Cost-Sharing and Adherence to Antihypertensives for Low and High Adherers

Published Online: November 06, 2009
Jean Yoon, PhD, MHS; and Susan L. Ettner, PhD

Objective: To examine how the influence of cost-sharing on adherence to antihypertensive drugs varies across adherence levels.

Study Design: Cross-sectional study using medical and pharmacy claims and benefits data on 83,893 commercially insured patients with hypertension from the 2000-2001 Medstat MarketScan Database.

Methods: We measured drug adherence using the medication possession ratio (MPR) for antihypertensive drugs over 9 months. Drug cost-sharing was measured as either copayments or coinsurance. Other patient characteristics included age, sex, comorbidity, health plan type, and county-level sociodemographics. We compared adherence for different cost-sharing categories with a bivariate test of equal medians and simultaneous quantile regressions predicting different percentiles of drug adherence.

Results: Median MPR was high (>80%) across all cost-sharing categories. Among the poorest adherers, the regression-adjusted MPR was 8 to 9 points lower among patients with the highest drug cost-sharing compared with patients with the lowest cost-sharing (copayment $5 or less). The effects of cost-sharing were smaller at the median (2-3 points lower) and nonsignificant at higher levels of adherence. Other significant factors influencing adherence at low adherence levels were drug class and comorbidity.

Conclusion: Cost-sharing had a substantial negative association with adherence among low adherers and little association at higher adherence levels. At a clinical level, physicians should closely monitor adherence to antihypertensive drugs, particularly for patients with multiple comorbidities and those taking multiple drugs. At a health system level, current benefit designs should encourage adherence while limiting the cost burden of drugs for patients with multiple chronic conditions taking multiple drugs.

(Am J Manag Care. 2009;15(11):833-840)


A cross-sectional study of a large sample of working-age adults was done to examine how cost-sharing affects adherence to antihypertensive drugs.

  • Cost-sharing had the most significant and negative relationship with adherence for low adherers; it did not have a significant relationship with adherence for high adherers.
  • Other predictors of worse adherence for low adherers were drug class and greater comorbidity.
  • Rising costs for prescription drugs are a particular concern for poor adherers.
As a result of the growing prevalence of chronic diseases due to the aging population and better detection and identification of individuals with such diseases, and the development of new drugs that are highly effective, the healthcare system relies heavily on prescription drugs to manage chronic conditions and reduce morbidity and mortality. Adherence to some drugs can be low, and concerns exist over whether patients are receiving the full benefit from their drugs. Among factors influencing patients’ adherence to drugs, one of the strongest relationships exists between higher out-of-pocket payments for drugs and less drug utilization, including lower adherence to drug prescriptions1-5 with modest price effects for drugs to treat asymptomatic conditions including hypertension.6,7

It is unknown, however, whether patients with low drug utilization are more likely to be responsive to drug prices compared with patients with higher utilization of drugs. Demand for prescription drugs may have heterogeneity in price responsiveness; for example, patients who experience adverse effects or do not feel the drug is working well for them may have more price-elastic demand for the drug than other patients.

Using claims data from employer-sponsored health plans for a large sample of working-age adults diagnosed with hypertension, we applied an innovative quantile regression approach to examine how drug cost-sharing and other patient and health plan characteristics affect adherence to antihypertensive drugs across the distribution of adherence rather than looking at mean adherence, which provides a more limited picture. We focused on hypertension because it is one of the most prevalent chronic conditions in the United States, can be treated effectively with prescription drugs, and is similar to other asymptomatic conditions for which long-term management is critical to reduce the risk of morbidity and mortality. Moreover, when comorbid with other conditions such as diabetes, hypertension leads to development of diseases and complications including renal and diabetic eye disease.8 The direct costs of care attributable to hypertension are estimated to be $569 per year per capita, and the total costs (including the cost of comorbid conditions) are estimated to be $4073 per year per capita, or $110.3 billion per year for the United States.9 Therefore, adherence to drugs and overall management of hypertension have significant cost implications.



This study used data from the 2000-2001 Medstat MarketScan Database. Claims data were analyzed for 83,893 patients age 18 to 64 years who were diagnosed with hypertension (International Classification of Diseases, Ninth Revision [ICD-9] codes 401-405) on any medical claim during 2000, were continuously enrolled in their health plan for the entire 2-year study period, and had at least 1 drug claim for an antihypertensive drug during an index period (January 2000-June 2000). Benefits information was available only from large firms, covering approximately 59% of patients.

Patient-level variables were measured in 2000 and obtained from enrollment and benefits information linked to medical and pharmacy claims data. County of residence was used to link to US Census data on household income, race, and education. This study was approved by the UCLA Institutional Review Board (approval #G06-09-089-02).


Antihypertensive drugs included all drugs identified in 7 classes of drugs: angiotensin-converting enzyme (ACE) inhibitors, beta blockers, angiotensin II receptor blockers, thiazide diuretics, calcium channel blockers, alpha-1 blockers, and sympathetic blockers. Users of antihypertensive drugs were identified by any prescription for an antihypertensive drug occurring within a 6-month period at the beginning of 2000, which was considered to be the index prescription. Immediately following the last days supply of the index prescription, refill information was used to calculate the medication possession ratio (MPR) over a 9-month period by drug class. (An earlier study using administrative data measured adherence to antihypertensive drugs over 250 days and concluded that this measure of adherence was valid as it correlated well with drug effects.10) The days supplied for all refills within a given class was summed and divided by 270 days to calculate the MPR, prorating the last refill when necessary. The MPR represents the amount of time for which a patient had his/her drug supply, and it could vary from 0% to more than 100% if patients had overlapping prescriptions. The MPR values were truncated for 2% of patients with an MPR above 120%.

A single MPR was calculated if patients switched from one drug class to another during the study period. If patients filled prescriptions for drugs in multiple classes simultaneously, then a separate MPR was calculated for each drug class, and the mean of all classes was calculated.


Patient cost-sharing for drugs was categorized as ≤$5 copayment per prescription (the reference group), $6 to $12 copayment, ≥$15 copayment, 10% coinsurance, or 20% coinsurance. Separate groups were created for copayment and coinsurance groups because the amount paid under coinsurance depends on the type of drug filled, and patients are more price responsive under coinsurance than under copayments.11 As 64% of study patients filled prescriptions for brand-name rather than generic drugs, and cost-sharing between brand-name and generic drugs was highly correlated (Pearson correlation coefficient = 0.44), only cost-sharing for brand-name drugs was used in analyses. Analyses originally included separate indicators of cost-sharing for generic and brand-name drugs and number of drug plan tiers, but a high degree of collinearity was found among these measures, so generic cost-sharing and number of drug plan tiers were dropped from analyses.

Comorbidity was measured using morbidity groupings from ICD-9 codes in medical claims called aggregated diagnosis groups (ADGs)12 using dummy codes for 32 ADGs (not mutually exclusive), count of major ADGs (conditions with greater severity) ranging from 0 to 8, and dummy variables for congestive heart  failure, diabetes, ischemic heart disease, hyperlipidemia, and depression—comorbidities associated with cardiovascular disease.

Dummy categories were created for each of the 7 classes of antihypertensive drugs with thiazide diuretics as the reference category.

Additional covariates included patient cost-sharing for outpatient visits, type of health plan, receipt of preventive care, age, sex, employee working status, county-level median household income, percentage of nonwhite residents, and percentage of residents with less than a high school education.

Statistical Analyses

A nonparametric test of equal medians was used to test for bivariate differences in the median MPR by patient and health plan characteristics. Simultaneous quantile regression then was used to estimate MPR. Quantile regression allows for differential effects of cost-sharing on the conditional median and other conditional quantiles of adherence. In quantile regression we regressed specific quantiles (or percentiles) of the dependent variable on patient characteristics.13 Predicting the conditional quantile in quantile regression is different from taking the unconditional distribution and dividing up the dependent variable to run separate regressions; the latter procedure is problematic because it artificially reduces the variation in the dependent variable.

For the regression model di = β0 + β1x1i + β2x2i + ∙ ∙ ∙ βkxki + ei , where d = drug adherence, to model the qth quantile, we defined the function

hi{ if ei −−−  > 0

(1 − q) otherwise

where q is any quantile ranging between 0 and 1.0.

The regression coefficients were chosen to minimize the function:


i = 1


The estimated quantile of the distribution of d, conditional on the values of the predictor variables, was:

Q[di|x1i, x2i ∙ ∙ ∙ xki]=β0 + β1x1i + β2x2i + ∙ ∙ ∙ βkxki

where Q(.) denotes the predicted quantile.

Five simultaneous quantile regression models were estimated setting q equal to 0.1, 0.25, 0.5, 0.75, and 0.9 (representing the 10th, 25th, 50th or median, 75th, and 90th percentiles of adherence). Coefficients are presented along with standard errors bootstrapped using the normal method with 500 repetitions. The coefficient in quantile regression represents how the specific quantile (q) changes with a unit change in the predictor, so the coefficients represent the difference in points of the MPR. Wald tests of the differences between coefficients in the median regression model versus the 0.10 and 0.90 quantile model estimates were conducted. A conservative significance level of <.01 was used in all analyses because of the large number of significance tests conducted. All analyses were conducted with Stata version 9.2 (Stata-Corp LP, College Station, TX).


PDF is available on the last page.


Recommended Reading

No Result Found