Machine learning model predicts prognosis in SARS-CoV-2-infected autoimmune encephalitis patients

To assess the neurological prognosis of patients with autoimmune encephalitis (AE) after severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and construct risk prediction models for a worse prognosis.

This was a prospective, multicenter, observational cohort study. Patients with AE, with or without SARS-CoV-2, were followed up every 3 months. Multivariate logistic regression was used to identify factors influencing neurological prognosis. Prediction models were created using nine machine-learning strategies.

SARS-CoV-2 infection was noted in 241 of 308 patients. Multivariate logistic regression showed inactivated immunizations plus recombinant or adenovirus vector vaccines protected against SARS-CoV-2 infection (odds ratio [OR] 0.05, 95% confidence interval [CI] 0.01-0.52, p = 0.01). Three months after infection, 12.4% of patients with stable AE had a worse neurological prognosis, 6.2% failed to recover their working conditions. Pre-infection modified Rankin Scale (mRS) = 1 (OR 4.06, 95% CI 1.11-14.91, p = 0.04) and mRS > 1 (OR 13.4, 95% CI 3.31-54.02, p < 0.01), immunotherapy during infection (OR 5.1, 95% CI 1.65-15.79, p = 0.01), and SARS-CoV-2-related drowsiness (OR 19.5, 95% CI 5.42-70.34, p < 0.01) and gastrointestinal symptoms (OR 4.4, 95% CI 1.58-11.96, p < 0.01) were identified as risk factors for worse prognosis. The ranger model(https://xingjieli1999.shinyapps.io/clinical_prediction_app/) using these four parameters showed a discrimination accuracy of 0.96 (95% CI 0.94-0.99).

Patients with AE could experience exacerbated neurological symptoms following SARS-CoV-2 infection. Machine-learning algorithms showed feasibility of predicting prognoses based on clinical information in patients with AE.