Related Experiment Videos
Development and validation of a risk-stratified prediction model for post-stroke vascular dementia: Clinical
Bing Wei1, Weimin Shi2, Guangcheng Ji1
1Center for Encephalopathy, The Third Clinical Hospital Affiliated to Changchun University of Traditional Chinese Medicine, Jilin 130117, China.
Background:
Post-stroke vascular dementia (VaD) affects 20-30% of ischemic stroke survivors within the first year, yet existing prediction models lack comprehensive integration of novel blood biomarkers and validated risk stratification strategies. This study aimed to develop and temporally validate a risk-stratified prediction model incorporating clinical, neuroimaging, and serum biomarker variables.
Methods:
This retrospective cohort study comprised a development cohort (n = 998, 2020-2022) and temporal validation cohort (n = 249, 2023-2024). Consecutive acute ischemic stroke patients aged ≥ 18 years with available baseline magnetic resonance imaging (MRI) and 12-month follow-up were included. Candidate predictors encompassed 25 variables: demographics, vascular risk factors, stroke severity assessed by the National Institutes of Health Stroke Scale (NIHSS), neuroimaging markers (Fazekas white matter hyperintensity score, brain atrophy index [BAI]), and serum biomarkers including neurofilament light chain (NFL) and glial fibrillary acidic protein (GFAP) measured by single-molecule array (Simoa). The primary outcome was incident VaD diagnosed by NINDS-AIREN criteria at 12 months.
Results:
Of 1247 included patients, 354 (28.4%) developed VaD. Least absolute shrinkage and selection operator (LASSO) selected 8 predictors: age, education level, NIHSS score, Fazekas score ≥ 2, BAI, previous stroke, plasma NFL, and plasma GFAP. In the development cohort, the model demonstrated excellent discrimination (C-statistic 0.89, 95% CI 0.86-0.92) and good calibration. Bootstrap validation yielded optimism-corrected C-statistic 0.88. In temporal validation, performance remained robust (C-statistic 0.85, 95% CI 0.81-0.89). Risk stratification revealed distinct cognitive trajectories: high-risk patients (25% of cohort) exhibited 67.3% VaD incidence and steep cognitive decline (mean Montreal Cognitive Assessment [MoCA] change -6.8 points), capturing 59.3% of all VaD cases.
Conclusions:
This biomarker-enhanced prediction model demonstrates excellent discrimination and calibration for post-stroke VaD. Risk stratification effectively identifies high-risk patients for targeted interventions, providing a practical tool for precision-based clinical management.
