Customize your JAMA Network experience by selecting one or more topics from the list below.
Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy Benefits and the Use of Drugs by the Chronically Ill. JAMA. 2004;291(19):2344–2350. doi:10.1001/jama.291.19.2344
Author Affiliations: RAND, Santa Monica, Calif (Drs Goldman, Joyce, and Escarce, Ms Pace, and Mr Solomon); Merck, West Point, Pa (Ms Landsman and Dr Teutsch); and California HealthCare Foundation, Oakland (Dr Lauori). Dr Lauori is now with Genentech, South San Francisco, Calif.
Context Many health plans have instituted more cost sharing to discourage use
of more expensive pharmaceuticals and to reduce drug spending.
Objective To determine how changes in cost sharing affect use of the most commonly
used drug classes among the privately insured and the chronically ill.
Design, Setting, and Participants Retrospective US study conducted from 1997 to 2000, examining linked
pharmacy claims data with health plan benefit designs from 30 employers and
52 health plans. Participants were 528 969 privately insured beneficiaries
aged 18 to 64 years and enrolled from 1 to 4 years (960 791 person-years).
Main Outcome Measure Relative change in drug days supplied (per member, per year) when co-payments
doubled in a prototypical drug benefit plan.
Results Doubling co-payments was associated with reductions in use of 8 therapeutic
classes. The largest decreases occurred for nonsteroidal anti-inflammatory
drugs (NSAIDs) (45%) and antihistamines (44%). Reductions in overall days
supplied of antihyperlipidemics (34%), antiulcerants (33%), antiasthmatics
(32%), antihypertensives (26%), antidepressants (26%), and antidiabetics (25%)
were also observed. Among patients diagnosed as having a chronic illness and
receiving ongoing care, use was less responsive to co-payment changes. Use
of antidepressants by depressed patients declined by 8%; use of antihypertensives
by hypertensive patients decreased by 10%. Larger reductions were observed
for arthritis patients taking NSAIDs (27%) and allergy patients taking antihistamines
(31%). Patients with diabetes reduced their use of antidiabetes drugs by 23%.
Conclusions The use of medications such as antihistamines and NSAIDs, which are
taken intermittently to treat symptoms, was sensitive to co-payment changes.
Other medications—antihypertensive, antiasthmatic, antidepressant, antihyperlipidemic,
antiulcerant, and antidiabetic agents—also demonstrated significant
price responsiveness. The reduction in use of medications for individuals
in ongoing care was more modest. Still, significant increases in co-payments
raise concern about adverse health consequences because of the large price
effects, especially among diabetic patients.
In recent years, many health plans have implemented policies to contain
drugs costs, including raising beneficiary co-payments, mandating use of generics,
requiring mail-order services, and expanding use of formularies, all of which
have large effects on total drug spending. For example, doubling co-payments
reduced total drug spending by 19% to 33% in one multiyear study of 25 companies.1 Such large responses often raise concerns about adverse
health consequences, particularly for chronically ill individuals. Indeed,
large changes in drug benefits are sometimes associated with substantial morbidity
and mortality in certain high-risk populations.2-4
In the privately insured nonelderly population, the patterns may differ.
Although there is also evidence that this population changes its patterns
of drug use when benefits change, there is less evidence of adverse health
consequences. One plausible explanation is that the health consequences manifest
over many years, and relevant longitudinal databases are scarce. In addition,
there may be different responses, depending on the diseases the drugs treat.
Several studies suggest that consumer sensitivity to cost sharing depends
on a drug's therapeutic class4-7 and
that increased cost sharing may decrease "nonessential" drug use more than
"essential" drug use.8,9 This
makes identifying the therapeutic classes most sensitive to benefit changes
We examined how changes in benefit design among privately insured populations
affect use of the most commonly used drug classes. In addition, we identified
populations more likely to be at risk for adverse health effects and isolated
their responsiveness to cost sharing.
We assembled a data set of pharmacy and medical claims from 1997 to
2000 for 30 large US employers covering 528 969 beneficiaries continuously
enrolled for up to 4 years (n = 960 791 person-years). The claims captured
all health care claims and encounters, including prescription drugs and inpatient,
emergency, and ambulatory services. Most drug claims included information
on the type of drug, drug name, national drug code, dosage, days supplied,
and place of purchase (retail or mail order). The medical claims included
the date of service, diagnosis and procedure codes, type of facility, and
We merged the claims data with information on the health insurance benefits
for each covered individual. Most of the companies offered employees a choice
of health insurance plans, yielding data from 52 health plans (n = 102 plan-years).
These plans varied across companies and by year because some of them changed
their benefits structure during the observation period. Although employees
typically had a choice of medical plans, only 2 companies had drug benefits
that differed across the plan options, thereby reducing potential bias from
selection of drug plans according to anticipated use.
For each plan-year, we obtained copies of the summary benefit design.
Salient benefit information was then abstracted by 2 members of the research
team. The few instances in which there were discrepancies were resolved with
a third opinion from one of the principal investigators. Abstracted drug-benefit
details included co-payments or coinsurance rates for retail and mail-order
pharmacies, generic substitution rules, and a list of drugs or drug classes
excluded from coverage. Characteristics of the medical benefit included deductibles
and patient cost-sharing arrangements for inpatient and ambulatory settings.
The result was a data set linking covered beneficiaries' drug use and insurance
benefits for approximately 1 million person-years.
We used a common classification scheme—the 2000 Red Book10—to associate each
drug with a therapeutic class. Table 1 shows
the 10 most common therapeutic classes (in terms of dollars spent) in our
sample for 2000. Together, these 10 classes accounted for more than $132 million
in sales across all plans in our sample and 51% of all drug spending. Not
surprisingly, the best-selling drugs include treatments for high cholesterol,
depression, gastrointestinal disorders, joint pain, allergies, and heart disease.
Our key independent variable was an index of plan generosity. In general,
it is difficult to translate the stated pharmacy benefit into actual prices
that consumers face. Multitier formularies are the standard for most private
plans, and they also offer discounts for purchases through mail-order or in-network
pharmacies. These added complexities mean that the price a consumer will pay
for a given drug depends not only on which tier it is placed in but also on
where it is dispensed. To address this issue, we constructed a price index
for the pharmacy benefit according to cost sharing for a standardized list
The index represents the expected out-of-pocket spending in each plan
for a standardized "market basket" of drugs. The market basket was generated
by drawing a representative sample of 100 enrollees from each plan who had
at least 1 drug claim, which was done separately for each year. We then compiled
a list of all the drugs purchased by these beneficiaries and aggregated this
list across plans to yield the market basket. To assign prices, we used the
average co-payment for each drug in each plan. In this way, our price for
a drug reflected not only the way a drug was tiered but also these other factors,
such as the accessibility of network pharmacies. Formally, the index is computed
as a weighted sum of average co-payments for each drug in the market basket,
where the weights reflect the per-beneficiary number of prescriptions filled
for each drug.
We modeled the likelihood that beneficiaries used any drugs in a therapeutic
class, as well as their spending, contingent on having filled at least 1 prescription
in that class. We estimated these outcomes by using a 2-part model estimated
separately for each therapeutic class. We used probit regression to estimate
the probability that a member had at least 1 pharmacy claim in the class.
The second part of the model used a generalized linear model to estimate drug
spending among members who were users. We chose a generalized linear model
with a logarithmic link function because it predicted days supplied better
than a 2-part model with log-linear regression in the second stage, but our
conclusions were insensitive to this choice.
Our therapeutic classes included most of the top-selling drugs, as shown
in Table 1. We combined several
classes used to treat the same chronic disease. For antihypertensives, we
included angiotensin-converting enzyme inhibitors, calcium channel blockers,
diuretics, β-blockers, and angiotensin II receptor blockers. Antidiabetes
drugs included sulfonylureas and other oral agents such as metformin and glitazones.
Insulin was not included because injectable drugs often have different benefits.
Antiulcerants included H2 receptor antagonists and proton pump
inhibitors, as well as other gastrointestinal drugs not elsewhere classified.
Antidepressants included selective serotonin reuptake inhibitor and tricyclic
drugs. Nonsteroidal anti-inflammatories included the cyclooxygenase 2 inhibitors.
Antiasthmatics included anticholinergics, anti-inflammatory asthma agents,
leukotriene modulators, oral steroids, steroid inhalers, sympathomimetics,
In addition to our index of plan generosity, we included binary indicators
identifying plans with coinsurance rates and policies requiring mandatory
generic substitution. The latter have been shown to have a large independent
effect on spending.1 Other independent variables
in our model included a set of variables to describe the medical benefits:
deductibles, co-payments, or coinsurance rates for physician office visits
and a binary indicator to identify plans with coinsurance rates for physician
office visits. The other covariates were binary indicators for age categories,
a binary indicator for male sex, median household income in the ZIP code of
residence, a binary indicator for active or retired status, a categoric variable
for urban residence, and binary indicators for years to control for time trends.
We also controlled for comorbid conditions by using a set of disease indicators
identified in the medical claims according to International
Classification of Diseases, Ninth Revision (ICD-9) diagnoses.1 (Full model results are available from the corresponding
For each therapeutic class, we used the results from the 2-part model
to simulate total days supplied. Specifically, we used estimates from the
first part of the model to predict the probability of nonzero days supplied
for each person under different values of the co-payment index. We used the
second part of the model to predict days supplied contingent on having at
least 1 claim. Total days supplied were predicted by using the product of
the 2. The predictions were then averaged over all individuals in the sample
for each value of the co-payment index. We predicted use for index values
of 168 and 336. These values correspond to moving from a 2-tier plan with
co-payments of $6.31 for generics and $12.85 for brand drugs to one with co-payments
of $12.62 and $25.70, respectively. Although the index value of 336 was not
observed in our sample, this level corresponds to co-payments observed recently
in the marketplace. Simulations of a 50% increase (using values within sample)
yielded proportionately similar results. All costs are in US dollars.
Patients were identified as having a chronic condition if their medical
claims included 2 or more office visits with the corresponding ICD-9 code (available on request) and they filled at least 1 prescription
for the listed therapeutic class. The chronic population varied by therapeutic
class: depression (for antidepressants), hypertension (antihypertensives),
hypercholesterolemia (antihyperlipidemics), gastric acid disorder (antiulcerants),
asthma (antiasthmatics), diabetes (antidiabetics), arthritis (nonsteroidal
anti-inflammatory drugs [NSAIDs]), and allergic rhinitis (antihistamines).
Patients with multiple conditions were included in each relevant subgroup
(eg, patients with diabetes and hypertension were included in the diabetes
and hypertension analyses). For each subgroup, we estimated 2 models: 1 for
disease-specific drug use and 1 for use of all other drugs. As an example,
for individuals who had hypertension and met the criteria above, we estimated
their use of antihypertensives and all other nonhypertensive medications.
We then predicted use for each group and each measure by following the methods
The RAND Human Subjects Protection Committee ruled that this research
was exempt from institutional review board approval.
Table 2 shows the characteristics
of the plans in our study sample. We broadly classified our plans into 1-tier,
2-tier, 3-tier, or coinsurance plans, with between 142 217 and 343 117
person-years for each type. The 1-tier plans are the most generous; these
plans require on average a $6.05 payment per retail prescription for a 30-day
supply. Co-payments are higher on mail-order prescriptions (averaging $9.60
per prescription), but they allow up to a 90-day supply. Average retail co-payments
in a 2-tier plan are $6.31 for generic drugs and $12.85 for brand drugs. Average
co-payments in a 3-tier plan range from $5.70 to $20.81. Mail-order co-payments
for 2-tier and 3-tier plans range from $8.91 for generics in a 3-tier plan
to $33.02 for nonpreferred brands.
The relative generosity of the plan is conveyed by the average price
index, which reflects average patient out-of-pocket spending for our market
basket of drugs. Our generosity index had an overall mean of $150, a median
of $136, and an interquartile range of $120 to $184 at the plan level (n =
102). Within plan type, 1-tier plans were the most generous (index value of
$124), followed by 2-tier plans ($168), 3-tier plans ($179), and then coinsurance
($181). The demographic characteristics of patients enrolled in each plan
type differed; for example, men made up 61% of the sample—not surprising,
given that this was a sample of primary beneficiaries—but the proportion
varied considerably by plan type. These differences underscore the importance
of our multivariate approach. On the other hand, the prevalence of chronic
disease was fairly similar across plan types, which suggests that unobserved
health differences were probably minimal.
Table 3 shows annual drug
use in each therapeutic class. On the whole, antihypertensives were used most
frequently (22% of the population). NSAIDs (19%), antihistamines (17%), antidepressants
(12%), and antihyperlipidemics (11%) were also used frequently. Antidiabetes
drugs (390 days) and antihypertensives (386 days) were used most heavily,
reflecting the chronic nature of these conditions and their treatment. In
contrast, NSAIDs (74 days), antihistamines (90 days), and antiasthmatics (117
days) were not used continuously throughout the year by most patients. These
findings suggest a classification of medications into 3 broad groups: drugs
that forestall disease progression and avoid long-term complications (antidiabetics,
antihypertensives, and antihyperlipidemics), medications that largely treat
symptoms or intermittent conditions (NSAIDS and antihistamines), and drugs
with characteristics of both (antidepressants, antiulcerants, and antiasthmatics).
Figure 1 shows the predicted
effects of doubling co-payments in each therapeutic class for the entire sample
and a subset of patients receiving ongoing treatment for a chronic illness.
We predicted the percentage change in annual days supplied in response to
a doubling of co-payments after adjusting for demographic and health characteristics.
The change was computed by predicting use for a drug plan with an index value
of 336, with predicted use for a plan with an index value of 168, which corresponds
to the effect of co-payments in a 2-tier plan increasing from $6.31 for generics
and $12.85 for brand drugs to $12.62 and $25.70, respectively. For the entire
study sample, we observed substantial reductions in spending for all classes
of drugs. The largest decreases occurred for NSAIDs (45%) and antihistamines
(44%). But the most striking feature is the much lower responsiveness among
chronically ill patients. For example, use of antidepressants by depressed
patients declined by only 8% when co-payments doubled compared with 26% overall.
Use of antihypertensives was reduced by 26% for the entire population compared
with only 10% among individuals with diagnosed hypertension. The lone exception
was patients with diabetes: their response to a doubling of co-payments (23%)
was virtually identical to that of the overall population (25%).
Figure 1 also compares use
within class and outside class for chronically ill patients only. Use of antidepressants
by depressed patients declined by 8% when co-payments doubled; however, their
use of all other drugs declined by 25%. Use of antihypertensives by patients
with high blood pressure declined by 10%, whereas their use of all other drugs
decreased by 27%. Similar but less dramatic differences emerged for antihyperlipidemics,
antiulcerants, and antiasthmatics. For antihistamines and NSAIDs, the pattern
reversed. Patients with arthritis and allergic rhinitis reduced their disease-specific
use by 27% and 31%, respectively, whereas their use of other drugs actually
declined by only 22%.
One might expect to see more price responsiveness for drugs with close
over-the-counter (OTC) substitutes or for higher-priced medications. Table 4 presents the predicted change in
annual days supplied when co-payments are doubled for people with 1 of 8 chronic
conditions. Medications with OTC substitutes included NSAIDs, antihistamines,
and antiulcerants (H2 receptor antagonists and proton pump inhibitors).
We found that a doubling of co-payments led to a 32% reduction in their use.
Use of medications without close OTC substitutes—defined as antidiabetics,
antiasthmatics, antihyperlipidemics, antihypertensives, and antidepressants—decreased
by only 15%. In contrast, we saw only modest differences by drug type (21%
for brand name vs 16% for generic). Taken together, the results suggest that
patients are more likely to forgo higher-priced medications and substitute
less-expensive OTC medications when possible as their out-of-pocket burden
In previous work, we found considerable price sensitivity in the demand
for prescription drugs among a working-age population with employer-provided
insurance.1 In that study, doubling co-payments
in a 2-tiered plan reduced overall drug spending by one third, but it is unclear
which therapeutic classes were most affected. The results presented here demonstrate
that doubling co-payments in a typical 2-tier plan is associated with significant
reductions in use across 8 of the most widely prescribed therapeutic classes.
The largest reductions were for drugs with close OTC substitutes that primarily
treat symptoms rather than the underlying disease. Doubling co-payments was
associated with reduced annual days' supply of antihistamines and NSAIDs of
about 45%; by comparison, use of antihypertensives and antidepressants decreased
Patients do not respond indiscriminately to co-payment increases. Individuals
receiving treatment for a specific condition are less likely to reduce their
use of disease-specific medications. For example, patients with diagnosed
high blood pressure reduced use of other drugs by 27% when co-payments doubled
but only by 10% for their antihypertensive medication (Figure 1). Because the average patient with high blood pressure
uses 386 days of antihypertensive medications annually (Table 3), these estimates imply that a doubling of co-payments would
reduce days supplied by more than 1 month (38.6 days).
This pattern—less responsiveness to price changes for disease-specific
medications relative to all other medications—was found in 5 of the
8 therapeutic classes we studied (antidepressants, antihypertensives, antihyperlipidemics,
antiulcerants, and antiasthmatics). Use of antihistamines and NSAIDs by people
with allergic rhinitis and arthritis, respectively, operated in the opposite
direction, meaning that these patients were price sensitive when taking their
disease-specific medications. Patients diagnosed with diabetes were a notable
exception; use of antidiabetes medications and nondiabetes medications decreased
by about one quarter in response to a 100% increase in co-payments. According
to Table 3, annual days supplied
decreased by more than 3 months when co-payments doubled (from 390 to 293
days). In supplemental analyses, we found that use of insulin, which we purposely
excluded from the antidiabetes class because it is often covered differently
by plans, was less responsive to benefit changes, with use decreasing by only
8% when co-payments doubled.
The populations most sensitive to price changes were the patients taking
long-term medications but who were not receiving ongoing care for the condition
(at least 2 medical visits per year for that condition). More research is
needed to determine whether this price-sensitive population consists of people
at risk for a disease or for whom the disease is well controlled and who do
not seek regular care or people with advanced disease who are not being treated
appropriately. Some of this price sensitivity may be beneficial for society
in the sense that it reduces excess consumption of drugs whose costs are greater
than the (monetized) health benefit. On the other hand, recent evidence suggests
that there may be substantial therapeutic benefits of lowering blood pressure
and serum cholesterol level for all individuals at risk, not simply those
with elevated rates.11 A recent clinical trial
also demonstrated that metformin can prevent or delay the onset of type 2
diabetes in patients at risk (although not as well as diet and exercise).12
When we examined the chronically ill population receiving routine care,
a group of patients who are most likely to benefit from drug treatment, we
still found that doubling co-payments is associated with reductions in drug
use of 8% to 23%. Although lower use of antihistamines or anti-inflammatories
is unlikely to affect patients' underlying health conditions, significant
reductions in the use of antidiabetic agents or medications to treat dyslipidemia
may have short- and clinical consequences.
Our findings raise concern that co-payment increases could lead to adverse
health consequences, at least for individuals with some conditions. In our
sample, we found evidence that co-payment increases led to increased use of
emergency department visits and hospital days for the sentinel conditions
of diabetes, asthma, and gastric acid disorder: predicted annual emergency
department visits increased by 17% and hospital days by 10% when co-payments
doubled (data not shown). However, because of limited information about the
full extent of medical benefit coverage and the choice of medical plans (which
are self-selected in a way that the drug benefits are not), these results
are not definitive. Other studies have found mixed evidence on this issue.3,13-15
There are several other limitations. First, our sample was drawn from
an insured working-age population, and thus our results are not necessarily
generalizable to other populations such as the poor or the elderly. Second,
we identified chronically ill patients from claims data. The main concern
with this approach is false positives if rule-out diagnoses are recorded on
the claims. We tried to minimize this error by restricting our analysis to
users of disease-specific drugs, requiring multiple physician visits or hospitalizations
for the condition and excluding laboratory claims from our diagnosis counts.
Finally, in all but 2 of our companies, beneficiaries did not have a choice
of drug benefits; most companies offered the same drug package to all employees
even as medical benefits varied. Excluding these companies from our analyses
did not appreciably affect the results.
Rapid changes in drug benefits have shifted a larger burden of pharmacy
costs onto beneficiaries. Beneficiaries have responded by reducing their use
of drugs, but their responsiveness varies substantially among the top-selling
therapeutic classes. The use of medications such as antihistamines and NSAIDs
that are taken intermittently to treat symptoms is sensitive to co-payment
changes. Other medications—antihypertensive, antiasthmatic, antidepressant,
antihyperlipidemic, antiulcerant, and antidiabetic agents—also demonstrate
significant price responsiveness. However, when one restricts attention to
subgroups with identifiable chronic illness, the reduction in use of medications
for those conditions is more modest. Still, significant increases in co-payments
do raise concern about adverse health consequences because of the large price
effects, especially among diabetic patients.