Methods/Statistics
Comparative analysis of methods for identifying multimorbidity patterns among people with opioid use disorder in Ontario, Canada Myanca Rodrigues* Myanca Rodrigues Glenda Babe Tea Rosic Brittany B. Dennis Richard Perez Claire de Oliveira Sameer Parpia Lehana Thabane Zainab Samaan
Background:
Multimorbidity is defined as the co-occurrence of two or more (2+) chronic conditions. As the prevalence of chronic conditions increases, identification of patterns of co-occurring illnesses is crucial to health system planning. Cluster analysis is a commonly used method to identify multimorbidity patterns. However, to-date, studies have focussed on hierarchical cluster analysis (HCA), with a paucity of research examining non-hierarchical cluster analytic methods, such as K-means analysis.
Objective:
Our primary aim was to compare multimorbidity patterns using two methods – HCA and K-means clustering – in our cohort of people with OUD in Ontario, Canada.
Methods:
We linked observational cohort data collected from 3,430 people receiving treatment for OUD between 2011 and 2021 in Ontario, Canada to provincial health administrative databases. We identified 18 chronic conditions, commonly used in multimorbidity studies in Ontario, using ICD-10-CA diagnostic codes and the diagnostic codes of physician billing claims, and followed the cohort over an eight-year period in the data holdings. We used HCA and K-means clustering to identify multimorbidity patterns. Analyses were stratified by sex, with results compared for each method.
Results:
HCA identified 4 clusters for males, and 2 for females. K-means identified 3 multimorbidity patterns for each sex. Although there were some differences by sex and method of analysis, two combinations of disease were observed consistently across sexes and both methods: (i) diabetes, hypertension and stroke, and (ii) asthma and chronic obstructive pulmonary disease.
Conclusions:
Our study findings illustrate that multimorbidity patterns vary depending on the method of analysis used (HCA vs. K-means). We found that HCA was useful for large research datasets for an in-depth examination of multimorbidity, whereas K-means may be better used to identify disease clusters found in clinical practice.