Clustering Analysis for Identification of Gastrointestinal Health Risk Profiles: A Community-Based Study Using Machine Learning Approaches

Resumen

Developing countries struggle with functional gastrointestinal issues, which lower quality of life and healthcare use. Traditional epidemiological methods often miss risk profile variability in afflicted groups. This study used unsupervised machine learning to determine Honduran community gastrointestinal health risk profiles for targeted intervention. From September to December 2025, 1,838 Honduran 12-49-year-olds from five departments participated in a cross-sectional survey. Data included sociodemographics, gastrointestinal symptoms, lifestyle, and medical history. K-means, DBSCAN, and hierarchical agglomerative clustering were used. Silhouette coefficient, Davies-Bouldin index, and Calinski-Harabasz index assessed cluster validity. PCA with t-SNE reduced dimension for visualization. The best partitioning was K-means clustering with K=3 (silhouette score: 0.162). There were three risk profiles: (1) High-risk lifestyle cluster (10.2%; n=187) with universal smoking (100%), elevated alcohol consumption (61%), and moderate symptom prevalence; (2) Low-risk cluster (57.3%; n=1,054) with younger individuals with healthy lifestyle habits and minimal symptoms (9.2% abdominal pain); and (3) High-symptom cluster (32.5%; n=597) predominantly female (71.9%) with significant gastrointestinal complaints (67% abdominal pain, 62.6% heartburn) and Machine learning-based clustering enabled individualized primary healthcare intervention design by stratifying the population into clinically meaningful risk profiles. These data support gastrointestinal health program resource allocation optimization.