Identification of Clusters in a Population With Obesity Using Machine Learning: Secondary Analysis of The Maastricht...

Area of Science:

• Data Science
• Machine Learning
• Personalized Medicine

Background:

• Modern lifestyles with physical inactivity and poor nutrition drive obesity and chronic diseases.
• Personalized interventions are effective for long-term behavior change.
• Machine learning (ML) can uncover complex relationships and population clusters in large datasets.

Purpose of the Study:

• To identify distinct clusters of individuals with obesity.
• To uncover relevant variables differentiating these clusters using a data-driven, hypothesis-free ML approach.

Main Methods:

• Utilized cross-sectional data from The Maastricht Study (n=4128) with 2971 variables.
• Applied the factor probabilistic distance clustering algorithm for high-dimensional data analysis.
• Employed the statistically equivalent signature algorithm to identify distinct, minimally redundant variables.

Main Results:

• Identified 3 distinct clusters within the obesity cohort.
• Cluster 1: Characterized by lower energy intake and higher unemployment.
• Cluster 2: Characterized by higher energy intake and predominantly male participants.
• Cluster 3: Characterized by higher cognitive functioning and higher educational attainment.

Conclusions:

• A data-driven, hypothesis-free ML approach successfully identified distinguishable clusters in a large, complex obesity dataset.
• Key differentiating variables (energy intake, occupation, sex, cognitive function, education) were identified.
• Findings support the development of targeted, personalized interventions for obesity management.