Assessment of Medication Adherence Using Pharmacy Data Before and After Percutaneous Coronary Intervention

Introduction

Non-adherence to cardiovascular medications is common and is associated with adverse outcomes and higher healthcare costs.^1–3 Adherence is particularly important for anti-platelet medications after percutaneous coronary intervention (PCI, also known as coronary stenting), since non-adherence may lead to stent thrombosis, myocardial infarction (MI), and death.^4,5 Hospitals and clinicians therefore invest significant resources in patient education and other interventions to promote adherence;⁶ however, these interventions may be costly, and non-adherence rates remain unacceptably high. Randomized trials have identified some effective strategies, but large improvements in adherence are uncommon and even fewer interventions alter clinical outcomes.^7,8

The impact and cost-effectiveness of interventions may be improved if health systems can identify patients at high risk for non-adherence. Unfortunately, clinicians are poor at assessing adherence,⁹ and patient self-report is often unreliable,¹⁰ highlighting the need for objective measures of adherence. The use of pharmacy fills for assessing adherence is well described,^11,12 and prior studies have demonstrated the utility of pharmacy data for predicting future medication-taking behavior.^13–15 There are increasing opportunities to integrate pharmacy data into electronic health systems,^16,17 but it is unknown if these technological investments will improve patient care. The potential role of pharmacy fills data for adherence risk prediction must be further investigated.

We therefore examined pre- and post-procedural adherence among patients treated with PCI in Veterans Affairs (VA) hospitals, the largest health system in the United States, which includes an integrated pharmacy system. We hypothesized that most patients would have an evaluable cardiovascular medication adherence history, that prior adherence to cardiovascular medications would be associated with post-PCI adherence, and that non-adherent patients would have worse clinical outcomes.

Methods

The VA Clinical Assessment, Reporting, and Tracking (CART) Program is a national quality and safety program for invasive cardiovascular procedures with universal participation by VA catheterization laboratories. Clinicians prospectively document patient characteristics and procedural details using standardized definitions derived from the National Cardiovascular Data Registry (NCDR). The VA CART Program has been previously described, including methods for ensuring data accuracy.^18,19 We linked the demographic, clinical, and procedural data from CART with VA pharmacy records for four cardiovascular medication classes: statins, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors/angiotensin II receptor blockers (ARBs), and P2Y₁₂ inhibitors. Aspirin usage is not consistently captured by VA pharmacy records because of a high rate of over-the-counter purchase, and so was not included in this analysis.

We examined 56,357 patients receiving PCI at a VA hospital from Jan 1, 2012 and Dec 31, 2018. Of these, 37,164 patients (66.0%) filled at least one cardiovascular medication (from any of the 4 classes listed above) for at least one year prior to the procedure and were therefore eligible for pre-PCI adherence assessment. Of these, 33,705 (90.7% of patients with pre-PCI adherence assessment) also had at least 1 year of pharmacy fill data for a P2Y₁₂ inhibitor after PCI (Supplementary Figure).

To assess pre-procedure adherence, we calculated a proportion of days covered (PDC) for each of the eligible medication classes, defined as the ratio of the number of days the patient had medication on hand over the evaluative period. A PDC of 0.8 or greater was considered adherent.¹² Days spent as an inpatient were not counted in the numerator or denominator. For patients on more than one class of medication prior to PCI, the highest of the individually calculated PDC values was used to classify overall pre-procedure adherence. An additional 6-month window was added to the 1-year pre-procedure evaluative period (18 months total) to capture supply on hand. Adherence to a P2Y₁₂ inhibitor after PCI was calculated similarly over a 1-year period. Any P2Y₁₂ inhibitor supply prior to PCI was accounted for when assessing post-PCI PDC.

In addition, we assessed clinical outcomes within 1 year of PCI. Death was assessed using VA mortality files. Readmission for myocardial infarction, stroke, and revascularization (PCI or CABG) were assessed using ICD and CPT codes (Supplementary Methods).

Statistical Analysis

We compared baseline demographic, clinical, and presentation characteristics among patients who were adherent vs non-adherent in the one year prior to presentation, using binomial test of proportions for categorical variables and Student’s t-tests for continuous variables.

To assess the association of pre-procedure adherence with post-procedure P2Y₁₂ adherence, an initial logistic regression with 54 candidate covariates was performed (Supplementary Methods). We then performed one elimination step to remove any covariate with a p-value < 0.1, leaving 26 covariates: age, race, heart failure (HF), hypertension, hyperlipidemia, prior myocardial infarction (MI), prior PCI, prior valve surgery, prior stroke/transient ischemic attack, peripheral arterial disease (PAD), dialysis status, sleep apnea, atrial fibrillation, deep vein thrombosis, prior cancer diagnosis, anxiety, depression, post-traumatic stress disorder (PTSD), current or former smoker, current or former alcohol abuse, illicit drug abuse, procedural status (elective, urgent, emergent, salvage), cardiogenic shock, diastolic blood pressure, high-density lipoprotein (HDL), and pre-procedure adherence. This final model had a c-statistic of 0.64 (Supplementary Methods). Exclusion of pre-procedural adherence from the model resulted in a c-statistic of 0.60.

Next, we assessed the association of pre-procedural adherence with death and the clinical composites death/MI/stroke and death/MI/stroke/revascularization within 1 year. We used a Cox proportional hazards models adjusting for the following covariates, selected a priori as potentially unbalanced between the groups or associated with the clinical outcomes: age, race, sex, body mass index (BMI), body surface area (BSA), HF, hypertension, hyperlipidemia, diabetes, prior MI, prior cardiac catheterization, prior PCI, prior coronary artery bypass graft surgery (CABG), prior valve surgery, prior PCI, prior stroke/transient ischemic attack, PAD, chronic lung disease, chronic kidney disease, dialysis status, prior renal transplant, sleep apnea, atrial fibrillation, deep vein thrombosis, prior cancer diagnosis, liver disease, anxiety, depression, post-traumatic stress disorder, current or former smoker, current or former alcohol abuse, illicit drug abuse, family history of coronary artery disease, ST-segment elevation MI presentation, procedural status (elective, urgent, emergent, salvage), cardiogenic shock, cardiac arrest, surgical turn down, blood pressure, creatinine, total cholesterol, low-density lipoprotein (LDL), and HDL. Patients who died were censored on the date of death and PDC was calculated from the time of PCI to the date of censoring. We then assessed the association of post-PCI adherence with clinical outcomes using the same methodology.

This analysis was approved by the VA Puget Sound Health Care System Institutional Review Board (IRB#01733). Analysis was performed with SAS version 9.4, SAS Institute Inc., Cary, NC, and with R version 4.2.1, The R Foundation for Statistical Computing.

Results

Overall, 6,974 patients (20.7%) were non-adherent to at least one cardiovascular medication prior to PCI (PDC ≥0.80). Mean PDC for those classified as adherent was 0.96 (S.D. 0.05), while those classified as non-adherent had mean PDC of 0.58 (S.D. 0.19). Non-adherent patients were younger, more likely to be black, have lower rates of medical comorbidities, higher rates of depression and substance abuse, and were more likely to present emergently or urgently for PCI (Table 1).

Table 1 Patient Characteristics Associated with Adherence or Non-Adherence to Cardiovascular Medications Within 1 Year Prior to PCI

Post-PCI non-adherence to P2Y₁₂ inhibitors was present for 6885 patients (20.4%) at 1 year. Patients who were non-adherent pre-PCI were more likely to be non-adherent post-PCI compared with pre-PCI adherent patients (33.0% vs 17.2%) (Figure 1). After adjustment for demographic and clinical factors, pre-PCI non-adherent patients had significantly lower odds of adherence post-PCI (OR 0.45 C.I. 0.41–0.46).

Figure 1 (Visual summary): Unadjusted rates of adherence to cardiovascular medications pre-PCI and adherence to P2Y12 inhibitors post-PCI.

Patients who were non-adherent before PCI had a higher adjusted mortality rate within 1 year of PCI (HR 1.17 C.I. 1.03–1.33), though rates of the composites death/MI/stroke (HR 1.11 C.I. 0.99–1.25) and death/MI/stroke/repeat revascularization (HR 1.04 C.I. 0.97–1.12) were not significantly different compared with patients who were adherent before PCI.

Patients who were non-adherent to P2Y₁₂ inhibitors after PCI had significantly higher adjusted rates of death (HR 1.18 C.I. 1.06–1.32), death/MI/stroke (HR 1.20 C.I. 1.09–1.32), and death/MI/stroke/revascularization (HR 1.15 C.I. 1.09–1.23) within 1 year (Table 2) compared with patients who were adherent to P2Y₁₂ inhibitors after PCI.

Table 2 1-Year Outcomes Among Patients Who are Adherent and Non-Adherent to P2Y12 Inhibitors After PCI

Discussion

In this study of consecutive patients receiving PCI at VA hospitals from 2012 to 2018, two thirds of patients were evaluable for adherence to cardiovascular medications prior to PCI using the VA’s integrated pharmacy system. Pre-procedure non-adherence was associated with post-procedure non-adherence to P2Y₁₂ inhibitors and increased rates of death. Automated assessment of medication use using pharmacy fills is a potential mechanism to identify patients at high-risk of non-adherence and poor outcomes.

Medication non-adherence following PCI and myocardial infarction remains persistently high despite several decades of research and significant attention from the clinical community. Suboptimal adherence has been documented for several important medication classes, including statins and anti-platelet medications.¹ Multiple reasons for non-adherence have been observed,²⁰ so diverse interventions have been tested in randomized trials.⁸ These include patient education, polypills, the reduction or elimination of medication copayments, and multifaceted interventions including pre-discharge education and post-discharge follow-up. Some of these interventions have improved patient adherence, though the overall impact has been modest and in most trials did not result in a reduction in clinical events.^21–26 Interventions may be more impactful if targeted to a high-risk population. Prospective adherence prediction is particularly important for PCI patients, since intervening after medication non-adherence occurs may be too late; the patient has already been exposed to higher risk of stent thrombosis and MI.

This study showed that patients who were adherent to their cardiovascular medications prior to PCI had 2.3 times greater odds of adhering to their anti-platelet therapy following PCI, compared with pre-PCI non-adherent patients. Importantly, a large, national, integrated health care system like the VA can provide evaluable adherence data for most patients at the bedside. A prior similar study of Medicare patients with MI could only assess 10% of patients due to reliance on Medicare Part D.¹⁵ However, outpatient pharmacy data are increasingly available to hospitals through commercial services,¹⁶ and with ongoing consolidation of healthcare systems and electronic health records in the US it is likely that point-of-care pharmacy data will be increasing available to clinicians. Our study provides one immediate application for these data.

However, our study also highlights challenges in predicting non-adherence and intervening. Our model, including demographic and clinical factors in addition to fill-based adherence, had only modest discriminatory ability (c-statistic 0.64), consistent with prior studies.^13,14 Even if non-adherence can be accurately predicted, it is unclear if or how clinical care should be modified. Providing adherence data to clinicians is not effective as a standalone strategy,²⁷ so adherence prediction must be paired with other interventions. Interventions such as copayment reduction or pharmacist-led education may be effective for some patients but not others, regardless of pre-PCI adherence status. One promising potential strategy is digital health; adherence data could be integrated with personal electronic devices to promote healthy medication-taking behaviors. There is increasing evidence of the impact of digital health interventions on adherence and outcomes for patients with cardiovascular disease.^28,29 Future studies may test integrated solutions using data obtained from pharmacy systems, personal devices, and patient-reported symptoms.

Our study has several limitations. There are many potential ways to assess fill-based adherence; our method may underestimate non-adherence relative to other algorithms,²⁸ identifying a relatively small high-risk group (20% of patients). Specifically, for patients taking multiple medications we assigned the highest individual medication PDC as the overall patient PDC, under the assumption that this demonstrates the best adherence the patient has achieved. In addition, we could not assess for interruptions in treatment directed by physicians, which could result in some adherence misclassification. Our risk model included clinical data available at the time of PCI, but we could not assess for other factors that are associated with adherence, such as socio-economic status and social support. Subjects were primarily older and male, and our findings may not be fully generalizable to non-VA populations. In addition, VA copayments are low relative to private insurance or Medicare, all patients can receive mail order prescriptions, and there are multiple options for reordering medications.²⁹ This may explain why adherence was high after PCI compared with other studies, though still suboptimal at approximately 80%. Associations of adherence with favorable clinical outcomes in our study may be biased by the well-known “healthy adherer” effect if adherent patients are also more likely to engage in other health-promoting activities.

Conclusions

Adherence to cardiovascular medications prior to PCI can be assessed for most patients using pharmacy fill data, and non-adherence is associated with continued non-adherence and mortality after PCI. Use of integrated pharmacy data to identify patients at high-risk for non-adherence could improve patient outcomes and the cost-effectiveness of adherence interventions. This strategy should be tested in prospective trials.