Advancing Diabetes Subtyping

Deep learning can capture the longitudinal heterogeneity in type 2 diabetes, which single-time point biomarkers miss.

Newly diagnosed type 2 diabetes patients have been widely classified into data-driven subphenotypes based on their clinical characteristics (e.g., age at diagnosis, body mass index, HbA1c, HOMA2-IR, HOMA2-B). These subphenotypes vary in their genetic profile, presentation, responses to antihyperglycemic medication and risk of diabetes-related complications. However, reasons such as unavailability of HOMA2 indices in routinely collected clinical data, scarcity of data on subphenotypes derived from non-European populations, non-replicable cut-points for clinical variables and hard cluster assignment ignoring multifactorial nature of diabetes, preclude the widespread translation of this phenotyping approach from cohort studies to clinical practice.

Our current work extends this paradigm by developing longitudinal subtypes that simultaneously improve our understanding of diabetes pathophysiology and risks of traditional & emerging complications.

This is ongoing work with Zhongyu Li.

Challenges in defining subphenotypes for type 2 diabetes.

Publications

  1. Varghese 2026 medRxiv on longitudinal subtypes from pooled cohorts.
  2. Li 2025 Diabetologia on translating subtypes from cohort studies to Epic Cosmos
  3. Guo 2025 Diabetes Care on pathophysiological risk factors of subtypes in cohort studies.
  4. Guo 2025 Journal of Diabetes and Its Complications on subtypes of newly diagnosed diabetes among youth
  5. Varghese 2023 Lancet Diabetes & Endocrinology Correspondence on Replicability of subphenotypes
  6. Varghese 2023 Primary Care Diabetes Brief Report on Ethnic Differences of subphenotypes
  7. Varghese 2021 Diabetes Technology & Therapeutics on CGM-derived profiles

Abstracts

2026

  1. Li 2026 ADA Scientific Sessions 2026 Oral Presentation on subtypes and cancer relative to adults without diabetes.
  2. Ekanayake 2026 ADA Scientific Sessions 2026 Poster Presentation on subtypes and dementia relative to adults without diabetes.
  3. Varghese 2026 ADA Scientific Sessions 2026 Oral Presentation on subtypes and cardiovascular complications relative to adults without diabetes.
  4. Hung 2026 ADA Scientific Sessions 2026 e-Poster Oral Presentation on subtypes of prediabetes and premature mortality.
  5. Hua 2026 ADA Scientific Sessions 2026 Poster Presentation on insulin deficiency, insulin resistance, and cancer outcomes.

2025

  1. Li 2025a ADA Scientific Sessions 2025 Oral Presentation on subtypes and cancer.
  2. Li 2025b ADA Scientific Sessions 2025 Oral Presentation on subtypes and dementia.
  3. Guo 2025 ADA Scientific Sessions 2025 Oral Presentation on pathophysiology of subtypes.
  4. Varghese 2025 ADA Scientific Sessions 2025 Poster Presentation on trajectories of biomarkers prior to diabetes incidence.
  5. Varghese 2025 AHA EPI/Lifestyle Oral Presentation on subtypes and cardiovascular disease.