Machine Learning–Based Prediction of Neurodegenerative Disease in Patients With Type 2 Diabetes by Derivation and Validation in 2 Independent Korean Cohorts: Model Development and Validation Study (Preprint)

• DOI: 10.2196/preprints.56922
• Publication Date: 2024-10-03T14:15:47Z
• Authors: Hyunji Sang; Hojae Lee; Jaeyu Park; Sunyoung Kim; Ho Geol Woo; Ai Koyanagi; Lee Smith; Sihoon Lee; You-Cheol Hwang; Tae Sun Park; Hyunjung Lim; Dong Keon Yon; Sang Youl Rhee
• Keywords: Preprint; Model Validation

<sec> <title>BACKGROUND</title> Several machine learning (ML) prediction models for neurodegenerative diseases (NDs) in type 2 diabetes mellitus (T2DM) have recently been developed. However, the predictive power of these is limited by lack multiple risk factors. </sec> <title>OBJECTIVE</title> This study aimed to assess validity and use an ML model predicting 3-year incidence ND patients with T2DM. <title>METHODS</title> We used data from independent cohorts—the discovery cohort (1 hospital; n=22,311) validation (2 hospitals; n=2915)—to predict ND. The outcome interest was presence or absence at 3 years. selected different ML-based hyperparameter tuning conducted area under receiver operating characteristic curve (AUROC) analysis cohort. <title>RESULTS</title> dataset included 22,311 (discovery) 2915 (validation) T2DM recruited between 2008 2022. observed 133 (0.6%) 15 (0.5%) cohorts, respectively. AdaBoost had a mean AUROC 0.82 (95% CI 0.79-0.85) dataset. When this result applied dataset, exhibited best performance among models, 0.83 (accuracy 78.6%, sensitivity specificity balanced accuracy 78.6%). most influential factors were age cardiovascular disease. <title>CONCLUSIONS</title> shows feasibility assessing suggests its potential screening patients. Further international studies are required validate findings.