OBJECTIVES:

Statin intolerance (SI) is a challenge for managing dyslipidemia patients. The understanding of SI is incomplete, and patient characteristics remain poorly understood. This study aimed to characterize clusters of similar SI patients based on diagnosis codes, to facilitate identification of groups of common comorbidities that coexist in the patient population.

METHODS:

A retrospective cohort study conducted using outpatient data from a high- dimensional Electronic Medical Record dataset in Germany (IQVIA™ Disease Analyzer). The study included 292,603 patients with high cardiovascular risk, atherosclerotic cardiovascular disease, and/or hypercholesterolemia between 2017 and 2020. Clustering was performed on SI patients (identified using expert and literature-informed rules such as statin down-titration and presence of statin-associated muscle symptoms [SAMs]) linking them via common diagnosis codes to build a graphical representation. Community detection algorithms (greedy modularity maximization) were applied to this graph, along with statistical tests about enrichment of data on certain comorbidities.

RESULTS:

The graph-based approach revealed clusters based on gender (male/female) and age strata (>= 60 years and <60 years). Three key patient clusters were observed for males >= 60 years with: a) presence of predominant musculoskeletal disorders; b) somatoform disorders (both a and b clusters have a higher incidence of SAMs and multiple statin use) and c) chronic kidney disease and cardiac conditions (lower incidence of SAMs and higher levels of down- titration). Patients with musculoskeletal disorders, anxiety, and obesity (higher incidence of SAMs); and those with obesity, chronic kidney disease, and hypertensive heart disease (lower incidence of SAMs) were the two prominent clusters for females >= 60 years. For male and female patients <60 years, depression and other psychiatric disorders were part of dominant clusters.

CONCLUSIONS:

Machine learning generated insights into distinct patient clusters that can be leveraged for the diagnosis and optimal treatment of SI patients.