A National Assessment of Medication Adherence to Statins by the Racial Composition of Neighborhoods

Abstract

Adherence to statins is lower in black and Hispanic patients, and is linked to racial/ethnic disparities in cardiovascular mortality. Poverty, education, and prescription coverage differentials are typically invoked to explain adherence disparities, but analyses at the level of neighborhoods and their pharmacies may provide additional insights. Among individuals filling new statin prescriptions in a national pharmacy chain (N=326,171), we compared adherence for patients residing in mostly minority neighborhoods to those living in mainly white areas. In analyses adjusting for patient-level factors associated with poor adherence, including age, insurance, payer, prescription cost and convenience, patients residing in black and Hispanic neighborhoods had 2-3 weeks less statin therapy over one year, a pattern not seen in Asian areas. In black and Hispanic neighborhoods, good adherence was associated with copays under $10, the use of 90-day refills, and payers other than Medicaid. Efforts to improve medication adherence for vulnerable populations may benefit from interventions at the level of local pharmacies, as well as medication benefit redesign.

Introduction

Chronic diseases affect more than half of the U.S. adult population and are associated with three quarters of all hospitalizations.(1-3) Adherence to medications is critical to the prevention of hospitalizations related to these chronic diseases.(4) In the area of cardiovascular disease, better adherence to HMG-Co-A reductase inhibitors, or “statins,” leads to significant reductions in cardiovascular hospitalization, revascularization, mortality,(5-9) and lower overall health care costs.(10, 11)

Medication adherence to cardiovascular medications is lower among black and Hispanic patients. (12-15) and likely contributes to a persistent 7 year lower overall life expectancy in blacks relative to whites.(16, 17) Interventions designed to improve cardiac risk factor control among minorities appear to mitigate racial and ethnic disparities in cardiovascular health outcomes in some studies.(18, 19)

Factors associated with better medication adherence have been extensively studied,(20-22) and include those linked to the patient, their medical condition, therapy, insurance, care providers, and socioeconomic status. (4, 23-28) The challenge of improving disparities in medication adherence has typically focused on individual patient characteristics, such as age, education, health literacy, and insurance, with socioeconomic factors less explored.(4, 29)

Efforts to mitigate differentials in cardiovascular health outcomes and reduce health disparities in the United States could benefit from insights derived from geospatial analysis, and greater attention to the potential role of local retail pharmacies in minority neighborhoods. Residential segregation of ethnic/racial groups in the United States is well-documented, with blacks and Hispanics disproportionately cared for by a relatively small group of physicians(30) and hospitals.(31) Recent work from the Multi-Ethnic Study of Atherosclerosis (MESA) suggests that greater racial/ethnic segregation is associated with greater incident cardiovascular disease, even after adjusting for individual social economic position and cardiac risk factors.(32) Less appreciated is that ethnic and racial minorities are often receiving prescriptions from a select group of pharmacies, because of this segregation. The identification of relationships between neighborhoods, pharmacies, and medication adherence would have important implications for efforts designed to improve medication adherence.

This study takes advantage of the fact that over 86% of all U.S. blacks and Hispanics live within 5 miles of a retail pharmacy belonging to one national network, and leverages block group level socioeconomic data from the recent U.S. census to characterize the neighborhoods where these pharmacy customers reside. This allowed us to investigate predictors of medication adherence in a large well-defined national sample of patients with an initial prescription for statin therapy in 2012.

Our study aims to answer two important questions. First, in a national cohort of new statin users, are individuals residing in predominantly minority communities less likely to adhere to therapy, even after accounting for previously described patient-level factors associated with poor adherence, including age, insurance, payer, prescription cost and convenience. Second, we sought to identify predictors of high statin adherence among patients living in predominantly black and Hispanic neighborhoods.

Methods

We conducted a retrospective analysis of one-year medication adherence among patients filling a statin prescription for the first time between 1/1/12 and 3/31/12 at Walgreens, a national chain pharmacy with over 8,200 drugstores in all 50 states, the District of Columbia, Puerto Rico and the U.S. Virgin Islands. The Walgreens enterprise data warehouse contains patient and prescription records for all of its pharmacy transactions. The pharmacy record contains data related to the patient (age, gender, address, and preferred language), drug (therapeutic class, dose, quantity, and number of days’ supply), primary payer type (private insurance, Medicare, Medicaid, and self-pay/uninsured), finance (sold date, cost, plan pay, and co-pay), and pharmacy (location and hours of operation). Race/ethnicity data were not routinely collected for individual patients.

We used the patient address from the pharmacy record as input to a geographic information system (GIS) to define the location as longitude, latitude and Federal Information Processing Standards (FIPS) code. The FIPS code was then merged with 2010 US census block level data. The census data included block-level demographic variables such as race/ethnicity (White, Black, Asian, and Hispanic), household income, education levels, and average household members. In addition, we used the Rural-Urban Commuting Area Codes (RUCAs) to determine urbanicity (metro, micro, rural, and small town). A typical block group has 1500 (600-3000) residents, a census area five-fold finer than the zip code or census tract, allowing more reliable attribution of neighborhood characteristics to each patient.

The percentage of incomes below the Federal poverty level for each patient’s neighborhood of residence was defined using 2010 U.S. Census Bureau Poverty Thresholds, which are derived from the average household income and the average number of household members in each block group. (www.census.gov/hhes/www/poverty/about/overview/measure.html )

Measures

Patient medication adherence was estimated based on prescription fill records, which are a common proxy for actual adherence.(33) In this study, we defined adherence as ‘patient days on therapy’ (PDOT) for their statin medication. PDOT is calculated by counting the number of days when a patient has medication available during the one-year observation period that spans from the index fill date to the index fill date plus 365 days. We also expressed our findings as the percentage of patients achieving 80% adherence.(34) For this calculation, we determined proportion of days covered (PDC) which is PDOT/365 and defined a patient as adherent if their ratio was greater than or equal to 80%. Patients were defined as new-to-therapy (NTT) if they had no record of a statin prescription in the prior 36 months in the pharmacy database.

A generalized linear model quantified the association between patient statin adherence (PDOT) at 12 months and block group racial composition (Black, Hispanic, and Asian), adjusting for individual factors of age, gender, payer (Medicaid, Medicare, private insurance, self-pay/uninsured), co-pay amount ($0, >$0.00-$5.00, >$5.00-$10.00, >$10.00-$15.00, >$15.00), use of 30-day vs. 90-day refills, as well as urbanicity, the neighborhood characteristics of urban (metro and micro) vs. other (rural and small town) location, educational attainment (% of residents with some college), and poverty level.

In addition, we used a logistic regression model to evaluate the predictors for adherent statin patients (PDC>=0.8) living in majority black or Hispanic neighborhood areas, and a generalized linear regression model to measure the relationship between adherence and copayment levels. All data analyses were performed with SAS 9.2 software (SAS Institute Inc, Cary, North Carolina). The level of significance for hypothesis testing was set at 0.05. The study protocol was approved by the Quorum Review IRB.

Results

Sample Description

There were 391,668 patients identified as new to therapy for a statin medication during the first 3 months of 2012; 332,193 of these patients had a valid home address that mapped to a FIPS code. A total of 326,171 patients had complete payer data and constituted the study cohort. Their mean age was 60.2 years, and 49.5% were female.

The study cohort closely paralleled U.S. census distributions. For example, of the block groups represented by the study cohort, 8.9% had at least 50% Black residency, compared to 8.8% of all U.S. residents in the 2010 US census; for Hispanic majority block groups, the figures were 9.9% of the cohort, compared to 10.0% in the census. When we analyzed the distribution by patient home address, rather than block group, 8.0 – 9.9% of study patients resided in block groups with at least 50% minority representation.

Descriptive Analysis

Characteristics of the study cohort are shown in table 1. Patients living in block groups with at least 50% black residency tended to be younger, female, have lower copays, use more Medicaid, have lower incomes, be less educated, and more likely reside in an urban area, compared to patients living in neighborhoods with lower proportions of black residents. A similar pattern was seen for patients residing in majority Hispanic block groups. For patients residing in majority Asian block groups, age was similar, but there were fewer females, lower copays, more private insurance, higher income, better education, and greater urbanicity, compared to study cohort patients residing in block groups with lower proportions of Asian population. Greater numbers of patients residing in majority Hispanic and Asian block groups expressed a preference for ‘communication in a language other than English’. Use of a 90-day supply at the time of the initial prescription fill was less likely for patients residing in majority black and Hispanic neighborhoods. We also examined differential continuity within the pharmacy chain as an alternative explanation for reduced adherence. Only modest differences were found, with 77% of those from Hispanic neighborhoods, and 79% of those from black majority neighborhoods filling prescriptions in the pharmacy chain in the following calendar year (2013), compared to a 81% rate in more white neighborhoods.

Table 1. Demographic Characteristics, Grouped by Census Characteristics in Block Group of Patient Residence.

	*Under 50% Black (n=300,315)	*>=50% Black (n=25,85 6)	*Under 50% Hispanic (n=293,867)	*>=50% Hispanic (n=32,30 4)	*Under 50% Asian (n=323,493)	*>=50% Asian (n=2,678)
Female %	48.73	58.45	49.25	51.78	49.53	46.27
Age (Y), mean	60.36	58.63	60.45	58.22	60.35	60.22
No. of Therapeutic classes, mean	3.31	3.73	3.33	3.40	3.34	3.04
initial fill copay group% (adjusted to 30-day)
0	21.19	22.68	20.87	25.36	21.28	24.91
>0 to 5	30.61	38.65	30.75	35.78	31.20	36.74
>5 to 10	22.26	19.01	22.49	17.52	22.04	17.10
>10 to 15	6.90	6.41	6.85	6.99	6.87	6.20
>15	19.04	13.25	19.04	14.36	18.61	15.05
Annual Total Copay,mean	49.71	29.82	50.0	31.23	48.23	36.75
Primary Insurance%
Medicaid	3.94	10.42	4.12	7.50	4.46	3.81
Medicare	31.90	32.14	32.00	31.32	31.97	26.44
Private	57.14	50.32	57.20	51.09	56.56	60.83
Self pay/uninsured	7.02	7.12	6.70	10.08	7.01	8.92
Language preference as English%	97.02	99.08	98.90	81.49	97.23	91.85
90-day as initial fill%	30.03	18.53	29.82	22.72	29.1	30.77
24h pharmacy Store for initial fill%	30.66	32.03	30.20	35.92	30.80	27.00
Distance(miles) between home to pharmacy%
<=0.5	15.96	20.61	15.29	25.84	16.14	39.21
>0.5 to<2	54.24	63.47	54.30	61.06	54.99	52.24
>=2 to <5	17.03	12.24	17.51	8.83	16.73	7.65
>=5	12.77	3.68	12.90	4.27	12.14	0.90
*Annual Household Income ($), mean	62,207	38,109	62,369	41,446	60,154	77,579
*Poverty-Yes%	0.82	6.01	0.73	5.80	1.24	0.41
*some college%	58.3	44.44	59.42	37.07	57.17	61.84
*Urbanicity
Urban	94.60	97.79	94.43	98.92	94.81	100.00
Other	5.30	2.20	5.56	1.07	5.19	0.00

^*Data fields are inferred by GIS block groupdata

Modeling Results

Patients residing in majority black and Hispanic block groups had lower statin adherence compared to patients residing in non-minority block groups, even after controlling for multiple confounders. For patients residing in block groups with at least 50% black population, the 12 month mean PDOT was 161.0 days compared to 184.9 for patients living in block groups with a lower proportion of blacks. Similarly, patients residing in block groups with at least 50% Hispanic residency had a 12 month mean PDOT of 165.2 days compared to 180.7 for patients in non-Hispanic block groups. There was no effect seen for those residing in predominantly Asian block groups (Table 2).

Table 2. Patient Days on Therapy (PDOT) by proportion of Black or Hispanic or Asian population in block group.

	Under 50%		>=50%		p*
	N	PDOT (95% confidence Interval	N	PDOT (95% confidence Interval
Black Unadjusted	300,315	203.4(202.9-203.8)	25,856	167.6(166.1-169.1)	<.01
Black Adjusted	300,315	184.9(167.9-201.9)	25,856	161.0(144.0-178.0)	<.01
Hispanic Unadjusted	293,867	203.8(203.3-204.3)	32,304	170.7(169.3-172.0)	<.01
Hispanic Adjusted	293,867	180.7(163.8-197.7)	32,304	165.2(148.2-182.2)	<.01
Asian Unadjusted	323,493	200.5(200.1-201.0)	2,678	199.5(194.9-204.6)	0.75
Asian Adjusted	323,493	173.2(156.3-190.0)	2,678	172.8(155.4-190.1)	0.84

p value shows the significance of difference between means

Restricting the analyses to patients living in block groups with higher concentrations (50% or more) of black or Hispanic residents, we compared patients with excellent adherence (defined as PDC >=80%) to the overall cohort of patients living in that same block group. In multivariate logistic regression models, more adherent patients residing in either majority black or Hispanic block groups were significantly older, male, on Medicare, taking more than one chronic medication, and less likely to pay cash, to have a 90-day refill policy and lower copays. More adherent patients residing in majority Hispanic block groups were also more likely to speak English (Tables 3, 4). The finding that patients residing more than 5 miles from the closest pharmacy had higher adherence was accounted for by higher rates of residence in rural and suburban areas, in areas with less poverty, as well as greater use of 90 day refills, and lower rates of Medicaid coverage.

Table 3. Predictors for Adherent Patients residing in majority Black census block group (N=25,856).

Predictors of Adherence>=80%	Odds Ratio	95% Wald Confidence Limits		P-Value
Age (Y), mean	1.021	1.019	1.024	<.0001
Gender
Male(reference)
Female	0.932	0.871	0.996	0.0376
Initial fill copay group %
0 (reference)
>0 to 5	0.835	0.769	0.906	<.0001
>5 to 10	0.558	0.501	0.622	<.0001
>10 to 15	0.359	0.301	0.427	0.0295
>15	0.072	0.058	0.088	<.0001
Annual Total Copay, mean	1.014	1.014	1.015	<.0001
No. of Therapeutic classes, mean	1.185	1.166	1.204	<.0001
Primary Insurance %
Private (reference)
Medicaid	0.764	0.669	0.872	0.015
Medicare	1.268	1.17	1.375	<.0001
Self pay/uninsured	0.583	0.496	0.684	<.0001
Language preference
English (reference)
Other than English	0.895	0.619	1.292	0.5532
90-day as initial fill %
Yes (reference)
No	0.464	0.43	0.501	<.0001
Distance(miles) between home to pharmacy %
<=0.5 (reference)
>0.5 to<2	1.13	1.006	1.27	0.7998
>=2 to <5	1.028	0.945	1.118	0.0008
>=5	1.465	1.209	1.775	0.0003
24h pharmacy Store for initial fill %
Yes (reference)
No	1.011	0.942	1.085	0.756
*Annual Household Income ($)	0.999	0.997	1.001	0.4465
*Poverty
No (reference)
Yes	1.061	0.921	1.224	0.1854
*Education: some college %	1.487	1.154	1.916	0.0021
*Urbanicity
Urban
Other	0.679	0.543	0.851	0.9326

^*Data fields are inferred by GIS block group data

Table 4. Predictors for Adherent Patients residing in majority Hispanic census block groups (N=32,304).

Predictors of Adherence>=80%	Odds Ratio	95% Wald Confidence Limits		P- Value
Age (Y), mean	1.015	1.012	1.017	<.0001
Gender
Male (reference)
Female	0.916	0.864	0.971	0.0031
initial fill copay group%
0 (reference)
>0 to 5	0.784	0.729	0.843	<.0001
>5 to 10	0.604	0.549	0.666	<.0001
>10 to 15	0.307	0.261	0.36	<.0001
>15	0.068	0.057	0.082	<.0001
Annual Total Copay, mean	1.013	1.012	1.014	<.0001
No. of Therapeutic classes, mean	1.237	1.219	1.256	<.0001
Primary Insurance%
Private (reference)
Medicaid	0.685	0.602	0.78	0.0003
Medicare	1.156	1.073	1.245	<.0001
Self pay/uninsured	0.571	0.503	0.648	<.0001
Language preference
English (reference)
Other than English	0.755	0.692	0.824	<.0001
90-day as initial fill%
Yes (reference)
No	0.508	0.475	0.542	<.0001
Distance (miles) between home to pharmacy%
<=0.5 (reference)
>0.5 to<2	1.013	0.946	1.085	0.2635
>=2 to <5	1.005	0.898	1.125	0.3553
>=5	1.173	0.997	1.378	0.0538
24h pharmacy Store for initial fill%
Yes (reference)
No	1.181	1.11	1.255	<.0001
*Annual Household Income ($)	1.003	1.001	1.006	0.0011
*Poverty
No (reference)
Yes	0.874	0.753	1.015	0.2685
*some college%	1.19	0.964	1.469	0.1051
*Urbanicity
Urban
Other	1.014	0.746	1.378	0.0473

Discussion

In this paper we have documented large and clinically significant differentials in medication adherence in a large national cohort of patients beginning statin therapy in 2012, for individuals residing in predominantly Black and Hispanic neighborhoods. These differentials remained even after adjusting for payer, co-pay amount, complexity of medication regimen, as well as two key census tract proxies associated with social disadvantage, median household income, and percent of residents with some college. These statistical adjustments are important to better understand racial and ethnic differentials in cardiovascular outcomes, given the complex associations of race and ethnicity with education and poverty.

Our findings are consistent with prior research evaluating the association of neighborhood characteristics and cardiovascular medication adherence and outcomes. For instance, one paper assessed adherence to cardiovascular therapy following admission for acute coronary syndrome and demonstrated clear neighborhood differentials in southeastern Michigan. (35) While a recent MESA study showing greater incident CVD for blacks living in racially segregated neighborhoods,(32) we demonstrate poorer medication adherence in both black and Hispanic predominant neighborhoods. Recent research has also described ‘pharmacy deserts’, areas with poorer access to pharmacy service, as more often located in segregated black communities, as well as in low-income communities and federally designated Medically Underserved Areas.(36)

Our study methodology provides a useful advance in addressing knowledge gaps around residential segregation and CVD disparities,(37) as these and the majority of adherence studies have been limited geographically, with many focusing only on insured populations. Race and ethnic data are not consistently collected in national prescription databases, by managed care, or in specialty society cardiovascular therapy quality collaboratives.(38) Pharmacy administrative database predictors of medication adherence on their own have not performed especially well in predicting statin adherence, (39) and a more nuanced approach which includes both pharmacy and socioeconomic variables is also supported by recent research linking statin adherence in peripheral artery disease(38) and the cardiovascular outcome of survival after cardiac surgery, to socioeconomic position,(40) rather than to race per se.

The higher levels of non-adherence to preventive medications by individuals residing in predominantly black and Hispanic neighborhoods may offer avenues to monitor and test strategies to mitigate racial and ethnic health disparities. Pharmacies in predominantly minority neighborhoods have been shown to offer less access to pain medications(41), and nicotine replacement therapy(42). Patients who have participated in brief face-to-face counseling sessions with a community pharmacist at the beginning of statin therapy demonstrate greater medication adherence and persistency(43), and more aggressive and tailored approaches to improve adherence are beginning to be explored(44-48). Neighborhood pharmacies located in minority neighborhoods, are potentially well positioned to serve as the locus for community level interventions designed to improve health, not only through appropriate stocking and staffing, and tailored adherence programs, but also through direct patient services. In the Walgreens network alone, there are currently over 2000 retail stores in the continental U.S. located in communities with at least 30% black or Hispanic population. Increasingly, the neighborhood retail pharmacy is being conceptualized as a neighborhood health asset and source of preventive services, for instance serving as a source of healthier foods in food deserts,(49, 50) and of vaccination programs coordinated with local hospitals.(51)

A number of policy and benefit design interventions have also shown value in improving adherence, including policy interventions to reduce copayments, systems interventions to offer case management, and patient-level educational interventions with behavioral support.(52) Cost sharing has been associated with increasing adverse events and the poor and elderly,(53) and elimination of copays has been associated with improved adherence in numerous studies. (23, 25, 54), without increasing total spending.(55) A greater proportion of patients in our study residing in predominantly black and Hispanic neighborhoods had co-pays under $10 for their statin prescription, and lower annual total copay amounts. Expanding the use of 90-day fills is a testable medication policy change that may hold promise in improving adherence for some chronic medications, as the prevalence of this approach was 30% lower for patients residing in black and Hispanic majority neighborhoods, relative to the full cohort. In our model Medicaid as a payer remained a strong predictor of poorer statin adherence in predominantly black and Hispanic neighborhoods, suggesting that national Medicaid expansion in itself may be insufficient to strongly reduce adherence disparities.

In a time major health care system change, our approach describes a practical and iterative process that can help assess whether the ‘natural experiments’ of state and national policy changes have differential impacts on adherence in predominantly minority communities. Twenty-six states have declined to participate in the Medicaid expansion promulgated under the Affordable Care Act. Commercial pharmacy chains with national reach and a strong presence in minority neighborhoods have the opportunity to evaluate differential trends as state policies shift. As an illustration, in 2000-2002 North Carolina Medicaid policy reduced the days supplied with prescriptions from 100 to 34 days, and significantly hurt adherence, though these results were not published until 2011.(56)

These study findings also have implications for the equity of national quality metrics related to medication adherence. Recent work has shown that the socioeconomic characteristics of enrollees have a significant effect on drug adherence performance measures for Medicare Part D contractors.(57) Moreover, the Centers for Medicare and Medicaid Services Five-Star Quality Rating System specifies that for health plans to achieve 5 stars, at least 75% of covered patients will need to obtain at least 80% of medication prescribed to them in the 3 classes of statins, anti-hypertensive medications, and hypoglycemic medications.(58) Clinical practices, networks, and health plans most involved in the care for vulnerable racial and ethnic populations may face real disadvantages as they are held accountable for coordinated care, which typically carry an expectation of better patient adherence to medication care plans.

Limitations

The socioeconomic variables are inferred from the individual patient’s home address, and may not completely capture the extent of neighborhood social disadvantage, or other unmeasured confounders. Prescriptions obtained from other sources were not available for analysis, and some ‘new-to-therapy’ prescriptions may represent chronic therapy transferred from another pharmacy source. While some patients may have filled prescriptions at other pharmacies, we believe our findings are representative of all large pharmacy chains which account for the majority of prescriptions dispensed in the U.S. Indeed, the use of data from a national chain may underestimate disparities in adherence, as state-level analysis suggests that the costs of retail prescriptions are highest for independent pharmacies in poorer neighborhoods.(59)

Adherence was inferred from prescription refill behavior derived from a single national pharmacy chain, though chain pharmacies provide the majority of U.S. prescriptions, this chain has strong presence in minority block groups, and prior research has suggested little affect in statin adherence between different types of community pharmacies.(60) While some patients may fill prescriptions at multiple pharmacies, in a large nationally representative sample 90% of patients filled statin and ACE/ARB prescriptions at a single pharmacy(61), and about 80% of our patients continued to fill prescriptions with the chain during the following year. While copays may vary by state, the consistency of these findings in a national sample is an important contribution to the literature. The correspondence between locally defined neighborhood and census block group is not precise, but permits a national analysis such as ours. The small area size of block group, relative to census tracts and zip codes, appears to reduce the risk of misclassification.(62) Finally other unmeasured neighborhood characteristics, such as resident-perceived social cohesion, may affect adherence and have been shown to ameliorate the risk of myocardial infarction.(63)

Conclusions

Clinically important disparities in statin adherence are found for individuals residing in predominantly black and Hispanic neighborhoods, even after controlling for age, co-pay, payer, and basic neighborhood socioeconomic factors. This process can also inform and assess programs targeted at the neighborhood level to improve adherence, and potentially reduce cardiovascular mortality. Local retail pharmacies in minority neighborhoods may be fruitful vehicles for disparity reduction interventions. Awareness and monitoring of national and local adherence patterns and exploration of their relation to various neighborhood characteristics can help assess the effects of national and state coverage and policy changes, in a relatively rapid and iterative manner. Expanding Medicaid coverage and co-pay policy changes alone will be insufficient to eliminate adherence differentials in predominantly black and Hispanic neighborhoods. These adherence differentials also call attention to the equity of national quality metrics for medication adherence, for clinicians and groups providing care in these settings.