Temporal multimorbidity patterns and cluster identification: a longitudinal analysis of administrative data.

Abstract

Background: Multimorbidity is analytically and clinically complex, involving multiple interactions between diseases each with unique implications for health. Identifying disease co-occurrence patterns at the population level could aid in disease prevention, management, and care delivery.

Methods: Here, we analyzed multimorbidity patterns using linked administrative data from a longitudinal cohort of 1,347,820 individuals with multimorbidity over 20 years in British Columbia, Canada. A directed network-based approach was used to assess disease patterns in multimorbidity by frequency (prevalence) and non-random association (lift). We applied a community detection algorithm to identify multimorbidity disease clusters.

Results: Mood and anxiety disorders and hypertension were the most common disease predecessors in prevalence networks, with differences between age groups. Lift networks revealed non-random disease associations. Some indicate potential etiological disease relationships (e.g., breast cancer preceding heart disease in young women), shared risk profiles (e.g., chronic obstructive pulmonary disease and lung cancer), or overlapping disease constructs (e.g., Parkinsonism and dementia). Disease clusters often centered around a single disease as a common predecessor or consequence, representing potential multimorbidity profiles, which may be relevant for patient subgrouping or management.

Conclusions: Insights from these analyses can complement traditional chronic disease surveillance methods, flagging disease patterns for further interrogation into their impacts on function, mortality, and health service utilization.