Zobraziť v EDS

Development and validation of a convenient dementia risk prediction tool for diabetic population: A large and longitudinal machine learning cohort study.

Uložené v:
Podrobná bibliografia
Názov: Development and validation of a convenient dementia risk prediction tool for diabetic population: A large and longitudinal machine learning cohort study.
Autori: Yang, Pei1,2 (AUTHOR), Xiao, Xuan1,2 (AUTHOR), Li, Yihui1,2 (AUTHOR), Cao, Xu1,2 (AUTHOR), Li, Maiping1,2 (AUTHOR), Liu, Xinting1,2 (AUTHOR), Gong, Lianggeng1,2 (AUTHOR) gong111999@126.com, Liu, Feng1,3 (AUTHOR) fengliu@tmu.edu.cn, Dai, Xi-jian1,2 (AUTHOR) daixjdoctor@126.com
Zdroj: Journal of Affective Disorders. Jul2025, Vol. 380, p298-307. 10p.
Predmety: *DISEASE risk factors, *WEB-based user interfaces, *SUBSET selection, *MACHINE learning, *PREDICTION models
Abstrakt: Diabetes mellitus has been shown to increase the risk of dementia, with diabetic patients demonstrating twice the dementia incidence rate of non-diabetic populations. We aimed to develop and validate a novel machine learning-based dementia risk prediction tool specifically tailored for diabetic population. Using a prospective from 42,881 diabetic individuals in the UK Biobank, a rigorous multi-stage selection framework was implemented to optimize feature-outcome associations from 190 variables, and 32 predictors were final retained. Subsequently, eight data analysis strategies were used to develop and validate the dementia risk prediction model. Model performance was assessed using area under the curve (AUC) metrics. During a median follow-up of 9.60 years, 1337 incident dementia cases were identified among diabetic population. The Adaboost classifier demonstrated robust performance across different predictor sets: full model with 32 predictors versus streamlined simplified model with 13 predictors selected through forward feature subset selection algorithm (AUC: 0.805 ± 0.005 vs. 0.801 ± 0.005; p = 0.200) in model development employing an 8:2 data split (5-fold cross-validation for training). To facilitate community generalization and clinical applicability, the simplified model, named DRP-Diabetes, was deployed to a visual interactive web application for individualized dementia risk assessment. Some variables were based on self-reported. A convenient and reliable dementia risk prediction tool was developed and validated for diabetic population, which could help individuals identify their potential risk profile and provide guidance on precise and timely actions to promote dementia delay or prevention. • We developed a novel dementia risk prediction model for the diabetic population. • Our model achieved a high predictive accuracy (97.1 %) and an high AUC value of 0.805. • The dementia risk prediction model stands out for its simplicity and practicality. • Cox model and SHAP analysis were consistent in showing protective or risk factors. • We deployed a simplified model, named DRP-Diabetes, as a visual web application. [ABSTRACT FROM AUTHOR]
Databáza: Academic Search Index
Popis
Abstrakt:Diabetes mellitus has been shown to increase the risk of dementia, with diabetic patients demonstrating twice the dementia incidence rate of non-diabetic populations. We aimed to develop and validate a novel machine learning-based dementia risk prediction tool specifically tailored for diabetic population. Using a prospective from 42,881 diabetic individuals in the UK Biobank, a rigorous multi-stage selection framework was implemented to optimize feature-outcome associations from 190 variables, and 32 predictors were final retained. Subsequently, eight data analysis strategies were used to develop and validate the dementia risk prediction model. Model performance was assessed using area under the curve (AUC) metrics. During a median follow-up of 9.60 years, 1337 incident dementia cases were identified among diabetic population. The Adaboost classifier demonstrated robust performance across different predictor sets: full model with 32 predictors versus streamlined simplified model with 13 predictors selected through forward feature subset selection algorithm (AUC: 0.805 ± 0.005 vs. 0.801 ± 0.005; p = 0.200) in model development employing an 8:2 data split (5-fold cross-validation for training). To facilitate community generalization and clinical applicability, the simplified model, named DRP-Diabetes, was deployed to a visual interactive web application for individualized dementia risk assessment. Some variables were based on self-reported. A convenient and reliable dementia risk prediction tool was developed and validated for diabetic population, which could help individuals identify their potential risk profile and provide guidance on precise and timely actions to promote dementia delay or prevention. • We developed a novel dementia risk prediction model for the diabetic population. • Our model achieved a high predictive accuracy (97.1 %) and an high AUC value of 0.805. • The dementia risk prediction model stands out for its simplicity and practicality. • Cox model and SHAP analysis were consistent in showing protective or risk factors. • We deployed a simplified model, named DRP-Diabetes, as a visual web application. [ABSTRACT FROM AUTHOR]
ISSN:01650327
DOI:10.1016/j.jad.2025.03.135