Data-Driven Patient clusters of Type 2 Diabetes May Help to Identify Patients at Risk and Predict Metabolic Surgery Outcomes on Diabetes Remission and NAFLD Pregression

Abstract

Abstract Background Precision medicine and customized medicine have gained enormous attention in recent years, especially in the treatment of type 2 diabetes (T2D). Different subgroups of diabetes have been identified by research employing data-driven cluster analysis, each with a unique diabetes progression and complication risk (Ahlqvist et al., Lancet Diabetes Endocrinol. 2018). Aims We aimed to apply the proposed cluster analysis to a patent population post metabolic surgery and investigate the association with T2D remission and presence of NAFLD. Methods We retrospectively linked newly defined clusters to metabolic surgery outcomes in 53 T2D patients. Utilizing k-means and hierarchical clustering, three clusters emerged based on glutamate decarboxylase antibodies, age, BMI, HbA1c, and homoeostatic model assessment estimates of β-cell function (HOMA2-B%) and insulin resistance (HOMA2-IR). Intraoperative liver biopsies assessed nonalcoholic fatty liver disease (NAFLD) presence differentiating between simple steatosis (NAFL) and steatohepatitis (NASH). Clinical and biochemical data were collected over two years, focusing on T2D remission and NAFLD improvement. Results Cluster 1, characterized by the lowest BMI, highest NASH rate, impaired beta-cell function, and increased insulin resistance, displayed a favorable response to surgery, indicating robust regeneration of beta-cells. Despite increased insulin production, T2D remission was surprisingly low at 44.0% after one year, in contrast to 75.00% in Cluster 2 and 100.00% in Cluster 3. Metabolic surgery notably reduced insulin resistance and promoted NASH remission, evidenced by a significant reduction in a non-invasive NASH detection score and liver enzyme levels across all three clusters. Conclusion Our findings suggest that patients in Cluster 1 already show a lack of beta-cell compensation being associated with a higher prevalence of NAFLD and poorer diabetic control and therefore especially might benefit from an earlier intervention. Data-driven classification might help to customize treatment plans and identify patients at higher risk of problems at diagnosis.