Unveiling the heterogeneity of T2DM: a narrative review of clustering algorithms in stratifying comorbidity and complication risk

Diabetes is a growing global health concern, affecting 537 million adults worldwide, significantly contributing to mortality and disability. Cluster analysis, a branch of Unsupervised machine learning (ML), has been increasingly used to identify distinct subgroups of patients with diabetes, enabling better disease stratification and personalized treatment strategies. Despite the growing burden of diabetes, clinical management often relies on generalized treatment approaches that may not account for patient heterogeneity. There is a need to better understand the underlying subtypes of diabetes to enable more personalized care. This narrative review evaluates the application of clustering in stratifying diabetes patients across diverse populations based on clinical, biomarker, and molecular characteristics and more. A literature search was conducted using PubMed, and Google Scholar following the Scale for the Assessment of Narrative Review Articles (SANRA) guidelines. A total of 22 articles were finalized to be included in the analysis where different clustering approaches on diabetes patients have been used. This review further classifies the studies into diabetes subgroups focusing specifically on type 2 diabetes patients where Severe Autoimmune Diabetes (SAID), Severe Insulin Deficient Diabetes (SIDD), Mild Obesity-Related Diabetes (MOD), and Mild Age-Related Diabetes (MARD) clusters were most commonly found, and biomarker-based classification such as inflammatory and molecular including only type 2 diabetes patients, where additional two clusters: Severe Insulin Resistant Diabetes with Relative Insulin Insufficiency (SIRD-RII), and Mild Age-Related Diabetes with Relative Insulin Insufficiency (MARD-RII), were found. This review underscores the potential of clustering in diabetes research and identifies gaps in standardization across studies. Future research should focus on refining clustering approaches, mainly introducing soft clustering to identify if patients fall in more than 1 cluster.