Treatment Selection and Medication Adherence for Stable Angina: The Role of Area-based Health Literacy

Abstract

Rationale, aims and objectives:

Clinical studies show equivalent health outcomes from interventional procedures and treatment with medication only for stable angina patients. However, patients may be subject to overuse or access barriers for interventional procedures and may exhibit sub-optimal adherence to medications. Our objective is to evaluate whether community-level health literacy is associated with treatment selection and medication adherence patterns.

Method:

The sample included Medicare fee-for-service beneficiaries (20% random sample) with stable angina in 2007-2013. We used an area-level health literacy variable due to the lack of an individual measure in claims. We measured the association between 1) area-based health literacy with treatment selection (medication only, percutaneous coronary intervention (PCI), or coronary artery bypass grafting (CABG) surgery) and 2) area-based health literacy with medication adherence. We controlled for other factors including demographics, comorbidity burden, dual eligibility, and area deprivation index.

Results:

We identified 8,300 patients of which 8.7% lived in a low health literacy area. Overall, 56% of patients received medication only, 28% received PCI, and 15% received CABG. Patients in low health literacy areas were less likely to receive CABG (−3.5 percentage points [95% CI: −6.8, −0.3]) compared to patients in high health literacy areas, but the significance was sensitive to specification. Overall, 81.5% and 71.5% of patients were adherent to anti-anginals and statins, respectively. Living in low health literacy areas was associated with lower adherence to anti-anginals (−3.3 percentage points [−6.1, −0.6]) but not statins.

Conclusions:

Low area-based health literacy was associated with being less likely to receive CABG and lower adherence, but the differences between low and high health literacy areas were small and sensitive to model specification. Individual factors such as dual eligibility status and race/ethnicity had stronger associations with outcomes than area-based health literacy, suggesting this area-based measure was inadequate to account for social determinants in this study.

INTRODUCTION

Recent years have seen increasing recognition of the importance of social determinants of health in medicine. Key social determinants including low individual health literacy have been linked with worse access to care, quality of care, and health outcomes for a wide array of conditions. However, gaps still exist in our understanding of the mechanisms through which social determinants affect these outcomes.^1-5

Social determinants may be measured at both an individual level and an area level, and both levels may affect health care use and outcomes.⁶ Social determinants of health typically cannot be measured at an individual level for population-level databases such as administrative claims since it would be infeasible to conduct social determinants assessments for such large populations. As an alternative, researchers may use area-based measures for social determinants of health to study associations with outcomes in such databases, even if an individual level measure would be more relevant for the analysis. However, it is unclear whether area-based measures are independently associated with individual level outcomes and can be used when individual-level social determinants measures are unavailable. Area-based measures may mask differences in individual status and weaken the signal between the social determinant and the outcome of interest. For example, some individuals living in areas with low community-level health literacy may actually have high individual-level health literacy. On the other hand, the status of one’s community may be an important social determinant measure regardless of an individual’s own status. Extreme poverty in a neighborhood may diminish health outcomes for everyone living in that area, and low average health literacy may lead to misconceptions spreading more easily in the community or greater distrust of physicians and the healthcare system. Prior research has found community-level effects for high neighborhood deprivation^6-10 and low health literacy.^11,12

The treatment for stable angina presents an opportunity to assess how area-based social determinants are associated with important individual-level medical behaviors including treatment selection and medication adherence. The most commonly used treatments for stable angina are medication only (treatment with only prescription medications), relatively less invasive surgery (percutaneous coronary intervention, or PCI) with medication, or relatively more invasive coronary artery bypass graft (CABG) surgery with medication.^13,14 Patients in all treatment groups receive anti-antianginal medications including beta-blockers, calcium channel blockers, nitrates, and ranolazine. These medications are primarily taken to relieve the symptoms of angina.¹⁴. Patients also typically receive statins. Unlike anti-anginals, statins are used for prevention.¹⁴ The treatment alternatives vary in terms of cost, how long it takes to receive symptom relief, and type and risk of complications. Medication only is the least expensive alternative and has comparable long-term symptom relief relative to PCI in patients with stable angina. However, patients may experience faster symptom relief from PCI than medication only.¹³ PCI is more costly¹⁵ and more invasive than medication only and requires adherence to dual antiplatelet therapy to prevent stent thrombosis.¹⁶ Two recent trials have found that PCI provides only minimal benefit relative to medication only for many stable angina patients.^13,17 The results from these trials have led to some concerns about overuse of PCI in the stable angina population.^18-20 CABG results in more complete symptom reduction than the other alternatives^21-23 CABG is the most invasive (with time needed for surgical recovery),²⁴ most costly,²⁵ and has a small risk of death or other serious complications.²⁶ However, CABG is often inappropriate for stable angina and is less commonly performed than PCI or medication only.¹⁴ A clinician may make a decision that CABG is an appropriate treatment for a patient presenting with stable angina who has other indications that are not typically available in claims data.¹⁴

Treatment selection is an important issue for stable angina since treatment depends, in part, on patient preferences.^14,27,28 Two prior studies found that poor comprehension of the treatment alternatives was strongly associated treatment preferences for stable angina.^29,30 Therefore, a patient’s ability to understand the tradeoffs between treatment options may influence the care received. Previous research has also found that low individual-level health literacy has been linked to less knowledge about treatment and conditions.^31,32 Individual health literacy has been linked to worse access to care and being less likely to receive treatment.^33-35 Prior systematic reviews examining the relationship between individual-level health literacy and medication adherence have either found inconsistent evidence^36,37 or a weak association between low health literacy and worse medication adherence.³⁸

We attempted to account for these established individual-level relationships between health literacy and our outcomes of interest through an area-level health literacy measure that has been previously validated in administrative claims.^39,40 Analyses that use individual health literacy measures have historically been difficult to conduct, since measuring health literacy skills on an individual-level is logistically difficult and involves conducting in-person assessments or questionnaires; such measurements are usually not plausible for large study samples. A census-based measure for health literacy enables analysis of the association of community-level health literacy with treatment selection and medication adherence with large, nationally representative claims databases.^40,41 In this analysis, we evaluated whether community level health literacy is associated with treatment selection and medication adherence among Medicare beneficiaries with stable angina. We additionally controlled for an area-deprivation index (ADI)^42,43 to assess whether the community-level health literacy variable measures variation that is distinct from the general deprivation captured by the ADI.

METHODS

Data Sources

The primary data source for this analysis was a 20 percent random sample of fee-for-service Medicare beneficiaries aged 65 and older who had at least one month of simultaneous enrollment in fee-for-service Parts A (hospital), B (outpatient medical), and D (prescription drug) between 2007-2013. This sample is nationally representative for Medicare beneficiaries meeting the sampling criteria. We used claims from 2006 to identify and exclude patients who had stable angina before 2007. We also used the following publicly available data sources: 1) the Area Health Resource File (AHRF),⁴⁴, 2) health literacy estimates at the census block group level,^39,40, and 3) an ADI at the 9-digit ZIP Code level.^42,43 The AHRF includes variables on regional medical supply, socioeconomic status, and health status.⁴⁴ The ADI was derived from 2013 American Community Survey data and it incorporated neighborhood-level measures of education, income, employment, and housing. ^42,45

We identified patients who met the criteria for a diagnostic algorithm for stable angina.⁴⁶ We excluded patients who met the diagnostic criteria before 2007, had fewer than 12 months of enrollment prior to the index date, had less than 12 months of follow up after the index date, did not receive a diagnostic angiogram or stress test, or had a nine digit ZIP code that could not be linked to the area-based measures. The full sample selection process appears in Figure 1. Information on the procedural and diagnosis codes used appear in Supplementary Tables S1-S2.

Figure 1.

Sample Flow Diagram Notes: CABG stands for coronary artery bypass grafting surgery, FFS stands for fee-for-service, MedPAR stands for Medicare Provider Analysis and Review, PCI stands for percutaneous coronary intervention

Variable Selection

The dependent variable for the first analysis was treatment selection within six months of the index angina claim: medication only, PCI, or CABG. If patients received both PCI and CABG, we assigned patients to whichever treatment occurred first. We categorized patients as medication only if they had no claims for PCI or CABG within 18 months of the index date. We excluded patients who received PCI and CABG between months 6-18 because we measured medication adherence during this period

The dependent variables for the second analysis were dichotomous indicators for whether the patient was adherent to medications. We measured adherence separately over months 6-12 and the 12-18 using proportion of days covered (PDC).⁴⁷ We considered a patient adherent on a given day if they had an active fill for a prescription to any anti-anginal. We considered a patient adherent for the period if they had a PDC of 80% or higher.⁴⁸ We also created separate measures for adherence for statins using the same approach. We assessed adherence to statins to evaluate how the results differed for medications that affect current symptoms (anti-anginals) versus preventive effects (statins). We included treatment selection as an explanatory variable for the medication adherence analysis.

We measured community health literacy using a previously developed area-based measure at the census block group level.³⁹ The measure is based on a predictive model derived using the 2003 National Assessment of Adult Literacy (NAAL).^40,41 This model included: gender, age, race/ethnicity, education, income, marital status, language spoken at home, rurality, and time living in the US.^40,41 The model was applied to demographic measures collected from the 2010 US census.³⁹ The model and the area-based measure have been validated in prior research.^39,40 The measure uses the NAAL categories of ‘below basic/basic’ and ‘above basic’ health literacy.^39,40 For clarity, we will use the terms ‘low health literacy’ (below basic/basic) and ‘high health literacy’ (above basic).

We included the following individual-level control variables: race/ethnicity (White, Black, Hispanic, and other) ⁴⁹, sex, age, comorbidity index,⁵⁰ programmatic status (full dual eligible, partial dual eligible, and recipient of the low-income subsidy), an indicator for whether the patient had diabetes (because diabetes is associated with worse treatment outcomes for PCI and CABG),⁵¹ and the ADI.^42,45 We excluded diabetes from the comorbidity index to avoid double-counting. We included county-level measures of primary care physicians, cardiologists, and hospital beds per 10,000 residents to account for differences in regional medical supply. We included year fixed effects to control for time trends and state fixed effects to control for geographic differences such as state policies.

We estimated three models for each dependent variable (treatment selection, adherence to anti-anginals, and adherence to statins). The first model included health literacy and all the control variables. The second model added the ADI because both the ADI and area-based health literacy measures use many of the same census demographic measures in their prediction models,^40-42,45 and low health literacy and low socioeconomic status are correlated.^40,41 We found the ADI and area-had a moderate inverse correlation (ρ = −0.55), which suggests that lower health literacy was moderately correlated with higher deprivation. Because of this correlation, it is possible that any associations between the area-based health literacy variable and the study outcomes may be due to the area-based health literacy variable being associated with general deprivation rather than anything specific to the average health literacy of the community. By controlling for ADI, we tested whether the area-based health literacy variable measured something distinct from neighborhood deprivation. The third model included health literacy, the control variables, ADI, and state fixed effects. We added state fixed effects separately because we expected that they would control for much of the variation in the area-based health literacy measure.

Statistical Methods

We used multinomial logistic regression to assess the relationship between the categorical treatment measure and community health literacy and treatment selection. We tested the independence of irrelevant alternatives (IIA) assumption using the Hausman-McFadden test.⁵² We found that the assumption was violated when medication only was the omitted category. We maintained the multinomial logit models as the main specification and included a sensitivity analysis to assess the violation of this assumption. We used logistic regression to assess the association between medication adherence and community health literacy We clustered the standard errors at the patient level to account for multiple observations (the first and second six-month periods). We did not find evidence to suggest violation of the assumption linearity of the independent variable with log odds and our sample was big enough to achieve the large sample properties of maximum likelihood.⁵³

We also tested to see whether the inclusion of treatment as an explanatory variable for the medication adherence analysis caused biased estimates. To test this possibility, we conducted an instrumental variables (IV) estimate using regional treatment patterns as the instrument to evaluate whether treatment selection was endogenous in the medication adherence analyses. However, we did not find any evidence for endogeneity.

Sensitivity Analyses

For the analysis of treatment selection, we re-estimated the relationship with treatment collapsed as a two-category variable. The categories were procedures (CABG or PCI) and medication only. This sensitivity analysis addressed the IIA violation.

RESULTS

Sample Characteristics

Table 1 shows descriptive statistics for the 8,300 patients included in the sample by health literacy status (8.7% in low health literacy areas and 91.3% in high health literacy areas). Before adjusting for other factors, patients living in low health literacy areas were more likely to receive medication only (69.6% vs. 54.9%) and less likely to receive PCI (22.1% vs. 29.1%) and CABG (8.3% vs. 16.1%) compared to patients living in high health literacy areas. Patients in low health literacy areas had lower adherence to anti-anginals (75.5% vs. 82.1%) and statins (62.1% vs. 72.4%) in months 6-12 relative to patients in high health literacy areas (Supplementary Table S3). Patients living in low health literacy areas were more likely to be a racial/ethnic minority, live in an area with high deprivation, be a full or partial dual eligible beneficiary, and to receive the low-income subsidy.

Table 1:

Descriptive Statistics by Community Health Literacy Level

Categorical Variables	Level	Low Health Literacy (N=720)	High Health Literacy N=(7,580)
		Freq. (Percent)	Freq. (Percent)
Treatment	Medication Only	501 (69.6%)	4,160 (54.9%)
	PCI	159 (22.1%)	2,203 (29.1%)
	CABG	60 (8.3%)	1,217 (16.1%)
Sex	Male	431 (59.9%)	3,926 (51.8%)
	Female	289 (40.1%)	3,654 (48.2%)
Age	65-70	278 (38.6%)	2,584 (34.1%)
	70-75	151 (21.0%)	1,803 (23.8%)
	75-80	144 (20.0%)	1,441 (19.0%)
	80+	147 (20.4%)	1,752 (23.1%)
Race	White	237 (32.9%)	6,890 (90.9%)
	Black	269 (37.4%)	235 (3.1%)
	Hispanic	164 (22.8%)	208 (2.7%)
	Other	50 (6.9%)	247 (3.3%)
Diabetes	Yes	270 (37.5%)	1,754 (23.1%)
Charlson Comorbidity Index	0	415 (57.6%)	4,747 (62.6%)
	1	150 (20.8%)	1,358 (17.9%)
	2	78 (10.8%)	896 (11.8%)
	3	54 (7.5%)	359 (4.7%)
	4+	23 (3.2%)	220 (2.9%)
Year	2007	234 (32.5%)	1,828 (24.1%)
	2008	135 (18.8%)	1,202 (15.9%)
	2009	83 (11.5%)	977 (12.9%)
	2010	67 (9.3%)	800 (10.6%)
	2011	58 (8.1%)	761 (10.0%)
	2012	57 (7.9%)	814 (10.7%)
	2013	86 (11.9%)	1,198 (15.8%)
Area Deprivation	Low Deprivation	389 (54.0%)	7,023 (92.7%)
	High Deprivation	331 (46.0%)	556 (7.3%)
Rural Status	Rural	127 (17.6%)	2,002 (26.4%)
Continuous Variables		Mean (SD)	Mean (SD)
Cardiologists per 10K		0.65 (0.59)	0.73 (0.52)
PCPs per 10K		7.15 (2.87)	6.88 (2.36)
Beds per 10K		33.47 (24.82)	37.51 (24.04)
Full Dual Eligible†		37.8% (47.3%)	9.9% (29.1%)
Partial Dual Eligible†		11.7% (31.1%)	5.4% (21.7%)
Receives Low Income Subsidy†		59.4% (48.3%)	18.8% (38.6%)

^†Full dual eligible, partial dual eligible, and receives Low Income Subsidy are measured as the percentage of months that beneficiaries meet the criteria in the year following the index date. Note: CABG stands for coronary artery bypass grafting, PCI stands for percutaneous coronary intervention, PDC stands for proportion of days covered.

Analysis One: Area-Based Health Literacy and Treatment Selection

While unadjusted treatment selection varies substantially by community health literacy level (Table 1), the differences were greatly attenuated after controlling for other factors. In the first specification (column M1; Table 1), living in a low health literacy area relative to a high health literacy area was associated with a 3.5 percentage point decrease in the likelihood of receiving CABG. However, the marginal effects for medication only and PCI were not statistically significant. After adding the ADI (column M2; Table 2), the marginal effect for CABG remained statistically significant (3.4 percentage point decrease). In the final specification (column M3; Table 2) that included state fixed effects, the marginal effect for CABG attenuated slightly to a 3.1 percentage point decrease and was no longer statistically significant.

Table 2:

Marginal Effects of Area-based Health Literacy and Treatment Selection

		M1: Full Model	M2: Full Model + ADI	M3: Full Model + ADI + State Fixed Effects
Variables	Level	Medication Only
Health Literacy	Low HL	3.08% (2.27%)	3.45% (2.36%)	3.54% (2.36%)
	High HL	Reference
Full Dual		4.49%** (1.51%)	4.49%** (1.51%)	5.04%*** (1.51%)
Partial Dual		2.13% (1.58%)	2.14% (1.58%)	3.48%* (1.59%)
Rural Status		1.27% (1.35%)	1.35% (1.35%)	1.35% (1.35%)
ADI	Low Depr.		Reference
	High Depr.		−0.97% (1.89%)	−0.45% (1.89%)
State Fixed Effects				✓
		PCI
Health Literacy	Low HL	0.46% (2.20%)	−0.07% (2.26%)	−0.44% (2.26%)
	High HL	Reference
Full Dual		−2.09% (1.48%)	−2.08% (1.48%)	−2.59% (1.48%)
Partial Dual		−0.64% (1.54%)	−0.65% (1.54%)	−1.58% (1.54%)
Rural Status		0.93% (1.27%)	0.81% (1.27%)	1.35% (1.35%)
ADI	Low Depr.		Reference
	High Depr.		1.54% (1.80%)	1.24% (1.80%)
State Fixed Effects				✓
		CABG
Health Literacy	Low HL	−3.54%* (1.64%)	−3.38%* (1.71%)	−3.10% (1.75%)
	High HL	Reference
Full Dual		−2.40% (1.24%)	−2.40% (1.24%)	−2.45%* (1.24%)
Partial Dual		−1.49% (1.30%)	−1.49% (1.30%)	−1.90% (1.31%)
Rural Status		−2.20%* (0.96%)	−2.16%* (0.97%)	−2.16%* (0.97%)
ADI	Low Depr.		Reference
	High Depr.		−0.57% (1.46%	−0.78% (1.46%)
State Fixed Effects				✓
Pseudo R2		0.0514	0.0514	0.0643

Standard errors in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05.

Note: ADI stands for area deprivation index. CABG stands for coronary artery bypass grafting, HL stands for health literacy, PCI stands for percutaneous coronary intervention.

While the area-based health literacy measure appeared to have a stronger relationship to outcomes than the ADI (which is also an area-based measure), the marginal effects for area-based health literacy were smaller in magnitude than individual characteristics. For example, being dual eligible was associated with a 4.5 to 5.0 percentage point increase in receiving medication only relative to non-dual eligible across the three specifications (Table 2). Also, Black patients were 7.5 percentage points less likely to receive CABG than white patients (Supplementary Table S4).

Analysis Two: Area-Based Health Literacy and Medication Adherence

For anti-anginals, living in a low health literacy area was associated with a lower percentage of patients being adherent. In the first specification (column AM1; Table 3), patients in low health literacy areas were 3.3 percentage points less likely to be adherent to anti-anginals. After adding in the ADI, the marginal effect attenuated slightly, but remained statistically significant (column AM2; Table 3). The addition of state fixed effects led to a greater attenuation and the marginal effect was no longer statistically significant (column AM3; Table 3). As with the treatment selection results, some individual characteristics were more strongly associated with non-adherence than the area-based health literacy measure. For example, Black patients had between 4.3 to 4.5 percentage points lower adherence and Hispanic patients had between 6.0 to 7.5 percentage points lower adherence compared to White patients (Supplementary Table S5). Adherence was also significantly higher for patients with index dates towards the latter end of the study period.

Table 3:

Marginal Effects of Medication Adherence to Anti-anginals and Statins

		Anti-anginals			Statins
Variables		AM1: Full Model	AM2: Full Model + ADI	AM3: Full Model + ADI + State Fixed Effects	SM1: Full Model	SM2: Full Model + ADI	SM3: Full Model + ADI + State Fixed Effects
Health Literacy	Low HL	−3.34%* (1.41%)	−3.06%* (1.47%)	−2.12% (1.45%)	−3.69% (1.88%)	−3.65% (1.95%)	−3.34% (1.96%)
	High HL	Reference			Reference
Full Dual		−0.64% (0.90%)	−0.65% (0.90%)	−0.74% (0.92%)	0.96% (1.18%)	0.96% (1.18%)	1.01% (1.20%)
Partial Dual		−1.19% (0.95%)	−1.19% (0.95%)	−1.30% (0.98%)	−1.15% (1.23%)	−1.15% (1.23%)	−0.93% (1.26%)
Rural Status		1.15% (0.85%)	1.20% (0.85%)	0.92% (0.89%)	−2.60%* (1.13%)	−2.59%* (1.13%)	−2.47%* (1.18%)
ADI	Low Depr.	Reference			Reference
	High Depr.		−0.75% (1.16%)	−0.41% (1.17%)		−0.13% (1.53%)	0.46% (1.54%)
State Fixed Effects				✓			✓
Pseudo R2		0.0188	0.0188	0.0253	0.0220	0.0220	0.0278

Note: HL = health literacy, ME = average marginal effect, SE = standard error

For statins, living in a low health literacy area was not significantly associated with adherence for any specification. However, the marginal effects were larger in magnitude across all three specifications (Table 3) and in the same direction as the marginal effects for anti-anginals. Similar to the other models, the magnitude of the marginal effects for low health literacy were larger than for the ADI in the models that included both area-based measures. Also, individual-level factors had marginal effects that were larger in magnitude than the marginal effects for low health literacy. Black patients were 13.4 percentage points less likely and Hispanic patients were between 10.3 to 10.5 percentage points less likely to be adherent to statins than White patients (Supplementary Table S6). Adherence also increased in the latter years of the study period.

Sensitivity Analysis

The sensitivity analysis (Supplementary Table S7) assessed whether the analysis of treatment selection was sensitive to collapsing CABG and PCI into a single category and re-running the analysis as a logistic regression model. The marginal effects for living in a low health literacy area with respect to receiving CABG/PCI were similar to the main analysis in that the marginal effect was negative, which suggests lower use of CABG/PCI. However, the marginal effects were no longer statistically significant in any of the specifications.

DISCUSSION

This study provides a thorough analysis of the association of area-based measure for health literacy with treatment selection and medication adherence. For treatment selection, living in communities with low health literacy was associated with a decrease in the percentage of patients receiving CABG, but not medication only or PCI. The marginal effect for living in a low health literacy area with respect to CABG was only slightly attenuated after including ADI but became non-significant after adding state fixed effects. For anti-anginals, living in a low health literacy area was associated with a significant decrease in the percentage of patients who were adherent. For statins, the marginal effects were in the same direction and larger in magnitude, but non-significant because of higher standard errors. Generally, the marginal effects for living in a low health literacy area were of a larger magnitude than the ADI across all outcomes. Individual factors such as dual eligibility status and race/ethnicity had stronger associations with treatment selection and medication adherence than area-based health literacy. Overall, these results provide limited evidence that area-based health literacy is a predictor for treatment selection and adherence to anti-anginals and statins.

Implications for Treatment Selection

This study assessed the relationship between an area-based measure for a social determinant (health literacy) and treatment selection for a condition in which less aggressive treatment may be preferred because of equivalent long-term outcomes and lower costs.^13,15 However, prior research on medical decision making for stable angina has assessed how individual comprehension is associated with treatment selection. Two prior studies analyzed how poor comprehension of the treatment alternatives affected treatment selection for stable angina.^29,30 These prior studies found that many patients who received PCI erroneously believed that the procedure would decrease their risk of death or myocardial infarction relative to medication only alone. This finding suggests that current patient-physician interactions are not resulting in patients being informed about the treatment alternatives.^29,30 A scoping review of the literature on the relationship between individual health literacy and treatment decision making was unable to form strong conclusions because of inconsistencies in the health literacy measures used and whether health literacy was associated with aspects of decision making.⁵⁴ Several studies that used individual-level health literacy measures have found that patients with low health literacy may prefer more aggressive treatment for end-of-life care.^55-57 Prior research has also found that patients with low individual health literacy have worse access to care.^33-35,58-60

In contributing to the existing literature, the results from this study suggest that patients living in low health literacy areas were less likely to receive CABG. The models included controls for other characteristics associated with poor access to care: rural status⁶¹ and variables related to low-income including full or partial dual eligibility, and the LIS; and racial/ethnic minorities.⁶² The marginal effects for area-based health literacy and these known predictors for poor access were all negative and statistically significant with respect to CABG. Therefore, it is possible that low area-based health literacy is an independent predictor for worse access to care and being less likely to receive surgery or more aggressive treatment, which is consistent with some prior literature.^33-35 In addition, the small magnitude and the sensitivity to model specification is consistent with the scoping review that found that the associations between health literacy and treatment selection were generally not strong.⁵⁴ The results are inconsistent with the finding that low health literacy patients preferred more aggressive care at the end-of-life.^55-57 It is possible that patient preferences are different for stable angina compared to end-of-life care or patient preferences may not be translated into the care they actually receive due to access barriers.

Implications for Medication Adherence

The previous research on area-based health literacy and medication adherence is limited. Prior systematic reviews examining the relationship between individual-level health literacy and medication adherence have either found inconsistent evidence^36,37 or a small effect with lower health literacy associated with worse adherence.³⁸

The findings in the current analysis are consistent with the prior literature in that we only observed limited evidence for a relationship between area-based health literacy and medication adherence. We found significant associations between adherence to anti-anginals, but not statins. While significant, the marginal effects were moderate in magnitude. The inconsistent findings for anti-anginals and statins may be due to the different uses for these medications. The relationship may vary by medication since anti-anginals are used mainly for symptom relief and statins are used for long-term prevention.¹⁴

It is unclear whether the area-based measure for health literacy would be helpful for informing interventions to improve medication adherence. Previous research has documented successful interventions to improve adherence for cardiovascular-related diseases including informational mailings,⁶³ telephone follow-up⁶⁴ and pharmacist-led interventions⁶⁵ Given the findings in the current study of the small magnitude of estimated associations, area-based health literacy does not appear to be strong predictors for medication adherence. More research is necessary to design interventions that promote safe medication use among all patients.

Limitations

This analysis has several limitations. First, data on individual assessments of health literacy are not available in claims data. While the area-based measure for health literacy has been validated, individual assessments like the REALM⁶⁶ would have less measurement error. Random measurement error biases estimates of associations towards zero (i.e., no effect). Second, many other potentially relevant individual characteristics are unobserved in claims data, including: symptoms, disease severity, anatomic variants, and the appropriateness of treatment. The analysis included the Charlson Comorbidity index and an indicator for diabetes status to control for the different treatment guidelines for patients with diabetes.¹⁴ However, no measures of the severity of the angina were available. Patients in the medication only group may have had less severe angina than patients in the PCI and CABG groups, so unmeasured differences in severity may have determined the treatment decision. Third, the IIA assumption that underlies multinomial logistic regression models was violated, which may bias parameter estimates, though the direction of the bias is not known. We found roughly similar results in a sensitivity analysis that combined the PCI and CABG categories. Fourth, it is possible that treatment selection and medication adherence were affected by affordability, which may differ systematically in low vs. high health literacy communities. While all patients had Medicare Parts A, B, and D coverage and we controlled for programmatic status, we were unable to control for other factors related to affordability including Medigap coverage and wealth.

Conclusions

In summary, we found that an area-based measures for health literacy had small associations with treatment selection and medication adherence for patients with stable angina. Individual factors including dual eligibility status and race/ethnicity had stronger associations with the outcomes and may be more relevant for predicting treatment selection and medication adherence. While the results were somewhat consistent with the prior evidence, the weak associations may mean that the area-based measure did not fully capture the role of health literacy in treatment selection and medication adherence for stable angina. Future research using individual assessments of health literacy, as well as measures of symptoms and angina severity could be insightful if such information becomes available in a survey with sufficient sample size. Future research could also extend this analysis to evaluate whether health outcomes are associated with community-level health literacy for stable angina patients following treatment selection.