Development and multi-center validation of a high-performance predictive model for early detection of cognitive impairment in older adults: data-based on communities in Northern China.

BACKGROUND

The prevalence of cognitive impairment (CI) is progressively rising. Insufficient awareness, complex diagnostic procedures, and the absence of effective treatment methods contribute to the challenge. Therefore, our objective is to develop and validate a CI screening model specifically for the elderly population in China.

METHODS

The multi-center cross-sectional study included 18 communities in four provinces, and data collection was performed on basic parameters, medical history, physical measurements, cognitive assessments, and blood sample tests. The least absolute shrinkage and selection operator (Lasso) and multiple-factor logistic regression analysis were used to select variables with best predictive performance to build the model that minimizes the mean squared error and build a nomogram of the CI screening model. The Receiver Operating Characteristic (ROC) curve, calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC) are utilized to assess the discriminative ability and clinical utility of the model. Both internal and external datasets are utilized for model validation.

RESULTS

The development set of the model included a total of 1,479 participants, among whom 278 were identified as having cognitive impairment (CI), with a prevalence of 18.8%. Eight predictive factors for CI were selected as the final model, including age, falls times in 1-year, educational level, place of residence, hypertension, stroke, protein intake, and subcutaneous fat loss. The area under the curve (AUC) of the model was 0.873, indicating good discriminative ability. The calibration curve demonstrated good consistency between predicted and observed results. DCA and CIC analysis showed that the model had favorable clinical utility. The AUC, calibration curve, DCA, and CIC of both internal and external validation sets showed consistent results with the development set, indicating good consistency.

CONCLUSIONS

We developed an accurate and valid nomogram of CI screening with all the required factors easily available and four of them modifiable.