425-P: Machine Learning Identifies Glypican 4 as Key Predictor of Five-Year Mortality in Heart Failure Patients with Prediabetes or Diabetes
Journal
Diabetes
ISSN
0012-1797
Type
journal article
Date Issued
2025-06-20
Author(s)
Leiherer, Andreas
;
Muendlein, Axel
;
Schnetzer, Laura
;
Mink, Sylvia
;
Heinzle, Christine
;
Brandtner, Eva Maria
;
Stella Gaenger
;
;
Thomas Plattner
;
Alexander Vonbank
;
Mader, Arthur
;
Larcher, Barbara
;
Christoph H. Saely
;
Peter Fraunberger
;
Drexel, Heinz
Abstract
Introduction and Objective: The rise of Big Data necessitates artificial intelligence-driven analyses to extract valuable insights, particularly for risk prediction in high-risk patient populations. This observational study applied machine learning (ML) algorithms to predict 5-year overall mortality in heart failure patients with type 2 diabetes mellitus (T2DM) or prediabetes
Methods: A cohort of 290 heart failure patients with T2DM or prediabetes was followed for 5 years, during which 54% of participants died. The dataset comprised 470 variables, e.g. anthropometric, clinical, social, family history, and lifestyle factors. After preprocessing, the data were analyzed using ML techniques implemented in R’s caret package. The dataset was split into training (75%) and test (25%) subsets.
Results: Among the ML models tested, the Random Forest algorithm demonstrated the best predictive performance, with a sensitivity of 82%, specificity of 89%, and overall accuracy of 85%. Clinical parameters were the most significant predictors, with the multimorbidity marker Glypican-4, hemoglobin, and glomerular filtration rate identified as the top three contributors.
Conclusion: In conclusion, ML-based Big Data analysis holds great potential for predicting mortality risk in pre-/diabetic heart failure patients, paving the way for personalized and timely interventions.
Methods: A cohort of 290 heart failure patients with T2DM or prediabetes was followed for 5 years, during which 54% of participants died. The dataset comprised 470 variables, e.g. anthropometric, clinical, social, family history, and lifestyle factors. After preprocessing, the data were analyzed using ML techniques implemented in R’s caret package. The dataset was split into training (75%) and test (25%) subsets.
Results: Among the ML models tested, the Random Forest algorithm demonstrated the best predictive performance, with a sensitivity of 82%, specificity of 89%, and overall accuracy of 85%. Clinical parameters were the most significant predictors, with the multimorbidity marker Glypican-4, hemoglobin, and glomerular filtration rate identified as the top three contributors.
Conclusion: In conclusion, ML-based Big Data analysis holds great potential for predicting mortality risk in pre-/diabetic heart failure patients, paving the way for personalized and timely interventions.
Language
English
Refereed
yes
Volume
74
Number
Supplement 1