Mixed‐Methods Clustering of Multi‐source Data to Identify Population Subgroups Based on BMI‐Associated Variables

This study aims to assess a healthcare informatics‐driven analytical approach for uncovering BMI‐related population subgroups in a large multi‐source health dataset using variable selection, dimensionality reduction, and clustering, to support tailored health behavior interventions.

A mixed methods approach combined Dutch Health Survey data with proximity‐to‐facilities and air pollution data, supplemented by expert evaluation. Statistically Equivalent Signature (SES) variable selection identified 47 BMI‐related variables. Clustering was performed using k‐means on principal components derived via PCAMIX. Silhouette scores guided the optimal number of clusters, and cluster stability was assessed using bootstrap‐based Jaccard similarity measures. An expert panel explored the practical relevance of the findings.

This study demonstrates the value of a healthcare informatics workflow for identifying stable BMI‐related population subgroups in heterogeneous health data. The findings suggest that variable selection enhances cluster interpretability and stability and highlight the multifaceted nature of BMI. Although the identified clusters are not yet suitable for direct clinical application, they provide a foundation for future research on personalized prevention and targeted interventions.

Beuken, M., K. van Berkel, M. Kleynen, I. Kanera, J. van den Brakel (2026). Mixed‐methods clustering of multi‐source data to identify population subgroups based on BMI‐associated variables. Discussion paper, Statistics Netherlands, The Hague/Heerlen.

Downloads