Establishment and validation of a 3-month prediction model for poor functional outcomes in patients with acute cardiogenic cerebral embolism related to non-valvular atrial fibrillation.

Abstract

Cardiogenic cerebral embolism (CCE) poses a significant health risk; however, there is a dearth of published prognostic prediction models addressing this issue. Our objective is to establish prognostic prediction models (PM) for predicting poor functional outcomes at 3 months in patients with acute CCE associated with non-valvular atrial fibrillation (NVAF) and perform both internal and external validations. We included a total of 730 CCE patients in the development cohort. The external regional validation cohort comprised 118 patients, while the external time-sequential validation cohort included 63 patients. Multiple imputation by chained equations (MICE) was utilized to address missing values and the least absolute shrink and selection operator (LASSO) regression was implemented through the glmnet package, to screen variables. The 3-month prediction model for poor functional outcomes, denoted as N-ABCD2, was established using the following variables: NIHSS score at admission (N), Age (A), Brain natriuretic peptide (BNP), C-reactive protein (CRP), D-dimer polymers (D), and discharge with antithrombotic medication (D). The model's Akaike information criterion (AIC) was 637.98, and the area under Curve (AUC) for the development cohort, external regional, and time-sequential cohorts were 0.878 (95% CI, 0.854-0.902), 0.918 (95% CI, 0.857-0.979), and 0.839 (95% CI, 0.744-0.934), respectively. The N-ABCD2 model can accurately predict poor outcomes at 3 months for CCE patients with NVAF, demonstrating strong prediction abilities. Moreover, the model relies on objective variables that are readily obtainable in clinical practice, enhancing its convenience and applicability in clinical settings.