Characterizing multimorbidity in ALIVE: comparing single and ensemble clustering methods.

Area of Science:

• Public Health
• Computational Biology
• Biostatistics

Background:

• Multimorbidity, the presence of two or more chronic conditions, presents a significant public health challenge.
• The heterogeneity of multimorbidity makes research complex, with varying numbers and combinations of conditions among individuals.
• Unsupervised machine learning clustering methods offer a potential solution for identifying distinct multimorbidity phenotypes.

Purpose of the Study:

• To evaluate and compare different clustering algorithms for identifying multimorbidity phenotypes.
• To assess the utility of a clustering ensemble approach in uncovering complex health patterns.
• To provide guidance on selecting appropriate clustering methods for multimorbidity research.

Main Methods:

• Application of three individual clustering algorithms: partition around medoids, hierarchical clustering, and probabilistic clustering.
• Utilization of a clustering ensemble approach to integrate results from multiple algorithms.
• Analysis of the AIDS Linked to the Intravenous Experience (ALIVE) cohort study data.
• Comparison of clustering results based on quality, interpretability, and predictive ability.

Main Results:

• Demonstration of multiple distinct multimorbidity clusters within the ALIVE cohort.
• Comparison of the performance of individual and ensemble clustering methods.
• Identification of key criteria for selecting the most effective clustering strategy for specific research aims.

Conclusions:

• Comparing multiple clustering algorithms and ensemble methods is essential for robust multimorbidity phenotype identification.
• The choice of clustering approach should align with the intended application of the identified patient subgroups.
• This study provides a framework for selecting optimal clustering techniques in complex health outcome research.