Methodological Considerations for Describing Medication Changes in Relation to Clinical Events and Death: An Applied Example in Patients with Type 2 Diabetes and Cancer.

Abstract

Certain clinical events reduce life expectancy and necessitate a reassessment of patient treatment. To describe medication changes in relation to a cancer diagnosis and the end of life and to highlight challenges and limitations with such descriptions. From a cohort with all Danish patients with type 2 diabetes, we matched patients with incident cancer during 2000-2021 (n = 41,745) with patients without cancer (n = 166,994) using propensity scores. We described their medication usage from cancer diagnosis until death. The 1- and 5-year mortality were 51% and 86%, respectively, in the cancer group, and 13% and 59% in the non-cancer group. In relation to cancer diagnosis and death, the use of symptomatic medications (e.g., opioids, benzodiazepines) increased (10-60 incident medications per 100 patient-months), and the use of preventive medications (e.g., antihypertensives, statins) decreased (5-30% fewer users). The changes in relation to the diagnosis were driven by patients with short observed lengths of survival (< 2years). In contrast, changes occurring within a year before death were less dependent on survival strata, and > 60% used preventive medications in their last months. Medication changes in relation to a cancer diagnosis were frequent and correlated to the length of survival. The results showcase the challenges and limited clinical utility of anchoring analyses on events or death. While the former diluted the results by averaging changes across patients with vastly different clinical courses, the latter leveraged information unavailable to the treating clinicians. While medication changes were common near death, preventive medications were often used until death.