2026 American Academy of Neurology Autoimmune Abstract Website

Objective:

To identify clinical features that distinguish immune-related encephalitis (irE) from alternative causes of encephalopathy in immune checkpoint inhibitor (ICI) exposed patients, and to characterize predictors of recovery among irE cases.

Background:

ICIs can cause serious immune-related adverse events (irAEs) including irE. Diagnosis of irE is difficult because of a wide range of mimics and lack of diagnostic biomarkers. Existing diagnostic criteria for autoimmune encephalitis were not empirically derived or validated in ICI-exposed patients.

Design/Methods:

Multicenter retrospective case-control study of ICI-exposed patients who developed encephalopathy and diagnosed with either irE (cases) or alternate cause (controls). We reviewed clinical presentations, diagnostics, treatments, and recovery. Graus autoimmune encephalitis criteria and APE2 score were systemically assessed for all patients. Firth penalized logistic regression was used for case-control discrimination and recovery prediction.

Results:

We analyzed 69 irE cases and 72 controls (median age 71 years; 40% female; most common cancers melanoma and lung adenocarcinoma; most common ICIs were pembrolizumab and nivolumab). 74% cases and 15% controls met Graus criteria (p<0.001). Median APE2 scores for cases and controls were 5 (IQR 3-5) and 3 (2-4) respectively (p<0.001). An APE2 score of ≥4 had 67% sensitivity and 72% specificity for irE. Each tier increase in Graus autoimmune encephalitis criteria strongly predicted irE diagnosis (adjusted OR 10.4, 95% CI 4.7–24.9, p<0.001). Augmenting Graus criteria with systemic irAE status and periodic discharges on EEG improved diagnostic discrimination (AUC 0.88 vs. 0.79). Among irE cases, prior CNS radiation (aOR 10.8, 95% CI 1.9–112.3, p=0.005) and each one-point rise in APE2 score (aOR 2.1, 95% CI 1.2–4.1, p=0.003) were both associated with higher odds of incomplete neurologic recovery.

Conclusions:

Empirically derived predictors such as EEG patterns and presence of systemic irAEs enhance existing autoimmune encephalitis diagnostic criteria. Prior CNS radiation and higher APE2 scores are associated with higher odds of incomplete recovery.

Generative AI Usage

Yes, used generative AI in the drafting or editing in this abstract.

Tool, version, and prompt(s) used, as well as area of the abstract affected
The authors used Claude (Anthropic) Sonnet 4.6 for AI-assisted R script generation and iterative refinement throughout the statistical analysis pipeline. Scripts were run independently by authors on R studio and output was verified for data integrity in context of the raw data. Results from these scripts are conveyed in the Results section. Given the iterative process, a series of prompts were used that followed this general structure: "I am building a statistical analysis pipeline in R for a multicenter retrospective case-control study. I have a REDCap-exported dataset in CSV format. Using the provided pre-specified Statistical Analysis Plan (SAP), generate an R script that implements [specific analysis] (e.g. Firth penalized logistic regression) for discrimination of cases from controls and for predictors of recovery status. The script should auto-detect the input file, apply pre-specified inclusion/exclusion criteria, implement the model as specified in the SAP, and export results as formatted tables to a timestamped output directory."