Related Experiment Video
Updated: Sep 17, 2025
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Development of a machine learning-based risk prediction model for mild cognitive impairment with spleen-kidney
Ya-Ting Ai1, Shi Zhou2, Ming Wang3
1School of Nursing, Hubei University of Chinese Medicine, Wuhan 430065, Hubei Province, China; Hubei Shizhen Laboratory, Hubei University of Chinese Medicine, Wuhan 430061, Hubei Province, China; Engineering Research Center of TCM Protection Technology and New Product Development for the Elderly Brain Health, Ministry of Education, Hubei University of Chinese Medicine, Wuhan 430065, Hubei Province, China.
Objective:
As an age-related neurodegenerative disease, the prevalence of mild cognitive impairment (MCI) increases with age. Within the framework of traditional Chinese medicine, spleen-kidney deficiency syndrome (SKDS) is recognized as the most frequent MCI subtype. Due to the covert and gradual onset of MCI, in community settings it poses a significant challenge for patients and their families to discern between typical aging and pathological changes. There exists an urgent need to devise a preliminary diagnostic tool designed for community-residing older adults with MCI attributed to SKDS (MCI-SKDS).
Methods:
This investigation enrolled 312 elderly individuals diagnosed with MCI, who were randomly distributed into training and test datasets at a 3:1 ratio. Five machine learning methods, including logistic regression (LR), decision tree (DT), naive Bayes (NB), support vector machine (SVM), and gradient boosting (GB), were used to build a diagnostic prediction model for MCI-SKDS. Accuracy, sensitivity, specificity, precision, F1 score, and area under the curve were used to evaluate model performance. Furthermore, the clinical applicability of the model was evaluated through decision curve analysis (DCA).
Results:
The accuracy, precision, specificity and F1 score of the DT model performed best in the training set (test set), with scores of 0.904 (0.845), 0.875 (0.795), 0.973 (0.875) and 0.973 (0.875). The sensitivity of the training set (test set) of the SVM model performed best among the five models with a score of 0.865 (0.821). The area under the curve of all five models was greater than 0.9 for the training dataset and greater than 0.8 for the test dataset. The DCA of all models showed good clinical application value. The study identified ten indicators that were significant predictors of MCI-SKDS.
Conclusion:
The risk prediction index derived from machine learning for the MCI-SKDS prediction model is simple and practical; the model demonstrates good predictive value and clinical applicability, and the DT model had the best performance. Please cite this article as: Ai YT, Zhou S, Wang M, Zheng TY, Hu H, Wang YC, Li YC, Wang XT, Zhou PJ. Development of a machine learning-based risk prediction model for mild cognitive impairment with spleen-kidney deficiency syndrome in the elderly. J Integr Med. 2025; 23(4): 390-397.
Insights
Machine learning models can predict mild cognitive impairment with spleen-kidney deficiency syndrome (MCI-SKDS) in older adults. The decision tree model demonstrated the best performance, offering a practical tool for early diagnosis in community settings.
Area of Science:
- Gerontology
- Neuroscience
- Traditional Chinese Medicine
Background:
- Mild cognitive impairment (MCI) prevalence increases with age, posing diagnostic challenges in community settings.
- Spleen-kidney deficiency syndrome (SKDS) is the most frequent MCI subtype in Traditional Chinese Medicine.
- A preliminary diagnostic tool for MCI-SKDS in older adults is urgently needed.
Purpose of the Study:
- To develop and evaluate machine learning models for predicting MCI-SKDS in elderly individuals.
- To identify key indicators for MCI-SKDS prediction.
- To assess the clinical applicability of developed models.
Main Methods:
- 312 elderly individuals with MCI were randomly assigned to training (3:1) and test datasets.
- Five machine learning methods (LR, DT, NB, SVM, GB) were employed to build diagnostic models.
- Model performance was evaluated using accuracy, sensitivity, specificity, precision, F1 score, AUC, and DCA.
Main Results:
- The Decision Tree (DT) model exhibited the best performance in the training and test sets (e.g., accuracy 0.904/0.845).
- Support Vector Machine (SVM) showed the highest sensitivity (0.865/0.821).
- All models achieved AUC > 0.9 (training) and > 0.8 (test), with good clinical applicability via DCA. Ten significant predictors were identified.
Conclusions:
- Machine learning-derived risk prediction models for MCI-SKDS are simple, practical, and possess good predictive value.
- The DT model demonstrated superior performance for MCI-SKDS prediction.
- The developed models show significant clinical applicability for early detection in community-dwelling elderly individuals.
