18) 0.77 (0.20) 0.78 (0.23) 0.14(0.18) +
Heath Utility Index
(HUI3)
0.85 (0.20) 0.81 (0.24) 0.83 (0.23) 0.80 (0.24)
SF-36 t-scores:
General Health
Perceptions (GH)
46.98 (10.42) 42.18 (11.31) 44.22 (10.96) 40.16 +
(11.42)
Role limitation due to
Physical problem (RP)
46.53 (11.83) 42.04 (12.77) 44.75 (10.56) 39.39 +
(12.79)
Bodily Pain (BP) 46.35 (10.75) 43.49 (11.42) 44.93 (11.21) 42.08 +
(11.66)
Mental Health (M H) 49.36 (9.97) 47.84 (10.85) 48.12 (10.33) 41.51 (10.69)
Physical Functioning
(PF)
46.36 (11.30) 41.65 (12.80) 44.93 (11.44) 38.43 +
(13.20)
Vitality (VT) M.30(10.16) 45.19 (11.23) 46.73 (10.56) 43.68 +
(11.50)
Role limitation due to
Emotional problem
(RE)
48.94 (10.94) 47.68 (11.92) 48.72 (11.11) 46.63 *
(12.31)
Social Functioning
(SF)
46.61 (11.33) 43.80 (12.74) 45.10 (12.29) 42.52 **
(13.05)
<0.05 ‘ *<0.01 -< 0.005
- Adjustment for multiple tests (0.05/10) = 0.005
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2 Validity of the Clinical Model of Severity of Illness
HRQoL
Initial results o f the clinical model of severity o f illness using 3 categories (mild,
moderate and severe as in table 3.3) indicated that 2 categories would better fit the
data: SEVERE was statistically significantly different than MILD for most domains
of interest and logged total cost. However. MODERATE failed to reach statistical
significance. Hence, the clinical model of severity o f illness categorized patients as
mild or severe according to the NAEPP guidelines delineated in table 3.4. This
clinical model was used for all analyses.
Based upon the results of the literature search in Section 2.10. the most sensitive
domains of the SF-36 in measuring HRQoL in respiratory disease are the Physical
Function (PF) and General Health (GH) domains. Consequently, although the results
include all 8 domains o f the SF-36 and the 2 component scales, the focus will remain
on the Physical Function (and by induction the Physical Function-related domains)
and General Health domains for the purposes o f validation in CRD. Due to the
inclusion o f all eight domains and two component scales, the significance level has
been adjusted for multiple tests.
Despite the fact that the model was based on clinical guidelines, none o f the
variables in table 4.4 reach clinical significance according to the metric proposed by
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ware, et al discussed above (>= 5 points). The clinical model of severity o f illness
was highly statistically significant, however, in explaining the variation in the
Physical Function domain and statistically significant in explaining the variation in
the General Health domain of the SF-36 (table 4.4). In addition, the direction o f the
effect was consistent with the a priori hypothesis: higher levels of severity were
associated with lower levels o f HRQoL (in particular Physical Function and General
Health). The Hausman specification test supported the use of the random effects
model formulation for both domains. These results are strong indicators o f the
construct validity of this model of severity of illness in respiratory disease.
In addition, the Physical Component Scale (PCS) and the Role Physical (RP) and
Social Function (SF) domains were statistically significant and in the expected
direction: Higher levels of severity were again associated with lower levels of
function. The Hausman specification test supported the use of the random effects
model formulation for the PCS and the Role Physical and Social Function domains.
The Mental Component Scale (MCS) and the Role Emotional (RE), Vitality (VT),
Mental Health (MH). and Bodily Pain (BP) domains, however, were not statistically
significant. For these dependent variables, all o f the parameter estimates for
SEVERE were in the expected direction (negative) except the Mental Health domain,
which was positive. O f these 4 domains and the MCS. the Hausman specification
test only supported the random effects model formulation for the Vitality domain.
The fixed effects model formulation was used for the other regressions.
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
As expected, the burden of respiratory morbidity is clearly captured by the
Physical Function - related domains. PF. RP. and PCS. as well as GH; but not by the
Mental Health - related domains, MH. RE and MCS. This provides further evidence
o f the construct validity o f the model in measuring severity o f respiratory disease.
The Non Respiratory' Chronic Disease Score (NRCDS) was highly significant and
negative in explaining the variation in all of the dependent variables, except the MCS
and the RE and MH domains. This result may be due to the fact that the burden of
illness in a CRD sample is predominantly captured by constructs related to physical
function or general health (such as the PF and GH domains o f the SF-36). Hence,
morbidity associated with mental health is likely not a significant burden in this
population. All three dependent variables for which the NRCDS was not statistically
significant are mental health-related constructs.
Other variables also demonstrated an impact on physical function-related
domains. Gender was statistically significant and negative in explaining the variation
in the Physical Function and Vitality domains, as well as the PCS. In addition, age
was statistically significant in explaining the variation in physical function-related
constructs: PF. RP and PCS. Income explained the variation in the PF. RP and SF
domains of the SF-36. None o f the variables had a statistically significant effect on
the Mental Component Scale (MCS). The impact o f SEVERE, NRCDS, gender, age
and income on physical-function related domains of the SF-36 appears to support the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
notion that the burden of illness in the CRD sample is related to physical function
rather than mental health.
In summary, the results confirm the a priori hypothesis that severity of illness is
associated with lower HRQoL (in particular the Physical Function - related and
General Health domains) and support the construct validity o f the clinical model as a
measure o f severity in respiratory disease.
Total Cost
The second step in establishing the validity o f the severity o f illness algorithm is
to examine the hypothesis that severity of illness is associated with greater total cost.
However, due to the non-normal distribution o f cost data, a monotonic, logarithmic
transformation has been applied to total cost resulting in logged total cost. From
table 4.5. the clinical model of severity of illness is highly statistically significant in
explaining the variation in logged total cost. As expected, the parameter estimates
are positive: higher levels of severity are associated with greater cost. Similar to the
CRD-sensitive domains of the SF-36. the NRCDS. age and gender are statistically
significant in explaining the variation in logged total cost. The Hausman
specification test supports the use of the random effects model formulation.
In addition, the adjusted R-squared is .24 suggesting that the clinical algorithm of
severity of CRD with sociodemographic covariates and the NRCDS explain a
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
substantial portion of the variation in logged total cost. (Clark et al. attained an R-
squared o f . 10 when regressing total cost on the Chronic Disease Score along with
sociodemographic covariates) (Clark et al. 1995).
As discussed previously, statistical significance does not necessarily imply
relevance or clinical significance. Smearing estimates have been calculated to
determine the relevance o f the statistical difference shown in the results in table 4.5
for the clinical model of severity. The smearing estimate is a nonparametric estimate
of the expected value of the response variable (total cost) on the untransformed scale
after fitting a linear regression model on the transformed scale (logged total cost)
(Duan 1983). The expected value of total cost after retransforming logged cost by
the smearing estimate results in a difference between mild and severe patients of
$1,973 per year (mild: $1,956: severe: $3,929). This difference is clearly
economically significant: $1,973 represents the cost of 28 office visits using Kaiser
Permanente's average cost o f an outpatient visit of $70 (Mccombs et al. 1997).
In summary, the results confirm the a priori hypothesis that severity o f illness is
associated with greater total cost and further support the construct validity o f the
clinical model as a measure of severity in respiratory disease.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.4 Estimated Effect of Clinical Severity on HRQoL (SF-36 Scores) Using
a Random Effects or Fixed Effects Model Formul
Variable PF GH PCS MCS RP RE VT MH SF BP
Intercept 55.14 43.54 54.86 44.30 51.99 56.71 46.31 44.19 46.80 66.26
- - -r
•
-r
* ••
-r T
SEVERE -2.52 -1.43 -2.07 -0.22 -1.73 -0.76 -0.89 0.04 -2.09 -1.09
-r
*
-
* **
NRCDS -1.51 -1.30 -1.62 -0.20 -1.37 -0.56 -1.25 -0.19 -1.47 -0.89
- -r -r -r - -r
*
Female -4.40 -1.70 -3.78 NA -3.04 NA -2.60 NA -2.52 NA
-
** * * *
Age -0.28 -0.06 -0.20 0.27 -0.17 0.01 -0.02 0.23 -0.02 -0.34
Income 1.09 0.56 0.69 NA 0.96 NA 0.68 NA 1.05 NA
** * *
Education 1.16
*
1.04 0.66 NA 0.37 NA 0.63 NA 0.17 NA
RSQ 0.18 0.06 0.14 0.70 0.09 0.60 0.05 0.73 0.06 0.75
Prob.>m 0.00 0.00 0.00 0.29 0.02 0.71 0.04 0.17 0.05 0.13
<0.05 ••< 0.01 -< 0.0045
- Adjustment for multiple tests (0.05/11) = 0.0045
Table 4.5 Estimated Effect of Clinical Severity on Logged Total Cost Using a
Random Effects Model Formulation (n=289, t=3) ______________________
Variable Parameter Estimate Probability > |T
Intercept
6.26051 0.0001
SEVERE
0.59517 0.0001
NRCDS
0.22984 0.0001
Female
0.25399 0.0016
Age
0.00730 0.0246
Income
0.00517 0.8482
Education
0.03318 0.3614
Model RSQ=0.2372 Prob. > m: 0.002
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3 Validity of the Chronic Respiratory Disease Scale (CRD Scale)
The median CRD scale score was 5. with a mean of 5.26. The maximum score in
this population was 14 out of a possible 17. while the minimum was 1. It is not
surprising that the minimum was not 0. given that patients were included in the
original CRD sample based upon prescription drug utilization (as discussed in
Section 3.5). The first and third quartile scores were 2 and 8. respectively. It is
evident from the comparison o f table 4.6 that there are very different patterns of
prescription drug utilization in the current CRD sample compared to the sample used
in Eisner's analysis. This could be the result of several factors, most importantly the
difference in time: the CRD sample was captured from 1992-1995 while Eisner's
study was published in 1998. In addition, the difference may be due in Part to
different sample selection criteria.
The distribution o f patients for each of the 10 items is evident in table 4.6. The
majority of patients have utilized systemic corticosteroids, inhaled beta agonists and
theophylline or oral beta agonists while a small minority have utilized inhaled
corticosteroids. This is in stark contrast to the current standard of treatment, but
reflects the utilization patterns at the time of data capture.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.6 Chronic Respiratory Disease Scale (CRD scale) and Distribution of
F req uencies (n=289 at t=2) _________________________________
Item Points N (%) Eisner (%)
Systemic corticosteroids:
1. Used at all during past 3 years
2
175 (61) (63)
2. Used at all in the past year
2
117(40) (35)
3. Used more than 2 months in the last
year
2
68 (24) (15)
Other medications used in the last year:
4. Inhaled beta agonists 1 174(60) (84)
5. Uncontrolled inhaled beta agonist
(>2 canisters per month)
2
26 (09) NA
6. Inhaled corticosteroids 1 60(21) (53)
7. Cromolyn 1 20 (07) (15)
8. Ipratropium 1 29(10) (9)
9. Long-acting beta agonists, oral beta
agonists or theophylline
1 162 (56) (35)
10. Hospitalization in last 3 years (for
respiratory disease)
4 76 (26) (34)
HRQoL
Similar to the clinical model o f severity of illness, the first step in validating the
CRD scale is to measure its association with HRQoL as measured by the SF-36 to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
confirm or disconfirm the hypothesis that severity is associated with lower HRQoL.
As discussed previously, the 2 most relevant domains in assessing respiratory-related
morbidity are the Physical Function and General Health domains.
As with the clinical model of severity, any discussion of significance must address
both statistical and clinical significance. Based on the parameter estimates for the
Physical Function domain, each 5.8 points on the CRD scale corresponds to a
clinically significant difference of 5 points on the Physical Function domain o f the
SF-36 (7 points for the General Health domain and 6.6 for the Physical Component
Scale). Thus the CRD scale can be classified into three separate groups, each
clinically significantly different than the others based on the total number o f points
received (mild 0-6. moderate 7-12 and severe 13-17. for example).
The CRD scale is highly statistically significant in explaining the variation in both
the Physical Function and the General Health domains o f the SF-36 (table 4.7). The
direction o f the effect is also consistent with a priori expectations: higher levels of
severity on the CRD scale are associated with lower levels of Physical Function and
General Health. In addition, the CRD scale is highly statistically significant in
explaining the variation in the Physical Component Scale (PCS) and the direction is
negative as expected. The Hausman specification test supports the use o f the random
effects model formulation for both domains and the PCS. These results confirm the
a priori hypothesis and support the construct validity o f the CRD scale as a measure
o f severity in CRD.
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Incidentally, the CRD scale was statistically significant in explaining the variation
in the Vitality (VT) and Social Function (SF) domains. The parameter estimates are
also negative: For the VT and SF domains, higher levels of severity on the CRD
scale are associated with lower levels o f function as expected. The Hausman
specification test supported the use of the random effects model formulation for the
PCS. but did not for the Vitality and Social Function domains. Hence, a fixed effects
model formulation was used for the VT and SF domains.
Similar to the clinical model of severity, results for the CRD scale suggest that the
burden of CRD is related to the morbidity associated with physical function rather
than mental health. Female and age were statistically significant in explaining the
variation in the PF domain and the PCS. but were not statistically significant for any
other domain. The Mental Component Scale (MCS) and the Mental Health (MH).
Role Emotional (RE), Vitality (VT) and Bodily Pain (BP) domains were not
statistically significant. All o f their parameter estimates were in the expected
direction (negative) except the Mental Health domain. O f these 4 domains and the
MCS. the Hausman specification test supported the random effects model
formulation for the Vitality domain: the fixed effects model formulation was used for
the regression of the MCS. RE. MH and BP domain on severity.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Total Cost
The second step in establishing the validity of the CRD scale is to measure the
association of the scale with logged total cost. The CRD scale is highly statistically
significant in explaining the variation in logged total cost (table 4.8). The direction
of the effect is consistent with the a priori hypothesis: Higher levels o f severity on the
CRD scale are associated with greater cost. In addition, the R-squared is even more
substantial than the clinical model of severity of illness (0.32). The Hausman
specification test also supports the use o f the random effects model formulation.
Again, statistical significance does not necessarily imply relevance when
examining logged total cost. Hence, smearing estimates have been calculated to
determine the relevance of the statistical difference shown in the results in table 4.8
for the CRD Scale. Results are shown in table 4.8a. The expected value of total cost
after retransforming logged cost by the smearing estimate is shown for each level of
the CRD Scale. Although each interval is not significantly higher than the previous,
there is a general trend to increasing total cost that correlates with the higher scores
on the CRD Scale. The largest cost ($11.819 for level 15) is significantly higher than
the lowest cost ($1,503 for level 1). This represents a substantial difference of
$10,316. In fact, an increase of greater than one point on the CRD Scale is
consistently associated with greater total cost. However, there are three values of the
CRD Scale that do correspond to a larger value of total cost compared to the value
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for the previous level (11,13 and 16); for example, receiving a 10 on the CRD scale
equates to $5,897. but receiving an 11 equates to $5,734. This is not surprising given
the variability in the source of a one-point difference in the CRD Scale outlined in
Table 3.5.
In summary, the results confirm the a priori hypothesis and provide additional
evidence of the construct validity o f the CRD scale as a valid measure o f severity in
respiratory disease.
Comparison of the Two Models of Severity of Illness
As a result of the common dependent variable and covariates for both models,
comparing the common and marginal explained sum of squares can elucidate the
relative robustness of each model. The common regression for both models using
Physical Function as the dependent variable and omitting the severity variable yields
an R-squared of 0.1626. Adding the severity variable from the clinical model
increases the R-squared to 0.1829. which corresponds to explaining 11% o f the
variation in the Physical Function domain. Adding the severity variable from the
CRD scale increases the R-squared to 0.2071. which corresponds to explaining
21.5% of the variation in the Physical Function domain. Hence, by the sum of
squares criteria, the CRD scale is a more robust model of severity in terms of
explaining variation in HRQoL scores.
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Using the same criteria for logged total cost, the common R-squared for both
models is 0.1359. Adding the severity variable for the clinical model increases the
R-squared to 0.2363. which signifies that the clinical model explains 42.5% o f the
variation in logged total cost. However, adding the variable for the CRD scale
increases the R-squared to 0.317. indicating that the CRD scale explains 59% of the
variation in logged total cost. These results again suggest that the CRD scale is a
more robust model.
Table 4.7 Estimated Effect of a Scale of Chronic Respiratory Disease on
HRQoL (SF-36) Using a Random Effects or Fixed Effects Model Formulation
(n=289, t-3) _____ _______________________________________________
Variable PF GH PCS MCS RP RE VT MH SF BP
Intercept 57.04 44.12 56.13 45.04
*
73.27 59.68
**
72.34 42.66
*
68.53
**
70.62
-r
CRD
SCALE
-0.86 -0.71 -0.76 -0.07 -0.20 -0.29 -0.49
*
0.13 -0.69
*
-0.42
NRCDS -1.44
**
-1.24 -1.57 -0.20 -0.73 -0.56 -0.60 -0.19 -0.57 -0.88
*
Female -4.03 -1.35 -3.85
**
NA NA NA NA NA NA NA
Age -0.26 -0.04 -0.21 0.27 -0.55 0.01 -0.21 0.23 -0.28 -0.35
Income 1.09
•*
0.58 0.76 NA NA NA NA NA NA NA
Education 1.10
*
0.98 0.77 NA NA NA NA NA NA NA
RSQ 0.21 0.07 0.16 0.70 0.71 0.60 0.79 0.73 0.72 0.75
Prob>m 0.04 0.00 0.01 0.24 0.14 0.87 0.11 0.06 0.12 0.12
<0.05 **<0.01 -< 0.0045
Adjustment for multiple tests (0.05/11) = 0.0045
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.8 Estimated Effect of a Scale of Clinical Severity on Logged Total Cost
Using a Random Effects Model Formulation (n=284>,t=3)
Variable Parameter Estimate Probability > |T|
Intercept 6.09412 0.0001
CRD SCALE 0.12349 0.0001
NRCDS 0.23124 0.0001
Female 0.22127 0.0017
Age 0.00519 0.0696
Income 0.00773 0.7432
Education 0.04566 0.1511
Model RSQ=0.3190 Prob. > m: 0.0395
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.8a Expected Value of Total Cost by Severity Using Smearing Estimates
from the Random Effects Linear Regression of Logged Total Cost (n=289, t=3)
CRD Scale Value Smearing Estimates: Expected Total Cost
1 $1,503
2
$1,569
j $1,844
4 $2,133
5 $2,675
6 $3,163
7 $3,470
8 $4,059
9 $4,669
10 $5,897
11 $5,734
12 $7,966
13 $7,653
14 $7,974
15 $11,819
16 $9,292
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.4 Construct Validity of the Health Utilities Index
With two valid models o f severity o f illness in CRD. we can begin to examine the
construct validity of the HUI2 and HUI3 in CRD. As discussed in the methods
Section, this will be accomplished by confirming or disconfirming the hypothesis
that severity of illness is associated with lower levels o f utility. In practice this will
be assessed by regressing HUI scores on the two models of severity o f illness to
measure their association.
The HUI2 and HUI3 do not appear to capture respiratory-related morbidity: HUI2
and HUD scores were not statistically significantly different for patients who have
severe respiratory disease compared to those who are mild (table 4.9). Furthermore.
HUD and HUD scores did not statistically significantly differ for varying levels o f
severity o f respiratory disease as measured by the CRD scale (table 4.10). Although
the direction of the parameter estimates for both SEVERE (table 4.9) and CRD
SCALE (table 4.10) are consistent with the a priori hypothesis, the parameter
estimates were not statistically significant for either the HUD or the HUD.
Other variables appear to have a statistically significant impact on utility scores:
the NRCDS. age. gender and education are statistically significant in explaining the
variation in HUD scores for both the clinical algorithm and the CRD scale model
specification; while the NRCDS. Age and Income were statistically significant in
explaining the variation in HUD scores.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In all four models, the F-test suggested a statistically significant model
specification. The adjusted R-squared for all four models varied from 0.11-0.14,
indicating that a significant portion o f the variation in HUI scores was left
unexplained by the models.
For the purposes o f comparison, a panel data analysis was performed regressing
the Linear Analogue Scale (LAS) on the 2 models of severity of illness. Although
the LAS is a measure o f value rather than utility, it provides a useful comparison o f
sensitivity to respiratory-related morbidity. Unlike the HUI. both the clinical model
of severity and the CRD scale are statistically significant in explaining the variation
in LAS (table 4.11 and table 4.12). The direction of the parameter estimates is also
negative as expected.
The LAS is not a methodologically sound Health Status Classification System
with a Multi-Attribute Utility Function consistent with Von Neumann Morgenstem
(VNM) expected utility theory like the HUI. It is simply a subjective measure of
general health. Nonetheless, it is interesting for the purposes of comparison to notice
that the measure enables patients to express their morbidity. In particular, the results
demonstrate that the LAS is capable o f measuring CRD-related morbidity.
In summary, the results do not confirm the a priori hypothesis that higher levels o f
severity are associated with lower levels o f utility. Hence, the results call into
question the construct validity of the HUI2 and HUI3 as measures o f utility in
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respiratory disease. Further research is needed before unambiguous conclusions can
be made.
These results appear to contradict those o f Revicki et al. discussed in section 2.6.
Their results supported the construct validity of the HUI2 by documenting a
statistically significant difference between severity classifications. However, the
authors used analysis o f variance methods to detect unadjusted mean differences
between severity groups. In fact, in the Descriptive Statistics section of the current
research, the unadjusted means were also statistically significantly different for HUI2
scores by severity group. The current research has argued that a more thorough,
multivariate analysis adjusting for possible confounders is necessary to properly
assess the relationship between utility and severity o f illness, thereby elucidating the
construct validity of the HUI2 and HUD. Indeed this section documents that this
type o f analysis produces results that contradict the unadjusted results found by
Revicki et al. and in the Descriptive Statistics section of this research. Thus the
different statistical methods used to assess the construct validity o f the HUD explain
the divergence in results between the current research and that o f Revicki et al.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.9 Estimated Effect of Clinical Severity on Utility (HUI2 and HUI3)
Using OLS at t=2 (n=609)_____ ______________ ______________ ________
Variable HU12
Parameter
Estimate
Probability >
|T|
HUI3
Parameter
Estimate
Probability
> |T|
Intercept 0.85411 0.0001 0.81558 0.0001
SEVERE -0.01886 0.2367 -0.01024 0.5614
NRCDS -0.02838 0.0001 -0.03390 0.0001
Female -0.00181 0.0095 -0.00222 0.9104
Age -0.04119 0.0241 -0.00194 0.0046
Income 0.00987 0.2269 0.02347 0.0003
Education 0.01341 0.0235 0.01600 0.0587
Model Adj R-sq:
0.1124
Prob.>F:
0.0001
Adj R-sq:
0.1374
Prob.>F:
0.0001
Table 4.10 Estimated Effect of a Scale of Clinical Severity (CRD Scale) on
Utility (HUI2 and HU13) Using OLS at t=2 _______________ _______
Variable HUI2
Parameter
Estimate
n=610; t=2
Probability
>|T|
HUI3
Parameter
Estimate
n=638; t=2
Probability
> |T|
Intercept 0.855 0.0001 0.826 0.0001
CRD SCALE -0.003 0.1757 -0.004 0.1025
NRCDS -0.029 0.0001 -0.033 0.0001
Female -0.002 0.0104 -0.003 0.8943
Age -0.042 0.0199 -0.002 0.0065
Income 0.010 0.2138 0.023 0.0003
Education 0.014 0.0218 0.016 0.0633
Model Adj. R-squared:
0.113
Prob.>F:
0.0001
Adj. R-squared:
0.141
Prob.>F:
0.0001
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.11 Estimated Effect
Effects Model Formulation
of Clinical Severity on LAS Using a Random
n=289, t=3)
Variable Parameter Estimate Probability > |T|
Intercept 66.84129 0.0001
SEVERE -3.10856 0.0090
NRCDS -2.54119 0.0001
Female -2.27655 0.2035
Age -0.03730 0.6047
Income 1.27516 0.0348
Education 0.13718 0.8658
Model R-squared=0.0632 Prob. >m : 0.017
Table 4.12 Estimated Effect
Effects Model Formulation
of Clinical Severity on LAS Using a Random
(n=289, t=3)
Variable Parameter Estimate Probability > |T|
Intercept 68.34271 0.0001
CRD Scale -0.90676 0.0001
NRCDS -2.47627 0.0001
Female -1.91653 0.2828
Age -0.01568 0.8277
Income 1.27277 0.0342
Education 0.02168 0.9786
Model RSQ= 0.0745 Prob. >m : 0.0180
4.5 Data Dropout
Comparison of Means for Dropouts
As previously discussed, patients are categorized as mild or severe based upon
prescription drug utilization (table 3.3). The motivation for this analysis is to
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
determine if patients are dropping out (not utilizing the same level of prescriptions)
because their symptoms are improving or because they are remaining severe (or
progressing) and becoming recalcitrant to treatment. In order to assess this dynamic,
the analysis will determine the extent to which dropout patients are milder than those
with severity classification at all 3 periods.
Based on the results from table 4.13, patients who are not classified as mild or
severe for all 3 periods (dropouts: < 3 periods) are statistically significantly milder
than those who are classified as mild or severe for all 3 periods consecutively (3
periods). In particular, the percentage o f dropouts who are mild ranges from 53% to
59% over the 3 periods compared to 36% to 41% for non-dropouts. This difference
is statistically significant for all 3 periods. In addition, the CRD scale, a measure of
severity o f respiratory disease, is highly statistically significantly lower for patients
who dropout compared to non-dropouts. Several other measures of health status
support the hypothesis that patients who are not classified as mild or severe for all 3
periods are milder: Logged cost is lower for dropouts; the Physical Function (PF) and
General Health (GH) domains o f the SF-36 and the linear analogue scale (LAS) are
higher for dropouts. These results are statistically significantly different between
patients who dropout and those who do not for t=0 and t= l. The direction is also
consistent for t=2, but does not reach statistical significance due to the limited
number o f dropouts.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, these results are based on unadjusted means. An analysis of adjusted
means elucidates the extent to which observed variables, in particular age. are driving
the differences in health status measures. Table 4.14 displays the results o f the
adjusted means analysis controlling for age. NRCDS. gender, education and income.
Again, controlling for observed covariates. it appears that patients who are not
categorized as mild or severe over all 3 periods are statistically significantly milder
than those who are. Dropouts are more likely to be mild (according to the NAEPP
guidelines in table 3.4). have better Physical Function and General Health scores, and
higher LAS scores over all 3 periods. In addition, dropouts are less expensive and
have milder respiratory disease according to the CRD scale over all 3 periods. All of
these adjusted means are statistically significantly different with the exception of
Physical Function and the Linear Analogue Scale in period 2.
In order to examine the effect of excluding milder patients from the panel data
analyses, dropouts were included in cross sectional regressions by period. The
results in table 4.15 indicate that the impact of including dropouts in cross sectional
regressions does not alter the conclusions drawn from the panel data analyses. The
variables of interest in the panel data analyses excluding dropouts remain statistically
significant and in the same direction in the cross sectional analyses including
dropouts. Although the magnitude of most o f the estimates is approximately
equivalent, the parameter estimates for the regression of SEVERE on PF vary
substantially. This is not surprising, however, given that the sample for the cross
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
sectional regressions (t=0: n= l.l 11: t= l: n=723; t=2: n=648) is substantially different
than the panel data sample (t =0. 1. 2: n=289). In addition, the estimation
methodology is not the same (OLS versus random/fixed effects). In summary, the
inclusion o f dropouts does not alter the conclusions drawn from the panel data
analyses excluding dropouts in the previous sections.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.13 Estimated Effect of Clinical Severity on LAS Using a Random
Effects Model Formulation (n=289, t=3)_____________________________
Variable Unadjusted Means
t=0 t=l 1= 2
3 periods <3
periods
3 periods <3
periods
3 periods <3
periods
MILD
0.41
n=677
0.53
n=655****
0.36
n=677
0.55
n=822****
0.41
n=677
0.59
n=535****
CRD scale
5.53
n=677
3.28****
n=655
5.80
n=677
3.32 ****
n=822
5.51
n=677
2.92 ****
n=535
CDS
2.56
n=646
2.33
n=610
3.42
n=645
2.96 ****
n=770
3.41
n=649
3.20
n=515
Log Cost
7.60
n=645
7.44
n=607**
7.62
n=644
7.40
n=768***
7.50
n=648
7.39
n=515
Utility
NA NA NA NA 0.77
n=369
0.75
n=278
LAS
59.87
n=632
61.72
n=591
59.85
n=441
63.09
n=481**
62.25
n=397
63.16
n=318
PF
42.13
n=642
44.31
n=600**
41.13
n=444
44.83
n=487****
41.30
n=404
42.10
n=320
RP
43.32
n=634
43.72
n=598
43.03
n=438
55.00
N=479*
42.20
n=397
41.85
n=316
RE
47.74
n=635
46.93
n=594
46.61
n=437
48.76
n=474**
47.80
n=397
47.53
n=315
VT
44.80
n=638
45.75
n=602
44.85
n=395
46.16
n=424
41.30
n=404
41.30
n=317
MH
47.41
n=638
46.81
n=602
47.44
n=394
47.39
n=424
48.03
n=403
47.60
n=317
SF
44.05
n=641
44.46
n=601
45.40
n=399
45.88
n=425
44.12
n=405
43.40
n=32l
BP
44.10
n=640
43.46
n=598
44.46
n=397
43.87
n=425
43.71
n=405
43.20
n=321
GH
41.75
n=636
44.31
n=596***
41.57
n=392
43.77
n=418**
41.44
n=399
43.11
n=313
Age
52.33
n=677
48.79
n=655****
53.28
n=677
50.18
n=822****
54.26
n=677
51.76
n=535**
Female
0.64
n=646
0.69
n=611
0.64
n=646
0.70
n=786*
0.65
n=652
0.70
n=515
Education
4.03
n=638
4.05
n=601
4.03
n=638
4.00
n=775
4.03
n=643
4.03
n=507
Income
3.42
n=612
3.44
n=554
3.42
n=612
3.43
n=739
3.41
n=617
3.46
n=488
*p<05 **p<01 ***p<001 ****p<000l
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.14 Adjusted Means Comparing Patients with Severity Classification at
All Three Periods and Those with Fewer than Three Periods
Variable Adjusted Means
t=0 t=l t=2
3
periods
<3
periods
3
periods
<3
periods
3
periods
<3
periods
MILD
0.41
n=608
0.54 ****
n=554
0.38
n=608
0.55 ****
n=738
0.41
n=613
0.59 ****
n=487
CRD
Scale
5.55
n=608
3.38 ****
n=554
5.84
n=608
3.35 ****
n=738
5.59
n=613
2.94 ****
n=487
Log cost
7.60
n=603
7.40 ***
n=547
7.62
n=603
7.37 ****
n=719
7.52
n=605
7.33 **
n=486
LAS
59.58
n=599
61.71 *
n=543
59.88
n=415
63.33 **
n=449
62.10
n=373
63.88
n=299
PF
42.66
n=602
43.90 *
n=544
41.62
n=414
44.74 ****
n=45I
41.61
n=377
42.47
n=299
GH
41.86
n=597
44.11 ***
n=544
41.32
n=364
44.05 ***
n=386
41.48
n=374
43.33 *
n=293
*p<.05 **p<.01 ***p<.001 ****p<.0001
Table 4.15 Selected Results of Cross Sectional Regressions by Time Including
Variable PF GH PCS Log Cost
t=0: n = l,l 11
Severe -2.71**** -3.50**** -2.61**** 0.72***
CRD Scale -0.73**** -0.97**** -0.73**** 0.13****
t=l; n=723
Severe -2.65*** -3.41**** -2.96*** 0.69****
CRD Scale -0.72**** -0.79**** -0.73**** 0.14****
t=2; n=648
Severe -4.02**** -3 17****
-4.21**** 0.55****
CRD Scale -0.75**** -0.75**** -0.71**** 0. 12****
*p<.05 **p<.01 ***p<.001 ****p<.0001
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.6 Response Rates
It appears from table 4.16 that the response rates of the CRD population are
similar to those of the general Kaiser Permanente/USC population. In addition, the
response rates for the HUI2. HUD and the SF-36 Physical Function and General
Health domains are similar.
Table 4.16 Complete Responses for SF-36 and HUI2 in General Population
Compared to CRD Population_____________ _________________________
Variable Kaiser/USC (n=6,921) CRD (n=2,188)
Complete Total %
Complete
Complete Total %
Complete
t=0
PF 13537 13682 99% 1242 1332 93%
GH 13457 13682 98% 1232 1332 92%
t=I
PF 8599 13682 63% 931 1499 62%
GH 7574 13682 55% 810 1499 54%
t=2
PF 7475 13682 55% 724 1212 60%
GH 7407 13682 54% 712 1212 59%
HUI2 6937 13682 51% 647 1212 53%
HUI3 7148 13682 52% 679 1212 56%
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.0 Conclusion
The current research has addressed the original three primary research questions
and. as a result, has significantly advanced the field in several ways. However, the
research has limitations. This section will discuss four issues. First, the conclusions
o f the three primary hypotheses will be summarized. Second, the manner and extent
to which this research has advanced the field will be discussed. Third, limitations of
the analysis will be presented: and fourth, promising areas o f future research will be
addressed.
5.1 Summary
There were two primary objectives of this research: One. to develop and validate
two models o f severity o f illness in Chronic Respiratory Disease; and two. to
evaluate the construct validity of the Health Utilities Index in CRD. Both o f these
objectives have been met. First, the research developed and validated a clinical
model of severity o f illness in CRD based on the NAEPP guidelines. In addition, the
CRD scale was developed based on pharmacy data and subsequently validated. Both
models appeared to be valid measures of severity o f respiratory disease.
Second, the research examined the construct validity o f the FTUI according to
established methods by comparing it to two models o f severity o f illness. As
discussed in Section 2.4, construct validation is an iterative process involving many
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
comparisons to multiple measures deemed theoretically appropriate. In accordance
with this process, additional research is needed to shed further light on the
measurement capabilities o f the HUI2 and HUI3 in respiratory disease before
unambiguous conclusions can be made. The current research has revealed the need
for additional studies examining the construct validity of the HUI within respiratory
disease. Furthermore, the results o f the current research suggest that future versions
o f the HUI. or other generic utility measures, may consider the inclusion o f
respiratory-specific items or questions to ensure that the instrument is a valid
measure of utility in respiratory disease.
5.2 Contribution to the Field
The current research contributes significantly to the field in two primary ways.
First, it elucidates the impact of disease-specific morbidity on a well-established and
theoretically valid measure of generic utility for respiratory-related disorders.
Second, the current research contributes to the development o f pharmacy-based risk
assessment methods by developing two original models of disease-specific,
pharmacy-based risk adjustment in respiratory disease.
Whether their research goals are to better explain the variation in cost or
utilization or to compare the effectiveness of two or more health interventions, the
methods developed in this research will aid researchers in developing new methods
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of pharmacy-based risk assessment. Health economists, health services researchers
and health policy makers from all disciplines are very often limited in the data that is
available to them. In many cases, they have only retrospective pharmacy claims data
along with some measures o f utilization. In order to conduct a rigorous scientific
analysis with this type o f non-randomized data, it is important to control for a variety
of differences between treatment groups. Normally, randomization would eliminate
any of these differences, but in retrospective claims analysis, researchers must use
statistical techniques and pharmacy-based risk assessment methods to control for
these differences. Severity o f illness is very often an important source of difference
between groups. Using the techniques developed in this research, health researchers
can control for severity of illness within specific diseases (like respiratory disease)
based on pharmacy utilization data. Similar to the advantages of the Chronic Disease
Score as a generic measure o f pharmacy-based risk assessment, this research has
explored the methodology for pharmacy-based risk assessment within specific
diseases.
In addition, the current research examined the validity of a very widely used
method of ascertaining Quality Adjusted Life Years for the purposes of cost-
effectiveness analysis. As discussed in the introduction, the advantage of the QALY
is its ability to combine measures of HRQoL and mortality, providing a
comprehensive measure of effectiveness in cost-effectiveness analyses. The
measurement o f QALYs is crucial for comparing the cost-effectiveness of different
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
health interventions. As discussed in Section 2.2. the basis o f the QALY is the
measurement o f utilities. If the HUI is not an accurate measure o f respiratory-
specific utility, the calculation o f QALYs will be biased. The result would be biased
estimates of the cost-effectiveness o f particular treatments for respiratory disorders.
This, in turn, would impact decisions at the policy level. Policy makers would base
decisions on inaccurate information and inappropriately allocate resources to less
effective treatments. Furthermore, the use o f the HUI as a generic, population-based
health status instrument would result in biased measurement due to the inability to
capture respiratory-related morbidity in a general population. The current research
has provided insight to the measurement capabilities of the HUI, an important step in
the process of a comprehensive measure o f effectiveness.
5.3 Limitations of the Current Research
The current research has many limitations. First, the analysis was restricted to the
retrospective data from the Kaiser/USC Patient Consultation Study. This poses
threats to the methods used and the generalizability o f the research. Given unlimited
resources, the question of the construct validity of the HUI in CRD might have been
assessed by comparing the HUI to both established clinical markers o f severity and
respiratory-specific measures of HRQoL or utility, in addition to the SF-36 and
pharmacy-based models of severity. However, many health researchers face data
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
limitations and the methods developed in the current research will benefit those with
access to limited resources.
In addition, the fact that the analysis was restricted to data from the Kaiser/USC
Patient Consultation Study may limit the generalizability of the research. The extent
to which the results based on a large HMO population in Southern California are
generalizable to national or even international populations is questionable. However,
results from Section 4.1 suggest that there is little difference between the US
population and the CRD sample.
Another limitation of the research resulting from the data source is the use of a
cohort from 1992-1995. The two models of severity o f illness would have been more
relevant to current pharmacy-based risk assessment if the cohort had been more
recent. For example, the date of the cohort necessitated the use o f the 1992 NAEPP
guidelines rather than the more recent 1997 revision. Prescription drug utilization
patterns are very different today than they were in 1995. Indeed this may have an
impact on the generalizability of the results to today. Given this limitation, the
methodology developed in the current research could easily be updated to include
more recent trends in pharmacy utilization, given a more current data set.
In addition, both of the models o f severity o f illness relied on patient medication
and utilization history and excluded patient symptoms. Although there was a
statistically significant correlation between these models o f severity and HRQoL and
total cost, there was not statistically significant correlation with utility as measured
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
by the HUI2 and HUI3. It is possible that patient symptoms impact utility more
directly than severity. Although respiratory symptoms very likely will drive
utilization, and thus severity, this relationship between respiratory symptoms and
utility was not directly addressed in the current research. In order to assess this
relationship, future research would need to measure respiratory symptoms directly
and compare them to HUI2 and HUI3 scores.
The duration of the study (3 years) limited the extent to which the random effects
model formulation could take advantage o f the variation over time for the purposes
o f efficient estimation. In addition, the sample size (n=289) does not attain the
asymptotic efficiency of large sample properties. Longer duration and a larger
sample size may have resulted in better estimation. However, given the current
duration and sample size, the random effects model formulation is methodologically
sound and results in unbiased and consistent parameter estimates.
The exclusion o f patients who were too mild to utilize pharmaceuticals for three
consecutive periods is a tradeoff, rather than a limitation. The use of the random
effects model formulation is theoretically correct given the inherent heterogeneity
between individuals, especially in terms o f estimating the impact of severity on
HRQoL and utility. As a result, it provides more accurate parameter estimates.
Cross sectional regression by period, although it allows the inclusion o f "dropouts”,
does not include this inherent heterogeneity in the model specification. The ideal
analysis would be a random effects model formulation that includes "dropouts.”
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, the very nature of these dropout patients precludes their utilization over
three periods, making this impossible. Hence, the current research has excluded the
dropouts in favor of better estimation and more theoretically justified model
specification.
5.4 Future Research
The most promising and immediate offshoot o f this research would be to update
the current research within CRD to reflect current pharmacy utilization patterns. The
clinical model of severity of illness could easily be updated to include the 1997
NAEPP (or other more recent) guidelines. The assessment of the construct validity
of the HUI would then be contemporaneous.
Future research could also readily focus on the application of the disease-specific,
pharmacy-based risk assessment methods developed in the current research to other
diseases. Consequently, these models o f severity o f illness could be compared to the
HUI to examine the construct validity of the HUI in other important diseases.
In addition, further research on the construct validity of the HUI in respiratory
disease is needed. This could be accomplished by using other measures as external
criteria to shed further light on the measurement capabilities of the HUI in respiratory
disease.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.0 Bibliography
Blanc. P. D.. M. Cistemas, S. Smith, and E. H. Yelin. 1996. Asthma, employment
status, and disability among adults treated by pulmonary and allergy
specialists. Chest 109. no. 3: 688-96.
Bousquet. J.. J. Knani, H. Dhivert. A. Richard, A. Chicoye, J. E. Ware, Jr., and F. B.
Michel. 1994. Quality o f life in asthma. I. Internal consistency and validity o f
the sf-36 questionnaire. A m JR espir Crit Care Med 149, no. 2 Pt 1: 371-5.
Boyle. M. H.. W. Furlong. D. Feeny. G. W. Torrance, and J. Hatcher. 1995.
Reliability of the health utilities index-m ark iii used in the 1991 cycle 6
Canadian general social survey health questionnaire. Oual Life Res 4. no. 3:
249-57.
Clark. D. O.. M. Von Korff. K. Saunders. W. M. Baluch. and G. E. Simon. 1995. A
chronic disease score with empirically derived weights. Med Care 33. no. 8 :
783-95.
Diemer FB. Horowitz E, Horton M et al. 1997. Validation of a questionnaire to
assess severity of asthma in adults. J. Allergy Clin Immunol 99. no. 560.
Duan. N. 1983. Smearing estimate: A nonparametric retransformation method.
Journal o f the American Statistical Association 78. no. 383: 605-610.
Eisner, M. D.. P. P. Katz, E. H. Yelin, J. Henke, S. Smith, and P. D. Blanc. 1998.
Assessment of asthma severity in adults with asthma treated by family
practitioners, allergists, and pulmonologists. Med Care 36. no. 11: 1567-77.
Feeny D, Torrance GW, Furlong W. 1996. Quality of life and pharmacoeconomics in
clinical trials:85-95. Philadelphia: Lippincott-Raven.
Feeny, D.. W. Furlong, M. Boyle, and G. W. Torrance. 1995. Multi-attribute health
status classification systems. Health utilities index. Pharmacoeconomics 7,
no. 6 : 490-502.
Feeny, D., W. Furlong, R. K. Mulhem. R. D. Barr, and M. Hudson. 1999. A
framework for assessing health-related quality o f life among children with
cancer. Int J Cancer Suppl 12: 2-9.
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Fishman, P. A. and D. K. Shay. 1999. Development and estimation o f a pediatric
chronic disease score using automated pharmacy data. M ed Care 37. no. 9:
874-83.
Froberg, D. G. and R. L. Kane. 1989. Methodology for measuring health-state
preferences— ii: Scaling methods. JC lin Epidemiol 42. no. 5: 459-71.
Gemke. R. J. and G. J. Bonsel. 1996. Reliability and validity of a comprehensive
health status measure in a heterogeneous population of children admitted to
intensive care. J Clin Epidemiol 49. no. 3: 327-33.
Global Initiative for Asthma: National Institutes of Health. (National Heart, Lung,
and Blood. Institute). 1995. Socioeconomics. Nih publication. No. 95-3659.
Bethesda. (MD): National Institutes o f Health, National Heart. Lung, and
Blood. Institute.
Gold. M.. P. Franks, and P. Erickson. 1996. Assessing the health o f the nation. The
predictive validity of a preference-based measure and self-rated health. Med
Care 34. no. 2: 163-77.
Gold. Marthe R.. ed. 1996. Cost-effectiveness in health and medicine. New York:
Oxford University Press.
Grootendorst, P.. D. Feeny. and W. Furlong. 2000. Health utilities index mark 3:
Evidence of construct validity for stroke and arthritis in a population health
survey. Med Care 38. no. 3: 290-9.
Johnson. R. E.. M. C. Hombrook. and G. A. Nichols. 1994. Replicating the chronic
disease score (cds) from automated pharmacy data. J Clin Epidemiol 47, no.
10:1191-9.
Juniper. E. F„ G. H. Guyatt. D. H. Feeny. L. E. Griffith, and P. J. Ferrie. 1997.
Minimum skills required by children to complete health-related quality of life
instruments for asthma: Comparison o f measurement properties. Eur Respir J
10. no. 10: 2285-94.
Juniper, E. F.. G. H. Guyatt. P. J. Ferrie. and L. E. Griffith. 1993. Measuring quality
of life in asthma. Am Rev Respir Dis 147, no. 4: 832-8.
Malone. D. C.. K. A. Lawson, and D. H. Smith. 2000. Asthma: An analysis o f high-
cost patients. Pharm Pract Manag Q 20, no. 1: 12-20.
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Mathias, S. D.. M. M. Bates. D. J. Pasta, M. G. Cistemas. D. Feeny, and D. L.
Patrick. 1997. Use of the health utilities index with stroke patients and their
caregivers. Stroke 28. no. 10: 1888-94.
McCombs, J. S.. M. Cody, K. Besinque. G. Borok, D. Ershoff. S. Groshen, J. Hay, K.
A. Johnson, M. B. Nichol, and M. T. Nye. 1995. Measuring the impact of
patient counseling in the outpatient pharmacy setting: The research design of
the kaiser permanente/usc patient consultation study. Clin Ther 17. no. 6 :
1188-206.
McDermott. M. F., D. G. Murphy. R. J. Zalenski. R. J. Rydman. M. McCarren, D.
Marder, B. Jovanovic. K. Kaur. R. R. Roberts, M. Isola, E. Mensah, R.
Rajendran. and L. Kampe. 1997. A comparison between emergency
diagnostic and treatment unit and inpatient care in the management of acute
asthma. Arch Intern Med 157. no. 18: 2055-62.
McDowell. Ian and Claire Newell. 1996. Measuring health : A guide to rating scales
and questionnaires. New York: Oxford University Press.
McHomey, C. A.. J. E. Ware, Jr., and A. E. Raczek. 1993. The mos 36-item short-
form health survey (sf-36): Ii. Psychometric and clinical tests o f validity in
measuring physical and mental health constructs. Med Care 31. no. 3: 247-
63.
Mundlak. Y. 1978. On the pooling o f time series and cross section data.
Econometrica 46: 69-85.
National Asthma Education and Prevention Program (National Heart Lung and
Blood Institute). 1997. Guidelines fo r the diagnosis and management o f
asthma : Expert panel report 2. Nih publication ; no. 98-4051. Bethesda,
Md.: U.S. Dept, of Health and Human Services Public Health Service
National Institutes o f Health National Heart Lung and Blood Insitute.
National Heart Lung and Blood Institute. National Asthma Education Program.
Expert Panel on the Management of Asthma. 1991. Guidelines fo r the
diagnosis and management o f asthma. Bethesda, Md.: National Asthma
Education Program Office o f Prevention Education and Control National
Heart Lung and Blood Institute National Institutes of Health.
Nunnally, Jum C. 1978. Psychometric theory. New York: McGraw-Hill.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Prescott. E.. P. Lange, and J. Vestbo. 1997. Effect of gender on hospital admissions
for asthma and prevalence o f self-reported asthma: A prospective study based
on a sample o f the general population. Copenhagen city heart study group.
Thorax 52. no. 3: 287-9.
Raina. P.. B. Bonnett. D. Waltner-Toews. C. Woodward, and T. Abernathy. 1999.
How reliable are selected scales from population-based health surveys? An
analysis among seniors. Can J Public Health 90. no. 1: 60-4.
Revicki. D. A.. N. K. Leidy. F. Brennan-Diemer. S. Sorensen, and A. Togias. 1998.
Integrating patient preferences into health outcomes assessment: The
multiattribute asthma symptom utility index. Chest 114. no. 4: 998-1007.
Serra-Batlles. J.. V. Plaza. E. Morejon. A. Cornelia, and J. Brugues. 1998. Costs of
asthma according to the degree of severity . Eur Respir J 12. no. 6: 1322-6.
Skobeloff. E. M.. W. H. Spivey. S. S. St Clair, and J. M. Schoffstall. 1992. The
influence o f age and sex on asthma admissions. Jama 268. no. 24: 3437-40.
Staquet. Maurice J.. Ron D. Hays, and Peter M. Fayers. 1998. Quality o f life
assessment in clinical trials : Methods and practice. Oxford medical
publications. Oxford : New York: Oxford University Press.
Stempel. D. A.. J. F. Durcannin-Robbins. E. C. Hedblom. R. Woolf. L. L. Sturm, and
A. B. Stempl. 1996. Drug utilization evaluation identifies costs associated
with high use o f beta-adrenergic agonists. Ann Allergy Asthma Immunol 76.
no. 2: 153-8.
Stewart. Anita L. and John E. Ware. 1992. Measuring functioning and well-being:
The medical outcomes study approach. Durham: Duke University Press.
Sunyer. J.. J. M. Anto. M. Kogevinas. M. A. Barcelo. J. B. Soriano. A. Tobias. N.
Muniozguren. J. Martinez-Moratalla. F. Payo. and J. A. Maldonado. 1997.
Risk factors for asthma in young adults. Spanish group o f the european
community respiratory health survey. Eur Respir J 10. no. 11: 2490-4.
Taylor. W. R. and P. W. Newacheck. 1992. Impact of childhood asthma on health.
Pediatrics 90, no. 5: 657-62.
Torrance. G. W. 1986. Measurement of health state utilities for economic appraisal. J
Health Econ 5. no. 1: 1-30.
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Torrance, G. W.. M. H. Boyle, and S. P. Horwood. 1982. Application of multi-
attribute utility theory to measure social preferences for health states. Oper
Res 30. no. 6: 1043-69.
Torrance, G. W.. D. H. Feeny. W. J. Furlong. R. D. Barr. Y. Zhang, and Q. Wang.
1996. Multiattribute utility function for a comprehensive health status
classification system. Health utilities index mark 2. M ed Care 34, no. 7: 702-
22.
Torrance, G. W.. W. H. Thomas, and D. L. Sackett. 1972. A utility maximization
model for evaluation of health care programs. Health Serv Res 7. no. 2: 118-
33.
Von KorfT. M.. E. H. Wagner, and K. Saunders. 1992. A chronic disease score from
automated pharmacy data. J Clin Epidemiol 45. no. 2: 197-203.
Von Neumann. John and Oskar Morgenstem. 1947. Theory o f games and economic
behavior. Princeton: Princeton Univ. Press.
Ware. J. E.. Jr. and C. D. Sherboume. 1992. The mos 36-item short-form health
survey (sf-36). I. Conceptual framework and item selection. Med Care 30. no.
6: 473-83.
Ware. John E. and New England Medical Center Health Institute. 1994. Sf-36
physical and mental health summary scales : A user’ s manual. Boston: The
Health Institute New England Medical Center.
Ware. John E.. Kristin K. Snow. Mark Kosinski, Barbara Gandek. and New England
Medical Center. Health Institute. 1993. Sf-36 health su rvey: Manual and
interpretation guide. Boston: The Health Institute New England Medical
Center.
Weiss. K. B. and S. D. Sullivan. 2001. The health economics o f asthma and rhinitis.
I. Assessing the economic impact. J Allergy Clin Immunol 107, no. 1: 3-8.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7.0 Appendix
7.1 Chronic Disease Comorbidity Using the American Hospital Formulary
Service (AHFS) Medication Coding System (Clark et a l, 1995) _____
Compffeidily Medication AHFS
Category
AHFS
Category
NUmoer
Coronary and
peripheral
Vascular disease
Anticoagulants
Pentoxifylline
Ticlopidine
Anticoagulants
Hemorrheologic agents
Unclassified therapeutic
agents
20:12.04
20:24
92:00
Cardiac disease
ASCVD
CHF
Beta-adrenergic blockers
Calcium channel blockers
Disopyramide
Vasodilator nitrates
Digitalis glycosides
Diuretics, loop
Procainamide
Quinidine
Class 1 A antiarrhythmics
Class 1 C antiarrhythmics
Class 1 I antiarrhythmics
Cardiac drugs 24:04
Hypertension ACE inhibitors
Alpha blockers
Antihypertensive
vasodilators
Beta-adrenergic blockers
Calcium channel blockers
Clonidine
Diuretics, thiazides
Ganglionic blockers
Guanethidine
Methyldopa
Rauwolfia alkaloids
Hypotensive agents 24:08
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Hyperlipidemia Antilipemics Antilipemic agents 24:06
Rheumatologic
Conditions
Systemic corticosteroids
Gold salts
Hydroxychloroquin
Adrenals
Gold compounds
Antimalarial Agents
68:04
60:00
8:20
Malignancies Antineoplastics
(Taxol, interleukins)
Colony-stimulating factors
Antinausea misc.,
ondansetron
Antineoplastic agents
Hematopoietic antiemetics
Antiemetics
10:00
20:16
56:22
Parkinson’s
disease
Autononics L-Dopa
Selegline
Antiparkinsonian agents
Unlisted in AMHS
12:08.4
Renal disease Potassium removing resins,
Kayexelate
Potassium removing
resins
40:18
Diabetes Insulins, sulfonylureas Antidiabetic agents 68:20
Liver failure Ammonia detoxicants Ammonia detoxicants 40:10
Acid peptic
disease
Histamine H2 blockers
Prostaglandin, misoprostil
Proton pump inhibitors,
Omeprazole
Misc. G! drugs 56:40
Respiratory
Illness,
Asthma
Beta agonist bronchodilators
Xanthines
Cromolyn
Inhaled corticosteroids
Sympathomimetic agents
Respiratory smooth
Muscle relaxants
Unclassified therapeutic
agents
Adrenals
12:12
86:16
92:00
68:04
Thyroid
disorders
Thyroid replacement
Antithyroid agents
Thyroid agents
Antithyroid agents
68:36.04
68:36.08
Crohn’s and
ulcerative
Colitis
Sulfasalazine
Olsalazine
Mesalamine
Sulfonamides
Misc. Gl drugs
8:24
56:40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Depression Tricyclic antidepressants
Monoamine oxidase inhibitors
SSRI, fluoxetine
Antidepressants 28:16.04
Psychotic
illness
Butyrophenones
Phenothiaznes
Antipsychotic misc.
Thiothixene
Tranquilizers 28:16.08
Bipolar
disorders
Lithium Antimanic agents 28:28
Anxiety and
tension
Benzodiazepines
Meprobamate
Antianxiety misc.
Benzodiazepines
Misc. anxiolytics,
sedatives,
and hypnotics
28:24.08
28:24.92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)

PDF

Effects of a formulary expansion of the use of SSRIs and health care services by depressed patients in the California Medicaid program

PDF

Physician adherence to national hypertension guidelines in an elderly Medicaid population

PDF

Compliance study of second-generation antipsychotics on patients with schizophrenia

PDF

Controlling for biases from measurement errors in health outcomes research: A structural equation modeling approach

PDF

A new paradigm to evaluate quality-adjusted life years (QALY) from secondary database: Transforming health status instrument scores to health preference

PDF

Assessing the cost implications of combined pharmacotherapy in the long term management of asthma: Theory and application of methods to control selection bias

PDF

The influence of drug copay change on drug utilization: The case of small-firm employees in California

PDF

Time dependent survival analysis of Kaiser Permanente/USC pharmacists' consultation intervention study

PDF

Performance management in health care in Iceland

PDF

Assessment of prognostic comorbidity in hospital outcomes research: Is there a role for outpatient pharmacy data?

PDF

Variations in physician practice patterns for eye care under the National Health Insurance of Taiwan

PDF

Care management for the uninsured: A force field analysis of the business case

PDF

Prescription drug profiles as health risk adjusters in capitated payment systems: An applied econometric analysis

PDF

Out -of -pocket health expenditures by older adults in relation to age, race, and insurance

PDF

An analysis of health risk selection and quality of care under Medicare fee -for -service and Medicare managed care health care systems

PDF

Using network perspective to examine the organization of community -based elder care systems across four communities

PDF

Regionalization of fire protection and emergency medical aid services: A comparative case study analysis of economic and socio-political impacts

PDF

Causal estimation of the effect of medication compliance on health outcomes

PDF

Leading change initiatives: communication and bounded agency in a health care organization

Asset Metadata

Creator Sullivan, Patrick William (author)

Core Title The construct validity of the Health Utilities Index in patients with chronic respiratory disease in a managed care population

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics and Policy

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics, general,health sciences, health care management,Health Sciences, Pharmacy,health sciences, public health,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-233247

Unique identifier UC11339219

Identifier 3073853.pdf (filename),usctheses-c16-233247 (legacy record id)

Legacy Identifier 3073853.pdf

Dmrecord 233247

Document Type Dissertation

Rights Sullivan, Patrick William

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA