Published online Aug 19, 2025. doi: 10.5498/wjp.v15.i8.105433
Revised: April 15, 2025
Accepted: June 13, 2025
Published online: August 19, 2025
Processing time: 198 Days and 21.7 Hours
Motoric cognitive risk (MCR) syndrome represents an “ultra-early” stage of dementia prevention, highlighting the need for effective screening tools.
To develop and validate a novel tool for MCR identification, comparing its effectiveness with existing methods.
As part of a community study on healthy aging, a cross-sectional study recruited 1189 Chinese participants aged 50 years and older between May 1, 2022, and March 15, 2023. The cohort was randomly split into training (70%) and testing (30%) datasets. Relevant features were selected for logistic regression (LR) and decision tree (DT) models using the training dataset, and their performance was subsequently assessed using the testing dataset to validate reliability and generalizability.
The prevalence of MCR was 13.12% among 1189 participants. DT models had the area under the curves (AUCs) of 0.834 and 0.821 for training and testing datasets, respectively, while LR models indicated AUCs of 0.840 and 0.859. Non-inferiority tests confirmed the DT model’s comparable effectiveness to the LR models in predicting MCR. Both models demonstrated good calibration and clinical utility. Seven modifiable risk factors were identified: Age, education level, social engagement, physical activity, nutritional status, depressive symptoms, and purpose in life. Notably, social engagement emerged as a novel factor compared to those previously identified. Both models are integrated into an easy-to-use, interpretable web-based user interface.
The interactive, web-based user interface of both models effectively identifies MCR, with the DT model recommended for its simplicity and interpretability, supporting community nurses and clinicians in triaging MCR.
Core Tip: The current study employed decision tree (DT) and logistic regression (LR) models to predict motoric cognitive risk (MCR) syndrome by identifying key modifiable risk factors spanning the demographic, lifestyle, and health-related domains. Feature selection was performed using the least absolute shrinkage and selection operator, with LR being used to develop a nomogram and DT being used to construct a classification tree. Both models were integrated into a user-friendly, web-based interface. This is the first known application of machine learning for predicting MCR, demonstrating that the DT model is effective in supporting community nurses and clinicians in triaging MCR due to its simplicity and interpretability.