Beyond prescriptions: chronic medication adherence predicts mortality risk in a large-scale cohort study

Abstract

Objectives: The Medication Adherence Risk Score (MARS) is a calculated score using pharmacy transactional data spanning 50% of the South African private pharmacy market. This study aims to demonstrate that the existing MARS model enhances risk stratification by identifying individuals at increased risk of mortality related to non-adherence to chronic medication.Methods: This was a retrospective cohort study in which an analysis of the relative mortality experience was compared to a standard fully underwritten base was performed for each of the MARS categories (low, medium, high and very high). The actual-to-expected ratio (AER) and relative risk (RR) for each category were compared across age groups and gender. The least absolute shrinkage and selection operator (LASSO) regression analysis method was applied to determine the most important variables within the dataset, providing insight into whether MARS offered more benefit than traditional risk rating factors. A time-to-event analysis by MARS categories was performed using the Cox proportional hazards model.Results: The mortality experience of the study population was higher than the expected fully underwritten base (AER = 175%). For the overall sample, increasing AER and RR did not correlate with increasing MARS categories. However, use of the MARS in addition to age band allowed for differentiation of risk within the 25 to 55 age bands, with a higher MARS score indicating a higher AER and RR. The time-to-event analysis showed a statistically significant difference in the mean number of months before death occurred between the different MARS categories (low = 26.53; medium = 8.93; high = 7.02; very high = 6.92; p < 0.001).Conclusion: The MARS is not generalisable across all groups, as evidenced by the absence of a monotonic trend in the overall sample. However, when combined with age, it effectively differentiated mortality risk for individuals aged 25–55. The standard fully underwritten model underestimated the number of deaths within this pharmacy population. The time-to-event analysis showed a significant inverse relationship between MARS category and survival time.