The predictive power of neuropsychological measures in MCI: early detection of dementia conversion.

BACKGROUND

With rising global life expectancy, early identification of dementia risk is crucial for implementing timely interventions. This study assesses the value of neuropsychological measures in stratifying the risk of mild cognitive impairment (MCI) progression over a 3 years period.

METHODS

This retrospective, longitudinal study included 349 MCI patients from the SPIN cohort who underwent an extended neuropsychological battery at baseline (2009-2021), enabling the calculation of composite cognitive domain scores; and follow-up clinical visits. A Ridge logistic regression following a data-driven selection of α within an Elastic Net tuning grid was utilized to identify key factors associated with dementia conversion.

RESULTS

Over 3 years, 37.8% of patients with MCI converted to dementia, 58.7% remained stable, and 3.4% reverted. At baseline, converters were significantly older, exhibited lower functional status, and demonstrated poorer performance across all cognitive domains compared to the stable and reverter groups. A regularized (Ridge) logistic regression, selected as the optimal model within an Elastic Net tuning grid; was used to identify the most robust predictors, yielding an Area Under the Curve (AUC) of 0.816 [95% CI: 0.769-0.862]. Predictor stability was assessed via 500 bootstrap iterations, which identified older age, male sex, higher education level, and lower baseline scores in MMSE, episodic memory, and executive functions as the most robust contributors to conversion. At the optimal probability threshold of 0.436 (Youden's Index, J = 0.536), the model demonstrated robust discriminative power, specifically achieving a good specificity of 83.9% and a moderate sensitivity of 69.7% in forecasting progression to dementia.

CONCLUSION

Findings underscore that baseline performance in episodic memory, and executive functions can effectively assist in screening for MCI progression. While the model shows good discriminatory potential, its clinical utility lies in risk stratification and identifying high-priority patients for monitoring, rather than serving as a definitive individual prognosis. Further research integrating longitudinal biomarkers is needed to refine these predictive frameworks and improve patient outcomes.