Predicting outcomes in patients with autoimmune glial fibrillary acidic protein astrocytopathy

Prognostic prediction and identification of key factors are essential in managing autoimmune glial fibrillary acidic protein astrocytopathy (A-GFAP-A), a rare, recently identified inflammatory neurological disorder. However, prognostic factors for functional outcomes remain unclear. This retrospective study of A-GFAP-A patients was conducted between 2021 and 2024. Based on 1-month modified Rankin scale score (mRS) after admission, patients were divided into good outcome group (mRS 0–2) and poor outcome group (mRS 3–6). Demographic variables, clinical data, brain MRI, and mRS scores were examined as predictors of prognosis in logistic regression and least absolute shrinkage and selection operator (LASSO) models, respectively. Multiple machine learning classification models were evaluated to select the optimal model, followed by the development of a logistic regression-based nomogram and Shapley Additive exPlanations (SHAP) interpretation for personalized risk assessment. 107 patients (73.8% male, mean age: 42.34; SD 14.69) were included. Common symptoms included headache (82.2%), fever (80.4%), limb weakness (75.7%) and impaired consciousness (50.5%); 97.2% of patients received acute immunotherapy and 62.6% received maintenance immunotherapy.. mRS scores at admission, meningeal enhancement, and brainstem involvement were identified as key predictors of poor outcomes. MRI showed parenchymal lesions in 72 (67.3%), meningeal enhancement in 61 (57%), myelitis in 44 (41.1%) patients. Among the various models evaluated, logistic regression and random forest demonstrated the best predictive performance and the former emerged as the optimal model. We developed and validated a logistic regression-based nomogram and a random forest model to predict A-GFAP-A prognosis, enhancing accuracy and guiding clinical decisions.