gms | German Medical Science

Background: Without sufficient adherence to therapy, the medicines cannot work. This is especially true for serious conditions such as acute myocardial infarction (MI) or heart failure (HF), where poor adherence can lead to hospitalization, readmissions, and death. According to the pertinent guidelines, the respective treatment regimens usually include drugs from different classes. Within these regimens, the impact of adherence to individual drugs has only rarely been investigated [1]. In particular, it is largely unclear what the overall impact of nonadherence to individual medications recommended in the guidelines (or the nonadherence to combinations of medications recommended in the guidelines) is on the risk of readmissions, which are a major consequence of potentially inadequate treatment.

Materials and Methods: A retrospective cohort was derived from health insurance claims data of AOK Baden-Württemberg between 2011 until the end of 2016. Specific readmissions were defined for MI and HF [2]. Starting one year before the index hospitalization and ending on the censoring date, the individual prescription durations of the medicines recommended in the guidelines at that time (MI: beta-blockers, angiotensin conversion enzyme inhibitors and AT1 receptor blockers (renin-angiotensin-system inhibitors, RASI), statins, and antithrombotic drugs; HF: beta-blockers, RASI, diuretics, and aldosterone antagonists) were derived from longitudinal reimbursement patterns [3]. The censoring date was defined by either the date of a specific readmission [2], patient death, or end of follow-up, whichever occurred first. Conceptually, for each patient and drug class, we determined whether the drug class was discontinued during the index hospitalization (no prescription within 90 d of discharge). Only patients who had at least three prescriptions in each drug class during the observation period (first prescription to censoring date) or who had discontinued treatment by definition were included to estimate adherence using the CMA6 method from the R package AdhereR. This yielded estimates of the proportion of days covered (PDC) to the corresponding drug class during the observation period (the adherence estimate was set to 0 in the case of treatment discontinuation). In predictive models, we defined a binary outcome of a specific readmission within 180 d (MI) and 90 d (HF), respectively. Guideline-adherent medication and covariates accepted in the literature were fixed terms in logistic regression models, while additional comorbidities were tested in a backward selection procedure before the final models were fitted by least absolute shrinkage and selection operator (LASSO) regression with interaction terms of the adherence estimates to each drug. In developing and validating the model, the split-sample procedure was used, in which every third patient was randomly assigned to the test set, to examine how well adherence to individual drug classes and different drug combinations predicted the risk for specific readmissions [2].

Results: The MI cohort included a total of 5611 patients (mean age: 78.9 years, median Elixhauser comorbidity score: 3, median beta blocker PDC: 0.58, median RASI PDC: 0.81, median statin PDC: 0.65, median antithrombotic PDC: 0.51). The HF cohort included a total of 20859 patients (mean age: 81.4 years, median Elixhauser comorbidity score: 5, median PDC for beta blocker: 0.56, RASI: 0.73, diuretics: 0.87, aldosterone antagonist: 0.26). There were positive linear correlations between the adherence estimates of the drug classes within the cohorts and adherence effects were generally more pronounced in HF patients than in MI patients (Figure 1 [Fig. 1]). Good predictive performance was indicated by high areas under the receiver operating characteristic curve (ROC) of 0.763 (MI) and 0.868 (HF). Explorative analyses revealed that inclusion of medication adherence into the prediction models significantly improved their performance (Figure 2 [Fig. 2]). Statin adherence had the greatest influence on the readmission risk for MI, followed by adherence to RASI and beta-blockers (Figure 3 A [Fig. 3]). In the HF cohort, adherence to beta-blockers had the greatest impact on the readmission risk, followed by adherence to RASI and adherence to diuretics (Figure 3 B [Fig. 3]). When combining adherence to individual drugs recommended by the guidelines into an overall adherence measure of all drugs, nonadherence was a predictor of readmission to hospital.

Conclusion: The clear and pronounced associations between adherence and readmissions in health insurance data confirm the relevance of the medications recommended in the guidelines at that time and underscore the potential importance of monitoring adherence. Adherence-promoting interventions for guideline-based therapies could therefore help prevent readmissions and thus burdensome time in hospital for the patient and considerable expenditures for the health care system. Therefore, using routine data to automatically estimate adherence could provide new opportunities to identify patients whose well-being could be improved by an intervention addressing nonadherence while concurrently reducing expenditures.