Background: Bedside prognostication after acute ischemic stroke remains challenging despite the availability of several composite scores, which often demand variables not routinely obtained at admission. Whether the combination of baseline neurological severity and early non-contrast computed tomography findings can support parsimonious, contemporaneous prediction of 3-month functional outcome is an open question, particularly given the well-documented overlap between clinical and imaging measures of stroke extent. Methods: We retrospectively analyzed 432 consecutive adults admitted with confirmed anterior-circulation acute ischemic stroke to a Turkish tertiary care university hospital between January 2022 and December 2025. The primary outcome was a poor functional outcome, defined as modified Rankin Scale (mRS) score 3–6 at 3 months. A prespecified multivariable logistic regression model incorporated three admission variables: baseline National Institutes of Health Stroke Scale (NIHSS) per five-point increment, Alberta Stroke Program Early CT Score (ASPECTS) per one-point increment, and age dichotomized at 65 years. Model performance was assessed with the area under the receiver operating characteristic curve (AUC), five-fold stratified cross-validation, and calibration testing. We additionally performed DeLong tests, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) analyses to evaluate the incremental contribution of ASPECTS, and compared the parsimonious model against a comprehensive multivariable model containing ten predictors. Results: At 3 months, 252 patients (58.3%) had poor functional outcome, with 177 deaths (41.0%). Baseline NIHSS emerged as the dominant predictor (adjusted odds ratio aOR 4.16 per five-point increase, 95% confidence interval CI 2.79–6.22, p < 0.001), whereas ASPECTS retained no independent prognostic value after adjustment (aOR 0.99 per point, 95% CI 0.73–1.34, p = 0.952). Age ≥ 65 years was likewise non-significant (aOR 0.79, 95% CI 0.46–1.37, p = 0.405). The parsimonious model achieved excellent discrimination (AUC 0.904, 95% CI 0.876–0.932; cross-validated AUC 0.899, 95% CI 0.873–0.926) and good overall calibration on internal validation (calibration slope 0.958; calibration-in-the-large 0.009), although the Hosmer–Lemeshow test was significant (p = 0.001), reflecting localized miscalibration in the intermediate-risk range. Adding ASPECTS to a reduced model containing only NIHSS and age produced no meaningful discrimination gain (ΔAUC + 0.0003, DeLong p = 0.50; IDI ≈ 0), consistent with substantial variance sharing between NIHSS and ASPECTS (Spearman ρ = −0.89). A comprehensive model including infarct volume, Glasgow Coma Scale, reperfusion therapy, and comorbidities did not improve discrimination (AUC 0.905) and yielded a worse Akaike information criterion, supporting the sufficiency of the parsimonious model. Conclusions: In this Turkish tertiary center cohort, baseline NIHSS captured most of the prognostic information available at admission, and early ASPECTS added minimal independent value once neurological severity was accounted for. Findings should be interpreted in the context of the high case-mix severity of this referral population; external validation in independent, multicenter cohorts is required before clinical implementation is considered.
Güzel et al. (2026) studied this question.