What question did this study set out to answer?

To develop and validate a machine learning model that predicts multi-system complications in diabetes using various clinical parameters.

June 7, 2026

1997-P: Beyond HbA1c: Machine Learning–Derived Triglyceride–BMI–Duration Axis as a Superior Predictor of Multisystem Diabetes Complications

Key Points

To develop and validate a machine learning model that predicts multi-system complications in diabetes using various clinical parameters.
Analysis of 400 T2DM patients from community screening camps with 47 clinical assessments.
Ensemble machine learning techniques (Random Forest, XGBoost, Neural Networks) used with SHAP analysis for feature importance.
Primary outcome measured was incidence of multi-system complications defined as two or more from specified conditions.
ML model achieved AUROC 0.87 for predicting complications, significantly outperforming the HbA1c-only model (AUROC 0.58, p<0.001).
TG-BMI-Duration Axis identified as primary predictor, with triglycerides contributing 28% of variance and increased complication odds by 8.4-fold when combined with obesity indicators.
Early intervention targeting TG/BMI reduced complication probability from 68% to 18%, demonstrating a significant 45-point decrease in Diabetes Complexity Score.

Abstract

Introduction and Objective: Diabetes risk stratification relies predominantly on HbA1c despite evidences that metabolic dysregulation beyond glycemia drives complications. We developed and validated a comprehensive machine learning model integrating 47 clinical parameters to predict multi-system complications and identify novel risk axes. Methods: Analysis of 400 T2DM patients from community screening camps with comprehensive assessment across 47 parameters. Primary outcome: multi-system complications (≥2 of: bone disease, neuropathy, dyslipidemia, pulmonary risk, hepatic steatosis). Ensemble machine learning (Random Forest, XGBoost, Neural Networks) with SHAP analysis for feature importance and interaction detection. Results: Integrated ML model achieved AUROC 0.87 (95% CI 0.83-0.91) for multi-system complications, significantly outperforming HbA1c-only model (AUROC 0.58, p0.001). SHAP analysis revealed novel “TG-BMI-Duration Axis” as primary driver: triglycerides contributed 28% of model variance, BMI 24%, diabetes duration 18%, while HbA1c contributed only 9%. Feature interactions showed synergistic effects: TG200 mg/dL + BMI30 increased complication odds 8.4-fold (OR 8.42, CI 4.67-15.18) versus either alone (OR 2.1-2.8). Model-derived Diabetes Complexity Score (DCS) stratified patients into low (DCS30, 12% complications), moderate (DCS 30-60, 48% complications), and high-risk (DCS60, 82% complications) categories, enabling precision intervention. Early achievement of targets within 5 years reduced DCS by 45 points and complication probability from 68% to 18% (NNT=2). Conclusion: Comprehensive multi-parameter machine learning identifies TG-BMI-duration axis as primary driver. Model-derived Complexity Score enables precision risk stratification (82% high-risk complication rate versus 12% low-risk) and demonstrates that early intensive metabolic intervention targeting TG/BMI yields greater benefit than HbA1c reduction alone. Disclosure A. Maheshwari: None. S. Patil: None. S. Singla: None. A. Tewari: None. N. Verma: None.

AI से पूछें

Bookmark

AI से पूछें

Bookmark

1997-P: Beyond HbA1c: Machine Learning–Derived Triglyceride–BMI–Duration Axis as a Superior Predictor of Multisystem Diabetes Complications

Key Points

Abstract

Cite This Study