Development and validation of standard-criteria and age-corrected nomograms for post-stroke cognitive impairment risk stratification

Objective The investigation aimed to develop and validate two complementary prognostic nomograms for post-stroke cognitive impairment (PSCI) among acute ischemic stroke patients. Methods In this prospective cohort study, 336 patients were enrolled for model development and internal validation, with 48 patients for external validation. Cognitive performance was evaluated using the Montreal Cognitive Assessment (MoCA) at six months post-stroke. The standard-criteria model defined PSCI as MoCA <26, while the age-corrected model applied age-specific cutoffs (<26 for <60 years, <25 for 60–69, <24 for 70–79, <23 for ≥ 80). Data on demographics, vascular risk factors, stroke features, neuroimaging, and biochemical markers were collected. The least absolute shrinkage and selection operator (LASSO) logistic regression was utilized to identify predictors and construct the nomogram. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, calibration curves, and decision curve analysis (DCA). Results Both models identified the same f ive independent predictors of PSCI: advanced age (standard-criteria: OR 1.18, 95% CI 1.11–1.26; age-corrected: OR 1.13, 95% CI 1.07–1.19), female gender (standard-criteria: OR 4.71, 95% CI 1.67–14.84; age-corrected: OR 4.88, 95% CI 1.86–12.83), elevated low-density lipoprotein cholesterol (LDL-C) (standard-criteria: OR 4.50, 95% CI 2.35–9.46; age-corrected: OR 3.25, 95% CI 1.78–5.91), key area cerebral infarction (standard-criteria: OR 6.22, 95% CI 2.58–16.25; age-corrected: OR 5.83, 95% CI 2.55–13.33), and global cortical atrophy (GCA) scale grade ≥ 2 (standard-criteria: OR 8.50, 95% CI 1.99–41.64; age-corrected: OR 5.39, 95% CI 1.42–20.52). The standard-criteria model demonstrated excellent discriminative ability in the training (AUC = 0.935, 95% CI 0.904–0.965), internal validation (AUC = 0.929, 95% CI 0.882–0.976), and external validation cohorts (AUC = 0.884, 95% CI 0.793–0.976), with precision of 0.88–0.91, recall of 0.91–0.93, specificity of 0.80-0.86, and F1-scores of 0.89–0.92. The age-corrected model showed comparable performance (AUC = 0.912 training, 0.905 internal validation, 0.776 external validation), with precision 0.86–0.89, recall 0.89–0.91, specificity 0.79-0.84, and F1-scores 0.88-0.90. Both models showed balanced performance, identifying both PSCI and non-PSCI patients effectively. Calibration plots confirmed strong agreement between predicted and observed outcomes, and DCA revealed substantial clinical net benefits for both models. Conclusion The developed dual nomograms, incorporating readily accessible clinical and imaging predictors, offer robust and practical tools for early risk stratification of PSCI in acute ischemic stroke survivors, facilitating targeted interventions.