We aimed to develop and externally validate a prognostic model for unfavorable disability outcome in patients with COVID-19-associated stroke based on multicenter data from the NIH-funded NEUROCOVID Databank. 

We obtained patient data from the NeuroCOVID Databank after receiving data access/IRB approvals. The derivation cohort consisted of stroke cases from 13 USA centers (n=205), and the validation cohorts of cases from Asia (n=68) and Egypt (n=123). We included adult cases with ischemic/hemorrhagic stroke and a hospital admission associated with COVID-19. We assessed 15 candidate predictors, including patient-, COVID-19-, and stroke-related factors. Unfavorable disability outcome was defined as modified Rankin Score (mRS) >3 at hospital discharge. We applied Median Aggregation of penalized Coefficients after Multiple Imputation (MALCoM) for variable selection and model development in multiply imputed datasets. Model Performance was evaluated using ROC curves (AUC), and accuracy was estimated in the derivation and test datasets. Subgroup analysis by stroke type was conducted.

396 patients were included: 205 in the USA development cohort, 123, and 68 in Egypt and Asia validation cohorts. Predictors’ distribution significantly varied between cohorts; however, the outcome distribution didn’t. In the development cohort, the model showed an AUC of 0.84 (0.78-0.9). Model accuracy was higher in ischemic versus hemorrhagic stroke subgroups (0.86 vs. 0.65). In the validation cohorts, receiver operating characteristic analysis yielded AUCs of 0.61 (Egypt ) and 0.51(Asia).

The model shows good performance in the development cohort and decreased performance in the validation cohorts, likely due to striking differences in patient characteristics between the large development and small validation cohorts. Extra validation in larger cohorts is needed before using the model in clinical practice.