Avorn J, Monette J, Lacour A, Bohn RL, Monane M, Mogun H, LeLorier J. Persistence of Use of Lipid-Lowering MedicationsA Cross-National Study. JAMA. 1998;279(18):1458–1462. doi:10.1001/jama.279.18.1458
From the Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, Mass (Drs Avorn, Monette, and Monane and Mss Bohn and Mogun); and the Centre de Recherche, Hôtel-Dieu de Montréal, Université de Montréal, Montreal, Quebec (Drs Lacour and LeLorier).
Context.— Although clinical trials have demonstrated the benefits of lipid-lowering
therapy, little is known about how these drugs are prescribed or used in the
Objective.— To estimate predictors of persistence with therapy for lipid-lowering
drug regimens in typical populations of patients in the United States and
Design.— A cohort study defining all prescriptions filled for lipid-lowering
drugs during 1 year, as well as patients' demographic and clinical characteristics.
Setting.— New Jersey's Medicaid and Pharmacy Assistance for the Aged and Disabled
programs and Quebec's provincial medical care program.
Patients.— All continuously enrolled patients older than 65 years who filled 1
or more prescriptions for lipid-lowering drugs (N=5611 in the US programs,
and N = 1676 drawn from a 10% sample in Quebec).
Main Outcome Measures.— Proportion of days during the study year for which patients had filled
prescriptions for lipid-lowering drugs; predictors of good vs poor persistence
Results.— In both populations, patients failed to fill prescriptions for lipid-lowering
drugs for about 40% of the study year. Persistence rates with 3-hydroxy-3-methylglutaryl
coenzyme A reductase inhibitors were significantly higher than those seen
with cholestyramine (64.3% vs 36.6% of days with drug available, respectively).
Patients with hypertension, diabetes, or coronary artery disease had significantly
higher rates of persistence with lipid-lowering regimens. In New Jersey, multivariable
analysis indicated that the poorest patients (those enrolled in Medicaid)
had lower rates of drug use than less indigent patients (those enrolled in
Pharmacy Assistance for the Aged and Disabled) after adjusting for possible
confounders, despite virtually complete drug coverage in both programs. When
rates of use were measured in the US population for the 5 years following
the study year, only 52% of surviving patients who were initially prescribed
lipid-lowering drugs were still filling prescriptions for this drug class.
Conclusion.— In all populations studied, patients who were prescribed lipid-lowering
drug regimens remained without filled prescriptions for over a third of the
study year on average. Rates of persistence varied substantially with choice
of agent prescribed, comorbidity, and socioeconomic status, despite universal
coverage of prescription drug costs. After 5 years, about half of the surviving
original cohort in the United States had stopped using lipid-lowering therapy
CLINICAL TRIALS have documented the clinical utility of lipid-lowering
regimens in reducing cardiac morbidity and mortality in a variety of populations.
However, evaluating the effectiveness of these drugs throughout the health
care system requires an understanding of drug use and outcomes in actual free-living
patient populations, as compared with rates seen in subjects recruited into
controlled clinical trials.1 While randomized
trials are invaluable in determining the efficacy of regimens under well-controlled
circumstances, by definition they include only volunteer subjects and physicians,
with medication provided and monitored in a setting of considerably greater
surveillance than exists in routine practice.
Although some studies have examined use of lipid-lowering medications
in health maintenance organizations,1,2
little is known about how these drugs are taken outside of managed care settings,
and among patients older than 65 years, minorities, and the poor. Given the
very large proportion of cardiac morbidity and mortality that occurs among
the older population and the disproportionate burden of such outcomes among
blacks and the poor, it is critical to understand how lipid-lowering drugs
are used in patient populations at highest risk for cardiac disease.
Persistence with long-term medication of various kinds has been found
to be approximately 50% in several studies.1,3,4
Lack of persistent use of a drug for a chronic condition may result from patient
noncompliance or from a physician's decision to discontinue therapy if adverse
effects are believed to outweigh benefits. Old age per se is not a risk factor
for nonpersistence, but the number of prescribed medications is.5
Controversy persists on the role of lipid-lowering agents in the very old,6- 8 but current evidence suggests
that older patients can also benefit from reduction of hypercholesterolemia,9 and lifelong treatment is generally necessary to manage
To measure actual use rates of lipid-lowering drugs in typical patients,
we evaluated filled prescription rates and the factors associated with such
use in 3 populations of patients receiving care in noncapitated settings,
yet who were covered by health insurance programs that provided medications
at little or no cost to enrollees.
The US study population was drawn from enrollees in the New Jersey Medicaid
program and that state's Pharmacy Assistance for the Aged and Disabled (PAAD)
program for the period from January 1, 1990, to July 1, 1991. During the study
period, eligibility for Medicaid required an annual income below the poverty
level. By contrast, the state's PAAD program provided drug coverage to a less
indigent population, with income ceilings of $15700 if single and $19250 if
married. Information on all filled prescriptions was extracted from the paid
claims files of these programs. Each such record contained information on
the specific drug dispensed, dosage, quantity dispensed, and number of days'
supply. Previous studies have reported a high degree of reliability and validity
of Medicaid prescription data.10- 12
We then linked drug use data with information on other health service use
to identify diagnoses and other clinical services used during the study period.
All personal identifiers were removed prior to analysis to protect the confidentiality
of program participants.
The Canadian population was drawn from a 10% random sample of the administrative
files of the universal health care system of Quebec, the Régie de l'Assurance
Maladie du Québec (RAMQ), covering the same time period. All Quebec
residents 65 years and older are automatically enrolled in this program, and
RAMQ administrative files contain demographic information on all enrollees.
As in Medicaid, detailed information is also available for all medical services
provided and prescriptions filled. Each claim record identifies the type of
service or procedure provided and the location and type of institution where
the service was provided. Each pharmacy record includes a drug code that identifies
the name of the product, dosage, manufacturer, and dosage form, as well as
data on cost, quantity dispensed, duration of the prescription, prescribing
physician, and dispensing pharmacist. These data have high reliability and
validity as well.13
During the study period, New Jersey's Medicaid and Quebec's RAMQ programs
had no deductible or maximum benefit for drugs, and charged no copayment for
prescribed medications, ensuring complete ascertainment of drug use. New Jersey's
PAAD program had no deductible and no maximum benefit but had a nominal $2
copayment for each prescription filled. None of the programs had any formulary
Lipid-lowering drugs included all those in use during the study period:
clofibrate, colestipol, cholestyramine, gemfibrozil, niacin, probucol, and
3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors, including
lovastatin, pravastatin, and simvastatin. We identified all enrollees in the
New Jersey Medicaid and PAAD programs and in the Quebec RAMQ program who were
65 years or older on January 1, 1989, and who filled at least 1 prescription
for any lipid-lowering drug. The index date for a given study subject was
defined as the first date during 1990 on which a patient filled a prescription
for 1 of these drugs. To ascertain ongoing program participation, subjects
were required to have filled at least 1 prescription for any drug (including
non–lipid-reducing drugs) during each of the 3 6-month periods preceding
the index date as well as in each of the 3 consecutive 6-month periods following
the index date. However, patients who died after the index date were retained
in the analysis until their date of death.
New users were defined as those who had no
prior lipid-lowering drug prescription filled during the 365 days preceding
the index date. Long-term users were those who had
filled 1 or more previous lipid-lowering drug prescriptions during the 365
days preceding the index date. The US subjects were required to have no nursing
home residency during the study period to ensure comparability with the Canadian
population, since data on medications dispensed to nursing home residents
are not available in the Quebec RAMQ database.
We reviewed all health care use by each subject to ascertain the presence
of markers for probable coronary artery disease (CAD), hypertension, and diabetes
in the year prior to the index date. Markers for CAD included diagnosis of
myocardial infarction or acute or chronic ischemic heart disease, coronary
artery bypass graft or angioplasty, or the use of nitroglycerine and related
nitrates. Markers for diabetes included the presence of that diagnosis or
its specific complications, or the use of insulin or oral hypoglycemic agents.
Markers for hypertension included the relevant diagnoses or complications
identified as directly related to this condition, or the prescription of thiazides,
angiotensin-converting enzyme inhibitors, or other common antihypertensive
drugs. Calcium channel blocker or β-blocker use was considered a marker
for hypertension if the patient did not have any markers for CAD. For each
patient, we also ascertained other factors potentially related to persistence,
including age, sex, number of prescriptions filled for other drugs, frequency
of outpatient visits, and days of acute hospitalization in the year preceding
the index date.
Persistence was measured by determining the quantity dispensed and number
of days' supply data recorded on each filled prescription for each patient.
This made it possible to measure the number of days a patient had lipid-reducing
medication available during the 365-day study period (days covered). For each
cohort member, the denominator represented the total duration of possible
drug use during the study year, generally 365 days. (For patients who died
within the 365 days following the index date, the denominator represented
the total duration of days from the index date to the date of death.) For
patients simultaneously taking 2 or more agents in this class, if 1 such prescription
was dispensed during the time interval of a previous prescription, the overlapping
number of days of available therapy was not duplicated. This analysis was
repeated for all prescriptions for lipid-lowering drugs filled by each patient
in the 365 days after the index date for each patient.
To estimate persistence with specific agents, we first classified subjects
according to the lipid-lowering prescription filled on their index date, and
excluded the small number of patients who on this date filled multiple prescriptions
from this class. Subsequent analyses measured the rate of switching from and
to each drug studied. Finally, to build a multiple regression model of predictors
of persistence, we categorized subjects into 2 groups, those whose filled
prescriptions were adequate to cover 80% or more of the study year vs those
whose filled prescriptions covered less than 80% of the study year.
For the US populations, we obtained complete filled prescription data
from January 1, 1990, through June 30, 1996, for the entire original cohort.
We then examined these data for the period July 1, 1995, to June 30, 1996,
to measure long-term drug use rates for all surviving subjects.
The likelihood of 80% or more persistence with lipid-lowering therapy
vs 80% or less persistence was estimated from odds ratios calculated through
unconditional logistic regression14 using the
SAS CATMOD procedure.15 Confidence intervals
for the estimated odds ratios and significance tests for differences from
the null value were calculated using the estimated SEs.16
Potential determinants of persistence in the regression model included age;
sex; presence of markers for hypertension, diabetes, and CAD; number of filled
prescriptions excluding lipid-lowering drugs; number of outpatient visits;
and number of days of hospitalization during the preceding year. Two-tailed P values less than .05 were considered significant.
In New Jersey, 5611 patients met study criteria, as did 1676 patients
in Quebec. During the 1-year study period, 79 patients died in the New Jersey
population and 47 in Quebec. Analysis of filled prescriptions revealed that
average persistence with lipid-lowering therapy, measured as the mean percentage
of days in the study year in which patients had filled prescriptions available,
was 59.0% (±30.3%; median, 62.2%) for the New Jersey population and
62.6% (±29.0%; median, 70.1%) for the Quebec population. In New Jersey
and in Quebec, respectively, 11.6% and 14.8% failed to fill any additional
prescriptions for a lipid-lowering drug after filling the prescription on
their index date, and 34.1% and 38.9% of patients had medication available
during 80% or more of the year.
The most commonly prescribed lipid-lowering drugs were, in decreasing
order, an HMG-CoA reductase inhibitor (New Jersey, 39.4%; Quebec, 28.6%) and
gemfibrozil (New Jersey, 35.9%; Quebec, 30.1%), followed by cholestyramine
in New Jersey (13.2%) and by nicotinic acid in Quebec (16.5%). Clofibrate
was used by 86 patients in the New Jersey population of lipid-lowering drug
users, and by 186 in Quebec. On the index date in the New Jersey population,
only 39 patients (0.7%) were simultaneously prescribed 2 agents from this
class, of whom 21 were receiving cholestyramine in combination with another
lipid-lowering drug. In Quebec, 20 (1.1%) were simultaneously prescribed 2
such agents on the index date, of whom 10 were receiving cholestyramine in
combination with another lipid-lowering drug.
In both the United States and Canada, the highest compliance was associated
with the use of an HMG CoA reductase inhibitor (64.3% ± 29.8% of days
covered), while the lowest was associated with the use of cholestyramine (36.6%
± 29.1% of days covered) (Figure 1).
No differences were seen among the reductase inhibitors. After controlling
for individual patients' demographic and clinical characteristics, US patients
who were prescribed an HMG-CoA reductase inhibitor had an odds ratio for good
persistence of about double that seen in patients prescribed other lipid-lowering
drugs, particularly bile acid sequestrants cholestyramine and colestipol (odds
ratio, 2.01; 95% confidence interval, 1.79-2.25). For each drug, persistence
was similar in both populations, but slightly higher in Quebec than in New
Table 1 presents the demographic
characteristics of patients with persistence of 80% or more compared with
patients with persistence of less than 80%. Logistic regression models revealed
the same associations for predictors of persistence found by univariate analysis
in both populations, so only results for the multivariate logistic regression
analyses are reported in Table 2.
In both populations, better levels of persistence were associated with the
presence of risk factors for future cardiac events, including hypertension,
diabetes, and CAD. As expected, long-term lipid-lowering drug users had significantly
higher persistence than new users (Table
2). These findings did not vary if cutoffs other than 80% were used.
Persistence did not differ by sex, but tended to decrease with increasing
age. In the New Jersey population, Medicaid enrollees were much less likely
than PAAD enrollees to have high rates of persistence, even after controlling
for the demographic and clinical factors described. New Jersey patients with
prescriptions for more than 16 drug products per year were less likely to
continue to fill prescriptions for lipid-lowering drugs; a similar trend was
seen in Quebec. Excluding all patients who filled only a single prescription
for a lipid-lowering drug did not alter the findings.
We classified patients according to the agent prescribed on their index
date and also identified those who were switched to another lipid-lowering
drug or had a second one added during the study year. In both populations,
about 15% of patients had an additional such agent prescribed after their
index date (Figure 2). In New Jersey,
a second drug was prescribed to 25.6% of patients taking cholestyramine, 25.6%
of those taking probucol, but only 10.0% of those taking an HMG-CoA reductase
inhibitor. These proportions were virtually identical in the Quebec population.
In both populations an HMG-CoA reductase inhibitor was the agent most commonly
added or substituted, followed by gemfibrozil and cholestyramine.
Of the 5611 patients originally studied in the US cohorts, 4092 (72.9%)
were still alive and receiving services from the Medicaid or PAAD programs
during the period from July 1, 1995, through June 30, 1996. Only about half
of these patients who were taking lipid-lowering drugs in the original study
year filled 1 or more prescriptions for any lipid-lowering drug during the
follow-up year of 1995-1996 (2111 or 51.6%).
Several methods are available to study actual drug use in typical practice
settings. Pill counts and drug level determinations (when available) are reliable
but impractical on a population-wide basis. Patient recall is frequently inaccurate
and biased by a reluctance to admit "improper" behavior. Subjective estimates
of persistence by physicians and nurses are often unreliable means of assessing
patients' medication use,17 with clinicians
often failing to detect markedly poor compliance in routine practice situations.18
Analyzing automated records of prescriptions actually filled makes it
possible to use a standardized measure from pharmacy data or a large fiscal
database to define continuity of medication use and gaps in therapy. Although
filling a prescription is not identical to consuming the drug, patterns of
ongoing prescription filling represent the most accurate way of estimating
actual medication use in large populations. These data from 3 large patient
populations in 2 countries present a picture of striking nonpersistence with
lipid-lowering therapy. Incomplete persistence with a prescribed medication
can contribute substantially to the variability observed in its therapeutic
outcome.19 It has been estimated that good persistence
will result in a 39% reduction in cholesterol level, while poor persistence
will result in a decrease of only 11%.20 The
data presented here come from diverse populations of typical patients and
physicians, enhancing their generalizability to routine health care practices.
The period of time described preceded the widespread dissemination of data
from several recent studies, and it is possible that physicians and patients
pursue lipid-lowering regimens more vigorously at present than in previous
years. However, the follow-up data through mid-1996 allow little room for
such optimism. Furthermore, in the case of hypertension, poor persistence
has been reported years after the publication of pivotal studies documenting
the efficacy of these drugs.21
The data demonstrate better persistence with HMG-CoA reductase inhibitors
and considerably worse persistence with the bile acid sequestrants, especially
in the US populations. These findings may reflect the greater convenience
of dosing regimens for the reductase inhibitors, or the differing side-effect
profiles of these drugs.22- 24
The need for more detailed patient counseling for the more demanding niacin
or bile acid sequestrant regimens may also have played a role. Even after
adjustment for health care resource use, patients with hypertension, diabetes,
and CAD adhered to their regimens more consistently than did those without
such risk factors.
Of particular concern is our observation that in the US data, even after
controlling for clinical and demographic differences, Medicaid enrollees,
who are indigent, were only 58% as likely to continue their therapy as PAAD
enrollees, who have higher annual incomes. Previous analyses have not consistently
found sociodemographic factors to be associated with lower rates of adherence
to drug regimens.25- 28
However, Medicaid patients, by virtue of greater poverty, may differ in their
access to health services as well as in level of education or basic health
knowledge, which may explain their lower rates of medication use.
These findings depict persistence of drug use in a set of fiscally generous
insurance systems. The Medicaid, PAAD, and Quebec systems all provide comprehensive
coverage of drug benefits for participating patients. Our data are a reminder
that adequate payment mechanisms are not in themselves sufficient to guarantee
appropriate use of covered services, and persistence is likely to be even
worse for these drugs in patients who must bear their considerable cost out-of-pocket.29 In a patient population without drug coverage, the
higher use rates seen with the newer agents compared with niacin might be
reduced because of their cost differential. Research is needed on drug-use
patterns in the large proportion of patients of all ages who receive little
or no reimbursement for pharmacy expenditures.
Discontinuing beneficial medication can cause preventable morbidity
and impose a considerable clinical and financial burden on the health care
system.30 As additional evidence emerges on
the utility of lipid reduction in reducing cardiac morbidity and mortality,
a comprehensive approach will be required to the care of patients with hyperlipidemia,
including monitoring of filled prescription patterns, patient and clinician
education, and reinforcement of patients' knowledge of their risk factor status
and its management.