Identification and epidemiological characterization of Type-2 diabetes sub-population using an unsupervised machine...

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• Type-2 Diabetes Mellitus (T2DM) exhibits diverse underlying pathologies, yet sub-population identification in epidemiological data remains underexplored.
• The National Family Health Survey-4 (NFHS-4) dataset, comprising 10,125 T2DM patients, offers a rich resource for analyzing T2DM heterogeneity.
• Key features analyzed include medical history, diet, addiction habits, socio-economic status, and lifestyle patterns.

Purpose of the Study:

• To detect and characterize distinct sub-populations (clusters) of Type-2 Diabetes Mellitus (T2DM) patients within the Indian population.
• To address the analytical challenges posed by diverse feature types in epidemiological datasets for T2DM sub-typing.
• To investigate the socio-demographic and dietary patterns associated with identified T2DM clusters.

Main Methods:

• A novel distributed clustering workflow was developed, integrating a dimension reduction tool (UMAP) with feature-type-specific similarity measures.
• UMAP was applied separately to continuous, ordinal, and nominal features to overcome conventional limitations with mixed data types.
• Reduced dimensions were integrated to achieve interpretable and unbiased clustering of the T2DM patient data.

Main Results:

• Four significant T2DM clusters were identified, with two predominantly comprising non-obese patients.
• The non-obese T2DM clusters were characterized by lower mean age and a higher proportion of rural residents.
• One obese cluster showed a high prevalence (90%) of non-vegetarian T2DM patients with low intake of plant-based protein.

Conclusions:

• A feature-type-distributed clustering approach using UMAP is effective for analyzing diverse epidemiological data, offering a novel methodology.
• The study confirms significant heterogeneity among Indian T2DM patients concerning socio-demographics and dietary habits.
• The identification of distinct non-obese, younger, and rural T2DM sub-populations necessitates the development of tailored screening criteria for T2DM in rural India.