University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THE INFLUENCE OF DRUG COPAY CHANGE ON DRUG UTILIZATION:
THE CASE OF SMALL-FIRM EMPLOYEES IN CALIFORNIA
Copyright 2004
by
Patrick Thiebaud
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
ECONOMICS
August 2004
Patrick Thiebaud
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3145302
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
®
UMI
UMI Microform 3145302
Copyright 2004 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.
TABLE OF CONTENTS
List of Tables iii
Abstract vi
Section 1: Introduction 1
Section 2: Previous research
2.A. Literature review 4
2.B. Evidence Table 10
Section 3: Analytical framework
3. A. Outline of a general model of drug demand 17
3.B. Specific model of drug demand
3.B.1 Health insurance in California 20
3.B.2 Selection of sample and final sample 24
3.B.3 Supply and demand of health insurance 28
3.B.4 Drug demand 45
3.C. Method
3.C.1 Cross-section 46
3.C.2 Fixed effect analysis 50
Section 4: Results
4.A. How representative is my sample? 56
4.B. Health insurance plans 61
4.C. descriptive statistics 64
4.D. Multivariate analysis 71
Section 5: Conclusion 76
References 80
Appendices A-E
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Evidence Tables 10
Table 1: Size of final sample, in number of enrollees and
number of firms employing them, and size of the excluded
subsamples
Table 2: Comparison between final sample and excluded
subsample 1: Age, gender, enrollment status, and selected
utilization measures over the three-year enrollment period
Table 3: Comparison between final sample and excluded
subsample 2: Age, gender, enrollment status, and selected
utilization measures over the two-year enrollment period
Table 4: Plans available in period one, with number of enrollees
Table 4b: Plans available in period two, with number of
enrollees
Table 5: Drug use in period one and two: means and proportions
Table 6: Demographic characteristics and enrollment status
Table 7: Health status profile for periods one and two:
percentage of enrollees who received a diagnosis in each ADG
category
Table 8: Demographic characteristics
Table 9: Drug use in period one
Table 9b: Change in service utilization between period one and
period two
Table 10: Comparison of drug use change for enrollees whose
copay changes and enrollees whose copay remains stable
Table 11a: WLS results for panel (without intercept): Effects of
copay change on enrollee drug utilization and on insurer drug
costs
58
59
61
63
63
65
66
67
68
69
70
70
73
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 1 lb: WLS results for panel (with intercept): Effects of
copay change on enrollee drug utilization and on insurer drug
costs
Table 11c: WLS results for panel (with intercept): Effects of
copay change on drug quantity, on number of drugs in the
formulary, and on insurer drug costs, by brand and generic
Table 12a: WLS results for cross section, period one: Effects of
copay levels on enrollee drug utilization and on insurer drug
costs
Table 12b: WLS results for cross section, period two: Effects of
copay levels on enrollee drug utilization and on insurer drug
costs
Table la (appendix A): Comparison of ADG frequencies
between the final sample and subsample 1, those who moved
among employers: Percent of enrollees receiving diagnosis in
each ADG
Table lb (appendix A):: Logistic: Probability that an enrollee
will remain in the same firm for three years or more
Table 2a (appendix A): Comparison of ADG frequencies
between the final sample and subsample 2, those who worked in
the same firm and could choose among different copay
structures: Percent of enrollees receiving diagnosis in each
ADG
Table 2b (appendix A):: Logistic: Probability to be in a firm
offering only one copay structure instead of a choice of copay
structure
Table 2c (appendix A):: Logistic: probability to be in a firm
offering one copay level only, measured using firm level
averages
Table 3 a (appendix B): Proportion of enrollees with multiple
ADGs
Table 3b (appendix B): Proportion of enrollees with multiple
diagnoses
74
74
75
75
87
83
89
90
91
93
94
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4a (appendix C): Drag utilization with the following
copay: $5 brand, $5 generic, no non-formulary copay
Table 4b (appendix C): Drag utilization with the following
copay: $10 brand, $5 generic, no non-formulary copay
Table 4c (appendix C): Drag utilization with the following
copay: $10 brand, $10 generic, no non-formulary copay
Table 4d (appendix C): Drag utilization with the following
copay: $10 brand, $5 generic, $20 non-formulary copay
Table 4e (appendix C): Drag utilization with the following
copay: $10 brand, $10 generic, $20 non-formulary copay
Table 4f (appendix C): Drag utilization with the following
copay: $15 brand, $10 generic, no non-formulary copay, period
two only
Table 4g (appendix C): Drag utilization with the following
copay: $20 brand, $10 generic, $25 non-formulary copay,
period two only
Table 8a (appendix D): WLS results for cross section, all
enrollees, period one
Table 8b (appendix D): WLS results for cross section, all
enrollees, period two
95
95
95
96
96
96
97
98
99
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
BACKGROUND: Drug expenditures have increased at a faster pace than any other
health care category over the last five years and account for more than 20% of total
health care inflation between 1999 and 2002. Employers and health plans are looking
for ways to reduce drug expenditures. One of the most effective ways to do this is to
increase cost sharing for health plan members, i.e. increase copays. OBJECTIVE:
Determine the effects of higher brand and generic drug copays on drug utilization.
DATA: Insurance claims from an administrative database made available by a major
West Coast pharmacy benefit management company. The database contains all
pharmacy and medical claims as well as plan benefit designs. The sample consists of
employees, their spouses, and their dependents, under age 65, who were enrolled
continuously for two years. Enrollees had to work for only one small California firm
(i.e. a firm with 2-50 employees) that offered only one health plan. The final sample
has 30,824 enrollees working in 4,554 small firms. METHOD: Drug utilization
variables and health status indicators are computed separately for each full year of
enrollment under a specific copay structure. Fixed effect analysis is performed with
the first difference estimator to estimate the effect of copay and health status change
on net drug cost and number of prescriptions. RESULTS: Higher generic drug
copays lead to a reduction in number of prescriptions (-0.23; p<0.0001) and total net
cost to the insurer (-7.06; pO.0001). Higher brand copays result only in lower total
net cost to the insurer (-2.5; p<0.0001). The generic price elasticity for number of
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
drugs is -0.2, the generic price elasticity for total net cost is -0.16, and brand price
elasticity for total net cost is -0.11. CONCLUSION: (1) Enrollees decrease their drug
use when faced with higher copays; (2) the response to higher generic copays is
stronger than the response to higher brand copays; and (3) health plans can
significantly reduce their outlays on drugs by raising drug copays.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
SECTION 1: INTRODUCTION
Employment-based health insurance covers nearly 175 million people in the US.
Between spring 2002 and spring 2003, the cost of employment-based health
insurance increased by 13.9 percent [Gabel et al. 2003]. The current premium
inflation, which has been accelerating since the mid-nineties, reflects a significant
increase in the cost of treatment due both to a higher quantity of services per person
and a higher price per service [Levit et al. 2003]. Drug expenditures have increased
at a faster pace than any other health care category over the last several years,
accounting for well over 20 percent of total health care inflation between 1999 and
2002. More than two thirds of this increase, holding age and gender constant, was
due to higher utilization intensity. [Davis and Cooper 2003].
Rising premiums - driven in large part by growing drug costs - are becoming
increasingly burdensome for employers, prompting them and the insurers they
contract with to look for cost containment mechanisms [Bymark and Waite 2001].
The search for effective cost control tools has produced a number of options: closed
formularies, higher copayments, generic substitutions, and narrow pharmacy
networks. A higher copayment shifts the financial burden of drug use from the
insurer to the enrollee. A narrow pharmacy network allows the insurer to negotiate
lower dispensing fees. A formulary identifies drugs as preferred treatment for
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
specific diseases and gives the insurer leverage to obtain discounts from drug
manufacturers. A closed formulary excludes certain drugs from coverage completely,
requiring the enrollee to pay the full cost of the prescription, whereas an incentive-
based formulary requires a higher copay for non-preferred drugs. Generic
substitution encourages the enrollee to buy generic drugs. It is implemented in one of
two ways: Either the enrollee pays a lower copayment for generic drugs than for
brand drugs, or the enrollee pays the difference in price.
These cost control tools have been used for years with mixed results. Higher
copayment is one of the most effective, but it is necessary to evaluate precisely its
effect on drug utilization in order to ensure that cost control is achieved without
jeopardizing the quality of treatment.
The relationship between drug copay and utilization - or drug demand - has not been
thoroughly analyzed since the 1970s. It is important to reevaluate it now, in light of
new market conditions and vastly improved drugs.
This paper aims to answer several questions: Do patients reduce their drug use when
faced with higher generic copays or higher brand copays? If they reduce their
utilization, how do they reduce it? Fewer days of supply? Fewer doses? Fewer
prescriptions? And do the plans realize significant savings?
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The hypotheses to be evaluated are:
• Enrollees decrease their drug use when faced with higher copays.
• Enrollees are more responsive to generic copay levels than to brand copay
levels.
• Health plans can reduce their outlays on drugs significantly by raising drug
copays.
This paper analyzes the effects of higher copays on enrollee drug use and determines
if and how it reduces insurer outlay. To this end, this paper investigates a model of
health insurance and prescription drug demand for people insured through their
employers. I discuss the different agents taking part in the health insurance market,
how they interact, and what factors influence an individual’s choice of employer,
health insurance, and drug purchase. I then construct a statistical model to estimate
the effects of higher copays on several measures of individual drug utilization over a
period of two years.
The plan of this article is as follows: Section 2 summarizes the literature on this
subject and compares this paper with previously published research. Section 3
presents the data, the model and method used to estimate drug demand. Section 4
presents descriptive statistics and results. Section 5 discusses the results and
comments on the limitations of this study
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
SECTION 2: PREVIOUS RESEARCH
2.A. LITERATURE REVIEW
Grossman’s human capital model of health care [Grossman 1972] provides a
foundation for the analysis of health care demand. A large body of literature, with
increasing levels of complexity, has developed on the premises of human capital
theory. These types of models formalize consumers’ choices as lifetime expected
utility maximization. Individuals use health inputs, such as prescription drugs or
physician services and time, to increase or maintain their depreciating health capital.
The utility function is maximized by choosing the optimal level of health capital over
time. Health care demand is therefore viewed as a derived demand. This model -
along with the assumptions that health care is a normal good and that production of
health has constant returns to scales - predicts that the quantity of drugs demanded
should decrease with higher drug prices, all else equal.
The market for health care differs from most other markets in the US: The consumer,
the patient, usually does not pay directly for the full cost of health care but only for
the residual cost determined by cost-sharing provisions in his/her health insurance
contract. What the patient pays and, by extension, his/her health care demand both
depend on the health plan. A comprehensive framework needs therefore to address
the interdependence of health insurance and health care demand.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Few of the empirical investigations of the effect of out-of-pocket price on consumer
drug demand refer to the Grossman model explicitly and even fewer to the details of
its theoretical constructs. There are, however, a number of studies that analyzed the
impact of cost sharing provisions on drug demand.
The literature on drug demand can be separated into 1) retrospective analyses of data
from insurance claims or of surveys of past use, and 2) randomized trials on financial
incentives to increase adherence to drug treatment. (Better adherence results in
higher drug use, therefore in a higher quantity of drugs demanded.) I will only
briefly touch on the randomized trials since their context and methods are different
from those in my research. Guifffida and Torgeson (1997) wrote a comprehensive
review of randomized experiments in which patients were offered such financial
incentives as vouchers, cash, lottery tickets, or coupons. They conclude that financial
incentives do increase patient compliance and are often more effective than other
interventions.
A survey1 of published retrospective analyses is presented in the evidence tables at
the end of this section. These studies reached the expected results: Higher out-of-
pocket costs lead to lower drug utilization. Only Johnson, Goodman, Hombrook, and
Eldredge (1997) qualify their findings; they also find a response but emphasize its
variability across therapeutic classes.
1 Studies published before 1980 w ere excluded.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Among the surveyed studies, four (Coulson and Stuart 1995, Grootendorst et al.
1997, Stuart and Grana 1998, and Lillard et al 1999) focused on elderly patients.
They show that the elderly increase their drug utilization when they gain access to
better drug insurance coverage, irrespective of the public or private nature of their
coverage. Thomas et al. (2002) focus on the effect of different levels of supplemental
employer-sponsored insurance coverage on drug use among the elderly. They
conclude that higher copays prompt elderly to switch to cheaper, generic, drugs
without altering the total number of drug used. Tamblyn et al. (2001) show that the
imposition of a 25 percent coinsurance on drugs that could previously be obtained
for free by low-income elderly and welfare recipients in Canada leads to a 9.1
percent to 14.4 percent reduction in demand for essential drugs.
The remaining 10 studies concentrate on non-elderly populations. Nelson et al.
(1984) find that Medicaid recipients reduce their drug use by 0.2 prescription per
month after the introduction of a $0.5 (in 1974) copay on all prescriptions that were
previously free. Leibowitz et al. (1985) summarize the results from the RAND
Health Insurance Experiment (HIE) regarding drug utilization with different
coinsurance rates: They find an elasticity of demand of about -0.25. The HIE results
are often considered benchmarks because they were obtained in a large-scale
randomized experiment. But their usefulness is limited - since the insurance design
used for the experiment is quite different from designs in the current market place.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The last eight studies were conducted on enrollees in employer-sponsored health
plans (Harris et al. 1990, Smith 1993, Motheral and Henderson 1999, Joyce et al.
2002, Ridley 2003, Huskamp et al. 2003, Rector et al. 2003, and Fairman et al.
2003). These studies analyze the same population as the present research and deserve
more detailed comments.
The majority of these studies (Harris et al 1990, Motheral and Henderson 1999,
Huskamp et al. 2003, Rector et al. 2003, and Fairman et al. 2003) use a quasi-
experimental intervention group vs. control group design. The control group was
either simply a group of enrollees in a different plan that did not have any change in
copay structure (Harris et al 1990, Huskamp et al. 2003, Rector et al. 2003, and
Fairman et al. 2003), a group of enrollees matched for age and gender (Motheral and
Henderson 1999), or a group of enrollees selected through clustering methods
(Huskamp et al. 2003). The remaining studies (Smith 1993, Joyce et al. 2002, and
Ridley 2003) rely on a comparison of average group use or a comparison of average
individual use among plans with different copay structures.
Results from these studies show that enrollees react to higher copays with lower
participation rates and fewer drugs bought. Harris et al. (1990) analyze drug use in
HMO enrollees and find that an increase from no copay to a $1.5 copay reduces
utilization by 10.7 percent and expenditures by 6.7 percent. A copay increase from
$1.5 to $3 further reduces utilization by 10.6 percent and expenditures by 5.2
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
percent. Smith (1993) evaluates drug utilization in employees of several large
employers and finds that increasing copays from $3 to $5 reduces utilization by 5
percent and expenditures by 10 percent. Joyce et al (2002) calculate that working-age
enrollees in a few dozen large-firm plans reduce their drug expenditures by 32.9
percent with a doubling of copays from $5 generic and $10 brand. Ridley (2003)
shows that the price elasticity for the number of prescriptions in a commercial
population is -0.3. The others focus on specific therapeutic classes with similar
results in terms of reduction in the number of prescriptions and lower expenditures.
The range of results is broad and the methods used to estimate price elasticity of
demand diverse. How do these studies compare to this one? In terms of population
studied, a number of papers focus on welfare or elderly populations, populations that
may be different from the working adults and their spouses and dependents that
make up my sample. Among papers dealing with working adults, the target
population is either enrollees from a few large employers or enrollees from a few
large managed care organizations. The present paper is unique in targeting the
enrollees of plans purchased by small firms. This has never been done before.
Another difference between my research and the published literature is my taking
into account the interactions between health plan choice and health care demand. An
example of such an interaction is the tendency of high health care users to prefer
extensive coverage, with lower drug copays. If these preferences are related to
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
unobserved individual characteristics, the estimated response to higher drug copays
will be inflated. This problem is called omitted variable bias. I develop a model of
simultaneous drug demand and health insurance demand and use it to thoroughly
discuss the potential biases that arise from the interactions between copay levels and
prior use or enrollees’ unobserved characteristics. Indeed, with the exception of
Joyce et al. (2002) and Ridley (2003), no paper attempts to evaluate the risk of
omitted variable bias. And even Joyce et al. (2002) and Ridley (2003) simply explain
why bias is unlikely and go on to estimate their models assuming the absence of bias.
Their estimates may therefore be unreliable.
Finally, no studies to my knowledge use fixed effect estimation. Fixed effect
estimation in this sample - small firms’ enrollees - should allow me to obtain more
reliable results than the treatment-control approach because it accounts for the
interaction between unobserved characteristics, copays, and drug use. To summarize,
the distinctive features of this research on drug demand are:
• a unique population - enrollees of plans offered by small firms;
• the development of a model that takes into account both drug demand and
insurance choice;
• an in-depth discussion of potential bias;
• fixed effect estimation.
2 This subject w ill be discussed in section 3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Nelson AA,
Reeder CE,
and
Dickson WM.
1984.
Sou th Carolina
Medicaid beneficiaries
(treatment group) and
Tennessee Medicaid
beneficiaries (control
group).
17,811 in South
Carolina and 27,841 in
Tennessee.
Four years
from January
1976 to
December
1979.
$0.5 copay
instituted in
January 1977
from previously
no copay.
Average
expenditure per
recipient per month
for drugs covered
by Medicaid before
and after policy
change. Drugs
divided into 10
therapeutic
categories.
OLS and
AR(1)
models for
non-seasonal
data
(differenced
by 12
months).
Significant reduction in
utilization rates and
expenditures: 0.2
prescription per month
decrease in S.C. vs.
Tennessee.
Leibowitz A,
Manning
WG,
and
Newhouse JP,
1985.
Random sample of
families. RAND Health
Insurance Experiment.
3,860 individuals.
Three- and
five- year
periods of
enrollment in
the
experiment.
5 different cost-
sharing
structures:
0%, 25%, 50%,
95%
coinsurance, or
a deductible
plan.
Drugs expenditures
per capita, number
o f prescriptions per
capita, and generic
fill rate.
2-part
model:
probit and
log linear.
Higher copays significantly
reduce utilization and costs.
Elasticity (number o f drugs)
= - 0.25.
Harris BL.,
Stergachis A.,
and
Ried LR.
1990.
HMO (Group Health
Cooperative o f Puget
Sound) enrollees under
65.
19,982 in treatment
cohort and 23,164 in
control cohort.
Four years
from July
1982 to June
1986.
$0 copay until
June 1983. $1.5
from July 1983
to June 1984. $3
from July 1984
to June 1985.
Number o f
prescriptions
dispensed, drug
ingredient costs to
HMO, average drug
ingredient cost per
prescription. Drugs
subdivided into
therapeutic
categories, i.e. into
essential and
discretionary.
Comparison
s of means
between
treatment
and control
group for
each one-
year period.
ANCOVA.
Reduction o f both utilization
and expenditure with
increase in per prescription
ingredient costs.
-10.7% total number o f
prescriptions with $1.5. -
10.6% with $3. -6.7%
ingredient costs with $1.5
and -5.2% with $3.
o
2.B. EVIDENCE TABLES
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Smith D.
1993.
Employees from
several employers
covered by a large
national MCO.
212 employer
group data.
Individual data
were aggregated at
employer level.
One year:
1989.
A single plan design
with variable levels
o f copays from $1 to
$8 for brand drugs.
Some plans have a
generic option, with
generic copays $2
lower than brand
copays.
Total net cost
for drugs for
the employer,
cost per
prescription,
and number of
drugs.
OLS on log
transformed
variables.
An increase in copay
from $3 to $5 leads to a
5% reduction in number
o f prescriptions, an
offsetting increase in
ingredient costs per
prescription, and a 10%
decrease in employer
costs.
Coulson NE
and
Stuart BC.
1995.
Medicare
beneficiaries in
Pennsylvania.
4,066
Survey
mailed in
summer 1990
and Medicare
claims for
1988 and
1989.
Elderly either have
Medicaid (mostly no
copay), employee-
sponsored
supplemental
insurance, or
Pharmaceutical
Assistance Contract
for the Elderly
(PACE) imposing a
$4 copay.
Number of
prescriptions
and probability
to have at least
one
prescription.
OLS and
probit. Test
conducted
for possible
endogeneity
o f
insurance.
PACE beneficiaries use
0.29 more drugs per 2-
week periods than
enrollees without drug
benefits.
2.B. Continued
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Johnson RE,
Goodman MJ,
Hombrook
MC,
and
Eldredge MB.
1997.
Elderly enrolled in two
Medicare risk-based
programs in Kaiser
Permanente Northwest.
4,220 Social HMO
enrollees and 16,960
Medicare Plus
enrollees.
Four
overlapping
two-year
analysis
periods from
1987 to 1991.
Social HMO:
increase copay
from $1 to $3 in
1988 and $3 to
$5 in 1989.
Medicare Plus:
50% coinsurance
increased to 70%
in 1990.
Drugs divided in
therapeutic classes,
essential and non-
essential.
Exposure to a
therapeutic class
(use/no use), days
o f supply, insurer’s
costs pre- and post
change.
Pre-test/post-test
intervention
groups compared
to control group.
ANCOVA.
No consistent effects
o f higher copay and
coinsurance on days
supply, costs, and
exposure across
therapeutic classes:
Some classes saw a
reduction in use
while others did not.
Grootendorst
PV,
O’Brien BJ,
and
Anderson
GM.
1997.
Persons 55-75 eligible
for the Ontario Drug
Benefit Plan (ODB). In
the ODB, people age
65 or above receive
first $ coverage for all
prescription drugs.
9,370
1990 Ontario
Health
Survey self-
reported drug
use over
previous four
weeks.
Comparison of
elderly on zero
copay plan with
younger
individuals on
non-zero copay
plans.
Number of different
drugs used and
likelihood o f any
drug use.
Two-part model;
probit for the
probability o f use
and negative
binomial for the
number o f drug in
the subsample of
drug users.
+20% to +30% drug
consumption after
age 65 depending on
health status (good to
poor).
Stuart B
and
Grana J.
1998.
Pennsylvania Medicare
beneficiaries.
4,066
1990 survey
o f a randomly
selected
group of
beneficiaries
Differences in
drug insurance
coverage among
beneficiaries:
Medicare only or
supplemental
private or public
insurance.
Number o f drugs
used to treat several
conditions.
Log linear OLS
and negative
binomial.
Medicare
supplementation
increases the odds o f
using drug treatment
by 6% to 17%
depending on the
condition. The effect
is often more
pronounced for less
serious conditions.
2.B. Continued
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Lillard LE,
Rogowski J,
and
Kington R.
1999.
Elderly persons
66 or over with
private
supplemental
insurance or
Medicaid.
910
Data from the
RAND Elderly
Health
Supplement to
the 1990 Panel
Study o f Income
Dynamics
(PSID). Annual
surveys since
1968. The 1990
survey collected
information for
1989.
Self-reported
out-of-pocket
expenditures on
drugs and
insurance status:
i.e. Medicare
only, private
insurance with
or without drug
coverage, or
Medicaid.
Self-reported
annual drug
expenditures in
1989.
2-part model
with probit
section
equation and
log normal
demand
equation.
Private health insurance
and private health
insurance covering drugs
increase the likelihood o f
use. Insurance variables
are not significant for total
expenditures on drugs.
Tamblynetal.
2001.
Low income
elderly and
adult welfare
medication
recipients in
Quebec,
Canada.
93,950 elderly
and 55,333
adults.
Four years from
August 1993 to
August 1997.
Public policy
change in
August 1996
from free drugs
to a 25%
coinsurance rate
up to CA$200
per year.
Mean daily
number o f drugs
before and after
policy change,
separated into
essential (improve
health) and less
essential (relieve
symptoms) drugs.
Random
effect pooled
time-series
based on 49
pre- and post
policy
months with
AR(1)
structure.
Essential drug use
decreases by 9.12%
among elderly and
14.42% among adults.
Less essential drug use
decreases by 15.14% for
elderly and 22.39% for
adults.
2.B. Continued
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Joyce GF,
Escarce JJ,
Solomon MD,
and
Goldman DP.
2002.
Working age (18
to 64) enrollees
in employer-
sponsored plans
from 25 large
firms.
420,786 enrollees
enrolled for 1, 2,
or 3 years.
Three
years from
1997 to
1999.
Different drug benefit
designs among plans.
Covariates include
copay in each tier or
coinsurance rates.
Average
annual drug
spending.
2-part model,
GLM with log
link used in the
second part (total
use conditional
on positive use).
Comparison of
simulated total
expenditures with
different copay
levels.
1-tier: an increase from $5 to
$10 lowers average total drug
cost by 22.3%
2-tier: an increase from $5
generic and $10 brand to $10
generic and $20 brand, lowers
cost by 32.9%
3-tier: adding a third tier of
$30 lowers total by 4%.
Thomas CP,
Wallack SS,
Lee S,
and
Ritter GA.
2002.
Persons 65 or
older in retiree
employer-
sponsored plans
administered by a
national PBM.
96 health plans
offering only one
prescription drug
plan for self-
insured
companies.
29,435 members
One year:
2001.
Variation in benefit
design among plans:
Flat-dollar copayment
plans: 1-tier, 2-tier, and
3-tier with varying
levels o f copay.
Coinsurance plans: 1-
tier or 3-tier with
different coinsurance
rates.
Number of
prescriptions
per enrollee,
annual
spending per
enrollee, and
annual out-of-
pocket costs.
Comparison o f
age-adjusted
averages and
proportions
among plans.
Higher cost-sharing
associated with higher
member out-of-pocket costs, a
switch to cheaper drugs
(generic or mail order), and a
lower total prescription
spending. There is no
apparent effect on use.
2.B. Continued
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Ridley D.
2003.
Drug claims data from
a PBM and drug
marketing data from a
consulting firm.
85 insurance groups
covering 3 million
enrollees. Data
aggregated monthly by
group
Two years o f
data from May
2000 to May
2002.
Change in copay for
PPI in insurance groups
over the two-year
period.
Log o f units of
each PPI drug
for each
insurance group
for each month.
Semilog
demand
model but
estimation
procedure
unclear (FE
panel?).
Copay elasticity -0.3
with increase in copay
for all drugs in a class
and -0.9 for drugs
whose price is the only
one to increase.
Huskamp
HA, Deverka
P,
Epstein AM,
Epstein RS,
McGuigan
KA, and
Frank RG.
2003.
Employees and
dependents who had
employer-sponsored
insurance from two
large employers
contracting with a large
health insurer.
Employer 1: 55,567 for
intervention group and
55,951 for control
group.
Employer 2:
11,653 for intervention
group and 27,051 for
control group
Three years
from January
1999 to
December 2001.
At least one
year o f data
before and after
intervention.
Employer 1: switched
from 1-tier $7 copay to
3-tiers, $8 (tier 1), $15
(tier 2), $30 (tier 3) in
2000.
Employer 2: switched
from 2-tiers, $6 (tier 1)
and $12 (tier 2), to
three-tiers, $6 (tier 1),
$12 (tier 2), and $24
(tier 3) in 2000.
Control group plans
matched initial copays
without change.
Comparison of
utilization in
three
therapeutic
categories:
statins, PPI, and
ACEI.
Difference in
rate o f change
between
intervention
and control.
Means
comparison
and two-part
model,
logistic in part
one for
participationa
nd OLS Ion
log
transormed
variable for
utilization.
Employer 1 (relative to
control)
-24% likelihood o f use
for statins and ACEI, -
34% for PPI.
-58.2% less plan
spending for ACEI, -
15.3% for PPI, and-
13.7% statins.
Employer 2 (relative to
control):
-5.6% less plan
spending for ACEI and
-2.3% for PPI.
2.B. Continued
Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission.
ON
Reference Population /
Sample Size
Period Treatment Outcome
measures
Method Results
Rector ST,
Finch MD.
Danzon PM,
Pauly MV,
and
MandaBS.
2003.
Enrollees in four
health plans
operating in the
same state and
using the same
formulary.
Limited to users of
PPI, ACEI and
statins in the first
quarter 1998 and
the third quarter
1999.
ACEI users:
1,347 intervention,
527 control.
21 months
from the
first
quarter
1998 to
third
quarter
1999.
Unit o f analysis is the
claim: Claims record
whether the drug is a
preferred drug and
whether the
enrollee’s plan is a 3-
tier plan.
Difference between
preferred and non
preferred: $15 ($5
and $20, $10 and
$25, $15 and $30) or
$18 ($7 and $25).
Use of
preferred
brands and
non-preferred
brands in three
therapeutic
classes
containing
both preferred
and non
preferred
drugs: PPI,
ACEI and
statins.
Logistic regression o f
the probability of
using a preferred drug
in a tiered plan or in a
no-tier plan.
Average net increase
in preferred brands
for ACEI, PPI, and
statins attributed to
tiered copay is
13.3%.
Fairman KA,
Motheral BR,
and
Henderson RR
2003.
Commercially
insured PPO
members 18 or over
in the Midwest.
3,577 in the
intervention group
and 4,132 in the
control group.
Four years
from
January
1997 to
December
2000.
Intervention group
started with an open
formulary 2-tier, with
$7 generic and $12
brand copays, and
switched to a 3-tier
plan in January 1998,
with $8 generic and
$15 brand copays.
Control group
remained on the 2-
tier plan for four
years.
Number o f
drugs used and
net drug costs
to the insurer.
12 months pre-switch
and 30 months post
switch are compared
in intervention and
control groups.
Bivariate analysis and
2-part model, logistic
and log transformed
dependent variables
for OLS controlling
for age, gender, and
chronic diseases.
Reduction in net cost
growth and lower
third tier utilization
in the intervention
group.
The intervention
reduces the number
o f drugs by 27.3%
and the number of
tier 3 drugs by
10.1%.
2.B. Continued
SECTION 3: ANALYTICAL FRAMEWORK
In this section, I introduce a theoretical framework to analyze individual drug
demand. The model presented below emphasizes the relationships between
employers, health insurance plans, and workers through a system of four linear
equations.
This section is followed by an empirical analysis that proceeds in three steps. I first
describe the law regulating health plans sold to small firms in California, the
rationale for focusing on small firms and the data available to me. I then introduce a
step-by-step simplification of the theoretical model to obtain an estimable model. I
conclude this section with the econometric method used to estimate drug demand.
3.A. OUTLINE OF A GENERAL MODEL OF DRUG DEMAND
I develop here a model outlining the different factors that affect both employees and
firms in the health insurance and health care purchase processes3.
3
This passage was partially inspired by a theoretical summary and a literature review written by
Chemew and Hirth (2002) on health insurance.
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the case of employer-sponsored health insurance, small firms decide whether or
not to provide health insurance to their employees and if they do, which plan to buy.
In economic theory, firms are usually assumed to be profit maximizers. They
purchase health insurance for their employees if they feel that a particular
combination of wages and benefits allows them to better attract and retain the
workers they need in a certain economic environment. Employers provide health
insurance if workers make it a precondition for employment.
Employees first decide which firm they want to join, an action that limits
considerably their health insurance options if they join a small firm. Once part of a
firm, workers can either take up the plan offered, subscribe to another insurance
(public or employer-sponsored through a spouse), or go without coverage. Health
insurance protects employees and their families from the expenses - possibly very
high - incurred for medical treatment. It reduces the risk of income shock. Risk
aversion to large losses, both in terms of income and in terms of health, is therefore a
major motivation for buying health insurance. For those taking up the plan offered
by the small firm, the next decision concerns the amount of health care to consume.
Following the concepts incorporated in the well-known Grossman model [Grossman
1972], individuals maximize utility derived from their own health and their
consumption of non-health care goods. They use health care to improve their health
status.
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Health insurance selection by firms and workers and health care utilization can be
schematically represented as follows (adapted and augmented from Bloomberg and
Nichols 2001):
Supply health insurance firm = f '(firm real premium, sector of activity, labor
market competition, owner characteristics, demand health
insurance w o r k e r)
Demand health insurance w o r k e r = f u(U(health insurance benefit), employee real
premium)
where
U(health insurance benefit) = f m(other sources of health insurance and
household characteristics, risk aversion, E(health care
expenditures) and health risks, demand health care w o r k e r)
and
Demand health care w o r k e r = f IV (income, education, gender, age, provider,
location, health status, time costs, reaction to treatment,
preferences for delivery system and treatment intensity,
copay)
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The quantity of health insurance demanded and supplied varies along three
dimensions: network breadth, number of services covered beyond the minimum, and
out-of-pocket price charged for use. The quantity is a combination of the number of
providers (large or narrow network), number of services covered, and percentage of
medical care costs covered.
Real premiums represent here either the nominal firm premiums minus any offsetting
reduction in workers’ wages or employee premiums plus any offsetting reduction in
wages.
Other variables will be discussed in turn in the following sections. The only
explanatory variables that are observed here are age, gender, location (of the
individual and of the firm), health status, and copay levels.
3.B. SPECIFIC MODEL OF DRUG DEMAND
3.B.1 Health insurance in California
The decision to focus this research on small firms and their employees allows me to
take advantage of the features of California’s health insurance market for small
firms. These features are critical to the empirical estimation strategy presented in
section 3.B.3.
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
California law imposes strict limitations on the design and marketing of plans sold to
small firms (defined as firms with 2 to 50 workers). The law regarding health
insurance offered to small firms is laid out in article 3.1 of the Knox-Keene Health
Care Service Plan Act of 1975 and its amendments. I summarize here only the
relevant portions: Premiums charged by an insurance carrier for a given service are
based on community ratings - namely the area in which a firm is located, whether
the plan purchased by the employee covers his/her spouse and dependents, and the
employee’s age. Insurance companies cannot use information on prior health care
use or adjust for the likelihood of future use. That is, experience rating is explicitly
banned. The region size used for calculating premiums is regulated; plans, for
instance, cannot be sold in an area smaller than an area covering a three digit zip
code. Waiting or an affiliation period of up to two months or, alternatively, exclusion
of reimbursement for preexisting conditions for up to six months is allowed. In
addition, premiums cannot be more than 110 percent or less than 90 percent of the
average charged for the plan. A plan may not reject an application from a small
employer or refuse to renew it except under very limited conditions.These are
referred to as guaranteed issue and renewal clauses. Also, the California Code of
Regulations provides a list of services that must be covered by a health plan
operating in California. These include: physician services, inpatient hospital services,
ambulatory care services, diagnostic lab services, home health services, preventive
health services, emergency services, and hospice services (§ 1300.67; California
Code of Regulations, Title 28, Division 1, Chapter 1).
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The law limits the ability of insurance carriers to tailor their plans to the
characteristics of a particular small firm and its workers.4 (Nor would this be a
profitable practice.) As a result, many small firms end up sharing the same plan. In
my data set, each plan is shared among 10 to 400 firms. The range of copay levels
offered has therefore little to do with the characteristics of a particular firm or its
workforce.
Further, the range of plans available to small firms is much narrower in comparison
to that of large firms. Surveys conducted by the General Accounting Office (GAO,
2001) and the Robert Wood Johnson Foundation (Lee 2002) illustrate the
discrepancy. The market for small-firm insurance is both smaller and far less
competitive than for large firms, resulting in fewer plans and more limited plan
designs from which to choose.5
Finally, there is a practical advantage to choosing small firms as a research subject.
The number of plans offered by small firms to their employees is limited: Many offer
only one plan, and even those who do offer multiple plans contract with a single
4 Carriers can, in contrast, tailor plans for large firms. Large firms, particularly if they are self-insured,
can purchase plans that are specifically designed for them depending on their or their workers’
characteristics; they can even dictate what services (vision, long-term care, etc.) should be covered
and what copays or deductibles should be charged.
5 There is some variation in the number of plans offered to small firms, explained by a firm’s
location. Large markets, like Los Angeles-Long Beach and the Bay Area, are served by more
insurance carriers than rural counties.
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
insurance carrier. This guarantees that all information about possible choices is
available to me and that enrollees have either only one or a few choice(s) of drug
copay structures. Larger firms, in contrast, often contract with several carriers. My
database has information regarding contracts with one large carrier in California.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.B.2 Data
Nature of data: sample selection and final sample
The sample consists of individual health care claims from the administrative
database of a single large health insurer in California. It includes claims from
workers in small firms, their spouses, and dependents who were enrolled
continuously for at least three years, from 10/1998 to 10/2001, and had worked for at
least one small firm during this period. For each enrollee, I kept two consecutive full
years of claims, corresponding to their companies’ contract periods.
From this population, I excluded enrollees 65 and older, i.e. people entitled to
Medicare, to focus on a working population. I then dropped enrollees who moved to
or from another firm, keeping only those who remained in the same firm for the
duration of observation. It’s important to mention here that I do not know the exact
dates employment began, ended, or the transition dates6 , but only the beginning and
ending dates of continuous enrollment. I have lull information on firms’ insurance
coverage, namely what plan they purchased for their employees, the coverage period,
and a complete description of the benefit package. As a result, I know with certainty
the plan workers were enrolled in at any date over the three years if and only if they
remained with the same firm in those three years. Finally, I excluded firms and their
enrollees if they had more than one plan with different copay levels between which
6 This is due to the data extraction algorithm that was used at Prescription Solutions.
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
they could choose. This step eliminated from my final sample those who could easily
switch between plans. The final sample is of people younger than 65, who worked in
the same small firm for at least two years and who had no choice of copay structure.
There were 30,824 enrollees working in 4,554 small firms.
Variables and variable definitions
The following variables were available in the database
On firms:
Names, locations, and health insurance plans with dates of enrollment.
On health insurance plans:
All information on benefit design, including copay, deductibles, and services
covered.
On enrollees:
Age, gender, zip code of residence, date of first enrollment, marital status,
enrollment status (employee, spouse, or dependent), past and present insurance
purchasers, i.e. employers (without dates of employment), beginning and ending
dates of continuous enrollment, and complete insurance claims history, including
drug claims, outpatient and inpatient services, copay and net insurance payments for
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
all medical goods and services, and ICD-9 codes7 for all services for which a
reimbursement claim was submitted.
From these data, I created a set of variables summarizing utilization of drugs and
medical services for each enrollee. Outcome measures are all yearly variables. The
first period is the first full year of data in a plan offered by a specific firm. The
second period is the second full year of data. Periods do not begin and end during the
same calendar year for all enrollees. The first period starts between 10/1998 and
10/1999 without exception. The second period ends between 10/2000 and 10/2001
without exception. The specific month at which a one-year period starts and ends
depends on the month that a firm began contracting with the insurer.
Dependent variables created:
• Insurer payment per drug
• Total insurer payment for drugs
• Number of prescriptions
• Total days of drug supply
• Total quantity of medication (number of pills dispensed)
• Number of drugs in formulary
7 International Classification of Diseases codes - an official list of categories of diseases issued by the
World Health Organization, For definitions of other clinical terms and measures, see appendix B.
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
All variables were also computed for generic and brand drugs separately.
Explanatory variables created:
• Age
• Gender
• County of residence
• Time to first enrollment
• Marital status
• Enrollment status (employee, spouse, or dependent)
• Health status (computed with ICD-9 codes; refer to appendix B for details on
the computation of health status)
Note: With the exception of health status, all variables are time constant. This is
mostly an artifact of the data extraction method: Data for these variables only show
individual status as of 2001, not personal event history, like marriage or change of
address.
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.B.3 Supply and demand for health insurance
The system of simultaneous equations of supply of health insurance, demand for
health insurance, and demand for health care presented in section 3.A. cannot be
estimated as such with available data. Too many explanatory variables are
unobserved for me to estimate the full model. Indeed, most of the variables in the
supply and demand of health insurance are unobserved. So are several variables in
the drug demand equation.
I will show in this and the following sections that a number of simplifications are
possible, reducing the size of the system and making the estimation of drug demand
possible.
The major issue I face in the development of an empirical model is how to deal with
a potentially serious problem called omitted variable bias. This bias would occur if
some unobserved variables affected both copay levels and the quantity of drugs
demanded through their influence on health insurance demand and supply.
I address the bias in two complementary ways. In this section, I discuss the
determinants of health insurance supply and demand, beginning with the supply of
health insurance by the firm. I show that most of them do not affect enrollees’ drug
utilization and argue that they can be ignored without creating any bias. I then turn to
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the matching of workers and insurance through the choice of employer, and assess a
worker’s chance of finding a health plan with the desired characteristics. Current
research supports the idea of imperfect job sorting with respect to insurance; i.e.
workers are not likely to find their ideal plan. As a result, the risk of endogeneity of
the copay variables is further reduced, even though I cannot positively rule out
possible interactions between copay and unobserved variables at this stage.
In the next section (section 3.B.4), I develop an econometric model of drug demand
using the reduced number of variables left from this section in the form of a simple
drug demand equation. I present a two-period panel approach that should allow me
to eliminate the problem of omitted variable bias.
Supply of health insurance benefit by the firm
Two categories of factors influence a firm’s decision to offer health insurance and its
selection of benefits: factors related to a firm’s characteristics and its economic
environment and factors related to workers’ demand for health insurance.
Cost: firm real premium
The cost of health insurance is in the first category. It is the most important factor
influencing the decision to provide health insurance benefits. Surveys (e.g. Fronstin
et al. 2003) clearly show that the overwhelming concern of small businesses is cost.
Excessive premiums and uncertain revenues are the main reasons not to offer health
insurance. This is why firms engage in industrial purchasing: They minimize costs
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for a certain level of quality [Maxwell and Temin 2002 (a)].
The total premium charged by the insurer for a plan depends first on firm size.
Larger firms have lower administrative costs and are more likely to have specialized
personnel to administer benefits. Benefit administrators can shop for the best deals
and use the company weight to obtain low premium prices. Smaller firms, on the
other hand, do not usually employ benefit administrators, find it harder to shop
around, and face higher search costs on a per-employee basis. Small firms pay higher
premium levels for poorer coverage, mostly because insurance administrative costs
are much higher for them: The average loading factor is 40 percent for small firms
compared to about 10 percent for large firms. Higher costs reduce the likelihood of
offering health insurance and prevent the firm from offering more generous coverage
[Bundorf 2001; Morrisey et al. 1994; Glover et al 2000].
Total premium also depends on combinations of copay provisions, network size, and
benefits covered. Community drug utilization has some influence on the pricing of a
plan. But, since experience rating is banned, the actual drug use in a given firm
barely affects the premium.
Further, firms seek to minimize their own share of the total premium cost. And there
is significant evidence that the share employers are willing to pay is independent
from their employees’ drug use. Employers’ contributions vary considerably among
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
small firms. The percentage of total premium paid may be 100 percent, some lower
percentage, or even nothing. In theory, the tax advantage given to employers for
contributing towards health insurance should induce firms to pay 100 percent of the
total premium. Practically, this is rarely observed. Much research on this subject is
still needed to explain this phenomenon, but the most credible models rely on the
concept of imperfect job sorting along with imperfect worker-by-worker shifting of
insurance costs to wages [Gruber and McKnight 2002]. I will discuss imperfect job
sorting in more detail in section 3.4.2. Imperfect wage offset makes it difficult to
accommodate workers’ divergent demands and to pass the costs of health insurance
to workers in the form of lower wages. For instance, low-wage firms tend to charge
higher employee premiums and are less likely to offer any health insurance in the
first place since they find it difficult to pass the cost of health insurance to employees
through even lower wages [Cunningham et al. 1999; Levy 2001; Blumberg and
Nichols 2001; Cutler 2002]. Gruber and McKnight (2002) showed that spousal labor
supply, Medicaid eligibility, and the unemployment rate were among the best
predictors of employer contribution: The higher they were, the lower the
contribution. They also demonstrated that employee medical costs had little
influence on employer contribution rates.
Sector of activity
Work environment differs from one sector to another, with some more likely to
contribute to stress-related conditions and others more likely to cause accidents and
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
disability. In other words, work environment may be related to health and,
consequently, to drug use. But whether work-caused health problems determine
availability of insurance or influence benefit design is another matter entirely.
Health insurance is more likely to be offered in some sectors (services, finance,
insurance, real estate, wholesale trade, utilities, transportation, communication, and
mining) than in others (retail, construction, agriculture, forestry, and fishing).
[National Center for Health Statistics 1997; Kaiser Family Foundation 2002;
Maxwell et al. 2002 b]. Mining, construction, agriculture, forestry, and fishing are all
arguably similar in their levels of physical risk and lead to more accidents and
disability than office jobs in the real estate or insurance sectors. But where mining is
likely to offer insurance, construction and fishing are not. In this particular case, the
discrepancy is a consequence of unionization. And in general, it is labor market
conditions (unionization, as well as wage levels and worker supply) that determine
whether a sector offers insurance and how generous that insurance is - not
differences in the sector’s health risks or its workers’ drug use [Marquis and Long
2001],
Furthermore, accidents do not particularly raise drug demand if they don’t result in
long-term disability. At any rate, accident-related disability is generally dealt with by
other non-health insurance mechanisms such as workers’ compensation.
32
permission of the copyright owner. Further reproduction prohibited without permission.
Labor market and competition for workers
Employers are aware of the potential advantages of health insurance for recruitment
and retention of employees [Fronstin and Hillman 2003; Peele et al. 2000]. The
decision to offer health insurance and the quality of the benefit package depend
primarily on the state of labor competition, namely how difficult it is to hire and
retain the right kind of employees. The harder it is to hire and keep a worker, the
higher the incentive to provide extensive benefits, including health insurance
[Chemew and Hirsh 2002; Bundorf 2001; Marquis and Long 2001]. Conversely,
high unemployment rates - even at the county and local levels - correlate with a
lower probability of offering health insurance. Also, health insurance benefits are
more generous if the small firm has to match benefits offered by other firms in its
sector of activity, and especially if it has to compete with larger firms for the same
workers. [Blumberg and Nichols 2001; Marquis and Long 2001], The scarcity of
employees, not their health care use, drives an industry’s insurance offer rate.
Owner characteristics
White, Asian, and educated business owners are more likely to offer health insurance
[Bernstein 2002]. No connection to coverage generosity has been shown. Blumberg
and Nichols (2001) claim that education makes one more likely to predict the
consequences of a lack of coverage and also helps one select better treatment.
However true for oneself, that does not necessarily mean an educated business owner
will encourage his/her workers to use more health care.
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Health insurance demand from workers
Firms’ actions follow foremost from a desire to attract and retain a certain kind of
worker. Workers’ careers and marketability thus play an important role in
determining the levels of benefits offered. Workers in high demand in a tight labor
market are more likely to get extensive health coverage for lower premiums from
their employers, regardless of their preferences. I determined above that neither
premium, sector of activity, competition, nor owner characteristics significantly
influences final drug consumption.
Since firms’ supply decisions are based mostly on market factors unrelated to
workers’ drug use, I will assume that firm characteristics can be ignored for my
analysis. The supply equation can therefore be reduced to:
Supply health insurance firm = f V (Demand health insurance w o r k e r)
Even this relationship is weak. Small firms may be aware of workers’ preferences
regarding drugs but the number of parameters involved - such as cost, limited range
of plans to choose from, and workers’ heterogeneous preferences regarding access,
services covered, and copay - and their complex interactions make it very difficult
for the firm to accommodate specific demands. The chances of a firm offering a plan
matching workers’ preferences specifically with respect to drug copay levels are
even slimmer.
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To check the validity of my assumption, I ran simple regressions of copay levels8 on
the firms’ employee characteristics - averaging their age, time to first enrollment,
proportion of female workers, proportion of married workers, health status, and
county of location. Results are consistent with my hypothesis. Health status does not
affect copay levels significantly. Copay is primarily related to location and
secondarily to marital status. Marital status may be significant as a proxy for
alternatives in health insurance, allowing married people to be more selective with
insurance take-up.
Employees’ demand for health insurance
A number of factors affect workers’ demand for health insurance: employee
premium, alternative sources of health insurance and household characteristics, risk
aversion, expected health care expenditures and health risks, and demand for drugs.
Here again, most of these factors can be ignored because they are essentially
unrelated to final drug use.
Employee premium
Employee premium is an important factor in the decision to take up a plan offered by
an employer [Feldman et al 1989]. Surveys show that employees are sensitive to
8 These regression results are not shown here (it would cover four lengthy tables for the regression on
brand and generic copays for periods one and two).
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
price (premium, followed to a lesser extent by copay), coverage breadth (what
services are covered), and physician choice [e.g. Booske et al. 1999]. Even though
employee premiums are a significant determinant of health insurance demand, they
do not relate systematically to drug use. The arguments used previously in the
section on firm premium also apply here. Since employee premium is just total
premium minus the employer’s share, and the employer’s share is independent from
drug use, so is the employee’s.
Access to other sources o f care and household characteristics
Decisions concerning health insurance are usually made in the context of a certain
family structure. The availability of other coverage and its generosity influence the
rate of insurance take-up: Being able to obtain better insurance through your spouse
or through public programs makes you less likely to take up the plan offered where
you work [Cutler 2002; Garrett et al. 2001]. The presence of dependents may also
increase the need for extensive coverage with low copay, as the financial risk
increases with family size. These variables, however, do not drive drug use: Being
married, taking care of dependents, or having some alternative source of coverage do
not, in themselves, change utilizatioa
Risk aversion
Risk aversion is related to the risk of income loss due to ill health. Risk-averse
individuals are more likely to seek complete and comprehensive coverage, possibly
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
with lower copay, even if that entails higher premiums. However the link between
risk aversion and drug use is weak: Aversion to taking risks with one’s money does
not translate into any particular attitude towards drug use. One might be afraid of
disease and use more treatment, or one might be afraid of treatment and its side
effects.
Health risks and expected health care expenditures
Health risks and expected health care expenditures are clearly taken into account
when people choose a health plan. Employees who expect high expenditures or know
that they are at risk for costly health problems may seek more extensive coverage,
possibly with the lowest drug copay if they expect treatment to be drug intensive.
There is a potential problem for my research here - selection of high users into low
drug copay plans could be a source of bias in my estimation - but it is mitigated by
the limited number of people who can predict fixture costs: Individual forecasts can
be accurate only for those who suffer from chronic stable conditions like asthma,
resulting in predictable treatment costs, or those affected with serious progressive
conditions like diabetes, resulting in predictably increasing costs.
More importantly for this research, expected drug use should not affect current use,
except in very rare cases. Consider preventive treatment: You may use some drugs
now for a condition, like hypertension or hypercholesterolemia, that could develop
into a serious disease, but the purpose is to prevent current problems and consequent
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
complications, not future drug use. In other words, drug use is related to current
diagnosis (i.e. health status), but not to expected drug use.
Employees’ health insurance choice
I have shown so far that most of the determinants of the demand for health insurance
- except for health care (drug) demand - can be assumed to be unrelated to current
drug utilization. Workers’ health insurance demand can thus be written as:
Demand health insurance w o r k e r = f V I (Demand health care w o r k e r)
Both supply and demand of health insurance are connected through drug demand;
supply depends very little on it, demand somewhat more. Can that be a source of
endogeneity of copay? It is unlikely, because of the overall mismatch between
demand and supply of health insurance caused by imperfect job sorting. An
employee’s demand for health insurance can be satisfied either by lobbying one’s
current employer or by finding employment in a different firm, in the case of small
firms at least. The two phenomena may overlap to a certain extent.
Lobbying current employer
Obtaining the ideal health insurance plan from one’s employer is difficult. First,
firms determine what level of health insurance benefit is necessary not only to satisfy
and retain current workers but also to attract new ones. Second, even among the
current workforce, individual preferences can vary widely and are diluted by the
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
number of people having a say in the decision. Blumberg and Nichols (2001) state
that “economists have yet to develop a satisfactory theory of how a firm actually
aggregates the heterogeneous preferences of its workers.” They mention two
alternative mechanisms, the median worker model and the average worker model,
adding that “neither option has been satisfactorily proven empirically, and alternative
theories clearly are possible. For example, the preferences of higher wage workers
may receive disproportionate weight. There is some empirical evidence that supports
the latter hypothesis, although further research in this area is clearly warranted
[Grubber and Lettau 2000]”. Bundorf (2001), in one of the few studies addressing
this problem, concludes that employee characteristics only marginally affect the
number of plans offered and plan generosity. The dearth of reliable theoretical
models along with the lack of applied analyses does not permit reliable predictions.
Current empirical results grant some influence to workers’ preferences, but it is
doubtful that these preferences relate to any one specific benefit or cost-sharing
provision, like drug copay, in any systematic way.
Finding the right job
Finding the right level of health insurance benefit, at an acceptable price (employee
premium and copay) may involve finding a new job, in a firm offering another health
insurance package. Individuals can choose to work for a certain firm based on their
preference for insurance and for particular plans. They could sort themselves into
different firms to get the appropriate level of coverage. This tendency to sort oneself
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
according to preferences is called job sorting. The related concept of job lock is also
helpful in this case. Two extreme cases can be considered here: perfect sorting and
no sorting. With perfect sorting, firms simply choose any health plan and wait for
workers to move between firms until they find a plan matching their preferences. In
the no-sorting case, individuals do not select their jobs based on health insurance
preferences at all. Perfect sorting is unlikely but so is the total absence of sorting
[Chemew and Hirth 2002]. There is some evidence for job sorting according to
whether health insurance is offered or not. A firm that does not offer it, for instance,
is more likely to attract workers with low preference for health insurance. But the
economic literature is divided on job sorting based on health insurance
characteristics. The number of papers supporting it for all workers or for specific
groups of workers is roughly equal to those that do not support it [Gruber and
Madrian 2002 for a review]. Statistics show that 79 percent of workers with high
value for health insurance are offered health insurance and 71 percent with low value
for health insurance are offered health insurance [Monheit and Vistnes 1999],
meaning that workers in the US are very likely to be offered health insurance
whether they want it or not. The participation rate, as measured in 1997, is 85
percent for eligible workers [Farber and Levy 2000]. Two-thirds of the remaining 15
percent have coverage elsewhere; one-third doesn’t want any coverage. If sorting
was perfect, the take-up rate should be 100 percent.9 Other statistics could be listed
without altering the conclusion. Blumberg and Nichols summarize current evidence
9 See Blumberg and Nichols 2001 for further examples.
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of job sorting as follows: “Facts suggest some workers sorting among jobs with
different compensation packages, they also suggest two other important differences:
(1) most workers are offered health insurance; and (2) there is some sorting among
jobs with more wage than health insurance, but it is highly imperfect and relatively
uncommon.”
Widespread job sorting based on health insurance is unlikely for several reasons.
Jobs vary in many dimensions besides health insurance, like career ladders, working
conditions, nature of the work itself, and complete benefit packages, including
retirement savings and life insurance [McLaughlin 1999]. Health insurance plans
also vary in numerous ways, from range of benefits to price (premium or copay) to
provider network. Searching for the ideal plan in a suitable firm is costly; collecting
all relevant information about health insurance plans offered by prospective
employers is burdensome and time-consuming, especially when added to
information on other work characteristics. These high search costs lead to “status quo
bias,” or the tendency people have to stick with what they know best - their current
health insurance plan - and to be reluctant to change. The significant persistence of
an individual’s choice of health insurance is clearly illustrated by research done on
job lock. Insurance transitions have negative effects; people with unstable private
coverage have a relatively low likelihood of having a usual source of care [Lee
2002]. Insurance contracts usually include provisions like waiting periods (no
coverage for up to two months) or exclusion of coverage for pre-existing conditions
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(up to six months) that increase switching costs and further contribute to status quo
bias and mismatch between individual preferences and insurance coverage. In
general, individuals with the highest stake in health insurance have the hardest time
finding the right match, as shown by the following evidence: Chronically ill workers
or workers with chronically ill dependents are more likely to experience job lock.
Older and sicker workers experience higher costs of switching, resulting in less job
mobility. Women tend to place a higher value on health insurance than men, partially
because of higher health care use. There is good evidence of job lock among women
(dual or sole earner), but little or no evidence of job lock among men (dual or sole
earner). Poor, older, or sicker enrollees are the least likely to switch insurance plans.
[Gilleskie and Lutz 1999; Brunetti et al. 2000; Stroupe et al. 2000; Royalty and
Solomon 1999; Buchmuller and Valetta 1996; Strombom et al. 2002]. Finally,
Cunningham and Kohn (2000) showed in a survey that less than 25 percent of all
workers change plans because of plan characteristics and fewer than 16 percent for
costs (all cost-sharing provisions and employee premiums). These results should be
interpreted with caution: The survey was conducted on employees of both small and
large firms. Switching because of plan characteristics may be higher in large firms
since it is easier to collect information and enroll in a different plan in a firm that
offers multiple plans. Furthermore, it did not specifically target drug benefits. Based
on that survey, up to 16 percent of people could have switched because of drug
copay alone, but the figures are most probably considerably lower.
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Caveat: high users
High drug users, such as people suffering from chronic conditions, are the most
likely to be sensitive to the differences in drug copay levels observed in my data set.
Since the distribution of drug expenditures is highly skewed, this is a cause for
concern. I can’t completely dismiss the idea that some heavy users shop around for
the job offering health insurance with the lowest drug copay but this risk is mitigated
by several factors. First, by small firms’ plan structure: Most of the plans in my data
have formularies. The list of drugs included in the formularies may vary from plan to
plan. Since actual out-of-pocket expenditure depends on whether the drugs used are
in the formulary or not, it may be difficult for high users to accurately forecast drug
costs in another plan. Second, high users concerned about out-of-pocket costs would
probably seek employment in a large firm, since large firms tend to offer better
coverage at lower cost, so they would not be in my population. This point is valid for
all enrollees in my sample, not just high users: Rational individuals with a high
preference for care and/or high medical care use should seek employment in larger
firms, all else the same. My population should therefore contain few such
individuals, with the exception of those who found it too costly to leave their jobs.
Third, high users are the most likely to experience job lock because of restrictions
imposed on insurance portability (like exclusion of coverage for preexisting
conditions).
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
It is worth noting the research of Strombom et al (2002). They found that after
controlling for employee premium, high health care users prefer less tightly managed
plans over low copay. This may apply to high drug users as well.
Supply and demand
So far, I have determined that small firm characteristics can reasonably be ignored in
my drug demand model because the most important factors are independent from
health care demand and because employees have great difficulties obtaining a plan
that matches their preferences closely. For the average employee, the number of
options regarding benefit levels, cost sharing provisions, and premiums is very
limited. The degree of mismatch between supply and demand of health insurance in
small firms has an important consequence: Workers are unlikely to find the level of
*
drug copay they desire. Employees faced with only a handful of options should be
more sensitive to large differences in the benefits offered or the premiums charged
than to the minor differences in copay as observed in my data set. This reduces
substantially the risk of endogeneity of copay, an important point for the estimation
of the model of drug demand presented in the next section.
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.B.4 Drug demand
Since only variables affecting both drug demand and health insurance choice should
concern us, the equation to be estimated becomes:
Demand Drug w o r k e r = f (income, education, gender, age, provider, location, health
status, time costs and reaction to treatment, preferences for delivery system and
treatment intensity, copay).
This equation provides the basis for the statistical model presented in the next
section.
3.C. METHOD
The demand equation can be written with the following linear model:
For enrollee i= l,...,n and period t=l,2
Yjt = p0 + PXit + yZj + E it
with
Yit drug expenditures, number of prescriptions, generic fill rates
Xjt time-varying characteristics: copay level and health status
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Z; time-invariant characteristics (in my data set): age, gender, marital status,
employment status, and location
S it unobserved disturbances
Throughout my analysis I will maintain that copay and health status affect all
individuals identically, i.e. the slope ( 3 is the same for all individuals. The question to
be answered to ensure consistent estimation is whether there is a correlation between
individual unobserved characteristics and copay levels, a potential problem since
employee drug demand is related to demand for health insurance. The omitted
variable bias due to unobserved heterogeneity could be introduced through the
simultaneous individual choice of drug copay and drug consumption.
3.C.1 Cross-section
I will show here that cross section analyses of periods one and two are likely to
produce inconsistent estimates. With possible interactions between copay and
unobserved individual characteristics, the assumption of strict exogeneity of copay
level cannot reasonably be maintained in cross section.
In this section, I introduce unobserved variables and elaborate on their relationship
with drug demand and copay in the context of cross section analysis. It will not only
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
help me discuss the sources and direction of bias in the OLS but also provide the
necessary background for panel analysis. In the next section, I will introduce the
panel analysis that should allow me to eliminate potential biases. (I present the OLS
results for comparison purposes in section 4.)
Drug demand depends on a number of factors, some of which are observed, like
gender and age, location (zip code, county), current health status, and copay, and
some of which are not, like income and wealth, education, provider, time costs,
reaction to treatment, preferences for delivery systems and treatment intensity. I
focus here on unobserved variables.
Income and wealth
Income and wealth can influence both the choice of health insurance and the amount
of drugs consumed. Ceteris paribus, high income workers can more readily afford
drug copay. They have also been shown to prefer more comprehensive benefits from
their health insurance, with more provider choice and fewer restrictions on use,
partly because of tax incentives (high wage workers have high preference for
insurance since they are not taxed for the firm’s contribution and can afford higher
employee premiums [Peele et al. 2000]. Lower income workers tend to be more
sensitive to out-of-pocket expenditure and to seek plans offering low premium and/or
low copay. Comprehensive, low restriction plans tend to result in higher premiums
and higher copays. In general, we can expect to see a slightly positive covariance
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
between income and copay and a positive covariance between income and drug use
(all else the same). This should introduce a slight upward bias on the estimates of
drug copay effects on drug use, towards zero.
Education
Education levels can also influence both the choice of health insurance and the
amount of drugs consumed. More educated people may better know the
consequences of insufficient health insurance coverage. They may be more
concerned about health, seek medical advice more readily and be more compliant
(since they might better understand the results of incomplete treatment). They may
also put more trust in the health care delivery system. [Blumberg and Nichols 2001].
This could possibly point to an increase in drug use and a desire for better health
insurance, with more comprehensive coverage and more freedom to choose
providers. Following the line of reasoning developed for income, higher education
levels may be positively correlated to higher copay. Hence, this unobserved variable
could introduce a small upward bias in my estimates.
Providers
Providers (e.g. physicians) follow specific prescribing patterns, and they have
considerable influence on the amount of medical care consumed in the US. Providers
recommend drugs and treatment intensity. The relationship between providers and
drug copay, however, is tenuous. In principle, the most restrictive plan could
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
constrain physicians’ prescribing patterns directly, say through prior authorization
requirements, mandatory generic substitution and treatment guidelines. This could
result in a positive correlation between high copay plans that give providers more
freedom and high intensities of treatment. This correlation would introduce an
upward bias in my estimates. The effect is extremely unlikely to occur, however.
First, there is no indication that the insurer I worked with uses any of the most
stringent forms of utilization management. Second, all plans available in my
sample - HMO and POS - use similar forms of utilization management, if any.
Time costs and reaction to treatment
Time costs and reaction to treatment can influence the number of prescriptions used.
The former by making it more or less difficult to obtain them, the latter by increasing
or lowering the amount needed. Their influence on the choice of health insurance is
less straightforward. If time is an issue in getting treatment, it might also prevent
careful selection of insurance and result in higher copay. This would bias the
estimate downward. Reaction to treatment is essentially unpredictable, unless the
enrollee has significant experience with a certain course of treatment. Predicting
reaction to treatment is then akin to forecasting costs accurately: For the enrollees
capable of this feat, high drug use might correlate with lower copay and induce a
downward bias.
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Preferences
Preferences are the most significant source of bias in cross section. Individual
preferences can be divided into two categories. Preferences concern either the choice
of providers - access to a specialist without referral, for example - or the intensity of
treatment. The first type mostly affects the choice of health insurance, with the
attending consequence, as I have assumed so far, that more freedom probably
correlates with higher copay. Since looser controls on utilization may lead to higher
use, the bias would be mostly upward. The second type, preference regarding
intensity of treatment, has the potential to introduce a sizeable downward bias in my
analysis. Enrollees, knowing they would rather use as much care as possible given a
certain condition may seek a low level of drug copay and consume more drugs than
the average enrollee. Contrary to the other potential sources of bias in cross section,
bias introduced by heterogeneous preferences for treatment intensity could be
quantitatively significant.
3.C.2 Panel: fixed effect estimation
As discussed in the previous paragraphs, cross section analysis is likely to be biased.
I will show here that fixed effect estimation in a two-period panel should eliminate,
or at least greatly minimize, potential biases.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Adding an unobserved individual effect to the equation presented in the previous
section, we obtain the following equation: for enrollee i=l,...,n and period t=l,2
Yit - Po + p X it + yZj + a* + Sit
with, as before,
Yjt drug expenditures, number of prescriptions, generic fill rates.
Xjt time-varying characteristics
Zi time-invariant characteristics
oij unobserved individual effects
E jt unobserved disturbances
Using the first difference transformation of the previous equation, the unobserved
individual effect disappears and the model becomes
AYi = AXj p+ Aej
with
A Yjt difference in drug expenditure, number of prescriptions
AXjt difference in copay level and health status
Asjt difference in unobserved disturbances
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Consistent estimation of the parameter of interest with the first difference estimator
depends critically on the assumption of strict exogeneity of copay levels. Namely,
E(CopayjS, e;t)=0 for all s and t = 1 ,2 must hold. This condition can also be written as
E(eit| Copayii, Copaya, otj) = 0 for t=l,2 . In essence, copay levels in each period
need to be uncorrelated with unobserved disturbances in each period, conditioning
on the individual effect, for the strict exogeneity assumption to hold.
In this section, I discuss evidence supporting this necessary assumption. I intend to
show that important simultaneous determinants of drug demand and health insurance
demand are time-constant individual effects. The question centers on which
unobserved factors change over a period of two years and which can be assumed to
be constant - in other words, whether they are part of the individual effect or are
random disturbances. Special attention must be paid to whether period one error
correlates with period two copay. Following the same layout as for cross section, I
discuss in turn the sources of variations - individual effect or idiosyncratic error - in
drug use, focusing on the possible correlation between idiosyncratic error and copay
levels.
Income
Income may fluctuate over a period of two years. Are income shocks during the first
year likely to induce an enrollee to alter his/her drug use and look for a different
plan? Following the permanent income hypothesis, temporary shocks to income are
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
unlikely to alter consumption behavior as consumption is far less variable over time
than income. Some subgroups, like heavy users of low means, may be more sensitive
to income shocks. However, these people (1) may be hard-pressed to spend anything
on drugs, (2) may well qualify for Medicaid, (3) are likely to have been already
enrolled in a plan with low copay, and (4) would find it difficult to quit their jobs to
enroll in a new plan somewhere else. OveraP mid be no correlation between
period one income shocks and
Educatic
Educatii
that sr
educati
Providers
Providers can be divided into usual sources of care (e.g. ,, 1
temporary sources of care (e.g. a specialist consulted for a specific problem). The
influence of the usual source of care would disappear in a fixed effect analysis.
Using a temporary source of care, with different prescribing patterns, should not
induce the enrollee to search for a new plan.
i years. It is unlikely
'an extra year of
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Preferences
Preferences pose the main difficulty. Economic theory usually assumes that
individual preferences are stable over time. Preferences related to unrestricted access
are unlikely to change quickly and be subject to shock. The situation is fuzzier for
the intensity of treatment preference. It has a permanent component, since one’s
attitude towards drug treatment is unlikely to change in a matter of weeks except in
unusual circumstances, like the death of a close relative. Are shocks to preference
lasting less than a year likely to prompt people to look for a different plan? Given the
considerable investment of time and energy required to find a new job with a
different drug copay structure, temporary shocks would need to be exceptionally
intense to justify the effort. (Indeed, if the shocks are life-altering, they are, by
definition, unlikely to be just “temporary.”) Preference changes that are significant
occur smoothly, over a period of time. Preferences can therefore be considered time
invariant (over a period of two years) for the purpose of estimation, because they are
not usually subject to shocks of sufficient intensity to trigger a job search.
Time costs and reaction to treatment
Small firm employees may occasionally have more time to find a new plan (by
looking for another job), but that does not give them any reason to do so. Hence,
there is little connection between fluctuation in time cost in period one and copay in
period two. I will assume that reaction to specific treatment is not subject to shocks.
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Fluctuation in use
Of the variables listed above, health status, provider, reaction to treatment, and
preference-based compliance are likely to explain most of the individual variations in
drug consumption in the short run. Because shocks to these variables are not
correlated to future copay levels, the assumption of strict exogeneity for fixed effect
estimation should hold.
Estimation
The fixed effect estimator is the OLS estimator of the regression of AY; on AX, for
n=l to 30,824. All fixed effect estimations were conducted with a two-stage
weighted least square (WLS) difference estimator, an efficient estimator. A White’s
test reveals heteroskedasticity in the data probably due to a significant increase in
drug use variation with worsening in health status. The equations estimated had
generic drug copay and brand drug copay as covariates of interest. Non-formulary
copay is never included since it is colinear with brand copay. Other covariates
include changes in health status.
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
SECTION 4: RESULTS
4.A. HOW REPRESENTATIVE IS THE SAMPLE?
A comparison of HMOs active in California
How does the sampled insurer compare to other HMOs active in the small group
business? Are the small firms it insures similar to other small firms in California?
Evidence suggests that both the plans and the small firms in my sample are
representative.
There are six insurers active in the small group market statewide: Aetna, Blue Cross,
Blue Shield, Health Net, Kaiser, and PacifiCare.
According to the California HMO report card, at
http://www.opa.ca.gov/report card/, the sampled insurer is often rated good, and
usually not worse than other HMOs. The counties it serves are relatively broad and
representative of both rural and urban counties in California. Health care utilization
was similar, on a per-member per-month basis, to utilization in other large HMOs.
For more specific information, refer to Baumgarten 2002. More importantly, the
range and average of drug copay levels are similar to those of other HMOs. Trauner
(2003) showed that 2001 plans (the only year data are available) share similar levels
of benefits across major HMOs. The California average is a $20 copay for brands
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and a $10 for generics. In 2000, the national average copay was $14 for brands and
$8 for generics (Kaiser/HRET 2001). In both cases, copay levels are similar to those
observed in my sample. There is thus little indication that the sampled insurer offers
plans that are fundamentally different from other plans sold in this market.
Along with comparing statewide insurers, it is important to compare the customers in
the small group market. Since I have information only on location and some
workforce characteristics, I compared these to published surveys [e.g. California
Healthcare Foundation/Mercer Survey 2000], Based on available information, there
is no indication that my sample is not representative of small firms’ enrollees in
California.
Comparison between final sample and excluded sub-samples
It is important to compare people in the final sample and those in the excluded
subsamples. The questions of interest are whether these samples are different,
particularly with respect to drug utilization, and whether these differences can be
explained by variations in observed variables.
Two subsamples were excluded from the final sample: enrollees who worked for
more than one firm over the duration of observation and enrollees who worked for a
small firm that offered plans with different copay structures. Both excluded
subsamples are smaller than the final sample. The following table shows the number
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of enrollees in the final sample and in each of the excluded subsamples.
Table 1
Size of final sample, in number of enrollees and number of firms employing them,
and size of the excluded subsamples
Final sample
One firm - one copay
structure
Excluded subsample 1
More than one firm
Excluded subsample 2
One firm - more than
one copay structure
Number of
enrollees
30,824 3,995 9,994
Number of firms 4,554 - 939
The first comparison is between the final sample and subsample 1, those who moved
among employers. Differences in means and frequencies are shown in table 2.
Age, the proportion of females, and enrollment as a spouse are similar between the
two groups. But those who switched employers are more likely to be married and
less likely to have dependents - probably because childless married people find it
easier to switch employers, relying on their spouse’s income and health insurance to
compensate for the loss of wage and coverage. This should not, however, affect drug
use (see arguments in section 3.B.3).
Differences in the use of medical services, illustrated by the number of office visits
and the number of hospital outpatient claims, are not statistically significant. Neither
are net payments on prescriptions. These figures suggest similar health care
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
utilization patterns. The number of prescriptions, however, is significantly higher in
subsample 1 - by about 0.74 prescription per year. This can explained by a
difference in health status; a higher proportion of people in subsample 1 have
medical problems (appendix A, table la).1 0 It should not pose any problem because
(a) as the logistic regression (appendix A, table lb) shows, health status has little
influence on the probability of changing employers and (b) health status will be
accounted for in the multivariate analysis.
Table 2
Comparison between final sample and excluded subsample 1: Age, gender,
enrollment status, and selected utilization measures over the three-year enrollment
period
Final sample
One firm - one copay
structure (a)
Excluded subsample 1
More than one
firm (b)
Difference (b)-(a)
(std. error)
Age (years) 35.3 35.7 0.4 (0.28)
% female 49.7 51.5 1.8
% spouse 17.0 15.8 -1.2
% dependent 29.6 25.8 -3.8
% married 35.5 42.0 6.5
Number of Rx 18.74 20.98 2.23 (0.18)
Net paid Rx $ 799.66 905.88 106.20 (53.56)
Number of office
visits
18.34 18.1 -0.24 (0.48)
N umber of hospital
outpatient services
2.43 2.40 -0.03 (0.16)
Table 3 shows the differences in means and frequencies between the final sample
and subsample 2, those who worked in the same firm and could choose among
different copay structures. The more-than-one copay structure group is slightly
1 0 For a detailed description of health status variables - adgl to adg32 - refer to appendix B.
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
younger and has a higher proportion of married people. Gender and enrollment status
are similar. There was no significant difference in the number of office visits and
hospital outpatient services. A small significant difference can be observed in drug
use: The final sample has more prescriptions per year (0.85) and higher net
prescriptions costs ($82.46). This can also be explained with a difference in health
status; a higher proportion of people in the final sample have medical problems
(appendix A, table 2a).
This shouldn’t be a problem - based on both individual-level and firm-level logistic
analyses. The individual-level logistic regression (appendix A, table 2b) shows that
little besides marital status and county of residence matters for the probability of
working in a one-copay-structure firm. The firm-level logistic regression (appendix
A, table 2c) - measuring the probability of offering plans with different copay levels
and using average enrollees’ characteristics as independent variables - confirms that
the availability of copay options is not related to enrollees’ characteristics. The only
significant coefficient was the number of employees a firm had enrolled in its plan;
the more employees, the more likely a firm was to offer a choice of copay structures.
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3
Comparison between final sample and excluded subsample 2: Age, gender,
enrollment status, and selected utilization measures over the two-year enrollment
period
Final sample
One firm - one
copay structure (a)
Subsample 2
One firm - more than
one copay structure (b)
Difference
(b)-(a) (std. error)
Age (years) 35.3 34.6 -0.53 (0.097)
% female 49.7 49.2 -0.5
% spouse 17.0 17.8 0.8
% dependent 29.6 29.9 0.3
% married 35.5 37.2 1.7
Number of Rx 11.85 10.15 -1.7 (0.23)
Total net paid for Rx $ 502.23 419.77 -82.46 (24.59)
Number of office visits 11.68 11.26 -0.42 (0.24)
Number of hospital
outpatient services
2.64 2.50 -0.14(0.19)
To conclude, the differences between the subsamples and the final sample are either
small, can be accounted for with observed variables, or can be ignored as unrelated
to drug utilization and should therefore not prevent me from generalizing my results.
4.B. SAMPLE HEALTH INSURANCE PLANS
Description
What did the sampled insurer offer to its small firms? This section introduces the
different plans and concludes with a description of the copay change over time.1 1
1 1 Information on drug utilization in each of the plans presented here can be found in appendix C.
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
77.6 percent of the enrollees in the final sample have an HMO plan, 22.4 percent a
Point-of-Service (POS) plan. For reference, HMO plans are managed care plans
contracting directly with participating providers where the enrollees bear the full cost
of services from non-network providers. POS plans are plans that allow enrollees to
obtain HMO services from in-network providers at a low copay rate, but also
reimburse claims from out-of-network providers at a higher rate.
Among the plans in this sample, there are only five combinations of brand and
generic copays in the first period, and seven in the second. Tables 4a and 4b present
each plan, its drug copay levels, and the number of people enrolled in it. Plan
numbering is arbitrary.
In period one, 55 percent of all enrollees are in the three-tier plan, with one rate for
brand drugs, one rate for generic drugs, and one rate, the highest, for non-formulary
drugs. Most of the others (about 45 percent) are in two-tier plans with identical brand
and generic copays and a higher non-formulary copay. A few enrollees (0.3 percent)
are in one-tier plans. The most common is the three-tier plan, with a $10 copay for
brands, a $5 copay for generics, and a $20 copay for non-formulary drugs.
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4a
Plans available in period one, with number of enrollees
Period 1
Brand copay $ Generic copay
$
Non-formulary
copay $
Number of
enrollees
Plan 1, one-tier 5 5 None 71
Plan 2, two-tier 10 5 None 97
Plan 3, one-tier 10 10 None 36
Plan 4, three-tier 10 5 20 16,958
Plan 5, two-tier 10 10 20 13,584
In period two, the distribution among one-, two- and three-tier plans remains
essentially the same.
Table 4b
Plans available in period two, with number of enrollees
Period 2
Brand copay $ Generic copay
$
Non-formulary
copay $
Number of
enrollees
Plan 1, one-tier 5 5 None 53
Plan 2, two-tier 10 5 None 110
Plan 3, one-tier 10 10 None 29
Plan 4, two-tier 15 10 None 7
Plan 5, three-tier 10 5 20 16,843
Plan 6, two-tier 10 10 20 13,588
Plan 7, three-tier 20 10 25 116
Change of Copay Over Time
As the above tables suggest, few people actually experienced a change in their level
of copay from period one to period two, probably because there was little overall
change in health insurance plans during 1999-2000. 256 people, however, did go
63
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
through a copay change, with the following breakdown (non-exclusive change
categories):
+$10 brand copay and +$5 non-formulary copay for 116 enrollees;
+$5 brand copay for 25 enrollees;
-$5 generic copay for 7 enrollees;
19
+$5 generic copay for 127 enrollees .
Since I worked only with benefit design descriptions and not with plans per se, I do
not know whether existing contracts saw an increase in copay or whether some firms
were switched to new contracts with higher copay.
4.C. DESCRIPTIVE STATISTICS
Drug utilization
All drug utilization summary variables are yearly totals. Period one refers to the first
full year of data in a plan offered by a specific firm. Period two refers to the second
full year of data. Periods do not begin and end during the same calendar year for all
enrollees. The first period starts between 10/1998 and 10/1999 without exception.
The second period ends between 10/2000 and 10/2001 without exception.
1 2 The exact breakdown was not recorded.
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Drag utilization variables include the net payment made by the insurer, the number
of prescriptions, and the total days of supply and quantity of medication. All of these
variables were evaluated in four categories: all drugs regardless of type, generic
1 ^
drugs, brand drags, and multisource drags.
Table 5 shows drug use over each period of the sample. Two-thirds of enrollees
filled at least one prescription, and the average number of prescriptions is low (5.7 in
period one and 6.2 in period two). Most variables show an increase in drug use
between periods one and two, and a relative shift away from generic drags toward
brand drugs. Also, all drags become more expensive for the insurer in period two.
Table 5
Drag use in period one and two: means and proportions
Period one Period two
Proportion of users (at least one rx) 66.4% 68.2%
Number of prescriptions 5.66 (std.dev. 10.7) 6.19 (std.dev. 11.7)
Number of generic drugs 1.72(4.29) 1.72(4.33)
Number of brand drugs 2.95 (6.16) 3.47 (6.98)
Number of multisource drugs 0.98 (3.21) 0.99 (3.18)
Average paid net of copay (all drugs) $ 20.86 (38.4) 24.13(43.54)
Total paid net of copay (all drugs) $ 225.93(1160.7) 276.34 (1266.4)
Average paid for generic drugs $ 3.91 (9.45) 4.40(11.04)
Average paid for brand drugs $ 22.63 (46.73) 26.57(55.15)
Average paid for multisource $ 7.41 (24.88) 7.76(27.41)
Days of supply (total) 154.90(310.45) 174.17(339.6)
1 3 “Multisource drugs” is here the default category for drugs that were neither generic nor single
brand.
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Demographic characteristics
Table 6 presents the final sample’s demographic characteristics, as recorded in 2001.
24.4 percent of this population is between ages 0 and 18; 63.9 percent are between
19 and 55, and 11.7 percent are in the 56 to 64 age range. Over half of the enrollees
are employees. 35.5 percent of the enrollees are married. 49.7 percent are females.
Table 6
Demographic characteristics and enrollment status
Variable Means and proportion
Age (years) 35.3
% age 0-5 2.7
% age 6-10 8.1
% age 11-15 8.8
% age 16-18 4.8
% age 19-20 3.1
% age 21-25 2.9
% age 26-30 4.1
% age 31-35 8.3
% age 36-40 11.2
% age 41-45 12.7
% age 46-50 12.0
% age 51-55 9.7
% age 56-60 7.6
% age 61-64 4.1
% femal e 49.7
% employee 53.4
% dependent 29.6
% spouse 17.0
% married 35.5
Health status
Table 7 shows the health status of enrollees. Health status is measured with a
classification system called ACG. An individual is assigned an ICD-9 code during
each medical encounter. ICD-9 codes correspond to ADG categories and the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
categories for the whole year are then aggregated into an ADG profile for that
individual, namely a profile of conditions that person suffers from. Note: ADG
categories are non-exclusive. Refer to appendix B for details about the method used
to assign individual ICD-9 codes in the ACG system and the method used to assign
ADG categories.
Table 7 focuses on the proportion of enrollees in each ADG category, with the same
individuals possibly appearing in several different categories. Conditions known for
their associated high drug use are worth noting; for instance, chronic medical stable
conditions, which affect 16.7 percent and 20.2 percent of the enrollees in periods one
and two, respectively. (For the proportion of enrollees with multiple ADGs and
diagnoses, refer to table 3a and 3b in appendix B.)
Table 7
Health status profile for periods one and two: percentage of enrollees who received a
diagnosis in each ADG category
ADG Content Period 1
%
Period 2
%
ADG-1 Time limited: minor 13.7 15.3
ADG-2 Time limited: minor-primary infection 24.4 25.6
ADG-3 Time limited: major 2.4 2.9
ADG-4 Time limited: major-primary infection 3.3 3.5
ADG-5 Allergies 5.3 6.3
ADG-6 Asthma 2.6 2.8
ADG-7 Likely to recur: discrete 8.8 10.5
ADG-8 Likely to recur: discrete-infection 12.3 13.5
ADG-9 Likely to recur: progressive 0.6 0.8
ADG-10 Chronic medical : stable 16.7 20.2
ADG-11 Chronic medical: unstable 5.9 7.1
ADG-12 Chronic specialty: stable orthopedic 1.4 1.7
ADG-13 Chronic specialty: stable ear, nose, throat 0.4 0.5
ADG-14 Chronic specialty: stable eye 4.8 5.9
67
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Table 7 Continued
ADG-16 Chronic specialty: unstable orthopedic 0.8 1.1
ADG-17 Chronic specialty: unstable ear, nose, throat 0.1 0.1
ADG-18 Chronic specialty: unstable eye 1.7 2.0
ADG-20 Dermatologic 7.6 9.7
ADG-21 Injuries or adverse effect: minor 9.9 10.6
ADG-22 Injuries or adverse effect: major 6.8 7.1
ADG-23 Psychosocial: time limited, not severe 1.2 1.6
ADG-24 Psychosocial: persistent or recurrent, stable 3.4 4.7
ADG-25 Psychosocial: persistent or recurrent, unstable 0.7 0.9
ADG-26 Signs or symptoms: minor 13.2 16.0
ADG-27 Signs or symptoms: uncertain 19.8 23.0
ADG-28 Signs or symptoms: major 12.3 14.3
ADG-29 Discretionary 6.5 7.3
ADG-30 See and reassure 2.0 2.5
ADG-31 Prevention or administrative 38.4 41.5
ADG-32 Malignancy 1.5 1.9
ADG-33 Pregnancy 2.5 2.3
Comparison of enrollees with copay change and enroliees without copay change
A brief comparison of basic demographic characteristics of the group of enrollees
whose plans had a rise in drug copay with the group whose copay remained constant
over two years is shown in table 8. There are no significant differences in age or
gender, but there are significantly more married people in the group that experienced
a change in copay.
Table 8
Demographic characteristics
Change in copay group No change in copay group
Age (years) 35.9 (std.err. 0.1) 35.4 (std.err. 1.06)
% married 46.1 35.4
% female 49.6 49.7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
There are very few differences in health status (not shown here) between enrollees
with change of copay and enrollees without change. The only significant difference
between the two groups was in period one in the ADG 8 (see appendix B), the ADG
for infections that are likely to recur. In period two, the difference disappeared.
Tables 9a and 9b compare the drugs and services used by the two groups. Table 9a
presents the baseline drug utilization. There are no significant differences in drug
use.
Table 9a
Drug use in period one
Change in
copay group (1)
No change in
copay group (2)
Difference
(2)-(l) (std.err.)
Number of prescriptions 4.74 5.67 0.93 (0.65)
Number of generic drugs 1.62 1.72 0.10(0.27)
Number of brand drugs 2.30 2.97 0.66 (0.39)
Number of multisource drugs 0.82 0.98 0.16(0.20)
Total paid for drugs $ 167.25 226.42 59.16(72.85)
Average net paid per prescription $ 16.46 20.90 4.43 (2.41)
Table 9b compares the changes for both groups in service utilization, namely the
changes in number of hospital outpatient services and office visits. The differences
are not statistically significant. These measures are consistent with expectations,
since neither group experienced any relative change in health status.
69
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Table 9b
Change in service utilization between period one and period two
Change in copay
group (1)
No change in
copay group (2)
Difference
(2)-(l) (std.err.)
Change in number of office
visits
0.12 0.59 0.46 (0.72)
Change in number of hospital
outpatient services
0.11 0.16 0.05 (0.19)
Because the demographics and health status of the two groups are similar as are their
baseline drug use and change in service utilization, there is no indication of enrollee
selection out of the change-in-copay group.
Table 10 summarizes changes in drug use for enrollees who kept the same copay
levels over two years and those who saw a change in their copay levels. The only
significant difference between the two groups is in the number of generic drugs used:
The change-in-copay group has a significant reduction relative to the no-change-in-
copay group.
Table 10
Comparison of drug use change for enrollees whose copay changes
and enrollees whose copay remains stable
Change in...
Change in copay
group (1)
No change in
copay group (2)
Difference
(2)-(l) (std.err.)
Number of prescriptions -0.11 0.53 0.64 (0.33)
Number of generic drugs -0.57 0.01 0.58 (0.17)
Number of brand drugs 0.33 0.51 0.18(0.23)
Average paid net of copay (all
drugs) $
2.24 3.28 1.04(2.38)
Total paid net of copay (all
drugs) $
18.85 50.65 31.8(41.29)
Days of supply (total) 3.3 19.4 16.1 (9.23)
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.D. MULTIVARIATE ANALYSIS
Results are presented in two parts: The first part introduces results from the panel
analysis and the second part compares these results with the cross section results.
Fixed effect estimation
Table 1 la contains results from the estimation without intercept, for all enrollees.
Tables lib and 11c contain results from the estimation with intercept, for all
enrollees. In all cases, independent variables include change in brand copay, change
in generic copay, and change in health status (i.e., the difference in ADG 1 through
ADG 33).
In tables 11a and 1 lb, the dependent variables include differences in mean insurer
payment, total insurer payment, number of prescriptions, and total days of supply.
The difference in mean payment refers to the difference in the actual cost -
ingredient cost plus dispensing fee plus sales tax - per drug for the insurer (net of
copay); the difference in total payment is the difference in total cost of drugs for the
insurer (net of copay). Total days of supply is the sum of the days of supply for all
prescriptions. In table 11c, the dependent variables include differences in quantity of
drugs (e.g. number of tablets), total insurer payment for generic drugs, total insurer
payment for brand drugs, and number of drugs that are in the formulary.
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The comparison of the estimates, with or without intercept, and the results presented
in table 10 point toward an underlying upward trend in drug use and in drug costs.
This may be partially due to my sample population getting older, but there has also
been significant drug price inflation over the past 10 years in the US, along with
increased drug use. Table 1 la shows that few coefficients are significant when the
underlying trend toward higher drug use per person is left in the data. I will therefore
focus on the more interesting table lib.
From table 1 lb, we can see that an increase in copay seems to reduce both the total
drug payment for the insurer (-7.06 for generic copay increase and -2.50 for brand
increase) and the total days of supply (-3.3 days for generic copay and -1.1 day for
brand copay increase). The coefficients for mean payment have the right sign, but
only the coefficient for generic copay is significant. Higher generic copay lowers
mean payment slightly, by -0.56.
Table 1 lc provides additional details that can be combined with the previous results
into this outline:
Higher brand copay has the following effects:
• It decreases brand drugs in quantity (e.g., capsules, tablets) and days of
supply but not in number of prescriptions.
• It decreases total insurer payment on brand drugs.
72
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Higher generic copay has the following effects:
• It lowers the number of drugs in formulary. (That is, it increases the number
of non-preferred drugs used.)
• It decreases the number of generic prescriptions.
• It decreases total insurer payment on generics.
Even relatively small changes in copay produce a significant degree of response. The
response is stronger to changes in generic copay. Higher generic copays lead to
greater savings for the insurer - both through the reduction in the number of generic
prescriptions and the increase in the number of non-preferred drugs used, for which
the insurer often charges a higher copay. Higher brand copays lead to lower savings,
mostly through a reduction of net payment on brand drugs. They do not induce
enrollees to alter significantly their drug use behavior.
Table 11a
WLS results for panel (without intercept)
Effects of copay change on enrollee drug utilization and on insurer drug costs
Difference AMean insurer ATotal insurer ANumber of ATotal days of
estimator payment payment prescriptions supply
Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr.
AGeneric copay
-0.13 0.10 -0.95 0.92 -0.16 0.02 -0.51 0.35
ABrand copay
0.122 0.064 0.30 0.51 0.04 0.01 0.72 0.21
73
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Table lib
WLS results for panel (with intercept)
Effects of copay change on enrollee drug utilization and on insurer drug costs
Difference AMean insurer ATotal insurer ANumber of ATotal days of
estimator payment payment prescriptions supply
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
AGeneric copay
-0.56 0.12 -7.06 0.46 -0.23 0.01 -3.3 0.42
ABrand copay
-0.02 0.05 -2.50 0.52 0.01 0.01 -1.1 0.27
Table 11c
WLS results for panel (with intercept)
Effects of copay change on drug quantity, on number of drugs in the formulary, and
on insurer drug costs, by brand and generic
Difference A Quantity of A Total insurer A Total insurer A Number of
estimator drugs payment for payment for drugs in the
generic drugs brand drugs formulary
Coef Std. Err. Coef. Std. Err. Coef. Std. Err. Coef Std. Err.
AGeneric copay
-5.02 0.95 -1.36 0.26 -5.44 1.74 -0.20 0.02
ABrand copay
-4.53 0.19 0.16 0.14 -2.02 0.49 0.01 0.01
Comparison of cross section and panel results
Cross section results are presented in tables 12a and 12b. More detailed results are
found in appendix D.
In both tables, log of mean insurer payment, log of total insurer payment, log of
number of prescriptions, and log of total days of supply are the natural logarithms of
the corresponding dependent variables for periods one and two and were computed
for non-users by adding 0.5 to all variables. Explanatory variables included the ones
listed in each table, namely generic and brand copays, but also gender, enrollment
status (employee, dependent, spouse), age, and ADG and county dummies.
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
An inspection of these tables shows that most coefficients have the same signs as
those obtained in panel analysis. As in the fixed effect estimate, generic copay is
everywhere significant. Brand copay is significant only in period two, for mean
insurer payment, total insurer payment, and total number of prescriptions. (Only the
mean insurer payment and total insurer payment are significant in both procedures.)
The main difference between the two procedures is in the magnitude of the drug
copay coefficients: Cross section coefficients imply a much larger response in drug
utilization to copay levels than do fixed effect coefficients. This may indicate that
cross section estimates suffer from significant omitted variable bias.
Table 12a
WLS results for cross section, period one
Effects of copay levels on enrollee drug utilization and on insurer drug costs
Period one L og o f m ean L og o f total L og o f number o f L og o f total days
insurer paym ent insurer payment prescriptions o f supply
Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr.
Generic copay -0.066 0.003 -0.118 0.003 -0.101 0.002 -0.16 0.003
Brand copay 0,007 0.03 0.037 0.03 0.025 0.02 0.009 0.04
Table 12b
WLS results for cross section, period two
Effects of copay levels on enrollee drug utilization and on insurer drug costs
Period two L og o f m ean L og o f total L og o f number o f L og o f total days
insurer paym ent insurer paym ent prescriptions o f supply
Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr.
Generic copay -0.055 0.003 -0.105 0.003 -0.087 0.002 -0.136 0.003
Brand copay -0.022 0.01 -0.032 0.01 -0.017 0.006 -0.029 0.02
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
SECTION 5: CONCLUSION
Results and comparison with prior findings
The hypothesized relationships between higher copay and drug utilization are largely
confirmed by the results of my analysis. Enrollees are sensitive to out-of-pocket drug
costs, reducing their drug use when faced with higher copays. And the response to
higher generic copays is more dramatic than the response to higher brand copays.
When charged higher copays on generics, patients reduce substantially their purchase
of generic drugs in favor of other drugs (either brand or non-formulary) and receive
prescriptions that cover fewer days of supply. Higher copays on brand drugs seem to
affect only the quantity of medication received and the number of days of supply.
Total number of prescriptions is not affected by higher copays on brand drugs. A
higher copay on brand drugs primarily causes enrollees to pay more out of pocket for
each prescription, without changing their drug mix much. This may be because 1)
there are fewer alternatives to brand drugs, 2) brand drugs are more effective, or are
perceived as more effective, than generic drugs, or 3) brand drugs are more heavily
promoted, increasing the chance of automatic renewal.
Insurance companies can effectively save money in their pharmacy benefit budgets
by raising copays on either brand drugs or generic drugs. The size of the savings,
relative to the increase in copay, depends on which category of drugs is targeted.
Raising generic copay leads to savings almost three times as large (at least in the
76
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
short period considered here) as those obtained through higher brand copays for the
same absolute increase.
The demonstrated response to change in copay raises some important issues with
regard to quality of care. Higher brand copays seem to have largely treatment-neutral
effects. Higher generic copays, on the other hand, induce patients both to switch
away from generic drugs and reduce their overall use. Further research is needed to
determine whether these effects are harmful to patients. If they are harmful and
undermine patients’ health in the long term, the savings realized by insurers could be
short-lived.
A comparison with previously published results shows that my estimates are very
close to those presented in studies targeting a similar population. I estimate the
generic price elasticity (PE) of number of drugs demanded to be -0.20, the generic
price elasticity of drug expenditures to be -0.16, and the brand price elasticity of drug
expenditure to be -0.11. Leibowitz et al (1985) found a PE of number of drugs
demanded of -0.25 in their analysis of the HIE results. Harris et al. (1990) found a
PE for number of drugs demanded of -0.11 and a PE for drug expenditures of -0.05.
Joyce et al (2002) estimated the PE for drug expenditures to be -0.39. Finally, Ridley
(2003) found a PE for number of prescriptions of -0.3. This comparison lends
support to the method used for my analysis.
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Data limitations
Several limitations should be mentioned here. I used an administrative database
containing mostly information on claims submitted to an insurance company for
reimbursement. Much information is missing from this kind of database both at the
individual level and at the firm level. For instance, I do not know enrollees’
educational achievement or income. I do not know the firms’ sector of activity either.
These problems are common in claim databases and represent one of their major
shortcomings, when compared to surveys such as the MCBS (Medicare Current
Beneficiary Survey) or the MEPS (Medical Expenditure Panel Survey). There are,
however, a few advantages in claim databases. Most importantly, I have full
information on the drugs and medical services used. This allows me to obtain a
clearer picture of individual response and enables me to track precisely the effects of
copay change on a battery of utilization measures.
Some other limitations are specific to my data set. First, I have to rely on just two
years of data for each individual, so I can’t follow enrollees over longer periods of
time. In addition, the two years available for most enrollees saw relatively little
variation in drug copay level; few people had a copay change over the 1998-2001
period. Finally, demographic information is updated regularly, but old records are
not kept. Hence, I only know whether an enrollee was married in 2001, but not
whether their status had changed over the preceding two years.
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These data limitations make a more detailed analysis of utilization change - for
instance, of drug use in different clinical categories - impossible to carry out. The
small number of enrollees with copay change reduces the chance that coefficients
can be found to be significant.
Further research
Further research is needed to better understand changes in individual drug use
patterns in response to variation in out-of-pocket costs and how these changes affect
a patient’s health. My research addresses the effects of higher copay on the use of all
drugs. It does not provide much insight into the effects of relative copay change
within a therapeutic class or within a group of medications.
Interesting questions also remain to be answered on how adherence relates to cost-
sharing levels and on how persistence in utilization of a particular brand relates to
cost-sharing levels. With a better knowledge of individual response, it might be
possible to use copay not only as a blunt instrument for cost-reduction but also as a
tool for improving treatment and health.
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
REFERENCES
Baumgarten A. California Managed Care Review 2002. California HealthCare
Foundation. 2002. Available at
http://www.chcf.org/doeuments/hospitals/CalManagedCareReview2002.pdf
Accessed June 7, 2003
Beeson Royalty A and Solomon N. Health Plan Choice: Price Elasticity in a
Managed Competition Setting. Journal of Human Resources. 1999; 34(1): 1-41.
Bernstein D. Fringe Benefit in small business: evidence from the federal reserve
board small business survey. Applied Economics. 2002; 34:2063-2067.
Blumberg, LJ, Nichols LM and Banthin J. Worker Decisions to Purchase Health
Insurance Int J Health Care Finance Econ. Sep-Dee 2001;l(3-4):305-25.
Blumberg LJ and Nichols LM. Why are so Many Americans Uninsured? A
Conceptual Framework, Summary of Evidence, and Delineation of the Gaps in Our
Knowledge. ERIU Working Paper 3, University of Michigan. July 2002. Available at
www.umich.edu/~eriu/pdf/wp3 .pdf. Accessed March 4, 2003.
Booske BC, Sainfort F, and Schoofs Hundt A. Eliciting Consumer Preferences for
Health Plans. Health Services Research. 1999; 34(4): 839-854.
Bowen Garrett, Len Nichols, and Emily Greenman, "Workers Without Health
Insurance: Who Are They and How Can Policy Reach Them," Washington, DC: The
Urban Institute and the W.K. Kellogg Foundation. August 2001. Available at
http://www.wkkf.org/pubs/Health/CommunityVoices/Pub712.pdf. Accessed March
4, 2003.
Braun W and Kious AG. Capitation: Fact or Fiction. The Health System Groups.
1996. Available at http://www.hlthsvs.com/pub/capfact.pdf. Accessed July 18,2003.
Brunetti MJ, Nayeri K, Dobkin CE, Brady HE. Health status, health insurance, and
worker mobility: a study of job lock in California. 2000. Presented at the California
Work and Health Survey Research Conference on December 8, 2000. Available at
http://ucdata.berkelev.edu/new web/proi ioblock.html. Accessed September 3,
2003.
Buchmuller TC and Valetta RG. The effect of employer-provided health insurance
on worker mobility. Industrial and Labor Relations Review. 1996; 49(3): 439-455.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bymark L and Waite K. Prescription Drug Use And Expenditures In California: Key
Trends And Drivers. Prepared by William M. Mercer, Inc. for California Health Care
Foundation, Oakland, California. May 2001. Available at
http://www.pharmacv.ca.gov/pdfs/pbm chef rx.pdf. Accessed March 4,2003.
California Healthcare Foundation / Mercer. Employer-sponsored Health Insurance
among Small Businesses: The 2000 California Healthcare Foundation / Mercer
Survey. Prepared by William M. Mercer Inc. for the California HealthCare
Foundation, Oakland, California. March 2002. Available at
http://www.cahu.org/broker images/cahuorg/forms/0 whvdontsmallbusinessessum
mary.pdf. Accessed May 23,2003.
Chemew ME and Hirth RA. Modeling the cause and consequences of lack of health
insurance coverage: gaps in the literature. ERIU Working Paper 3, University of
Michigan. Jan 2002. Available at http://www.umich.edu/~eriu/pdf/wp 1 .pdf.
Accessed September 3,2003.
Coulson NE and Stuart BC. Insurance choice and the demand for prescription drugs.
Southern Economics J. 1995; 61(4): 1146-1157.
Cunningham PJ, Schaefer E, Hogan C. Who declines employer-sponsored health
insurance and is uninsured? Issue Brief No. 22, Center for Studying Health System
Change, Washington D.C. October 1999. Available at
http://www.hschange.org/CONTENT/46/46.pdf. Accessed January 29,2003.
Cunningham PJ and Kohn LT. Health Plan Switching: Choice or Circumstance?
Health Affairs. May/June 2000; 19(3): 158-164
Cutler DM. Employee costs and the decline in health insurance coverage. NBER
Working Paper 9036. July 2002.
Davis K and Cooper BS. American Health Care: Why So Costly? 2003. The
Commonwealth Fund, New York, NY 10021. Available at
http://www.cmwf.org/programs/quality/davis senatecommitteetestimonv 654.pdf.
Accessed July 24,2003.
Dranove D, Spier KE, Baker L. Competition among employers offering health
insurance. Journal of Health Economics. 2000; 19:121-140.
Fairman KA, Motheral BR, and Henderson RR. Retrospective Long-Term Follow
Up Study Of The Effect Of A Three-Tier Prescription Drug Copayment System On
Pharmaceutical And Other Medical Utilization And Costs. Clin Therapeutics. 2003;
25(12): 3147-3161.
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Farber HS and Levy H. Recent trends in employer-sponsored health insurance
coverage: are bad jobs getting worse? Journal of Health Economics. 2000; 19: 93-
119.
Feldman R, Finch M, Dowd B, Cassou S. The demand for employment-based health
insurance plans. Journal of Human Resources. 1989; 24(1): 117-142.
Fronstin P and Helman R. Small Employers and Health Benefits: Findings From the
2002 Small Employer Health Benefits Survey. EBRI. Issue Brief Number 253,
Employee Benefit Research Institute, Washington. January 2003.
http://www.ebri.org/sehbs/1000ib.pdf. Accessed January 29,2003.
Gabel J, Claxton G, Holve E, Pickreign J, Whitmore H, Dhont K, Hawkins S, and
Rowland D. health Benefits in 2003: Premiums Reach Thirteen-Year High As
Employers Adopt New Forms Of Cost Sharing. Health Affairs. 2003; 22(5): 117-126.
Gabel J.. Job-Based Health Insurance, 1977-1998: The Accidental System Under
Scrutiny. Health Affairs. 1999; 18(6):62-74.
General Accounting Office. Private Health Insurance: small employers continue to
face challenges in providing coverage. United States General Accounting Office,
Washington, D.C. October 2001. Available at
http://www.openminds.com/indres/gaoprivateinsurancel2-20.pdf. Accessed Mrach
10,2003.
Gilleskie and Lutz 1999. The impact of employer-provided health insurance on
dynamic employment transitions. NBER Working Paper 7037.1999.
Guiffrida A and Torgeson DJ. Should we pay the patient? Review of financial
incentives to enhance patient compliance. BMJ. 1997; 315:703-707.
Glover SH, Stoskopf C, Brown TE, Wheeler F, Kim Y, Xirasagar S. Small Business
and Access to Health Insurers, Particularly HMOs. Prepared for the Office of
Advocacy of the U.S. Small Business Administration by Consult Inc., Orangeburg,
South Carolina. August 2000. Available at
http://www.sba.gov/advo/research/rs202tot.pdf. Accessed March 15,2003.
Grootendoorst PV, O’Brien BJ, and Anderson MA. On becoming 65 in Ontario. Med
Care. 1997; 35(4): 386-398.
Grossman. On the concept of health capital and the demand for health. Journal of
Political Economy. 1972; 80:223-255.
82
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Gruber J and Lettau M, “How Elastic is the Firm’s Demand for Health Insurance?”
NBER Working Paper Number 8021, Cambridge MA, National Bureau of Economic
Research. November 2000.
Gruber J and Madrian BC. Health insurance, labor supply, and job mobility: a
critical review of the literature. NBER Working Paper 8817. 2002.
Gruber J and McKnight R. Why did employee health insurance contribution rise?
NBER Working Paper 8878. April 2002.
Harris BL, Stergachis A, Ried LD. The effect of drug co-payments on utilization
and cost of pharmaceuticals in a health maintenance organization. Med Care. 1990;
28:907-917.
Huskamp HA, Deverka PA, Epstein AM, Epstein RS, McGuigan KA, and Frank RG.
The Effect of Incentive-Based Formularies on Prescription-Drug Utilization and
Spending.
N Engl J Med. 2003; 349: 2224-2232.
Hillman AL, Pauly MV, Escarce JJ, Ripley K, Gaynor M, Clouse J, and Ross R.
Financial incentives and drug spending in managed care. Health Affairs. 1999;
18(2): 189-200
Johnson RE, Goodman MJ, Hombrook MC, Eldredge MB. The effect of increased
prescription drug cost-sharing on medical care utilization and expenses of elderly
health maintenance organization members. 1997. Med Care; 35:1119-1131.
Joyce GF, Escarce JJ, Solomon MD, and Goldman DP. Employer drug benefit plan
and spending on prescription drugs. JAMA. 2002 ; 288:1733-1739.
Kaiser Family Foundation and Health Research And Educational Trust (HRET).
Employer Health Benefits, 2001 Annual Survey. September 2001. Available at
http://www.kff.org/insurance/ehbs-archives.cfm. Access April 5,2003.
Kaiser Family Foundation. Trends and indicator in the changing health care
marketplace. Chartbook May 2002. Available at
http://www.kff.org/insurance/loader.cfm?url=:/commonspot/securitv/getfile.cfm&Pag
eID=14967. Accessed April 5,2003.
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Lee J. Are health insurance premiums higher for small firms? Research Synthesis
Report No.2. The Robert Wood Johnson Foundation, Princeton, New Jersey.
September 2002. Available at
http ://www.rwi f.org/publications/svnthesis/reports and briefs/pdf/
no2 researehreport.pdf, Accessed January 12,2003.
Lebowitz A, Willard G, and Newhouse JP. The demand for prescription as a function
of cost-sharing. Social Science and Medicine. 1985; 21(10):1063-1069.
Levit K, Smith C, Cowan C, Lazenby H, Sensing A, and Catlin A. Trend in US
Health care Spending, 2001. Health Affairs. 2003; 22(1): 154-164.
Levy H and Feldman R. Does the Incidence of Group Health Insurance Fall on
Individual Workers? International Journal of Health Care Finance and Economics.
2001; 1(3): 227-247.
Lillard LA, Rogowski J, and Kington R. Insurance Coverage for Prescription Drugs:
The Effect on Use And Expenditures in the Medicaid Population. Med Care. 1999;
37(7):926-936.
Madden CW, Mackay BP, and Skillman SM. Measuring health status for risk
adjusting capitation payment. Working paper. Center for Health Care Strategies, Inc.,
New Jersey. July 2001. Available at
http://www.chcs.org/usr doc/riskadiustment.pdf. Accessed August 2,2003.
Marquis MS and Long SH. Employer Health Insurance and Local Market
Conditions. International Journal of Health Care Financing and Economics. 2001; 1:
273-292.
Maxwell J and Termin P. Managed Competition versus Industrial Purchasing of
Health Care among the Fortune 500. Journal of Health Politics, Policy, and Law.
2002; 27(1): 5-30. (a)
Maxwell J, Termin P, and Saminaz S. The Benefits Divide: Health Care Purchasing
In Retail Versus Other Sectors. Health Affairs. 2002; 21(5): 224-233.
McLaughlin GG. Health Care Consumers: Choice and Constraints. Medical Care
Research and Review. 1999; 56 Supplement 1:24-59.
Monheit CA and Vistnes JP. Health insurance availability at the workplace: how
important are worker preferences? Journal of Human Resources. 1999; 34(4): 771-
785.
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Morrisey MA, Jensen GA, and Morlock RJ. Small employers and the health
insurance market. Health Affairs Winter 1994; 13(4):149-161.
Motheral BR and Henderson R. The Effect of Copay Increase on Pharmaceutical
Utilization, Expenditures, and Treatment Continuation. Am J Manag Care. 1999;
5(11):1383-1394.
National Center for Health Statistics. Employer-sponsored health insurance: State
and national estimates. Hyattsville, Maryland. 1997. Available at
http://www.cdc.gov/nchs/data/misc/employer.pdf. Accessed September 17, 2003.
Nelson AA, Reeder CE, and Dickson WM. The Effect of a Medicaid Drug
Copayment Program on The Utilization and Cost of Prescription Services. Med Care.
1984; 22(8):724-735.
Peele PB, Lave JR, Black JH, Evans JH. Employer sponsored health insurance:
Employers' choice and employee preferences. Milbank Quarterly. 2000; 78:5-21.
Ridley D. Payment, promotion, and the purple pill. 2003. Working paper. August
2003. Available at www.missouri.edu/~econwww/seminar/papers/Ridlev-Rx-Elas-
2003.pdf. Accessed February 2,2004
Rector TS, Finch MD, Danzon PM, Pauly MV, and Manda BS. Effect of Tiered
Prescription Copayments on the Use of Preferred Brand Medications. Med Care.
2003.41(3): 398-406.
Smith DG. The effects of copayments and generic substitution on the use and costs
of prescription drugs. Inquiry. 1993; 30:189-198.
Strombom BA, Buchmuller TC, and Feldstein PJ. Switching costs, price sensitivity
and health plan choice. Journal of Health Economics. 2002; 21: 89-116.
Stroupe KT, Kinney ED, and Knieser TJJ. Does Chronic Illness Affect the Adequacy
of Health Insurance Coverage. Journal of Health Politics, Policy, and Law. 2000;
25(2):309-41.
Stuart B and Grana J. Ability To Pay And Decision To Medicate. Med Care. 1998;
36(2):202-211.
Tamblyn R, Laprise R, Hanley JA, Abrahamowicz M, Scott S, Mayo N, Hurley J,
Grad R, Latimer E, Perrault R, McLeod P, Huang A, Larochelle P, Mallet L.
Adverse events associated with prescription drug cost-sharing among poor and
elderly persons. JAMA. 2001; 285:421-429.
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Thomas CL, Wallack SS, Lee S, and Ritter GA. Impact Of Health Plan Design And
Management On Retirees’ Prescription Drug Use And Spending, 2001. Health
Affairs. Dec 4 2002; 408-419.
Trauner J. Tracking California’s Individual and Small Group Health Insurance
Markets. California Health Care Foundation Oakland, California. 2003. Available at
www.chcf.org/topics/view.cfm?itemID: : =20246. Accessed December 3,2003.
Schauffler HH and Brown ER. The State of Health Insurance in California, 1999.
Berkley, CA: Regents of the University of California, January 2000.
Available at http ://chpps.berkelev. edu/publications/000-HIPP99Complete. pdf.
Accessed January 22,2003.
Van de Ven, WPMM. and Ellis, RP. Risk Adjustment in Competitive Health Plan
Markets. In Culyer, A.J. and Newhouse, J.P. (eds.), Handbook of Health Economics.
Amsterdam: Elsevier. 2000; volume 1A: 755-845.
Weiner JP, Starfield BH, Steinwachs DM, and Mumford LM. Development and
application of a population oriented measure of ambulatory care case mix. Med Care.
1991; 29(5):452-472
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX A
Table la
Comparison of ADG frequencies between the final sample and subsample 1, those
who moved among employers: Percent of enrollees receiving diagnosis in each ADG
ADG Final sample
One firm — one
copay structure
Excluded
subsample 1
More than one
firm
Likelihood
Ratio Chi-
Square
(probability)
Time limited: minor 33.8 35.5 0.0336
Time limited: minor-primary infection 49.6 50.2 0.4569
Time limited: major 7.1 7.3 0.5855
Time limited: major-primary infection 9.0 9.7 0.1499
Allergies
14.2 15.4 0.0305
Asthma 6.1 6.4 0.4742
Likely to recur: discrete 22.7 22.9 0.7953
Likely to recur: discrete-infection 28.6 29.8 0.1178
Likely to recur: progressive 1.9 1.7 0.4975
Chronic medical: stable 34.3 35.9 0.0403
Chronic medical: unstable 13.8 14.6 0.1269
Chronic specialty: stable orthopedic 4.0 4.6 0.1165
Chronic specialty: stable ear, nose, throat 1.2 1.4 0.3244
Chronic specialty: stable eye
13.3 14.6 0.0180
Chronic specialty: unstable orthopedic
2.3 2.8 0.0511
Chronic specialty: unstable ear, nose, throat 0.2 0.3 0.0855
Chronic specialty: unstable eye 4.1 4.5 0.2202
Dermatologic
20.4 21.5 0.1112
Injuries or adverse effect: minor 25.1 26.0 0.1820
Injuries or adverse effect: major 17.5 18.8 0.0441
Psychosocial: time limited, not severe 3.6 3.3 0.3342
Psychosocial: persistent or recurrent, stable 9.3 10.0 0.1461
Psychosocial: persistent or recurrent,
unstable
1.8 2.0 0.2733
Signs or symptoms: minor
34.0 36.7 0.0010
Signs or symptoms: uncertain
44.6 46.7 0.0128
Signs or symptoms: major
30.0 33.0 0.0001
Discretionary 17.4 17.8 0.5738
See and reassure
6.0 6.1 0.8754
Prevention or administrative
68.2 69.7 0.0593
Malignancy
3.4 4.0 0.0620
Pregnancy
5.1 5.7 0.1134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table lb
Logistic: Probability that an enrollee will remain in the same firm
for three years or more
Parameter Estimate Std. Err.
Sex (female=l) -0.1189 0.0395
Marital status (married=l) -0.1139 0.0431
Spouse(yes=l) 0.3412 0.2375
Dependent (yes=l) 0.2691 0.0533
Age 6-10 -0.0567 0.1316
Age 11-15 0.2874 0.1356
Age 16-18 0.2928 0.1477
Age 19-20 0.4904 0.1681
Age 21-25 0.2836 0.1727
Age 26-30 0.0148 0.2748
Age 31-35 0.1803 0.2716
Age 36-40 0.2632 0.2708
Age 41-45 0.4063 0.2714
Age 46-50 0.4345 0.2724
Age 51-55 0.4924 0.2742
Age 56-60 0.6062 0.2772
Age 61-64 0.5811 0.2836
Time limited: minor -0.01% 0.0383
Time limited: minor-primary infection -0.00339 0.0374
Time limited: major 0.0188 0.07
Time limited: major-primary infection -0.044 0.06
Allergies -0.1008 0.05
Asthma -0.0401 0.0728
Likely to recur: discrete 0.0888 0.045
Likely to recur: discrete-infection -0.0347 0.0413
Likely to recur: progressive 0.2153 0.139
Chronic medical: stable -0.00912 0.0428
Chronic medical: unstable -0.0302 0.0544
Chronic specialty: stable orthopedic -0.0484 0.0871
Chronic specialty: stable ear, nose, throat -0.0451 0.1551
Chronic specialty: stable eye -0.0283 0.051
Chronic specialty: unstable orthopedic -0.202 0.1108
Chronic specialty: unstable ear, nose, throat -0.5342 0.3188
Chronic specialty: unstable eye -0.0911 0.0866
Dermatologic -0.023 0.0441
Injuries or adverse effect: minor -0.00843 0.0423
Injuries or adverse effect: major -0.0611 0.0473
Psychosocial: time limited, not severe 0.155 0.0988
Psychosocial: persistent or recurrent, stable -0.0239 0.0615
Psychosocial: persistent or recurrent, unstable -0.1345 0.1282
Signs or symptoms: minor -0.0584 0.0398
Signs or symptoms: uncertain -0.00519 0.0398
Signs or symptoms: major -0.0946 0.0419
Discretionary 0.0447 0.0479
See and reassure 0.0726 0.0744
Prevention or administrative -0.0225 0.0421
Malignancy -0.114 0.0933
Pregnancy -0.00543 0.0822
Variables
description:
Health status
variables refer
to the ADG
dummies as
explained in
appendix B.
Age is
introduced as
a dummy
variable for an
age range.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2a
Comparison of ADG frequencies between the final sample and subsample 2, those
who worked in the same firm and could choose among different copay structures:
Percent of enrollees receiving diagnosis in each ADG
ADG Final sample
One firm - one
copay structure
Excluded
subsample 1
More than one
firm
Likelihood
Ratio Chi-
Square
(probability)
Time limited: minor 25,19 24.94 0.6260
Time limited: minor-primary infection 39.66 39.38 0.6226
Time limited: major 4.94 4.67 0.2715
Time limited: major-primary infection 6.24 5.97 0.3429
Allergies 10.10 10.56 0.1944
Asthma 4.40 4.44 0.8649
Likely to recur: discrete 16.72 16.49 0.5849
Likely to recur: discrete-infection 21.73 21.91 0.6997
Likely to recur: progressive 1.29 1.03 0.0361
Chronic medical: stable 27.67 26.69 0.0541
Chronic medical: unstable 10.36 9.78 0.0913
Chronic specialty: stable orthopedic 2.83 2.68 0.4352
Chronic specialty: stable ear, nose, throat 0.79 0.67 0.2194
Chronic specialty: stable eye 9.70 9.99 0.4041
Chronic specialty: unstable orthopedic 1.66 1.53 0.3570
Chronic specialty: unstable ear, nose, throat 0.11 0.12 0.7358
Chronic specialty: unstable eye 3.07 2.71 0.0675
Dermatologic 15.04 14.23 0.0473
Injuries or adverse effect: minor 18.23 17.73 0.2597
Injuries or adverse effect: major 12.64 11.99 0.0837
Psychosocial: time limited, not severe 2.49 2.16 0.0633
Psychosocial: persistent or recurrent, stable 6.96 6.21 0.0094
Psychosocial: persistent or recurrent
unstable
1.26 1.08 0.1525
Signs or symptoms: minor 25.33 24.14 0.0174
Signs or symptoms: uncertain 34.81 34.03 0.1556
Signs or symptoms: major 22.61 21.32 0.0072
Discretionary 12.42 12.52 0.7953
See and reassure 4.22 3.72 0.0264
Prevention or administrative 57.88 56.26 0.0046
Malignancy 2.54 2.25 0.0987
Pregnancy 3.90 3.84 0.7966
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2b
Logistic: Probability to be in a firm offering only one copay structure instead of a
choice of copay structure
Parameter Estimate Std. Err.
Sex 0.0433 0.0267
Marital status -0.1232 0.0298
Spouse -0.0595 0.0356
Dependent 0.0158 0.1472
Age 6-10 -0.1535 0.0842
Age 11-15 -0.0967 0.0849
Age 16-18 -0.1442 0.0928
Age 19-20 -0.0263 0.1026
Age 21-25 0.00662 0.1108
Age 26-30 -0.0908 0.1725
Age 31-35 0.0704 0.1698
Age 36-40 -0.0452 0.1688
Age 41-45 0.0341 0.1689
Age 46-50 0.081 0.1696
Age 51-55 0.1135 0.1709
Age 56-60 0.177 0.1728
Age 61-64 0.2391 0.1785
Time limited: minor -0.0263 0.0286
Time limited: minor-primary infection 0.0227 0.0262
Time limited: major 0.0145 0.0578
Time limited: major-primary infection -0.00367 0.0502
Allergies -0.0561 0.0399
Asthma 0.0141 0.0586
Likely to recur: discrete -0.0243 0.0342
Likely to recur: discrete-infection -0.0046 0.0309
Likely to recur: progressive 0.1948 0.1162
Chronic medical: stable 0.0217 0.031
Chronic medical: unstable 0.00752 0.043
Chronic specialty: stable orthopedic -0.0212 0.0744
Chronic specialty: stable ear, nose, throat 0.08 0.1433
Chronic specialty: stable eye -0.0705 0.0406
Chronic specialty: unstable orthopedic 0.0577 0.0988
Chronic specialty: unstable ear, nose, throat -0.2416 0.3443
Chronic specialty: unstable eye 0.1031 0.0738
Dermatologic 0.0387 0.0348
Injuries or adverse effect: minor 0.0228 0.0326
Injuries or adverse effect: major 0.0335 0.038
Psychosocial: time limited, not severe 0.1051 0.0812
Psychosocial: persistent or recurrent, stable 0.1327 0.0506
Psychosocial: persistent or recurrent, unstable 0.0531 0.1146
Signs or symptoms: minor 0.024 0.0298
Signs or symptoms: uncertain -0.0177 0.0283
Signs or symptoms: major 0.0422 0.0319
Discretionary -0.0592 0.0375
See and reassure 0.0919 0.0631
Prevention or administrative 0.0534 0.0267
Malignancy 0.0401 0.0809
Pregnancy 0.0779 0.0662
Variables
description:
Health status
variables
refer to the
ADG
dummies as
explained in
appendix B.
Age is
introduced as
a dummy
variable for
an age range.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2c
Logistic: probability to be in a firm offering one copay level only, measured using
firm level averages
Parameter Estimate Std. Err.
Intercept 1.4717 0.398
Time limited: minor -0.3224 0.1616
Time limited: minor-primary infection -0.00117 0.1492
Time limited: major 0.271 0.3099
Time limited: major-primary infection 0.0209 0.2753
Allergies -0.1806 0.2264
Asthma 0.5075 0.3871
Likely to recur: discrete -0.1754 0.1826
Likely to recur: discrete-infection -0.2713 0.1797
Likely to recur: progressive -0.0119 0.5175
Chronic medical: stable 0.0293 0.164
Chronic medical: unstable 0.091 0.2177
Chronic specialty: stable orthopedic 0.0717 0.3977
Chronic specialty: stable ear, nose,
throat -0.1254 0.6918
Chronic specialty: stable eye -0.0916 0.2201
Chronic specialty: unstable orthopedic 0.4982 0.5525
Chronic specialty: unstable ear, nose,
throat -2.58 2.3064
Chronic specialty: unstable eye -0.0296 0.3485
Dermatologic -0.0883 0.1842
Injuries or adverse effect: minor 0.4279 0.199
Injuries or adverse effect: major -0.352 0.2095
Psychosocial: time limited, not severe -0.00345 0.4146
Psychosocial: persistent or recurrent,
stable 0.5259 0.2928
Psychosocial: persistent or recurrent,
unstable -0.1586 0.5802
Signs or symptoms: minor 0.1602 0.174
Signs or symptoms: uncertain -0.2483 0.1587
Signs or symptoms: major 0.2768 0.1764
Discretionary -0.3607 0.1892
See and reassure 0.7487 0.3589
Prevention or administrative -0.0854 0.1536
Malignancy 0.0691 0.3592
Pregnancy 0.6024 0.3757
Avg age o f enrollees 0.0113 0.00648
Avg proportion o f dependents 0.2057 0.0793
Avg proportion o f married enrollees -0.2481 0.1276
Avg proportion o f spouses -0.0746 0.1223
Avg proportion o f females -0.0427 0.1628
Number o f employees -0.044 0.00414
Avg time to first enrollment (in year) 0.0168 0.0291
!
Variables description:
All variables are
intra-firm averages
(except county
dummies), i.e.
averages o f all
employees’
characteristics.
Health status
variables refer to
the ADG dummies
as explained in
appendix B.
Avg age is the
average age among
individuals enrolled
in a firm’s plan.
Avg proportion o f
dependents,
spouses, females,
and married
enrollees is the
average proportion
o f these variables
in each firm.
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX B
This appendix introduces the method used to assess the health status of individual
enrollees and to quantify it. This method is called the Ambulatory Care Group
(ACG®) risk assessment system. The ACG system is a cornerstone of this analysis
since patients’ severity of illness is a critical element of health care demand. The
decision to adopt the ACG for severity assessment is based on several criteria. First,
this system has been extensively studied, used, and validated. The most recent
version of the ACG system has been shown to be accurate in the context of risk
adjustment, either forecasting cost based on past diagnosis (prospective adjustment)
or adjusting payment based on current diagnosis (retrospective adjustment). It is
particularly accurate for low cost groups since the ACG gives more weight to
outpatient services (hence, lower cost) than other risk adjustment systems.
Furthermore, the population used to develop the ACG is similar to the population
used in the present study. Finally, the ACG software is relatively easy to obtain.
Refer to van de Ven and Ellis (2000), Weiner et al. (1991) and Madden et al. (2001)
for more detail.
The ACG system is a method of categorizing patients’ illnesses. Based on the pattern
of these morbidities, the ACG approach assigns each individual to a single ACG
category. Thus, an ACG category captures the specific clustering of morbidities
experienced by a person over a given period of time, such as a year.
The ACG Case-Mix System assigns all International Classification of Diseases
(ICD-9) codes to one of 32 diagnosis clusters known as Aggregated Diagnostic
Groups (ADGs). Individual diseases or conditions are placed into a single ADG
cluster based on five clinical dimensions:
■ Duration of the condition (acute, recurrent, or chronic): How long will
healthcare resources be required for the management of this condition?
■ Severity of the condition (e.g., minor and stable versus major and unstable):
How intensely must healthcare resources be applied to manage the condition?
■ Diagnostic certainty (symptoms versus documented disease): Will a
diagnostic evaluation be needed or will services for treatment be the primary
focus?
■ Etiology of the condition (infectious, injury, or other): What types of
healthcare services will likely be used?
■ Specialty care involvement (medical, surgical, obstetric, hematology, etc.):
To what degree will specialty care services be required?
All diseases, even those yet to be discovered, can be classified along these
dimensions and categorized into one of these 32 ADG clusters. Because most
management applications for population-based case-mix adjustment systems require
that patients be grouped into single, mutually exclusive categories, the ACG System
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
uses a branching algorithm to place people into one of 934 discrete categories based
on their assigned ADGs, their age, and their gender. The result is that individuals
within a given ACG have experienced a similar pattern of morbidity and resource
consumption over the course of a given year. ACGs can be assigned to individuals
using readily available diagnostic information derived from outpatient or
“ambulatory” physician visit claims records, encounter records, inpatient hospital
claims, and computerized discharge abstracts. A patient/enrollee is assigned to a
single ACG based on the diagnoses assigned by all clinicians seeing them during all
contacts, regardless of setting.
Refer to appendix E for examples of ICD-9 codes in each ADG category. Further
information can be found a www.hsr.ihsph.edu.
Tables 3a and 3b show the proportions of enrollees receiving multiple ADGs and
multiple unique diagnoses. Few people received no diagnosis in period one or in
period two. A majority of enrollees received between one and four different
diagnoses or were classified into one to four ADG categories.
Table 3a
Proportion of enrollees with multiple ADGs
Percent enrollees with Period 1 Period 2
0 ADG 0.9 0.1
1 ADG 25.5 22.4
2 ADGs 22.9 21.8
3 ADGs 18.0 17.0
4 ADGs 12.6 13.1
5 ADGs 7.8 9.3
6 ADGs 5.0 6.3
7 ADGs 3.1 4.0
8 ADGs 1.9 2.5
9 ADGs 1.1 1.6
10 + ADGs 1.1 2.0
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3b
Proportion of enrollees with multiple diagnoses
Percent enrollees w ith Period 1 Period 2
0 unique diagnosis code 0.9 0.1
1 unique diagnosis code 21.4 18.2
2 unique diagnosis codes 19.2 17.5
3 unique diagnosis codes 14.8 14.0
4 unique diagnosis codes 11.6 11.3
5 unique diagnosis codes 8.7 9.0
6 unique diagnosis codes 6.2 7.3
7 unique diagnosis codes 4.4 5.3
8 unique diagnosis codes 3.5 3.9
9 unique diagnosis codes 2.5 3.2
10 + unique diagnosis codes 6.9 10.4
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
APPENDIX C
This appendix presents the breakdown of drug utilization by insurance plan in
periods one and two. Each copay combination has its own table, from 4a to 4g.
Tables 4a to 4e show drug use in plans having the same copay structure in periods
one and two. (These plans are not necessarily identical.) Plans in tables 4f and 4g
have a copay combination that is unique to period two.
Table 4a
Drug utilization with the following copay:
$5 brand, $5 generic, no non-formulary copay
Period one Period two
Total net paid 113.4 85.2
A verage net per Rx 18.1 15.2
Num ber o f R x 4.4 3.3
Num ber o f brand drugs 1.42 1.19
Num ber o f generic drugs 2.01 1.45
Table 4b
Drug utilization with the following copay:
$10 brand, $5 generic, no non-formulary copay
Period one Period two
Total net paid 203.7 257.9
A verage net per Rx
19.2 24.5
Number o f R x
5.8 7.3
Num ber o f brand drugs 2.97 3.43
Num ber o f generic drugs
1.41 2.67
Table 4c
Drug utilization with the following copay:
$10 brand, $10 generic, no non-formulary copay.
Period one Period two
Total net paid
158.3 81.0
A verage net per Rx
19.5 13.1
Number o f Rx
4.1 3.4
Num ber o f brand drugs
2.44 2.14
Number o f generic drugs
0.69 0.66
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4d
Drug utilization with the following copay:
$10 brand, $5 generic, $20 non-formulary copay.
Period one Period two
Total net paid
248.8 296.2
Average net per R x
20.8 24.0
Number of R x
6.4 6.9
Number of brand drugs
3.09 3.63
Number of generic drugs
2.26 5.03
Table 4e
Drug utilization with the following copay:
$10 brand, $10 generic, $20 non-formulary copay.
Period one Period two
Total net paid
199.1 253.5
Average net per R x
20.9 24.4
Number of Rx
4.8 5.3
Number of brand drugs
2.81 3.30
Number of generic drugs
1.06 1.06
Table 4f
Drug utilization with the following copay:
$15 brand, $10 generic, no non-formulary copay, period two only.
Period two
Total net paid
663.9
Average net per R x
55.6
Number of Rx
6.9
Number of brand drugs
4.29
Number of generic drugs
0.86
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4g
Drug utilization with the following copay:
$20 brand, $10 generic, $25 non-formulary copay, period two only.
Period two
Total net paid
190.1
A verage net per Rx
20.6
Num ber o f Rx
4.8
Num ber o f brand drugs
2.91
Num ber o f generic drugs
0.89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX D
Table 8a
WLS results for cross section, all enrollees, period one
Period one Log o f mean
insurer payment
Log o f total
insurer payment
Log o f number o f
prescriptions
Log o f total days
o f supply
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. En.
Generic copay -0.066 0.003 -0.118 0.003 -0.101 0.002 -0.157 0.003
Brand copay 0.007 0.027 0.037 0.031 0.025 0.021 0.009 0.044
Female 0.229 0.015 0.503 0.017 0.396 0.010 0.668 0.018
Marital status 0.048 0.016 0.010 0.018 -0.046 0.011 -0.062 0.020
Dependent -0.874 0.047 -1.642 0.062 -1.013 0.030 -1.748 0.053
Spouse 0.028 0.017 0.096 0.021 0.090 0.012 0.101 0.021
Age 6-10 -0.223 0.043 -0.488 0.059 -0.336 0.030 -0.545 0.049
Age 11-15 -0.112 0.048 -0.287 0.058 -0.264 0.029 -0.349 0.048
Age 16-18 0.061 0.052 -0.092 0.068 -0.179 0.036 -0.170 0.059
Age 19-20 0.506 0.054 0.523 0.071 0.207 0.033 0.533 0.061
Age 21-25 0.234 0.054 0.128 0.073 -0.047 0.029 0.030 0.054
Age 26-30 -0.494 0.046 -1.073 0.067 -0.685 0.035 -1.089 0.064
Age 31-35 -0.388 0.042 -0.982 0.057 -0.660 0.029 -1.019 0.053
Age 36-40 -0.302 0.039 -0.844 0.054 -0.586 0.028 -0.928 0.052
Age 41-45 -0.254 0.039 -0.793 0.055 -0.573 0.028 -0.896 0.050
Age 46-50 -0.069 0.039 -0.495 0.054 -0.393 0.028 -0.644 0.051
Age 51-55 0.131 0.036 -0.024 0.053 -0.050 0.028 0.002 0.050
Age 56-60 0.306 0.038 0.346 0.053 0.238 0.029 0.461 0.052
Age 61-64 0.398 0.040 0.566 0.058 0.351 0.029 0.663 0.056
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 8b
WLS results for cross section, all enrollees, period two
Second period Log o f mean
insurer payment
Log o f total
insurer payment
Log o f number o f
prescriptions
Log o f total days
o f supply
Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr. Coef. Std. E rr.
Generic copay -0.055 0.003 -0.105 0.004 -0.087 0.002 -0.136 0.003
Brand copay -0.022 0.011 -0.032 0.013 -0.017 0.007 -0.029 0.015
Female 0.173 0.014 0.414 0.018 0.348 0.009 0.539 0.016
Marital status 0.080 0.015 0.047 0.020 -0.003 0.010 -0.025 0.018
Dependent -1.057 0.048 -2.034 0.056 -1.249 0.023 -2.107 0.054
Spouse -0.060 0.017 -0.039 0.021 0.006 0.010 0.016 0.019
Age 6-10 -0.217 0.050 -0.259 0.049 -0.164 0.021 -0.317 0.047
Age 11-15 -0.029 0.050 -0.009 0.050 -0.052 0.022 -0.043 0.051
Age 16-18 0.259 0.054 0.339 0.057 0.147 0.028 0.315 0.055
Age 19-20 0.687 0.059 1.011 0.057 0.451 0.024 0.936 0.064
Age 21-25 0.249 0.048 0.371 0.060 0.125 0.027 0.259 0.056
Age 26-30 -0.479 0.044 -1.039 0.064 -0.700 0.032 -1.113 0.058
Age 31-35 -0.483 0.043 -1.117 0.053 -0.743 0.026 -1.161 0.053
Age 36-40 -0.313 0.041 -0.853 0.052 -0.609 0.025 -0.990 0.049
Age 41-45 -0.286 0.040 -0.803 0.049 -0.543 0.025 -0.872 0.049
Age 46-50 -0.177 0.039 -0.556 0.050 -0.421 0.026 -0.680 0.049
Age 51-55 0.021 0.038 -0.063 0.052 -0.068 0.024 -0.075 0.050
Age 56-60 0.283 0.039 0.374 0.050 0.217 0.026 0.382 0.049
Age 61-64 0.376 0.041 0.588 0.054 0.339 0.026 0.607 0.052
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX E
Sample listing of the common ICD-9-CM diagnosis codes assigned to each ADG
cluster. This listing presents assignment for selected ICD-9-CM codes. Given that the
software assigns over 14,000 codes, this exemplary list is provided to give users a sense
of the logic behind the system. Copyright © The Johns Hopkins University, 1995, 1997,
1998, 1999, 2000, 2001. All Rights Reserved.
ADG 1 Time Limited: Minor
351.0 Bell’s Palsy
373.2 Chalazion
380.4 Impacted Cerumen
386.30 Labyrinthitis
558.9 Noninfectious Gastroenteritis and Colitis
627.2 Menopausal or Female Climacteric States
681.11 Onychia and Paronychia of Toe
690 Erythematosquamous Dermatosis
691.0 Diaper or Napkin Rash
692.8 Other Dermatitis
692.9 Contact Dermatitis and Other Eczema, Unspecified Cause
703.0 Ingrowing Nail
709.9 Unspecified Disorder of Skin and Subcutaneous Tissue
726.10 Disorders of Bursae and Tendons in Shoulder Region
726.5 Enthesopathy of Hip Region
726.71 Achilles Bursitis or Tendinitis
726.73 Calcaneal Spur
726.79 Other Enthesopathy of Ankle and Tarsus
726.9 Unspecified Enthesopathy
726.90 Enthesopathy of Unspecified Site
727.3 Other Bursitis Disorders
728.71 Plantar Fascial Fibromatosis
728.8 Other Disorders of Muscle, Ligament, and Fascia
782.1 Rash and Other Nonspecific Skin Eruption
ADG 2 Time Limited: Minor-Primary Infections
008.8 Intestinal Infection due to Other Organism, Not Elsewhere Classified
009.3 Diarrhea of Presumed Infectious Origin
041.1 Staphylococcus Infection in Conditions Classified Elsewhere
041.9 Bacterial Infection, Unspecified
053.9 Herpes Zoster without Mention of Complication
079.9 Unspecified Viral Infection
098.0 Gonococcal Infection (Acute) of Lower Genitourinary Tract
112.0 Candidiasis of Mouth (Thrush)
131.0 Urogenital Trichomoniasis
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
132.0 Pediculus Capitis (Head Louse)
136.9 Unspecified Infectious and Parasitic Diseases
373.11 Hordeolum Externum
380.1 Infective Otitis Externa
460 Acute Nasopharyngitis (Common Cold)
461.9 Acute Sinusitis, Unspecified
462 Acute Pharyngitis
463 Acute Tonsillitis
464.0 Acute Laryngitis
464.4 Croup
465.0 Acute Laryngopharyngitis
465.9 Acute Upper Respiratory Infections of Unspecified Site
466.0 Acute Bronchitis
481 Pneumococcal Pneumonia
486 Pneumonia, Organism Unspecified
487.1 Influenza with other Respiratory Manifestations
490 Bronchitis
511.0 Pleurisy without Mention of Effusion or Current Tuberculosis
597.8 Urethritis, Unspecified
684 Impetigo
686.9 Unspecified Local Infection of Skin and Subcutaneous Tissue
ADG 3 Time Limited: Major
245.9 Thyroiditis, Unspecified
286.5 Hemorrhage Disorder due to Circulating Anticoagulants
348.2 Pseudotumor Cerebri
357.9 Unspecified Inflammatory and Toxic Neuropathies
360.3 Hypotony of Eye
361.0 Retinal Detachment with Retinal Defect
365.22 Acute Angle-Closure Glaucoma
370.0 Corneal Ulcer
401.0 Malignant Essential Hypertension
411.1 Intermediate Coronary Syndrome
415.1 Pulmonary Embolism and Infarction
444.9 Arterial Embolism and Thrombosis
451.2 Phlebitis and Thrombophlebitis of Lower Extremities
512.8 Other Spontaneous Pneumothorax
518.0 Pulmonary Collapse
560.3 Impaction of Intestine
560.9 Unspecified Intestinal Obstruction
574.0 Calculus o f Gallbladder with Acute Cholecystitis
584.9 Acute Renal Failure, Unspecified
596.1 Intestinovesical Fistula
641.1 Hemorrhage From Placenta Previa
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
749.0 Cleft Palate, Unspecified
ADG 4 Time Limited: Major-Primary Infections
011.9 Unspecified Pulmonary Tuberculosis
038.0 Streptococcal Septicemia
047.9 Unspecified Viral Meningitis
053.19 Herpes Zoster with Other Nervous System Complications
054.7 Herpes Simplex with Other Specified Complications
070.3 Viral Hepatitis B without Mention of Hepatic Coma
091.3 Secondary Syphilis of Skin or Mucous Membranes
112.4 Candidiasis of Lung
360.0 Purulent Endophthalmitis
391.8 Other Acute Rheumatic Heart Disease
424.90 Endocarditis, Valve Unspecified, Unspecified Cause
466.1 Acute Bronchiolitis
487.0 Influenza with Pneumonia
572.0 Abscess of Liver
573.3 Hepatitis, Unspecified
604.9 Orchitis and Epididymitis, Unspecified
608.4 Other Inflammatory Disorders of Male Genital Organs
614.0 Acute Salpingitis and Oophoritis
614.1 Chronic Salpingitis and Oophoritis
614.9 Unspecified Inflammatory Disease of Female Pelvic Organs and Tissues
682.6 Cellulitis and Abscess of Leg, Except Foot
686.0 Pyoderma
711.0 Pyogenic Arthritis
727.06 Tenosynovitis of Foot and Ankle
730.2 Unspecified Osteomyelitis
790.7 Unspecified Bacteremia
998.5 Postoperative Infection, Not Elsewhere Classified
ADG 5 Allergies
446.2 Hypersensitivity Angiitis
477.0 Allergic Rhinitis due to Pollen
477.9 Allergic Rhinitis, Cause Unspecified
478.0 Hypertrophy of Nasal Turbinates
478.8 Upper Respiratory Tract Hypersensitivity Reaction
708.0 Allergic Urticaria
708.2 Urticaria due to Cold and Heat
708.9 Unspecified Urticaria
995.3 Allergy, Unspecified, N ot Elsewhere Classified
ADG 6 Asthma
493.0 Extrinsic Asthma
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
493.1 Intrinsic Asthma
493.9 Asthma, Unspecified
ADG 7 Likely to Recur: Discrete
263.9 Unspecified Protein-Calorie Malnutrition
266.2 Other B-Complex Deficiencies
274.9 Gout, Unspecified
280.9 Iron Deficiency Anemia, Unspecified
283.9 Acquired Hemolytic Anemia, Unspecified
354.0 Carpal Tunnel Syndrome
386.11 Benign Paroxysmal Positional Vertigo
427.0 Paroxysmal Supraventricular Tachycardia
427.61 Supraventricular Premature Beats
443.0 Raynaud’s Syndrome
451.9 Phlebitis and Thrombophlebitis of Unspecified Site
454.0 Varicose Veins of Lower Extremities with Ulcer
478.1 Other Diseases of Nasal Cavity an Sinuses
531.9 Gastric Ulcer, Unspecified as Acute or Chronic
533.9 Peptic Ulcer
535.5 Unspecified Gastritis and Gastroduodenitis
535.6 Duodenitis
536.8 Dyspepsia and Other Specified Disorders of Function of Stomach
562.11 Diverticulitis of Colon
564.0 Constipation
592.0 Calculus of Kidney
601.9 Prostatitis, Unspecified
691.8 Other Atopic Dermatitis and Related Conditions
724.2 Lumbago
724.4 Thoracic or Lumbosacral Neuritis or Radiculitis, Unspecified
724.5 Backache, Unspecified
ADG 8 Likely to Recur: Discrete-Infections
034.0 Streptococcal Sore Throat
054.1 Genital Herpes
112.9 Candidiasis of Unspecified Site
381.1 Chronic Serous Otitis Media
381.2 Chronic Mucoid Otitis Media, Simple or Unspecified
382.9 Unspecified Otitis Media
472.0 Chronic Rhinitis
472.1 Chronic Pharyngitis
474.0 Chronic Tonsillitis
523.0 Acute Gingivitis
590.80 Pyelonephritis, Unspecified
590.9 Infection of Kidney, Unspecified
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
595.0 Acute Cystitis
595.2 Other Chronic Cystitis
599.0 Urinary Tract Infection, Site Not Specified
601.0 Acute Prostatitis
616.0 Cervicitis and Endocervicitis
616.1 Vaginitis and Vulvovaginitis
646.6 Infections of Genitourinary Tract in Pregnancy
ADG 9 Likely to Recur: Progressive
250.10 Adult-Onset Type Diabetes Mellitus with Ketoacidosis
251.0 Hypoglycemic Coma
386.2 Vertigo of Central Origin
410.0 Acute Myocardial Infarction of Anterolateral Wall
410.7 Subendocardial Infarction
411.8 Other Acute and Subacute Forms of Ischemic Heart Disease
415.0 Acute Cor Pulmonale
427.1 Paroxysmal Ventricular Tachycardia
434.0 Cerebral Thrombosis
434.1 Cerebral Embolism
435.1 Vertebral Artery Syndrome
435.9 Unspecified Transient Cerebral Ischemia
514 Pulmonary Congestion and Hypostasis
571.1 Acute Alcoholic Hepatitis
572.2 Hepatic Coma
577.0 Acute Pancreatitis
ADG 10 Chronic Medical: Stable
094.0 Tabes Dorsalis
240.9 Goiter, Unspecified
244.9 Unspecified Acquired Hypothyroidism
250.00 Adult-Onset Type Diabetes Mellitus without Mention of Complication
256.4 Polycystic Ovaries
272.0 Pure Hypercholesterolemia
272.4 Other and Unspecified Hyperlipidemia
278.0 Obesity
281.0 Pernicious Anemia
289.9 Unspecified Diseases of Blood and Blood-Forming Organs
342.9 Hemiplegia, Unspecified
343 Infantile Cerebral Palsy, Unspecified
343.2 Congenital Quadriplegia
345.9 Epilepsy, Unspecified
389.9 Unspecified Hearing Loss
401.9 Unspecified Essential Hypertension
424.0 Mitral Valve Disorders
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
426.1 Atrioventricular Block, Other and Unspecified
426.4 Right Bundle Branch Block
429.2 ASCVD
440.9 Generalized and Unspecified Atherosclerosis
491.0 Simple Chronic Bronchitis
600 Hyperplasia of Prostrate
607.84 Impotence of Organic Origin
610.1 Diffuse Cystic Mastopathy
617.9 Endometriosis, Site Unspecified
625.6 Stress Incontinence, Female
642.0 Benign Essential Hypertension Complicating Pregnancy, Childbirth & Puerperium
715.9 Osteoarthrosis, Unspecified Whether Generalized or Localized
729.1 Myalgia and Myositis, Unspecified
733.0 Osteoporosis
745.4 Ventricular Septal Defects
758.0 Down’s Syndrome
780.5 Sleep Disturbances
797 Senility Without Mention of Psychosis
ADG 11 Chronic Medical: Unstable
135 Sarcoidosis
242.9 Thyrotoxicosis without mention of Goiter
277.0 Cystic Fibrosis
279.3 Unspecified Immunity Deficiency
282.6 Sickle-Cell Anemia
331.4 Obstructive Hydrocephalus
332.0 Paralysis Agitans
340 Multiple Sclerosis
413.9 Other and Unspecified Angina Pectoris
414.0 Coronary Atherosclerosis
414.9 Chronic Ischemic Heart Disease, Unspecified
424.1 Aortic Valve Disorders
424.3 Pulmonary Valve Disorders
427.3 Atrial Fibrillation and Flutter
427.9 Cardiac Dysrhythmia, Unspecified
428.0 Congestive Heart Failure
429.0 Myocarditis, Unspecified
429.9 Heart Disease, Unspecified
433.1 Occlusion and Stenosis of Carotid Artery
440.2 Atherosclerosis of Arteries of the Extremities
443.9 Peripheral Vascular D isease, Unspecified
491.2 Obstructive Chronic Bronchitis
492.8 Other Emphysema
496 Chronic Airway Obstruction
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
518.3 Pulmonary Eosinophilia
571.5 Cirrhosis of Liver without Mention of Alcohol
586 Renal Failure
707.1 Ulcer of Lower Limbs, Except Decubitus Ulcer
710.0 Systemic Lupus Erythematosus
710.9 Unspecified Diffuse Connective Tissue Disease
714.0 Rheumatoid Arthritis
722.6 Degeneration of Intervertebral Disc, Site Unspecified
746.85 Congenital Coronary Artery Anomaly
V43.3 Heart Valve Replaced by Other Means
V45.0 Postsurgical Cardiac Pacemaker in situ
ADG 12 Chronic Specialty: Stable-Orthopedic
717.7 Chondromalacia of Patella
718.8 Other Joint Derangement, Not Elsewhere Classified
721.0 Cervical Spondylosis without Myelopathy
721.9 Spondylosis of Unspecified Site
726.0 Adhesive Capsulitis of Shoulder
735.2 Hallux Rigidus
735.4 Other Hammer Toe (Acquired)
735.8 Other Acquired Deformities of Toe
736.1 Mallet Finger
756.1 Congenital Anomalies of Spine
ADG 13 Chronic Specialty: Stable-Ear, Nose, Throat
380.5 Acquired Stenosis of External Ear Canal, Unspecified as to Cause
385.3 Cholesteatoma
385.9 Unspecified Disorder of Middle Ear and Mastoid
387.9 Otosclerosis, Unspecified
389.1 Sensorineural Hearing Loss
476.0 Chronic Laryngitis
478.31 Partial Unilateral Paralysis of Vocal Cords or Larynx
ADG 14 Chronic Specialty: Stable-Eye
362.51 Nonexudative Senile Macular Degeneration of Retina
362.60 Peripheral Retinal Degeneration, Unspecified
366.1 Senile Cataract
366.16 Senile Nuclear Cataract
366.50 After-Cataract, Unspecified
367.0 Hypermetropia
367.1 M yopia
367.21 Regular Astigmatism
367.4 Presbyopia
367.9 Unspecified Disorder of Refraction and Accommodation
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
368.03 Refractive Amblyopia
369.0 Blindness, bilateral
371.57 Endothelial Corneal Dystrophy
372.9 Unspecified Disorder of Conjunctiva
377.21 Drusen of Optic Disc
379.31 Aphakia
ADG 16 Chronic Specialty: Unstable-Orthopedic
719.26 Villonodular Synovitis Involving Lower Leg
721.91 Spondylosis of Unspecified Site with Myelopathy
722.2 Displacement of Intervertebral Disc, Site Unspecified, Without Myelopathy
723.0 Spinal Stenosis in Cervical Region
724.02 Spinal Stenosis of Lumbar Region
730.1 Chronic Osteomyelitis, Site Unspecified
732.2 Nontraumatic Slipped Upper Femoral Epiphysis
732.4 Juvenile Osteochondrosis of Lower Extremity, Excluding Foot
732.7 Osteochondritis Dissecans
733.4 Aseptic Necrosis of Bone, Site Unspecified
ADG 17 Chronic Specialty: Unstable-Ear, Nose, Throat
383.1 Chronic Mastoiditis
386.0 Meniere’s Disease
ADG 18 Chronic Specialty: Unstable-Eye
361.3 Retinal Defects without Detachment
362.01 Background Diabetic Retinopathy
362.3 Retinal Vascular Occlusion, Unspecified
362.5 Degeneration of Macula and Posterior Pole of Retina
365.04 Ocular Hypertension
365.11 Primary Open Angle Glaucoma
365.9 Unspecified Glaucoma
366.9 Unspecified Cataract
370.33 Keratoconjunctivitis Sicca, Not Specified as Sjogren’s
374.1 Ectropion, Unspecified
375.15 Tear Film Insufficiency, Unspecified
378.1 Exotropia
379.0 Scleritis
ADG 20 Dermatologic
078.0 Molluscum Contagiosum
078.1 Viral Warts
110.9 Dermatophytosis
380.0 Perichondritis of Pinna, Unspecified
448.1 Nevus, Non-Neoplastic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
696.1 Other Psoriasis and Similar Disorders
697.0 Lichen Planus
701.1 Keratoderma, Acquired
702 Other Dermatoses
704.0 Alopecia
704.8 Other Specified Diseases of Hair and Hair Follicles
705.81 Dyshidrosis
706.1 Other Acne
707.8 Chronic Ulcer of Other Specified Sites
709.1 Vascular Disorders of Skin
709.3 Degenerative Skin Disorders
ADG 21 Injuries/Adverse Effects: Minor
846.0 Lumbosacral (Joint) (Ligament) Sprain
847.0 Neck Sprain
847.2 Lumbar Sprain
848.9 Unspecified Site of Sprain and Strain
910.0 Abrasion or Friction Bum of Face, Neck, and Scalp
918.1 Superficial Injury of Cornea
920 Contusion of Face, Scalp, and Neck
924.9 Contusion of Unspecified Site
949.0 Bum
959.1 Other and Unspecified Injury to Trunk
ADG 22 Injuries/Adverse Effects: Major
507.0 Pneumonitis due to Inhalation of Food or Vomitus
805.0 Closed Fracture of Cervical Vertebra
815.0 Closed Fracture of Metacarpal Bones
818.0 Fractures of Upper Limb
820.8 Fracture of Unspecified Part of Neck of Femur, Closed
823.8 Closed Fracture of Unspecified Part of Tibia
824.8 Unspecified Closed Fracture of Ankle
844.9 Sprain of Unspecified Site of Knee and Leg
854.0 Intracranial Injury
972.1 Poisoning by Cardiotonic Glycosides and Drugs of Similar Action
977.9 Poisoning by Unspecified Drug or Medicinal Substance
984.9 Toxic Effect of Unspecified Lead Compound
995.2 Unspecified Adverse Effect of Drug, Medicinal and Biological Substance
998.2 Accidental Puncture or Laceration During a Procedure
999.5 Other Serum Reaction
ADG 23 Psychosocial: Time limited, Not severe
300.1 Hysteria
305.1 Tobacco Use Disorder
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
307.4 Specific Disorders of Sleep of Nonorganic Origin
308.3 Other Acute Reactions to Stress
309.0 Adjustment Reaction with Brief Depressive Reaction
309.4 Adjustment Reaction with Mixed Disturbance of Emotions and Conduct
799.2 Nervousness
V40 Mental and Behavioral Problems
V61.2 Parent-Child Problems, Unspecified
V62.4 Social Maladjustment
ADG 24 Psychosocial: Persistent/Recurrent, Stable
295.5 Latent Schizophrenia
300.0 Anxiety State
300.4 Neurotic Depression
300.7 Hypochondriasis
301.0 Paranoid Personality Disorder
307.2 Tics
307.51 Bulimia
307.6 Enuresis
309.1 Adjustment Reaction with Prolonged Depressive Reaction
311 Depressive Disorder
312.9 Unspecified Disturbance of Conduct
314.01 Attention Deficit Disorder of Childhood with Hyperactivity
317 Mild Mental Retardation
319 Unspecified Mental Retardation
ADG 25 Psychosocial: Persistent/Recurrent, Unstable
290.0 Senile Dementia
291.0 Delirium Tremens
292 Drug Psychoses
292.0 Drug Withdrawal Syndrome
294.9 Chronic Organic Brain Syndrome
295.0 Simple Type Schizophrenia
295.4 Acute Schizophrenic Episode
296.3 Major Depressive Disorder, Recurrent Episode
296.7 Bipolar Affective Disorder, Unspecified
298.9 Unspecified Psychosis
300.14 Multiple Personality
301.83 Borderline Personality
303.9 Alcohol Dependence
304.6 Other Specified Drug Dependence
ADG 26 Signs/Symptoms: Minor
458.9 Hypotension, Unspecified
626.4 Irregular Menstrual Cycle
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
709.0 Dyschromia
729.5 Pain in Limb
729.8 Swelling of Limb
780.4 Dizziness and Giddiness
780.6 Fever
780.8 Hyperhidrosis
783.1 Abnormal Weight Gain
784.0 Headache
784.7 Epistaxis
786.2 Cough
786.52 Painful Respiration
787.0 Nausea and Vomiting
788.4 Frequency of Urination and Polyuria
ADG 27 Signs/Symptoms: Uncertain
285.9 Anemia, Unspecified
388.3 Tinnitus
427.6 Premature Beats
458.0 Orthostatic Hypotension
599.7 Hematuria
623.0 Dysplasia of Vagina
716.9 Unspecified Arthropathy
719.06 Effusion of Lower Leg Joint
719.4 Pain in Joint
719.7 Difficulty in Walking
723.1 Cervicalgia
728.4 Laxity of Ligament
780.3 Convulsions
780.7 Malaise and Fatigue
785.0 Tachycardia, Unspecified
785.6 Enlargement of Lymph Nodes
786.5 Chest Pain
786.6 Swelling, Mass or Lump in Chest
787.2 Dysphagia
788.1 Dysuria
788.3 Incontinence of Urine
790.6 Other Abnormal Blood Chemistry
ADG 28 Signs/Symptoms: Major
256.8 Other Ovarian Dysfunction
379.9 Unspecified Disorder o f Eye and Adnexa
381.8 Other Disorders of Eustachian Tube
429.3 Cardiomegaly
565.1 Anal Fistula
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
578.1 Melena
602.9 Unspecified Disorder of Prostate
611.7 Mastodynia
621.3 Endometrial Cystic Hyperplasia
626.8 Disorders of Menstruation and Abnormal Bleeding
628.9 Infertility, Female, of Unspecified Origin
733.9 Other and Unspecified Disorders of Bone and Cartilage
780.2 Syncope and Collapse
782.3 Edema
783.2 Abnormal Loss of Weight
783.4 Lack of Expected Normal Physiological Development
785.9 Other Symptoms Involving Cardiovascular System
786.0 Respiratory Abnormalities
795.0 Nonspecific Abnormal Papanicolaou Smear of Cervix
ADG 29 Discretionary
353.4 Lumbosacral Root Lesions
441.4 Abdominal Aneurysm without Mention of Rupture
454.9 Varicose Veins of Lower Extremities
455.6 Unspecified Hemorrhoids
470 Deviated Nasal Septum
474.9 Unspecified Chronic Disease of Tonsils and Adenoids
526.1 Fissural cysts of jaw
550.9 Inguinal Hernia
598.9 Urethral Stricture
700 Corns and Callosities
701.4 Keloid Scar
706.2 Sebaceous Cyst
727.0 Synovitis and Tenosynovitis
727.1 Bunion
735.0 Hallux Valgus
736.7 Other Acquired Deformities of Ankle and Foot
749.1 Cleft Lip, Unspecified
938 Foreign Body in Digestive System
ADG 30 See and Reassure
259.4 Dwarfism, Not Elsewhere Classified
278.1 Localized Adiposity
282.5 Sickle-Cell Trait
368.5 Color Vision Deficiencies
372.7 Conjunctival Vascular Disorders and Cysts
379.21 Vitreous Degeneration
379.5 Nystagmus, Unspecified
575.6 Cholesterolosis of Gallbladder
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
611.1 Hypertrophy of Breast
621.6 Malposition of Uterus
709.2 Scar Conditions and Fibrosis of Skin
728.5 Hypermobility Syndrome
734 Flat Foot
785.1 Palpitations
V65.5 Person with Feared Complaint in whom no Diagnosis was Made
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ADG 31 Prevention/Administrative
V06.3 Need for Prophylactic Vaccination with DTP + Polio Vaccine
V20.2 Routine Infant or Child Health Check
V24.2 Routine Postpartum Follow-Up
V25.9 Unspecified Contraceptive Management
V42.0 Kidney Replaced by Transplant
V57.1 Care Involving Other Physical Therapy
V58.3 Attention to Surgical Dressings and Sutures
V61.1 Marital Problems
V61.5 Multiparity
V67.0 Follow-Up Examination Following Surgery
V68.1 Issue of Repeat Prescriptions
V70.0 Routine General Medical Examination at a Health Care Facility
V72.0 Examination of Eyes and Vision
V72.3 Gynecological Examination
V72.4 Pregnancy Examination or Test, Pregnancy Unconfirmed
V72.5 Radiological Exam
V72.6 Laboratory Exam
ADG 32 Malignancy
153.9 Malignant Neoplasm of Colon
162.9 Malignant Neoplasm of Bronchus and Lung
174.9 Malignant Neoplasm of Breast (Female)
180.9 Malignant Neoplasm of Cervix Uteri
185 Malignant Neoplasm of Prostate
200.1 Lymphosarcoma, Unspecified Site
201.9 Hodgkin’s Disease, Unspecified Type
202.1 Mycosis Fungoides
202.8 Other Malignant Lymphomas
208.9 Unspecified Leukemia
V58.0 Radiotherapy Session
ADG 33 Pregnancy
640.0 Threatened Abortion
642.4 Mild or Unspecified Pre-eclampsia, Unspecified as to Episode of Care
644.1 Other Threatened Labor, Unspecified as to Episode of Care
646.9 Unspecified Complication of Pregnancy
650 Delivery in a Completely Normal Case
651.0 Twin Pregnancy, Unspecified as to Episode of Care
669.7 Cesarean Delivery, Without Mention of Indication
V22.2 Pregnant State, Incidental
V23.9 Supervision of Unspecified High-Risk Pregnancy
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

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

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

The design and synthesis of novel ligands as possible mimics of methane monooxygenase

PDF

The role of information in economic and public policies: Empirical applications for measuring human capital and water regulatory contracts

PDF

Physician profiling and clinical pathways: Combining the tools to change physician resource utilization

PDF

Essays on regulation of public utilities and the provision of public goods

PDF

Compliance study of second-generation antipsychotics on patients with schizophrenia

PDF

TfR-mediated oral delivery of protein drugs: Oral delivery of recombinant G -CSF -Tf fusion protein and its spacer optimization

PDF

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

PDF

Early adolescent drug use among multiethnic males: A prospective examination of the influences of psychological distress, relationship with family and school, law abidance, guilt and peer drug use

PDF

Labor contracts under general equilibrium: Three essays on the comparative statics of employment

PDF

The implementation of endangered species policy: An analysis of the 'takings' issue in specific contexts

PDF

Structure-function studies on the mammalian intestinal dipeptide transporter hPEPT1

PDF

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

PDF

Two essays on international economics: Purchasing power parity in the long run and the effect of real exchange rates on foreign direct investment

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

Asset Metadata

Creator Thiebaud, Patrick (author)

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

Degree Doctor of Philosophy

Degree Program Economics

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

Tag economics, general,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-c16-562884

Unique identifier UC11340321

Identifier 3145302.pdf (filename),usctheses-c16-562884 (legacy record id)

Legacy Identifier 3145302.pdf

Dmrecord 562884

Document Type Dissertation

Rights Thiebaud, Patrick

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