Development and validation of a risk score (CHANGE) for cognitive impairment after ischemic stroke

Abstract

Post-stroke cognitive impairment (PSCI) warrants early detection and management. We sought to develop a risk score for screening patients at bedside for risk of delayed PSCI. Ischemic stroke survivors with PSCI and no cognitive impairments (NCI) 3-6 months post-stroke were studied to identify candidate variables predictive of PSCI. These variables were used to develop a risk score using regression models. The score, and the best identified clinical cutoff point, underwent development, stability testing, and internal and external validation in three independent cohorts from Singapore and Hong Kong. Across 1,088 subjects, the risk score, dubbed CHANGE, had areas under the receiver operating characteristics curve (AUROC) from 0.74 to 0.82 in detecting significant risk for PSCI, and had predicted values following actual prevalence. In validation data 3-6 and 12-18 months post-stroke, subjects with low, medium, and high scores had PSCI prevalence of 7-23%, 25-58%, and 67-82%. CHANGE was effective in screening ischemic stroke survivors for significant risk of developing PSCI up to 18 months post-stroke. CHANGE used readily available and reliable clinical data, and may be useful in identifying at-risk patients for PSCI.

Figure 1

Final CHANGE risk score for clinical use.

Figure 2

Calibration graphs plotting the predicted probabilities of the risk score (dashed) against the prevalence of PSCI (solid and marked) at each point of the risk score in model development (A), internal validation (B), and external validation (C). There were no observations at score = 14 in all three cohorts, none at score = 9 and 13 in internal validation, and two at score = 13 in external validation.

Figure 3

Decision curve analysis demonstrating the utility CHANGE in the development (A), internal validation (B), and external validation (C) cohorts at 3–6 months post-stroke. In each subfigure, the utility of using CHANGE (red, green, blue lines) to screen for PSCI for interventional purposes is compared against the utility of adopting a “treat all” approach (broken black line), a “treat none” approach (solid black line), and an approach to treat based on age and education screening (orange line).