Predicting the functional outcomes of anti-LGI1 encephalitis using a random forest model

Area of Science:

• Neurology
• Immunology
• Artificial Intelligence in Medicine

Background:

• Anti-leucine-rich glioma-inactivated 1 (LGI1) encephalitis is a severe autoimmune neurological disorder.
• Predicting functional outcomes in LGI1 encephalitis is crucial for patient management and treatment strategies.

Purpose of the Study:

• To develop a predictive model for functional outcomes in patients with anti-LGI1 encephalitis.
• To identify key factors influencing neurological recovery using a random forest algorithm.

Main Methods:

• Retrospective review of 79 patients with confirmed anti-LGI1 antibodies (January 2015 - July 2020).
• Functional outcomes assessed using modified Rankin Scale (mRS); model developed with random forest algorithm.
• Model performance compared against logistic regression, Naive Bayes, and Support Vector Machine (SVM) using AUC and accuracy.

Main Results:

• A random forest model utilizing 16 variables predicted poor functional outcomes in anti-LGI1 encephalitis with 83% accuracy and 60% F1 score.
• The random forest model showed superior predictive performance (AUC 0.90) compared to logistic regression (0.80), Naive Bayes (0.70), and SVM (0.64).
• 25% of patients experienced poor functional outcomes after a median follow-up of 24 months.

Conclusions:

• The random forest model effectively predicts poor functional outcomes in anti-LGI1 encephalitis.
• This AI-based model offers enhanced accuracy and reliability over conventional statistical methods for predicting LGI1 encephalitis prognosis.