- Email: [email protected]
Associations Between Prescription Copayment Levels and β-Blocker Medication Adherence in Commercially Insured Heart Failure Patients 50 Years and Older
Clinical Therapeutics/Volume 33, Number 5, 2011
Associations Between Prescription Copayment Levels and -Blocker Medication Adherence in Commercially Insured Heart Failure Patients 50 Years and Older Mark E. Patterson, PhD, MPH1; Susan J. Blalock, PhD, MPH2; Andrew J. Smith, PharmD, BCPS1; and Michael D. Murray, PharmD, MPH3,4 1
University of Missouri-Kansas City School of Pharmacy, Kansas City, Missouri; 2Eshelman School of Pharmacy, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina; 3Purdue University College of Pharmacy, West Lafayette, Indiana; and 4Regenstrief Institute, Inc, Indianapolis, Indiana ABSTRACT Background: High prescription copayments may create barriers to care, resulting in medication nonadherence. Although many studies have examined these associations in commercially insured patients with chronic disease, few have examined -blocker effects in heart failure patients. Objective: Associations between -blocker prescription copayment levels and medication nonadherence were examined within commercially insured beneficiaries with a diagnosis of heart failure. Methods: Heart failure patients were identified as those with at least 1 inpatient claim or 2 outpatient claims with an associated International Classification of Diagnosis, 9th Edition (ICD-9) code of 428.x, in addition to those with at least 2 -blocker claims. Copayment levels were defined in using $5.00 (USD) interval categories, and adherence was defined using the medication possession ratio (MPR). Ordinary least squares (OLS), fixed effects (FE), and random effect (RE) models were used to estimate associations between copayment level and MPR. Logistic regression was used to estimate the probability of nonadherence (MPR ⬍ 0.80) conditional upon copayment level. Regressions controlled for patient demographics, health status, prior hospitalizations, and concomitant medication use. Results: The highest -blocker copayment level ($26⫹) had an average MPR that was 0.07 (95% CI, – 0.11 to – 0.03), 0.08 (95% CI, – 0.12 to – 0.04), and 0.09 (95% CI, – 0.17 to – 0.02) units lower than -blocker copayment level ($0 to $1) in the OLS, RE, and FE models, respectively. Copayment levels $21– $25 and $26⫹ were significantly associated with an increased risk of medication nonadherence (OR ⫽
608
1.64; 95% CI, 1.1–2.4; and OR ⫽ 2.5; 95%, CI 1.6 – 4, respectively). Conclusions: Commercially insured heart failure patients aged ⱖ50 years who are prescribed higher costing -blockers may have up to an average 9% decrease in annual -blocker medication supply as well as an increased risk of nonadherence (MPR ⬍0.80). Results need to be interpreted with caution given the potential of selection bias due to selective prescribing. Associations between copayment levels and nonadherence need to be further explored given the adverse health consequences of nonadherence to -blockers. (Clin Ther. 2011;33:608 – 616) © 2011 Published by Elsevier HS Journals, Inc. Key words: beta-adrenergic blockers, cost sharing, heart failure, medication adherence
INTRODUCTION Health insurance companies use cost-containment strategies, such as prescription copayments, as a means of decreasing utilization and minimizing health care costs. Previous studies have demonstrated that costsharing reduces utilization and expenditures within commercially insured populations.1-5 Given the ever increasing prevalence and incidence of congestive heart failure (CHF) and the economic burden associated with the disease in conjunction with the adverse clinical consequences of heart failure medication nonadherence, more analyses specific to heart failure medications are needed. Since decreased rates of utilization of Accepted for publication April 26, 2011. doi:10.1016/j.clinthera.2011.04.022 0149-2918/$ - see front matter © 2011 Published by Elsevier HS Journals, Inc.
Volume 33 Number 5
M.E. Patterson et al. essential medications have been associated with increased adverse events6-8 and increased hospitalization,9 it remains important to examine copayment effects within heart failure patients. As such, this study is relevant to both policy makers and clinicians. A few recent studies10-12 have examined associations between adherence and clinical outcomes and copayment and adherence in patients with heart failure. A Canadian retrospective cohort study examining the impact of increased restrictions on medication found that, compared with patients residing in Quebec (a province with a less restrictive plan), patients residing in either Ontario (ON) or British Columbia (BC) (provinces with more restrictive plan), were less likely to be prescribed -blockers (OR ⫽ 0.53; 95% CI, 0.46 – 0.60; and OR ⫽ 0.36; 95% CI, 0.29 – 0.44, for ON and BC, respectively).10 A retrospective cohort study examining associations between CHF medication copayments and utilization among 5259 commercially insured beneficiaries in the United States reported that a $10 copayment (USD) increase was associated with a 1.8% (95% CI, 1.4 –2.2) decrease in the medication possession ratio (MPR), which in turn was associated with an 8.7% (95% CI, 3.8 –13.8) increase in the risk of heart failure-related hospitalizations.11 Another retrospective cohort study examining the associations between CHF medication adherence and utilization and costs among 37,408 Medicaid beneficiaries in the United States revealed that, compared with nonadherent beneficiaries, adherent beneficiaries had 13% fewer hospitalizations (P ⬍ 0.01), and were 3% less likely (P ⬍ 0.01) to have an emergency room visit.12 In summary, results from these studies suggest that higher prescription copayments for heart failure medications are associated with an increased risk of nonadherence as well as hospitalizations and emergency room use. The present study examines the associations between -adrenergic blocker copayment levels and medication adherence levels within commercially insured patients diagnosed with heart failure. We extend prior research by including a more generalized sample of commercially insured heart failure patients residing in the United States, and by comparing ordinary least square (OLS), random effect (RE), and fixed effect (FE) estimators to account for unobserved individual variables that may otherwise result in biased estimates. The breadth of the population analyzed increases the external validity of this study, and the sophisticated model-
May 2011
ing approach provides insight into how statistical methods affect the results and interpretation of such studies. We hypothesized that heart failure patients who are prescribed -blockers with low copayment levels would have higher medication adherence compared with those who had higher copayment levels.
PATIENTS AND METHODS De-identified data originated from a proprietary database containing administrative claims from more than 30 different US health care plans across 8 different census regions. Eligibility, inpatient, outpatient, and prescription utilization information was available for over 38 million unique individuals enrolled between 1998 and 2005. A retrospective cohort design was used to measure the associations between -blocker copayment level and medication adherence. Heart failure patients were identified by requiring at least 1 inpatient or 2 outpatient claims having a primary diagnosis of heart failure (International Classification of Diseases, 9th Edition [ICD-9] code 428.x), and at least 1 angiotensin-converting enzyme (ACE) inhibitor, -blocker, or diuretic claim. This combined disease and medication algorithm is a variation of 1 previously reported by Rector et al13 with a sensitivity of 49% and a specificity of 94% by using at least 2 heart failure claims (ICD-9 code 428.x) in addition to 1 -blocker claim (eg, carvedilol, metoprolol, or bisoprolol) over a 2-year follow-up period in order to identify heart failure cases.13 Although Rector et al found that adding pharmaceuticals indicated for other diseases increased the sensitivity of the algorithm (at the expense of decreased the specificity),13 the current analysis also included ACE inhibitors and diuretics in addition to -blockers in the algorithm since all 3 medications are clinically recommended for heart failure. The cohort was restricted to patients aged ⱖ50 years who had at least 3 years of continuous health insurance and prescription drug coverage. We further required that patients have at least 3 years of continuous -blocker use without significant treatment discontinuation. Treatment discontinuation was defined as a gap in therapy with a length equal to or greater than 3 times the magnitude of the days’ supply. For example, a beneficiary would have been classified as discontinued if a 30-day supply was followed by a 90day gap in therapy. This was an adaptation of an algorithm used by Hamilton et al,14 defining a therapy gap
609
Clinical Therapeutics as equivalent to 2 times the days’ supply.14 Re-defining therapy gaps with longer periods of time ensures that individuals have indeed discontinued therapy. Unique follow-up dates were created for each individual. The beginning of follow-up was defined as the date of the first -blocker refill following the accumulation of either 1 inpatient or 2 outpatient heart failure diagnoses. The end of follow-up was defined as the date of medication discontinuation or the date at which the claims records ended, whichever occurred first. After applying these criteria, 2359 individuals were included in the final analytic dataset, representing a cohort of relatively compliant and continuously enrolled heart failure patients. The original administrative claims assigned a copayment level to each prescription refill in categorical ranges of $5 increments starting at $0 and proceeding upward to $50. In order to facilitate making comparisons across refills with no equivalent day supply amounts, copayments were standardized to 30-day supply amounts. Because the refills only had copayment ranges associated with them and calculations needed to be made with a numeric value, the standardized copayment was calculated by dividing the midpoint of the copayment range by the day supply associated with the medication refill. For example, a 90-day supply of medication originally categorized as an $11 to $15 refill was assigned a value of $13, and then divided by 3 in order to obtain a standardized 30-day copayment amount of $4.30. Person-year level copayment categories were constructed by averaging the standardized 30-day copayment amounts. The MPR was used to capture medication adherence. The MPR captures the percentage of medication supply coverage within a defined interval of time, based upon the day supply of medication and the number of calendar days elapsing between 2 refills. It is computed by dividing the total day supply of medications by the total number of days that elapse before an individual returns to the pharmacy for the next prescription refill. For example, an individual who obtains a refill 30 days after obtaining a 30-day supply of medications would have an MPR of 1 and hence be classified as 100% compliant. In contrast, an individual who obtains a refill 60 days after obtaining a 30-day supply of medications would have an MPR of 0.5 and hence be classified as 50% compliant. This computation as well as other varia-
610
tions have been described and validated using managed care administrative claims data.15 MPRs were constructed at the person-year level by dividing the cumulative days of medication supply by the cumulative number of calendar days over any particular follow-up year. Any MPRs over 1 were truncated to 1 to reflect an assumption that those with an oversupply of medications were still compliant. The Shapiro-Francia test16 was conducted to test for normal distribution of MPR values after this truncation, and MPR was found to be normally distributed. To account for possible nonlinear relationships between copayment levels and adherence, MPR was also defined as a binary variable, above or below an MPR of 0.80, a clinically relevant threshold used in a past study that examined the effects of heart failure medication adherence on hospitalizations.9 Control variables included individual level demographic characteristics, health status, concomitant drug utilization, health plan level characteristics, and either census regions or metropolitan statistical areas. Demographic information included age, gender, and geographical region. Health status variables included baseline comorbidities and heart failure severity. The number of comorbidities at baseline was constructed using the Charlson comorbidity index17 adapted to the most current ICD-9-CM (clinical modification) codes based upon a previous study.18 Prior hospitalizations served as a proxy measure for heart failure severity,9 and was defined as whether an individual had any type of hospitalization in the year prior to baseline. Annual concomitant ACE inhibitor, diuretic, and angiotensinreceptor blocker utilization was measured using binary indicator variables. Because history of adherence is a strong predictor of future adherence, average personlevel MPRs were used to capture -blocker adherence in the previous year. The use of other concomitant medications apart from -blockers was measured using 2 variables: an annual total count of the number of unique concomitant medications and a total annual sum of copayment amounts of concomitant medications. An individual’s health plan was classified as commercial, managed Medicare, or managed Medicaid, as well as health maintenance organization (HMO), preferred provider organization (PPO), pointof-service (POS), or independent plan. Models also included calendar year variables to account for time trend main effects.
Volume 33 Number 5
M.E. Patterson et al. Multivariate linear models were used to estimate the marginal effect of copayment category on medication adherence. In order to explore how bias could potentially affect the associations between copayment levels and medication adherence, these marginal effects were computed using 3 different linear estimators: OLS, RE, and FE models.19 These models have particular advantages over standard linear regression. RE may be preferred over OLS because of the ability of RE estimates to account for repeated refills per person, which would otherwise result in biased OLS estimates because these refills are technically not independent observations. FE may be preferred over both RE and OLS because of the ability of FE estimates to account for unobserved time-invariant factors, such as race, which could be correlated with copayment levels and medication adherence. The Bruesch-Pagan20 and the Hausman21 specification tests were performed to determine the best fit model. Multivariate logistic regression was used to estimate the effect of copayment category on the probability of nonadherence (MPR ⬍0.8). These models used robust errors in order to account for clustering within individuals. All analyses were conducted using SAS version 9 (SAS, Inc, Cary, North Carolina). This research protocol was reviewed and granted a waiver by the Institutional Review Board of University of North Carolina School of Public Health.
RESULTS The sample was predominantly male (59.6%) with an average age of 67 years (Table I). Data indicated most beneficiaries residing in the Northeastern or Mid-Atlantic regions of the United States. But given that 27% of the sample had censored information on census region for the purpose of protecting patient confidentiality, these regional estimates are most likely not representative of the observed sample. More than two thirds (70%) of the sample were enrolled in commercial plans, compared with only 30% and 0.6% for Medicare and Medicaid managed care plans, respectively (Table II). Within these plans, almost half of the patients (47%) were enrolled in an HMO and about 25% were enrolled in a PPO. Approximately 41% of the sample had a Charlson comorbidity index score of either 0 or 1, and about one third (31%) had experienced at least 1 any-cause hospitalization prior to baseline. Average follow-up time per patient was 4.2 years. The mean beta-blocker MPR
May 2011
Table I. Baseline characteristics of commercially insured beneficiaries diagnosed with heart failure regularly using -blockers (N ⫽ 2345). Data are percentage of patients unless otherwise indicated. Characteristic Age, mean (SD), y Male sex US Census region Northeast/Mid-Atlantic South Midwest Pacific/Mountain National*
Value 67 (8.5) 59.6 64.3 2.5 2.9 3.4 27.0
Commercial health plan type Commercial Medicare managed care Medicaid managed care
69.9 29.5 0.6
Managed care organization type HMO Independent PPO POS Other
46.6 21.7 25.1 4.5 2.1
Charlson index17 0 1 2 3 ⬎3
10.4 30.6 24.9 13.8 20.3
Any prior hospitalizations Number of follow-up days, mean (SD)
31.3 1524 (329)
HMO ⫽ health maintenance organization; POS ⫽ pointof-service; PPO ⫽ preferred provider organization. *Missing census information to preserve patient confidentiality.
was 0.98 at baseline (Table II), which reflects an annual average undersupply of 2%. Over 70% had at least some concomitant diuretic use, whereas about 40% and 13% had at least some concomitant use of ACE inhibitors and aldosterone antagonists, respectively. The mean monthly copayment of all medica-
611
Clinical Therapeutics
Table II. Medication use characteristics at baseline of commercially insured beneficiaries diagnosed with heart failure regularly using -blockers (N ⫽ 2345). Data are percentage of patients unless otherwise indicated. Characteristic MPR, mean (SD) Mean 30-day copayment levels per month $0 $1 to $5 $6 to $10 $11 to $15 $16 to $20 $21 to $25 $26 to $30 Any concomitant medication use ACE inhibitors Diuretics Aldosterone antagonists Number of non–-blocker unique medications per month, mean (SD) Other out-of-pocket nonindex drug expenses per month, mean (SD), US $ Monthly -blocker out-of-pocket expenses per month, mean (SD), US $
Value 0.98 (0.21)
14.9 49.4 15.5 11.8 5.3 2.1 1.1 43.8 72.8 13.1 4.9 (2.2)
46.4 (33.4)
5.9 (6.3)
ACE ⫽ angiotensin-converting enzyme; MPR ⫽ medication possession ratio.
tions was $52, of which $6 and $46 were attributed to beta blockers and all other concomitant medications, respectively. OLS models found that, compared with patients prescribed -blockers having a 30-day refill copayment under $1, individuals prescribed -blockers having a 30-day refill copayment of ⱖ$26 had an average MPR 0.07 units lower (Table III). This indicates patients with this highest copayment level compared with the lowest copayment level had on average a 7% less medication supply of -blockers over the course of a year. RE and FE models also found consistent results, reveal-
612
ing average differences of 0.08 and 0.09 MPR units in RE and FE models, respectively (Table III). The Bruesch-Pagan test favored the RE estimator over the OLS estimator, and the Hausman test favored the FE estimator over the RE estimator. Multivariate logistic regressions found that compared with patients prescribed -blockers having a 30-day refill copayment under $1, patients prescribed -blockers having a 30day refill copayment between $21 and $25 and between $26 to $30 had a 1.6 greater risk (OR ⫽ 1.6; 95% CI, 1.1–2.4) and 2.5 greater risk (OR ⫽ 2.5; 95% CI, 1.6 – 4.0), respectively, of medication nonadherence (MPR ⬍0.8) (Table IV).
DISCUSSION Commercially insured heart failure patients aged ⱖ50 years prescribed -blockers at higher levels of copayment may be at higher risk of medication nonadherence, as suggested by the significant associations found between high annual average copayment levels and lower MPR in both the linear and the logistic models. Because specification tests preferred the FE linear models over both the RE and OLS, the most valid conclusions would be drawn from the FE model estimates. These inverse associations between -blocker copayment levels and medication adherence are consistent with a previous study that also found higher -blocker copayments to be associated with decreased MPR.11 In addition, it is important to note that associations between copayment levels and adherence were only statistically significant in the FE models at the highest copayment level ($26⫹) in the linear regressions, and the 2 highest copayment levels ($21 to $25; $26⫹) in the logistic regressions. This implies that -blocker copayments may be raised as high as $20 before adherence levels are affected. These results are encouraging for both clinicians and patients interested in maintaining optimal medication adherence because the copayment effect on adherence may only be present in patients prescribed -blockers with the highest copayment levels. Other noteworthy findings include the differences in parameter estimates between the OLS, RE, and FE models. The fact that the OLS estimate for the highest copayment (ⱖ$26) was 2% points lower than the FE estimate for the highest copayment (ⱖ$26) category, in conjunction with the Bruesch-Pagan and Hausman test favoring the FE model, suggests that unobserved individual-level variables may be underestimating the as-
Volume 33 Number 5
M.E. Patterson et al.
Table III. Associations between annual average 30-day -blocker copayment levels and annual adherence in commercially insured patients with heart failure (N ⫽ 7686 person-years). Copayment Level*
OLS 
95% CI
(n ⫽ 1158)
0.02†
(0.0004 to 0.04)
$6 to $10 (n ⫽ 363) $11 to $15 (n ⫽ 276) $16 to $20 (n ⫽ 124) $21 to $25 (n ⫽ 49) $26⫹ (n ⫽ 25)
0.035‡ 0.01 0.02 –0.03 –0.07‡
RE  0.01
(0.33 to 0.37) 0.03† (–0.03 to 0.05) 0.01 (–0.02 to 0.05) 0.01 (–0.07 to 0.01) –0.03 (–0.11 to –0.03) –0.08‡
95% CI
FE 
95% CI
(–0.01 to 0.03)
–0.01
(–0.03 to 0.01)
(0.01 to 0.05) (–0.01 to 0.03) (–0.03 to 0.05) (–0.07 to 0.01) (–0.12 to –0.04)
0.02 0.02 0.01 –0.02 –0.09†
(–0.02 to 0.06) (–0.02 to 0.06) (–0.03 to 0.05) (–0.08 to 0.04) (–0.17 to –0.02)
FE ⫽ fixed effects estimator; OLS ⫽ ordinary least squares estimator; RE ⫽ random effects estimator. *Reference copayment level ⬍$1 (USD). † P ⬍ 0.05. ‡ P ⬍ 0.01.
sociations between copayment level and -blocker adherence. What is a plausible scenario that could result in this underestimation in the OLS model given the presence of a time-invariant unobserved variable in the error term? If racial status is negatively associated with both copayment level and medication adherence, which may happen if race is associated with lower income, educational level, or resources to assist in medication access, differencing this race effect out using FE
Table IV. Probability of average annual -blocker therapy nonadherence conditional upon average annual -blocker prescription copayment level in commercially insured patients with heart failure (N ⫽ 7686 person-years). Copayment Level* $1 to $5 (n ⫽ 1158) $6 to $10 (n ⫽ 363) $11 to $15 (n ⫽ 276) $16 to $20 (n ⫽ 124) $21 to $25 (n ⫽ 49) $26⫹ (n ⫽ 25)
OR (95% CI)† 1.01 (0.79–1.29) 1.00 (0.75–1.34) 1.00 (0.72–1.37) 1.06 (0.72–1.53) 1.64 (1.12–2.42)‡ 2.54 (1.61–4.00)§
*Reference copayment level ⬍$1 (USD). † Robust standard errors. ‡ P ⬍ 0.05. § P ⬍ 0.01.
May 2011
would result in the FE estimate being more negative compared with the OLS estimate. Because FE estimates have the advantages over both the RE and the OLS models in allowing for correlation between unobserved, time-invariant variables (in this case, race) and explanatory variables and because FE models effectually difference out any unobserved heterogeneity due to time-invariant effects,22 the most appropriate conclusions regarding associations between copayment levels and adherence would be drawn using the FE estimators. This is the first known study examining whether OLS, RE, or FE estimators are more appropriate for measuring associations between -blocker copayment levels and adherence in a sample of commercially insured heart failure patients. The differences found between the models warrants the need for caution in interpreting the parameter estimates or risk ratios between copayment levels and adherence in previous studies. If models used are missing important variables that confound the associations, an underestimation of the associations may result in policymakers or clinicians minimizing the potential risks involved in raising copayment levels or prescribing more expensive -blockers. In addition to comparing associations with different model estimates, this study also adds to the literature by drawing a sample from over 38 million commercially insured individuals from 30 different plans across the United States. This leads to results that are
613
Clinical Therapeutics more generalizable to heart failure patients nationwide compared with prior studies focusing on Medicaid beneficiaries12 or just 1 commercial plan.11 In addition to improved external validity compared with prior studies, this analysis has a longer follow-up of 4 years, which allows sufficient time to detect associations between copayment levels and adherence, if associations were, in fact, present during follow-up. This study has several limitations. First, the results cannot be generalized to either less compliant or noncontinuously enrolled patients because analysis was restricted to continuously enrolled, commercially insured heart failure patients with relatively high baseline medication adherence rates. This selection issue was difficult to avoid, given that capturing prescription copayment levels in this particular database required an individual to have at least 1 prescription claim. Hence, inherent in the design was a requirement that the copayment measure itself was dependent upon a minimum level of -blocker utilization. Despite this limitation, the results can be generalized to more adherent, continuously enrolled beneficiaries diagnosed with heart failure. A second limitation is that associations between copayment levels and adherence may be due to selection factors. As such, it cannot be inferred that higher copayments are causing the lower adherence. Because this was a cross-sectional study and, therefore, limited to measuring associations between copayment levels and adherence as opposed to associations between changes in copayment levels and changes in adherence, it cannot be ruled out that lower medication adherence was due to a lower cost medication requiring several doses per day, causing a provider to prescribe a higher cost medication limited to 1 dose per day in an attempt to improve adherence. This scenario would also create an association between higher copayments and lower adherence, but would reflect a situation in which lower adherence is causing higher copayments. The opposite association, as found in the RE model, may occur if patients in the $6 –$10 copayment group compared with other upper copayment level groups were uncharacteristically healthier as well as more adherent than patients in the $0 copayment group. Interestingly, this association was attenuated in the preferred FE model, again demonstrating the potential of FE in accounting for unobserved heterogeneity. To improve the ability to infer causality, an optimal study design in future studies would measure associations between copay-
614
ment changes and medication adherence changes as a result of a particular policy change. Unfortunately, the data in this study only contained copayment level information at the prescription level and did not include formulary changes at the beneficiary level. As such, it was not possible to eliminate all sources of selection bias. A third limitation is the small sample size contained within the upper-most copayment level may have lowerbound MPR outliers potentially resulting in false-positive associations between copayment and adherence levels. Despite this, we have reason to believe internal validity is still preserved given that all individual 30day refill MPR values were normally distributed before aggregating them to the person-year level measurements used in the linear and logistic regressions. Furthermore, even if outliers did exist, there is reason to believe these would reflect true medication levels and would be appropriate to retain in the analysis. Given this limitation, results should be interpreted with caution, especially since this group of 25 patients in the uppermost group may not be representative of all older heart failure patients who are commercially insured. Despite these methodological limitations, this analysis contributes to the literature by extending recent studies of heart failure medication copayments and adherence to a more diverse sample of commercially insured patients. The fact that significant associations were found in even a more compliant, continuously enrolled commercially insured population suggests that this relationship may be even stronger in a less compliant population. Additionally, this analysis demonstrates how conclusions may be erroneously drawn based on biased estimates if specification tests are not conducted. The differing magnitudes found between the OLS, RE, and FE models imply that adopting the appropriate methodology remains paramount, most especially if pharmaceutical polices are formulated based on results of studies based on administrative data. In this study, the preferred FE estimates are valuable in accounting for potential unobserved time-invariant factors, such as race, which would otherwise result in biased estimates if policymakers relied upon RE or OLS estimates. As such, it must be emphasized that conclusions from this study, as any study, should be drawn on the most unbiased estimators before formulating policies that may impact patients’ health or access to medications or health services.
Volume 33 Number 5
M.E. Patterson et al. Future research in this area should take this into consideration in order to prevent minimizing the risk higher copayments may have on adherence to essential medications. In addition to the significant associations found, the null associations found in lower copayment levels suggests that commercially insured patients are not at risk of nonadherence in moderate copayment levels.
CONCLUSIONS Although high copayment levels of -blocker medications may be creating a barrier to optimal adherence in even relatively compliant commercially insured heart failure patients, low to moderate copayment levels are not affecting -blocker adherence. Although these results should give pause to the clinician prescribing a higher cost medication because a patient may be at more risk of nonadherence, the results should also provide encouragement to the clinician prescribing a low to moderate cost mediation since most -blockers have copayments within this range. In addition to the results differing between copayment levels, the results differing between the linear regressions should give pause to health services researchers examining associations between medication adherence and copayment levels. As demonstrated by the differing results among the linear regressions in conjunction with the specification tests, associations between adherence and copayment may be biased if potential confounders are not accounted for in the analyses.
ACKNOWLEDGMENTS Dr. Patterson’s source of funding and support originated from an American Foundation of Pharmaceutical Education (AFPE) pre-doctoral fellowship 2005– 2006, the year during which most of the analysis and writing was completed. As primary author, Dr. Mark Patterson contributed significantly to the literature review, the study design, the data analysis, interpretations, and conclusions. He also oversaw the revisions and contributions of (1) Dr. Susan Blalock, who served as a dissertation committee member and advised on study design, data interpretation, and manuscript revisions; (2) Dr. Michael Murray, who served as chairman of the dissertation committee and provided overall guidance to study design, medication adherence issues, data interpretation, and writing; and (3) Dr. Andrew Smith, who contributed
May 2011
expertise in clinical outcomes and heart failure pharmacotherapy. No other authors other than those listed contributed specifically to the writing of this manuscript. The authors would like to recognize Dr. Herb Patterson of University of North Carolina Eshelman School of Pharmacy, Dr. Richard Hansen of the Auburn University Harrison School of Pharmacy, and Dr. David Ridley of the Duke University Fuqua School of Business for their significant contributions and input as dissertation committee members. During the writing of this manuscript, Dr. Susan Blalock served as a consultant for Alco, Inc, and Dr. Patterson served as a consultant for GlaxoSmithKline, Inc, and Dr. Michael Murray served in a unpaid capacity as chair of the Safe Medication Use expert panel at the United States Pharmacopeia (USP). The authors have indicated that they have no conflicts of interest regarding the content of this article.
REFERENCES 1. Fairman KA, Motheral BR, 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 Ther. 2003;25: 3147–3161; discussion 3144 –3146. 2. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefits and the use of drugs by the chronically ill. JAMA. 2004;291: 2344 –2350. 3. 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. 4. Huskamp HA, Deverka PA, Epstein AM, et al. The effect of incentive-based formularies on prescription-drug utilization and spending. N Engl J Med. 2003;349:2224 –2232. 5. Joyce GF, Escarce JJ, Solomon MD, Goldman DP. Employer drug benefit plans and spending on prescription drugs. JAMA. 2002;288:1733–1739. 6. Soumerai SB, McLaughlin TJ, Ross-Degnan D, et al. Effects of a limit on Medicaid drug-reimbursement benefits on the use of psychotropic agents and acute mental health services by patients with schizophrenia. N Engl J Med. 1994;331:650 – 655. 7. Soumerai SB, Ross-Degnan D, Avorn J, et al. Effects of Medicaid drug-payment limits on admission to hospitals and nursing homes. N Engl J Med. 1991;325:1072–1077. 8. Tamblyn R, Laprise R, Hanley JA, et al. Adverse events associated with prescription drug cost-sharing among poor and elderly persons. JAMA. 2001;285:421– 429.
615
Clinical Therapeutics 9. Stroupe KT, Teal EY, Weiner M, et al. Health care and medication costs and use among older adults with heart failure. Am J Med. 2004;116: 443– 450. 10. Thanassoulis G, Karp I, Humphries K, et al. Impact of restrictive prescription plans on heart failure medication use. Circ Cardiovasc Qual Outcomes. 2009;2:484 –90. 11. Cole JA, Norman H, Weatherby LB, Walker AM. Drug copayment and adherence in chronic heart failure: effect on cost and outcomes. Pharmacotherapy. 2006;26:1157–1164. 12. Esposito D, Bagchi AD, Verdier JM, et al. Medicaid beneficiaries with congestive heart failure: association of medication adherence with healthcare use and costs. Am J Manag Care. 2009;15:437– 445. 13. Rector TS, Wickstrom SL, Shah M, et al. Specificity and sensitivity of claims-based algorithms for identifying members of Medicare plus choice health plans that have chronic medical conditions. Health Serv Res. 2004; 39:1839 –1857. 14. Hamilton RA, Briceland LL. Use of prescription-refill records to assess patient compliance. Am J Hosp Pharm. 1992;49:1691–1696. 15. Steiner JF, Koepsell TD, Fihn SD, Inui TS. A general method of compliance assessment using centralized pharmacy records: Description and validation. Med Care. 1988;26:814 – 823. 16. Shapiro SS, Francia RS. An approximate analysis of variance test for normality. J Am Stat Assoc. 1972;67: 215–216. 17. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. 18. Sundararajan V, Henderson T, Perry C, et al. New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality. J Clin Epidemiol. 2004;57:1288 –1294.
616
19. Kennedy P. A Guide to Econometrics. 4th ed. Cornwall, United Kingdom: TJ International; 1998. 20. Breusch TS, Pagan AR. A simple test for heteroscedasticity and random coefficient variation. Econometrica. 1979;47:1287–1294.
21. Hausman JA. Specification tests in econometrics. Econometrica. 1978; 46:1251–1271. 22. Wooldridge J. Introductory Econometrics: A Modern Approach. 2nd ed. Boston, Massachusetts: South-Western College Publishing; 2002.
Address correspondence to: Mark E. Patterson, PhD, MPH, Assistant Professor, University of Missouri-Kansas City School of Pharmacy, Division of Pharmacy Practice and Administration, 4245 Health Sciences Building, 2464 Charlotte Street, Kansas City, MO 64108-2718. E-mail: [email protected]
Volume 33 Number 5
Associations Between Prescription Copayment Levels and -Blocker Medication Adherence in Commercially Insured Heart Failure Patients 50 Years and Older Mark E. Patterson, PhD, MPH1; Susan J. Blalock, PhD, MPH2; Andrew J. Smith, PharmD, BCPS1; and Michael D. Murray, PharmD, MPH3,4 1
University of Missouri-Kansas City School of Pharmacy, Kansas City, Missouri; 2Eshelman School of Pharmacy, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina; 3Purdue University College of Pharmacy, West Lafayette, Indiana; and 4Regenstrief Institute, Inc, Indianapolis, Indiana ABSTRACT Background: High prescription copayments may create barriers to care, resulting in medication nonadherence. Although many studies have examined these associations in commercially insured patients with chronic disease, few have examined -blocker effects in heart failure patients. Objective: Associations between -blocker prescription copayment levels and medication nonadherence were examined within commercially insured beneficiaries with a diagnosis of heart failure. Methods: Heart failure patients were identified as those with at least 1 inpatient claim or 2 outpatient claims with an associated International Classification of Diagnosis, 9th Edition (ICD-9) code of 428.x, in addition to those with at least 2 -blocker claims. Copayment levels were defined in using $5.00 (USD) interval categories, and adherence was defined using the medication possession ratio (MPR). Ordinary least squares (OLS), fixed effects (FE), and random effect (RE) models were used to estimate associations between copayment level and MPR. Logistic regression was used to estimate the probability of nonadherence (MPR ⬍ 0.80) conditional upon copayment level. Regressions controlled for patient demographics, health status, prior hospitalizations, and concomitant medication use. Results: The highest -blocker copayment level ($26⫹) had an average MPR that was 0.07 (95% CI, – 0.11 to – 0.03), 0.08 (95% CI, – 0.12 to – 0.04), and 0.09 (95% CI, – 0.17 to – 0.02) units lower than -blocker copayment level ($0 to $1) in the OLS, RE, and FE models, respectively. Copayment levels $21– $25 and $26⫹ were significantly associated with an increased risk of medication nonadherence (OR ⫽
608
1.64; 95% CI, 1.1–2.4; and OR ⫽ 2.5; 95%, CI 1.6 – 4, respectively). Conclusions: Commercially insured heart failure patients aged ⱖ50 years who are prescribed higher costing -blockers may have up to an average 9% decrease in annual -blocker medication supply as well as an increased risk of nonadherence (MPR ⬍0.80). Results need to be interpreted with caution given the potential of selection bias due to selective prescribing. Associations between copayment levels and nonadherence need to be further explored given the adverse health consequences of nonadherence to -blockers. (Clin Ther. 2011;33:608 – 616) © 2011 Published by Elsevier HS Journals, Inc. Key words: beta-adrenergic blockers, cost sharing, heart failure, medication adherence
INTRODUCTION Health insurance companies use cost-containment strategies, such as prescription copayments, as a means of decreasing utilization and minimizing health care costs. Previous studies have demonstrated that costsharing reduces utilization and expenditures within commercially insured populations.1-5 Given the ever increasing prevalence and incidence of congestive heart failure (CHF) and the economic burden associated with the disease in conjunction with the adverse clinical consequences of heart failure medication nonadherence, more analyses specific to heart failure medications are needed. Since decreased rates of utilization of Accepted for publication April 26, 2011. doi:10.1016/j.clinthera.2011.04.022 0149-2918/$ - see front matter © 2011 Published by Elsevier HS Journals, Inc.
Volume 33 Number 5
M.E. Patterson et al. essential medications have been associated with increased adverse events6-8 and increased hospitalization,9 it remains important to examine copayment effects within heart failure patients. As such, this study is relevant to both policy makers and clinicians. A few recent studies10-12 have examined associations between adherence and clinical outcomes and copayment and adherence in patients with heart failure. A Canadian retrospective cohort study examining the impact of increased restrictions on medication found that, compared with patients residing in Quebec (a province with a less restrictive plan), patients residing in either Ontario (ON) or British Columbia (BC) (provinces with more restrictive plan), were less likely to be prescribed -blockers (OR ⫽ 0.53; 95% CI, 0.46 – 0.60; and OR ⫽ 0.36; 95% CI, 0.29 – 0.44, for ON and BC, respectively).10 A retrospective cohort study examining associations between CHF medication copayments and utilization among 5259 commercially insured beneficiaries in the United States reported that a $10 copayment (USD) increase was associated with a 1.8% (95% CI, 1.4 –2.2) decrease in the medication possession ratio (MPR), which in turn was associated with an 8.7% (95% CI, 3.8 –13.8) increase in the risk of heart failure-related hospitalizations.11 Another retrospective cohort study examining the associations between CHF medication adherence and utilization and costs among 37,408 Medicaid beneficiaries in the United States revealed that, compared with nonadherent beneficiaries, adherent beneficiaries had 13% fewer hospitalizations (P ⬍ 0.01), and were 3% less likely (P ⬍ 0.01) to have an emergency room visit.12 In summary, results from these studies suggest that higher prescription copayments for heart failure medications are associated with an increased risk of nonadherence as well as hospitalizations and emergency room use. The present study examines the associations between -adrenergic blocker copayment levels and medication adherence levels within commercially insured patients diagnosed with heart failure. We extend prior research by including a more generalized sample of commercially insured heart failure patients residing in the United States, and by comparing ordinary least square (OLS), random effect (RE), and fixed effect (FE) estimators to account for unobserved individual variables that may otherwise result in biased estimates. The breadth of the population analyzed increases the external validity of this study, and the sophisticated model-
May 2011
ing approach provides insight into how statistical methods affect the results and interpretation of such studies. We hypothesized that heart failure patients who are prescribed -blockers with low copayment levels would have higher medication adherence compared with those who had higher copayment levels.
PATIENTS AND METHODS De-identified data originated from a proprietary database containing administrative claims from more than 30 different US health care plans across 8 different census regions. Eligibility, inpatient, outpatient, and prescription utilization information was available for over 38 million unique individuals enrolled between 1998 and 2005. A retrospective cohort design was used to measure the associations between -blocker copayment level and medication adherence. Heart failure patients were identified by requiring at least 1 inpatient or 2 outpatient claims having a primary diagnosis of heart failure (International Classification of Diseases, 9th Edition [ICD-9] code 428.x), and at least 1 angiotensin-converting enzyme (ACE) inhibitor, -blocker, or diuretic claim. This combined disease and medication algorithm is a variation of 1 previously reported by Rector et al13 with a sensitivity of 49% and a specificity of 94% by using at least 2 heart failure claims (ICD-9 code 428.x) in addition to 1 -blocker claim (eg, carvedilol, metoprolol, or bisoprolol) over a 2-year follow-up period in order to identify heart failure cases.13 Although Rector et al found that adding pharmaceuticals indicated for other diseases increased the sensitivity of the algorithm (at the expense of decreased the specificity),13 the current analysis also included ACE inhibitors and diuretics in addition to -blockers in the algorithm since all 3 medications are clinically recommended for heart failure. The cohort was restricted to patients aged ⱖ50 years who had at least 3 years of continuous health insurance and prescription drug coverage. We further required that patients have at least 3 years of continuous -blocker use without significant treatment discontinuation. Treatment discontinuation was defined as a gap in therapy with a length equal to or greater than 3 times the magnitude of the days’ supply. For example, a beneficiary would have been classified as discontinued if a 30-day supply was followed by a 90day gap in therapy. This was an adaptation of an algorithm used by Hamilton et al,14 defining a therapy gap
609
Clinical Therapeutics as equivalent to 2 times the days’ supply.14 Re-defining therapy gaps with longer periods of time ensures that individuals have indeed discontinued therapy. Unique follow-up dates were created for each individual. The beginning of follow-up was defined as the date of the first -blocker refill following the accumulation of either 1 inpatient or 2 outpatient heart failure diagnoses. The end of follow-up was defined as the date of medication discontinuation or the date at which the claims records ended, whichever occurred first. After applying these criteria, 2359 individuals were included in the final analytic dataset, representing a cohort of relatively compliant and continuously enrolled heart failure patients. The original administrative claims assigned a copayment level to each prescription refill in categorical ranges of $5 increments starting at $0 and proceeding upward to $50. In order to facilitate making comparisons across refills with no equivalent day supply amounts, copayments were standardized to 30-day supply amounts. Because the refills only had copayment ranges associated with them and calculations needed to be made with a numeric value, the standardized copayment was calculated by dividing the midpoint of the copayment range by the day supply associated with the medication refill. For example, a 90-day supply of medication originally categorized as an $11 to $15 refill was assigned a value of $13, and then divided by 3 in order to obtain a standardized 30-day copayment amount of $4.30. Person-year level copayment categories were constructed by averaging the standardized 30-day copayment amounts. The MPR was used to capture medication adherence. The MPR captures the percentage of medication supply coverage within a defined interval of time, based upon the day supply of medication and the number of calendar days elapsing between 2 refills. It is computed by dividing the total day supply of medications by the total number of days that elapse before an individual returns to the pharmacy for the next prescription refill. For example, an individual who obtains a refill 30 days after obtaining a 30-day supply of medications would have an MPR of 1 and hence be classified as 100% compliant. In contrast, an individual who obtains a refill 60 days after obtaining a 30-day supply of medications would have an MPR of 0.5 and hence be classified as 50% compliant. This computation as well as other varia-
610
tions have been described and validated using managed care administrative claims data.15 MPRs were constructed at the person-year level by dividing the cumulative days of medication supply by the cumulative number of calendar days over any particular follow-up year. Any MPRs over 1 were truncated to 1 to reflect an assumption that those with an oversupply of medications were still compliant. The Shapiro-Francia test16 was conducted to test for normal distribution of MPR values after this truncation, and MPR was found to be normally distributed. To account for possible nonlinear relationships between copayment levels and adherence, MPR was also defined as a binary variable, above or below an MPR of 0.80, a clinically relevant threshold used in a past study that examined the effects of heart failure medication adherence on hospitalizations.9 Control variables included individual level demographic characteristics, health status, concomitant drug utilization, health plan level characteristics, and either census regions or metropolitan statistical areas. Demographic information included age, gender, and geographical region. Health status variables included baseline comorbidities and heart failure severity. The number of comorbidities at baseline was constructed using the Charlson comorbidity index17 adapted to the most current ICD-9-CM (clinical modification) codes based upon a previous study.18 Prior hospitalizations served as a proxy measure for heart failure severity,9 and was defined as whether an individual had any type of hospitalization in the year prior to baseline. Annual concomitant ACE inhibitor, diuretic, and angiotensinreceptor blocker utilization was measured using binary indicator variables. Because history of adherence is a strong predictor of future adherence, average personlevel MPRs were used to capture -blocker adherence in the previous year. The use of other concomitant medications apart from -blockers was measured using 2 variables: an annual total count of the number of unique concomitant medications and a total annual sum of copayment amounts of concomitant medications. An individual’s health plan was classified as commercial, managed Medicare, or managed Medicaid, as well as health maintenance organization (HMO), preferred provider organization (PPO), pointof-service (POS), or independent plan. Models also included calendar year variables to account for time trend main effects.
Volume 33 Number 5
M.E. Patterson et al. Multivariate linear models were used to estimate the marginal effect of copayment category on medication adherence. In order to explore how bias could potentially affect the associations between copayment levels and medication adherence, these marginal effects were computed using 3 different linear estimators: OLS, RE, and FE models.19 These models have particular advantages over standard linear regression. RE may be preferred over OLS because of the ability of RE estimates to account for repeated refills per person, which would otherwise result in biased OLS estimates because these refills are technically not independent observations. FE may be preferred over both RE and OLS because of the ability of FE estimates to account for unobserved time-invariant factors, such as race, which could be correlated with copayment levels and medication adherence. The Bruesch-Pagan20 and the Hausman21 specification tests were performed to determine the best fit model. Multivariate logistic regression was used to estimate the effect of copayment category on the probability of nonadherence (MPR ⬍0.8). These models used robust errors in order to account for clustering within individuals. All analyses were conducted using SAS version 9 (SAS, Inc, Cary, North Carolina). This research protocol was reviewed and granted a waiver by the Institutional Review Board of University of North Carolina School of Public Health.
RESULTS The sample was predominantly male (59.6%) with an average age of 67 years (Table I). Data indicated most beneficiaries residing in the Northeastern or Mid-Atlantic regions of the United States. But given that 27% of the sample had censored information on census region for the purpose of protecting patient confidentiality, these regional estimates are most likely not representative of the observed sample. More than two thirds (70%) of the sample were enrolled in commercial plans, compared with only 30% and 0.6% for Medicare and Medicaid managed care plans, respectively (Table II). Within these plans, almost half of the patients (47%) were enrolled in an HMO and about 25% were enrolled in a PPO. Approximately 41% of the sample had a Charlson comorbidity index score of either 0 or 1, and about one third (31%) had experienced at least 1 any-cause hospitalization prior to baseline. Average follow-up time per patient was 4.2 years. The mean beta-blocker MPR
May 2011
Table I. Baseline characteristics of commercially insured beneficiaries diagnosed with heart failure regularly using -blockers (N ⫽ 2345). Data are percentage of patients unless otherwise indicated. Characteristic Age, mean (SD), y Male sex US Census region Northeast/Mid-Atlantic South Midwest Pacific/Mountain National*
Value 67 (8.5) 59.6 64.3 2.5 2.9 3.4 27.0
Commercial health plan type Commercial Medicare managed care Medicaid managed care
69.9 29.5 0.6
Managed care organization type HMO Independent PPO POS Other
46.6 21.7 25.1 4.5 2.1
Charlson index17 0 1 2 3 ⬎3
10.4 30.6 24.9 13.8 20.3
Any prior hospitalizations Number of follow-up days, mean (SD)
31.3 1524 (329)
HMO ⫽ health maintenance organization; POS ⫽ pointof-service; PPO ⫽ preferred provider organization. *Missing census information to preserve patient confidentiality.
was 0.98 at baseline (Table II), which reflects an annual average undersupply of 2%. Over 70% had at least some concomitant diuretic use, whereas about 40% and 13% had at least some concomitant use of ACE inhibitors and aldosterone antagonists, respectively. The mean monthly copayment of all medica-
611
Clinical Therapeutics
Table II. Medication use characteristics at baseline of commercially insured beneficiaries diagnosed with heart failure regularly using -blockers (N ⫽ 2345). Data are percentage of patients unless otherwise indicated. Characteristic MPR, mean (SD) Mean 30-day copayment levels per month $0 $1 to $5 $6 to $10 $11 to $15 $16 to $20 $21 to $25 $26 to $30 Any concomitant medication use ACE inhibitors Diuretics Aldosterone antagonists Number of non–-blocker unique medications per month, mean (SD) Other out-of-pocket nonindex drug expenses per month, mean (SD), US $ Monthly -blocker out-of-pocket expenses per month, mean (SD), US $
Value 0.98 (0.21)
14.9 49.4 15.5 11.8 5.3 2.1 1.1 43.8 72.8 13.1 4.9 (2.2)
46.4 (33.4)
5.9 (6.3)
ACE ⫽ angiotensin-converting enzyme; MPR ⫽ medication possession ratio.
tions was $52, of which $6 and $46 were attributed to beta blockers and all other concomitant medications, respectively. OLS models found that, compared with patients prescribed -blockers having a 30-day refill copayment under $1, individuals prescribed -blockers having a 30-day refill copayment of ⱖ$26 had an average MPR 0.07 units lower (Table III). This indicates patients with this highest copayment level compared with the lowest copayment level had on average a 7% less medication supply of -blockers over the course of a year. RE and FE models also found consistent results, reveal-
612
ing average differences of 0.08 and 0.09 MPR units in RE and FE models, respectively (Table III). The Bruesch-Pagan test favored the RE estimator over the OLS estimator, and the Hausman test favored the FE estimator over the RE estimator. Multivariate logistic regressions found that compared with patients prescribed -blockers having a 30-day refill copayment under $1, patients prescribed -blockers having a 30day refill copayment between $21 and $25 and between $26 to $30 had a 1.6 greater risk (OR ⫽ 1.6; 95% CI, 1.1–2.4) and 2.5 greater risk (OR ⫽ 2.5; 95% CI, 1.6 – 4.0), respectively, of medication nonadherence (MPR ⬍0.8) (Table IV).
DISCUSSION Commercially insured heart failure patients aged ⱖ50 years prescribed -blockers at higher levels of copayment may be at higher risk of medication nonadherence, as suggested by the significant associations found between high annual average copayment levels and lower MPR in both the linear and the logistic models. Because specification tests preferred the FE linear models over both the RE and OLS, the most valid conclusions would be drawn from the FE model estimates. These inverse associations between -blocker copayment levels and medication adherence are consistent with a previous study that also found higher -blocker copayments to be associated with decreased MPR.11 In addition, it is important to note that associations between copayment levels and adherence were only statistically significant in the FE models at the highest copayment level ($26⫹) in the linear regressions, and the 2 highest copayment levels ($21 to $25; $26⫹) in the logistic regressions. This implies that -blocker copayments may be raised as high as $20 before adherence levels are affected. These results are encouraging for both clinicians and patients interested in maintaining optimal medication adherence because the copayment effect on adherence may only be present in patients prescribed -blockers with the highest copayment levels. Other noteworthy findings include the differences in parameter estimates between the OLS, RE, and FE models. The fact that the OLS estimate for the highest copayment (ⱖ$26) was 2% points lower than the FE estimate for the highest copayment (ⱖ$26) category, in conjunction with the Bruesch-Pagan and Hausman test favoring the FE model, suggests that unobserved individual-level variables may be underestimating the as-
Volume 33 Number 5
M.E. Patterson et al.
Table III. Associations between annual average 30-day -blocker copayment levels and annual adherence in commercially insured patients with heart failure (N ⫽ 7686 person-years). Copayment Level*
OLS 
95% CI
(n ⫽ 1158)
0.02†
(0.0004 to 0.04)
$6 to $10 (n ⫽ 363) $11 to $15 (n ⫽ 276) $16 to $20 (n ⫽ 124) $21 to $25 (n ⫽ 49) $26⫹ (n ⫽ 25)
0.035‡ 0.01 0.02 –0.03 –0.07‡
RE  0.01
(0.33 to 0.37) 0.03† (–0.03 to 0.05) 0.01 (–0.02 to 0.05) 0.01 (–0.07 to 0.01) –0.03 (–0.11 to –0.03) –0.08‡
95% CI
FE 
95% CI
(–0.01 to 0.03)
–0.01
(–0.03 to 0.01)
(0.01 to 0.05) (–0.01 to 0.03) (–0.03 to 0.05) (–0.07 to 0.01) (–0.12 to –0.04)
0.02 0.02 0.01 –0.02 –0.09†
(–0.02 to 0.06) (–0.02 to 0.06) (–0.03 to 0.05) (–0.08 to 0.04) (–0.17 to –0.02)
FE ⫽ fixed effects estimator; OLS ⫽ ordinary least squares estimator; RE ⫽ random effects estimator. *Reference copayment level ⬍$1 (USD). † P ⬍ 0.05. ‡ P ⬍ 0.01.
sociations between copayment level and -blocker adherence. What is a plausible scenario that could result in this underestimation in the OLS model given the presence of a time-invariant unobserved variable in the error term? If racial status is negatively associated with both copayment level and medication adherence, which may happen if race is associated with lower income, educational level, or resources to assist in medication access, differencing this race effect out using FE
Table IV. Probability of average annual -blocker therapy nonadherence conditional upon average annual -blocker prescription copayment level in commercially insured patients with heart failure (N ⫽ 7686 person-years). Copayment Level* $1 to $5 (n ⫽ 1158) $6 to $10 (n ⫽ 363) $11 to $15 (n ⫽ 276) $16 to $20 (n ⫽ 124) $21 to $25 (n ⫽ 49) $26⫹ (n ⫽ 25)
OR (95% CI)† 1.01 (0.79–1.29) 1.00 (0.75–1.34) 1.00 (0.72–1.37) 1.06 (0.72–1.53) 1.64 (1.12–2.42)‡ 2.54 (1.61–4.00)§
*Reference copayment level ⬍$1 (USD). † Robust standard errors. ‡ P ⬍ 0.05. § P ⬍ 0.01.
May 2011
would result in the FE estimate being more negative compared with the OLS estimate. Because FE estimates have the advantages over both the RE and the OLS models in allowing for correlation between unobserved, time-invariant variables (in this case, race) and explanatory variables and because FE models effectually difference out any unobserved heterogeneity due to time-invariant effects,22 the most appropriate conclusions regarding associations between copayment levels and adherence would be drawn using the FE estimators. This is the first known study examining whether OLS, RE, or FE estimators are more appropriate for measuring associations between -blocker copayment levels and adherence in a sample of commercially insured heart failure patients. The differences found between the models warrants the need for caution in interpreting the parameter estimates or risk ratios between copayment levels and adherence in previous studies. If models used are missing important variables that confound the associations, an underestimation of the associations may result in policymakers or clinicians minimizing the potential risks involved in raising copayment levels or prescribing more expensive -blockers. In addition to comparing associations with different model estimates, this study also adds to the literature by drawing a sample from over 38 million commercially insured individuals from 30 different plans across the United States. This leads to results that are
613
Clinical Therapeutics more generalizable to heart failure patients nationwide compared with prior studies focusing on Medicaid beneficiaries12 or just 1 commercial plan.11 In addition to improved external validity compared with prior studies, this analysis has a longer follow-up of 4 years, which allows sufficient time to detect associations between copayment levels and adherence, if associations were, in fact, present during follow-up. This study has several limitations. First, the results cannot be generalized to either less compliant or noncontinuously enrolled patients because analysis was restricted to continuously enrolled, commercially insured heart failure patients with relatively high baseline medication adherence rates. This selection issue was difficult to avoid, given that capturing prescription copayment levels in this particular database required an individual to have at least 1 prescription claim. Hence, inherent in the design was a requirement that the copayment measure itself was dependent upon a minimum level of -blocker utilization. Despite this limitation, the results can be generalized to more adherent, continuously enrolled beneficiaries diagnosed with heart failure. A second limitation is that associations between copayment levels and adherence may be due to selection factors. As such, it cannot be inferred that higher copayments are causing the lower adherence. Because this was a cross-sectional study and, therefore, limited to measuring associations between copayment levels and adherence as opposed to associations between changes in copayment levels and changes in adherence, it cannot be ruled out that lower medication adherence was due to a lower cost medication requiring several doses per day, causing a provider to prescribe a higher cost medication limited to 1 dose per day in an attempt to improve adherence. This scenario would also create an association between higher copayments and lower adherence, but would reflect a situation in which lower adherence is causing higher copayments. The opposite association, as found in the RE model, may occur if patients in the $6 –$10 copayment group compared with other upper copayment level groups were uncharacteristically healthier as well as more adherent than patients in the $0 copayment group. Interestingly, this association was attenuated in the preferred FE model, again demonstrating the potential of FE in accounting for unobserved heterogeneity. To improve the ability to infer causality, an optimal study design in future studies would measure associations between copay-
614
ment changes and medication adherence changes as a result of a particular policy change. Unfortunately, the data in this study only contained copayment level information at the prescription level and did not include formulary changes at the beneficiary level. As such, it was not possible to eliminate all sources of selection bias. A third limitation is the small sample size contained within the upper-most copayment level may have lowerbound MPR outliers potentially resulting in false-positive associations between copayment and adherence levels. Despite this, we have reason to believe internal validity is still preserved given that all individual 30day refill MPR values were normally distributed before aggregating them to the person-year level measurements used in the linear and logistic regressions. Furthermore, even if outliers did exist, there is reason to believe these would reflect true medication levels and would be appropriate to retain in the analysis. Given this limitation, results should be interpreted with caution, especially since this group of 25 patients in the uppermost group may not be representative of all older heart failure patients who are commercially insured. Despite these methodological limitations, this analysis contributes to the literature by extending recent studies of heart failure medication copayments and adherence to a more diverse sample of commercially insured patients. The fact that significant associations were found in even a more compliant, continuously enrolled commercially insured population suggests that this relationship may be even stronger in a less compliant population. Additionally, this analysis demonstrates how conclusions may be erroneously drawn based on biased estimates if specification tests are not conducted. The differing magnitudes found between the OLS, RE, and FE models imply that adopting the appropriate methodology remains paramount, most especially if pharmaceutical polices are formulated based on results of studies based on administrative data. In this study, the preferred FE estimates are valuable in accounting for potential unobserved time-invariant factors, such as race, which would otherwise result in biased estimates if policymakers relied upon RE or OLS estimates. As such, it must be emphasized that conclusions from this study, as any study, should be drawn on the most unbiased estimators before formulating policies that may impact patients’ health or access to medications or health services.
Volume 33 Number 5
M.E. Patterson et al. Future research in this area should take this into consideration in order to prevent minimizing the risk higher copayments may have on adherence to essential medications. In addition to the significant associations found, the null associations found in lower copayment levels suggests that commercially insured patients are not at risk of nonadherence in moderate copayment levels.
CONCLUSIONS Although high copayment levels of -blocker medications may be creating a barrier to optimal adherence in even relatively compliant commercially insured heart failure patients, low to moderate copayment levels are not affecting -blocker adherence. Although these results should give pause to the clinician prescribing a higher cost medication because a patient may be at more risk of nonadherence, the results should also provide encouragement to the clinician prescribing a low to moderate cost mediation since most -blockers have copayments within this range. In addition to the results differing between copayment levels, the results differing between the linear regressions should give pause to health services researchers examining associations between medication adherence and copayment levels. As demonstrated by the differing results among the linear regressions in conjunction with the specification tests, associations between adherence and copayment may be biased if potential confounders are not accounted for in the analyses.
ACKNOWLEDGMENTS Dr. Patterson’s source of funding and support originated from an American Foundation of Pharmaceutical Education (AFPE) pre-doctoral fellowship 2005– 2006, the year during which most of the analysis and writing was completed. As primary author, Dr. Mark Patterson contributed significantly to the literature review, the study design, the data analysis, interpretations, and conclusions. He also oversaw the revisions and contributions of (1) Dr. Susan Blalock, who served as a dissertation committee member and advised on study design, data interpretation, and manuscript revisions; (2) Dr. Michael Murray, who served as chairman of the dissertation committee and provided overall guidance to study design, medication adherence issues, data interpretation, and writing; and (3) Dr. Andrew Smith, who contributed
May 2011
expertise in clinical outcomes and heart failure pharmacotherapy. No other authors other than those listed contributed specifically to the writing of this manuscript. The authors would like to recognize Dr. Herb Patterson of University of North Carolina Eshelman School of Pharmacy, Dr. Richard Hansen of the Auburn University Harrison School of Pharmacy, and Dr. David Ridley of the Duke University Fuqua School of Business for their significant contributions and input as dissertation committee members. During the writing of this manuscript, Dr. Susan Blalock served as a consultant for Alco, Inc, and Dr. Patterson served as a consultant for GlaxoSmithKline, Inc, and Dr. Michael Murray served in a unpaid capacity as chair of the Safe Medication Use expert panel at the United States Pharmacopeia (USP). The authors have indicated that they have no conflicts of interest regarding the content of this article.
REFERENCES 1. Fairman KA, Motheral BR, 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 Ther. 2003;25: 3147–3161; discussion 3144 –3146. 2. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefits and the use of drugs by the chronically ill. JAMA. 2004;291: 2344 –2350. 3. 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. 4. Huskamp HA, Deverka PA, Epstein AM, et al. The effect of incentive-based formularies on prescription-drug utilization and spending. N Engl J Med. 2003;349:2224 –2232. 5. Joyce GF, Escarce JJ, Solomon MD, Goldman DP. Employer drug benefit plans and spending on prescription drugs. JAMA. 2002;288:1733–1739. 6. Soumerai SB, McLaughlin TJ, Ross-Degnan D, et al. Effects of a limit on Medicaid drug-reimbursement benefits on the use of psychotropic agents and acute mental health services by patients with schizophrenia. N Engl J Med. 1994;331:650 – 655. 7. Soumerai SB, Ross-Degnan D, Avorn J, et al. Effects of Medicaid drug-payment limits on admission to hospitals and nursing homes. N Engl J Med. 1991;325:1072–1077. 8. Tamblyn R, Laprise R, Hanley JA, et al. Adverse events associated with prescription drug cost-sharing among poor and elderly persons. JAMA. 2001;285:421– 429.
615
Clinical Therapeutics 9. Stroupe KT, Teal EY, Weiner M, et al. Health care and medication costs and use among older adults with heart failure. Am J Med. 2004;116: 443– 450. 10. Thanassoulis G, Karp I, Humphries K, et al. Impact of restrictive prescription plans on heart failure medication use. Circ Cardiovasc Qual Outcomes. 2009;2:484 –90. 11. Cole JA, Norman H, Weatherby LB, Walker AM. Drug copayment and adherence in chronic heart failure: effect on cost and outcomes. Pharmacotherapy. 2006;26:1157–1164. 12. Esposito D, Bagchi AD, Verdier JM, et al. Medicaid beneficiaries with congestive heart failure: association of medication adherence with healthcare use and costs. Am J Manag Care. 2009;15:437– 445. 13. Rector TS, Wickstrom SL, Shah M, et al. Specificity and sensitivity of claims-based algorithms for identifying members of Medicare plus choice health plans that have chronic medical conditions. Health Serv Res. 2004; 39:1839 –1857. 14. Hamilton RA, Briceland LL. Use of prescription-refill records to assess patient compliance. Am J Hosp Pharm. 1992;49:1691–1696. 15. Steiner JF, Koepsell TD, Fihn SD, Inui TS. A general method of compliance assessment using centralized pharmacy records: Description and validation. Med Care. 1988;26:814 – 823. 16. Shapiro SS, Francia RS. An approximate analysis of variance test for normality. J Am Stat Assoc. 1972;67: 215–216. 17. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. 18. Sundararajan V, Henderson T, Perry C, et al. New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality. J Clin Epidemiol. 2004;57:1288 –1294.
616
19. Kennedy P. A Guide to Econometrics. 4th ed. Cornwall, United Kingdom: TJ International; 1998. 20. Breusch TS, Pagan AR. A simple test for heteroscedasticity and random coefficient variation. Econometrica. 1979;47:1287–1294.
21. Hausman JA. Specification tests in econometrics. Econometrica. 1978; 46:1251–1271. 22. Wooldridge J. Introductory Econometrics: A Modern Approach. 2nd ed. Boston, Massachusetts: South-Western College Publishing; 2002.
Address correspondence to: Mark E. Patterson, PhD, MPH, Assistant Professor, University of Missouri-Kansas City School of Pharmacy, Division of Pharmacy Practice and Administration, 4245 Health Sciences Building, 2464 Charlotte Street, Kansas City, MO 64108-2718. E-mail: [email protected]
Volume 33 Number 5