Construction of a predictive model for cognitive impairment among older adults in Northwest China.

Abstract

Background: Cognitive impairment is most common in older adults and seriously affects their quality of life. Early prediction of cognitive impairment could be beneficial for identifying vulnerable individuals and planning primary and secondary prevention to reduce the incidence of cognitive impairment. The aim of this study is to combine the advantages of machine learning and logistic regression to construct a risk prediction model for cognitive impairment among older adults in Northwest China to identify individuals at increased risk.

Methods: A cross-sectional study was conducted. The participants and data included in this study were from the National Key Research and Development Project "Intelligent Elderly Disability Monitoring and Early Warning Network System Construction." Older adults in Northwest China were assessed between March 2022 and January 2023 using a multistage sampling method. We used random forest algorithms to select important features from potential predictors. The features identified using the random forest model were subjected to logistic regression analysis to develop a cognitive impairment prediction model. Model performance was evaluated on the basis of the area under the curve, sensitivity, specificity, accuracy, F1 score, precision, and recall.

Results: A total of 12,332 older adults were recruited and screened with the Mini-Mental State Examination Scale. The detection rate of cognitive impairment was 24.86%. The random forest algorithm and multifactorial logistic regression analysis revealed that the independent predictive factors for cognitive impairment among older adults in Northwest China were advanced age, high BMI, low literacy, low gait speed, primary financial resources from children or labor, freelance work, less exercise, low scores on instrumental activities of daily living, low walking test scores, low levels of activities of daily living, and irregular participation in social activities, all of which were used to create the nomogram. The model established with the above 12 independent predictors achieved an area under the curve of 0.816 (95% CI: 0.807∼0.824); the risk prediction value of 0.211 was the best cut-off value and showed good sensitivity (75.50%), specificity (72.40%), accuracy (73.14%), F1 score (0.802), precision (89.91%), and recall (72.38%).

Conclusion: The prevalence of cognitive impairment in older adults is high in Northwest China. The combination of machine learning and logistic regression yielded a practical cognitive impairment prediction model and has great public health implications for the early identification and risk assessment of cognitive impairment among older adults in Northwest China.