Personalized long-term prediction of cognitive function: Using sequential assessments to improve model performance.

Area of Science:

• Neurology
• Gerontology
• Biostatistics

Background:

• Accurate prediction of cognitive decline and dementia onset is crucial for understanding disease progression and planning healthcare for at-risk populations.
• Existing prediction models often rely on single-time-point assessments, limiting their accuracy for long-term cognitive status prediction.
• Incorporating time-varying risk factors assessed sequentially over multiple time points is essential for robust predictive models.

Purpose of the Study:

• To develop and test pilot models for predicting cognitive status over extended periods using time-varying risk factors.
• To evaluate the feasibility of using both estimated and observed risk factor data for cognitive decline prediction.
• To compare the predictive performance of models utilizing sequentially estimated versus directly observed time-varying risk factors.

Main Methods:

• Developed two distinct predictive models, each comprising multiple base prediction units (BPUs) trained on the same dataset.
• Model 1: Utilized sequentially estimated risk factors, initially obtained 8 years prior and subsequently optimized.
• Model 2: Employed observed time-varying risk factors from follow-up visits, enabling real-time data integration.
• Model performance was rigorously evaluated using 10-fold cross-validation across diverse patient subgroups.

Main Results:

• Both developed pilot models demonstrated supporting evidence of predictive capability for cognitive status.
• The model using observed time-varying risk factors yielded superior prediction performance compared to the estimated risk factor model.
• The models provide upper and lower bounds for predictive performance, offering insights into prediction uncertainty.

Conclusions:

• Pilot models incorporating time-varying risk factors show promise for predicting cognitive decline and dementia progression.
• Observed, sequentially assessed risk factors offer enhanced predictive accuracy over estimated factors.
• Future refinements will integrate both estimated and observed data for flexible, comprehensive prediction of dementia status and changes over time.