https://www.ajmc.com/journals/issue/2009/2009-07-vol15-n7/ajmc_09jul_esposito_437to445/
Medicaid Beneficiaries With Congestive Heart Failure: Association of Medication Adherence With Healthcare Use and Costs

Dominick Esposito, PhD; Ann D. Bagchi, PhD; James M. Verdier, JD; Deo S. Bencio, BS; and Myoung S. Kim, PhD

Congestive heart failure (CHF) is a leading cause of hospitalization and mortality in the United States, affecting more than 5 million people at an expected cost of $34.8 billion in 2008.1 The Centers for Medicare & Medicaid Services (CMS) has prioritized improved treatment of CHF, among other chronic conditions, through demonstrations and pilot programs for its beneficiaries.2-4 The prevalence of CHF is as high as 2.6% among Medicaid beneficiaries and 10.7% among those dually enrolled in Medicare and Medicaid (dual eligibles).5 Patients with CHF account for a disproportionate share of CMS spending. In 1999, 14% of fee-for-service Medicare beneficiaries with CHF accounted for 43% of total spending.2

Patients with CHF are generally at increased risk for heart attack, stroke, emergency department (ED) visits, hospitalization, and death.6-8 To minimize their risk, most patients with CHF should use 1 or more drugs from different therapeutic subclasses, including loop diuretics, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers, and β-blockers.9-11 However, medication nonadherence is common among patients with CHF, and Medicaid beneficiaries’ drug use is often inconsistent with practice guidelines.12-17

Despite evidence that poor adherence leads to higher hospitalization rates, few studies18-20 have examined the relationship between adherence and healthcare costs for patients with CHF, although hospitalization accounts for their highest share of expenditures. If higher CHF drug adherence is associated with lower hospitalization risk, it stands to reason that it is also associated with lower healthcare costs.

This study had 3 primary objectives. The first objective was to examine the association of CHF medication adherence with healthcare use and costs in a Medicaid population. The second objective was to investigate whether the association between drug adherence and outcomes was a graded one. Throughout the literature, the primary threshold used to represent adherent behavior is a medication possession ratio (MPR) of 80%, but we hypothesized that the relationship was more likely graded. The third objective was to estimate the potential savings to Medicaid based on any findings that suggested an association between CHF medication adherence and healthcare costs.

METHODS
Data and Sample Selection
This study used medical and pharmacy claims data from the 1998 State Medicaid Research Files, the 1999 Medicaid Analytic eXtract, and the 1999 Medicare Standard Analytic File for Medicaid beneficiaries residing in Arkansas, California, Indiana, and New Jersey. We selected these states because they are geographically and demographically diverse and because they had limited or no capitated managed care for the disabled and older Medicaid population in 1998 and 1999. In addition, CHF drug utilization differences across these states were not due to differences in copayments or benefit designs.17

The research sample included noninstitutionalized beneficiaries with at least 1 CHF drug claim in 1999, medical claims for CHF, and continuous enrollment in fee-for-service Medicaid with pharmacy benefit coverage. The CHF medications were identified using First DataBank’s Master Drug Data Base21 therapeutic classification system and included the following drug groups: antianginals, β-blockers, calcium channel blockers, antiarrhythmics, antihypertensives, and diuretics. Beneficiaries were identified as having CHF if they were hospitalized with a CHF diagnosis in 1998 or had at least 2 ambulatory visits in 1998 with a CHF diagnosis (International Classification of Diseases, Ninth Revision, Clinical Modification codes 402.xx, 404.xx, and 428.x).

Medication Adherence
We used the MPR to measure CHF medication adherence in 1999.22 Using all CHF drug claims, the MPR was calculated by dividing a patient’s total days’ supply of medication by the number of days between the date of the patient’s first fill and the last day on which the patient had medication available. Days during which a patient stayed in a hospital are excluded from the calculation, and days for which more than 1 CHF drugs were available are counted only once. Using multiple CHF drug subclasses to examine adherence is more lenient than focusing on 1 subclass and is appropriate for a Medicaid population, as research indicates considerable underutilization of CHF drugs from any single subclass.15,17

This study considers multiple MPR specifications. First, the threshold of 80% is used to deem patients as adherent (≥80%) or as nonadherent (<80%). This value is borrowed from established literature on cardiovascular disease and from previous adherence research.6,23-31 However, there is no clinical evidence to support using this ratio or any other value as the threshold for medication adherence. Second, in a sensitivity analysis we also specified the MPR as a continuous variable. Third, we specified the MPR as 4 different ordinal variables. The first ordinal variable has 3 levels, segmenting patients with an MPR of 95% or higher from patients with an MPR between 80% and less than 95% and from patients with an MPR below 80%. We also specified 3 different 5-level ordinal variables. First, we examined the MPR by adherence quintile (eg, the first quintile is 0%-20%, and the last quintile is 80%-100%). Second, we segmented the MPR by quintile such that roughly 20% of the sample fell into each group. Third, we specified an ordinal variable to examine adherence for patients with a nearperfect MPR (with each subgroup containing ≥10% of the sample). The 5 levels were 99% or higher, 95% to less than 99%, 80% to less than 95%, 50% to less than 80%, and less than 50%.

Outcome Variables and Regression Analyses
We examined healthcare costs and utilization in 1999. Cost outcomes included total healthcare (including and excluding drug costs) and drug, inpatient, outpatient, and other medical costs (skilled nursing facility, hospice, ED, and durable medical equipment). Utilization included any hospital use, the number of hospital admissions, the number of hospital days, any ED use, and the number of ED visits. Regression analyses examined the association in 1999 between CHF drug adherence and outcomes.

The distribution of costs dictated regression specifications for models in which costs were the dependent variables. For cost data with only nonzero values (total costs, including drug costs), we estimated a generalized linear model (GLM). For skewed data with many zero values, we used a 2-stage procedure.32,33 We first estimated a logistic regression to model the likelihood of having a nonzero cost and then estimated costs with a GLM, multiplying cost estimates by the predicted probability of having nonzero costs to obtain final cost estimates. For all GLM equations, we used the modified Park test to determine the appropriate link function.33 We estimated costs through the method of recycled predictions, setting all sample members as adherent or as nonadherent, while keeping all other individual characteristics constant. For models in which adherence was specified as a 3-level or 5-level variable, we estimated costs for each of the 3 to 5 subgroups separately.

We estimated logit models for hospital admissions and ED visits and least squares regressions for the number of hospitalizations, the number of hospital days, and the number of ED visits. All utilization outcomes were estimated through recycled predictions. We estimated all regressions using commercially available statistical software (STATA, release 9; StataCorp LP, College Station, TX).34

Independent Variables
The independent variable of interest was the MPR. Regression analyses also included demographic characteristics, health risk factors, and CHF comorbidities.  demographic characteristics included a dual-eligible indicator, age (≤64, 65-74, 75-84, and ≥85 years), sex, state of residence, and race/ethnicity (white, African American, or other). Health risk factors and comorbidities included whether the beneficiary also had diagnosed coronary artery disease or diabetes mellitus and whether the beneficiary had any hospitalizations related or unrelated to CHF during 1998, as well as a diagnostic risk adjustor based on the Chronic Illness and Disability Payment System35 using 1998 medical claims data.

Potential Savings to Medicare
To estimate the potential savings to Medicare from higher CHF medication adherence, we extrapolated study findings on an aggregate level. This was based on published estimates of the number of beneficiaries with CHF and on assumptions about their mean medication adherence.36

RESULTS
In the 4 study states, 37,408 of Medicaid beneficiaries met the inclusion criteria (Table 1). About 36% were younger than 65 years, and 15% were 85 years or older. Slightly more than half were white, and roughly a quarter were African American. Almost three-fourths were female, 72% were dual eligibles, and about half were classified as disabled (and about half of these were also dual eligibles [data not shown]). Many beneficiaries had medical claims for other cardiovascular conditions in addition to CHF, including coronary artery disease (29%) and diabetes (30%). In 1998, 38% of the sample had a hospitalization for CHF, and 38% had a hospitalization for other conditions. In 1999, beneficiaries averaged 1.2 CHF drug claims per month. The most common CHF drug subclasses in the sample were diuretics (59% of patients), ACE inhibitors (45%), and antianginals (35%) (data not shown). Among adherent sample members as specified by an MPR threshold of 80%, the demographic profile was similar to that of the entire research sample, but the number of drug fills was higher.

Hospital and ED Use
Hospital and ED outcomes were always lower for adherent beneficiaries compared with nonadherent beneficiaries, and all differences were significant at P = .01 (Table 2). Adherent beneficiaries were less likely to have a hospitalization (0.4 percentage points), had fewer hospitalizations per beneficiary (13%), had in excess of 2 fewer days spent in the hospital (25%), were less likely to have an ED visit (3%), and had fewer ED visits per beneficiary (10%). When medication adherence was specified as an ordinal variable (at 3 or 5 levels), all healthcare utilization outcomes were generally least likely or lowest for beneficiaries with the highest MPR (data not shown).

Healthcare Costs
Except for total drug costs, healthcare costs were lower for adherent beneficiaries than for nonadherent beneficiaries (P <.01 for most comparisons) (Table 2). Total healthcare costs (including drug costs) were $5910 (23%) less per year. When the MPR was specified with 3 or more levels, the relationship between adherence and healthcare costs was graded (Table 3). For example, beneficiaries with adherence rates of 95% or higher had about 15% lower total healthcare costs, including drug costs, than those with adherence rates between 80% and less than 95% ($17,665 vs $20,747, P <.01). The same pattern was evident when the sample was split into quintiles by adherence level (20% intervals of the MPR) or by sample size (20% of the sample in each quintile). A specification by decile was also used, but the data are not shown.

The relationship between medication adherence rates and total healthcare costs was stronger when the most adherent beneficiaries were segmented into finer subgroups (Table 3). Beneficiaries with adherence rates of 99% or higher (nearperfect adherence) had 6% lower total healthcare costs, including drug costs, than patients with adherence rates between 95% and less than 99% ($16,989 vs $18,141, P <.01). The association of CHF medication adherence with costs was higher in absolute US dollars for dual-eligible beneficiaries adherent patients had annual costs (Table 4). Among dualeligible beneficiaries, adherent patients had annual costs (including drug costs) that were $7913 lower than annual costs of nonadherent patients, or 24% of the nonadherent mean (P <.01). However, the difference between adherent and nonadherent beneficiaries among non–dual-eligible beneficiaries was only $2859 or 19% of the nonadherent mean (P <.01).

Potential Savings to Medicare From Improved Adherence In 2002, approximately 13% of community-dwelling Medicare beneficiaries had CHF, and their mean healthcare costs were about $24,000.36 Because more than 90% of Medicare enrollees reside in the community and the total number of enrollees in 2002 was about 40 million, roughly 5 million community-dwelling beneficiaries had CHF. Based on the association between medication adherence and healthcare costs for dual-eligible beneficiaries in this study, we estimated total costs to Medicare assuming that a fixed proportion of enrollees were adherent (≥80% MPR).

Because the mean annual healthcare costs for nonadherent dual eligibles were 23% higher than those for adherent dual eligibles, we estimated that the mean annual healthcare costs among nonadherent beneficiaries were $28,374 compared with $21,750 for adherent beneficiaries. If 60% of enrollees with CHF were adherent and that percentage rose to 80%, Medicare costs would be $6.6 billion lower, or about 2% of total Medicare spending. This estimate is sensitive to the initial proportion of beneficiaries who are presumed to be adherent. If 65% are adherent, then savings are about $5 billion. Moreover, these savings assume that Medicare could achieve higher mean patient adherence at little or no cost. However, because we had only 1 year of data, it is impossible to estimate the effect of persistent medication adherence from one year to the next.

Sensitivity Analyses
We conducted sensitivity analyses by specific drug subclass and by the number of distinct subclasses filled by beneficiaries. First, we specified the MPR as a continuous variable and estimated costs at various MPR levels (50%, 75%, 80%, 85%, 95%, and 99%). Consistent with our primary results, healthcare costs decrease monotonically as the MPR rises (Table 5). Second, we estimated regressions for the top 4 CHF drug subclasses (ranked by the proportion of patients with ≥1 fill) in the sample (ACE inhibitors, antianginals, βblockers, and diuretics) using only the MPR calculated for that drug subclass. Results for this analysis were qualitatively similar to those of the main analysis.

A potential analytical limitation is that our measure of mean adherence depends on the number of CHF medications a patient fills. To test whether the association between healthcare costs and medication adherence varied among patients with differing numbers of unique drugs filled, we estimated regressions for the subgroups of patients with 1, 2, 3, and 4 or more unique CHF drugs filled. Across all 4 groups, results were qualitatively consistent with those of the main analysis (Table 5).

The final sensitivity analysis examined the decision to estimate the relationship between medication adherence and healthcare costs contemporaneously. Estimating models in this way cannot account for the potential of reverse causality that healthcare outcomes cause changes in medication adherence rather than vice versa. Although the results of this research do not suggest that better adherence results in fewer adverse health events and lower healthcare costs, the inclusion of healthy sample members who adhere regularly to medications and have few medical problems other than CHF might bias our results. To test this hypothesis, we examined 1998 healthcare use and estimated 4 separate models for patients with a (1) a Chronic Illness and Disability Payment System score of 1 or higher, (2) a diabetes diagnosis, (3) a diagnosis of coronary artery disease, and (4) hospitalization for any condition. For all 4 models, the association between the MPR and healthcare expenditures was qualitatively the same as that in the primary analysis.

DISCUSSION
In our study, higher medication adherence among Medicaid beneficiaries with CHF and those dually enrolled in Medicare was associated with a lower likelihood of hospitalization and ED use. This study’s finding that adherent patients were slightly less likely to have a hospitalization is lower in magnitude than previous results among patients with CHF in which magnitudes were 8 to 10 percentage points6 and 6.1% to 8.7%.23 Findings on ED use were also lower in magnitude, although qualitatively similar.8 Unlike other research on patients with CHF that did not find or did not examine other outcomes, this study also finds an association between CHF drug adherence and the number of hospitalizations, hospital days, and ED visits. That nonadherent beneficiaries are more likely than adherent beneficiaries to experience more of these adverse health events is likely important to state Medicaid agencies and to the federal government, as these events are expensive. Among patients in this research sample, the mean inpatient costs in 1999 among those with at least 1 visit were $19,432, or more than $6000 per visit. Given the persistent financial problems plaguing Medicare and the high mean cost of inpatient visits, improvement of CHF drug adherence among its beneficiaries (particularly dual eligibles) could result in considerable savings.

The relationship between CHF drug adherence and total costs was stark. When the MPR threshold of 80% was used, total costs for adherent patients were almost $6000 lower per year (Table 2). Although no other research has reported such a relationship for patients with CHF, one other study6 found differences for commercially insured patients with hypertension and hypercholesterolemia.

This study also finds that the association between total healthcare expenditures and patient adherence is a graded one, challenging the 80% threshold used throughout the literature on medication adherence. Total healthcare costs of patients with adherence rates of 95% or higher were more than $3000 lower (almost 15%) than those of patients with adherence rates between 80% and less than 95% (Table 3). This result suggests that Medicaid agencies and the CMS could benefit substantially from interventions that improve beneficiaries’ adherence to CHF drug therapy (as long as the cost of these interventions does not exceed their potential savings).

Total healthcare costs of patients with near-perfect medication adherence (≥99%) compared with patients whose medication adherence was slightly lower (95% to <99%) were about $1150 per year (6%) less than those of patients with slightly lower medication adherence (Table 3). Whether it would be cost-effective for Medicaid agencies or the CMS to encourage near-perfect adherence compared with adherence at least 95% of the time is dependent on how much more costly it is to these agencies to achieve near-perfect adherence rates among their beneficiaries. Future research should consider quantifying how much it might cost these agencies to improve medication adherence for patients with CHF who are already very adherent.

There are some limitations related to the use of administrative claims data and reverse causality. First, using pharmacy data to measure adherence can inform us that a prescription was filled but cannot confirm that patients take medications as directed. As in other medication adherence studies, we cannot account for this bias. Further research is needed on the association of patients’ estimated medication adherence from claims with their reported adherence, possibly from surveys. Second, it was impossible to determine the severity of illness among patients with CHF by any means other than proxy measures. The association of medication adherence with healthcare utilization and costs might be different among patients having lower CHF severity compared with patients having higher severity. Future research should carefully address the association of CHF severity to inform policy makers of the risks of medication nonadherence among beneficiaries in the poorest of health. Third, our data did not allow us to account for important socioeconomic factors such as income or years of education. Because these factors are likely associated with drug adherence, their inclusion may have explained some of the variation in medication adherence across the sample. In particular, some research suggests that adherence to physician-recommended drug regimens (including adherence to a placebo) is associated with enhanced patient outcomes, indicating that researchers should attempt (whenever feasible) to examine as many factors as possible when estimating the association between adherence and patient outcomes.37-40

An additional limitation to this study concerns our inability to determine whether it is truly medication adherence that is the only factor associated with lower healthcare utilization and costs. It is possible that patients who are adherent to medications are also adherent to other types of treatments (such as exercise and diet), but it is impossible with these data to assess adherence to these treatments. Further research should attempt to compare adherence to pharmaceutical and nonpharmaceutical therapies versus their joint association with healthcare utilization and costs.

Finally, because we measured medication adherence, healthcare use, and healthcare costs contemporaneously, our results might be biased by reverse causality that high healthcare costs could cause low medication adherence. However, the intent was not to suggest a direction of causality but merely an association. Moreover, if the primary results were biased in some way, we should expect to find no significant association between medication adherence and healthcare costs among patients in poor health at baseline. Yet, sensitivity analyses dispute this hypothesis.
Print | AJMC Printing...