Multicenter Study Uncovers Distinct Prognostic Patterns in Autoimmune Encephalitis

A retrospective multicenter cohort study across eleven tertiary medical centers in China has revealed that patients diagnosed with autoimmune encephalitis experience fundamentally distinct prognostic pathways depending on autoantibody detection and initial clinical presentation. The investigation, which evaluated clinical trajectories in 498 patients admitted between January 2016 and October 2024, showed that while one-year functional outcomes in antibody-positive cases correlated directly with neurological disability metrics at hospital discharge, outcomes in antibody-negative cases were governed primarily by advanced age at onset and atypical initial manifestations. Furthermore, the analysis established that patient vulnerability to delayed immunotherapeutic intervention was tightly restricted to those presenting with atypical clinical features rather than classic canonical manifestations, challenging prevailing assumptions that all subtypes of autoimmune encephalitis respond uniformly to treatment timing.

Autoimmune encephalitis represents a severe and rapidly progressive encephalopathy with heterogeneous clinical trajectories spanning from full functional recovery to persistent disability or death. In recent clinical practice, standardized scoring frameworks such as the anti-N-methyl-D-aspartate receptor Encephalitis One-Year Functional Status score and the Clinical Assessment Scale in Autoimmune Encephalitis have offered clinicians standardized instruments to evaluate initial severity and predict recovery. However, these clinical assessment tools were derived predominantly from antibody-positive patient populations. Consequently, their applicability to antibody-negative cases—an increasingly recognized diagnostic subgroup diagnosed primarily by exclusion based on established consensus criteria—has remained poorly characterized, leaving clinicians with minimal guidance for early risk stratification and escalation of immunomodulatory treatments.

To rigorously examine these clinical disparities, the participating research group analyzed 498 verified cases, comprising 382 seropositive individuals representing 76.7 percent of the cohort and 116 seronegative individuals accounting for 23.3 percent. Among the antibody-positive group, the most prevalent autoantibody identified was against the N-methyl-D-aspartate receptor in 156 cases, followed by leucine-rich glioma-inactivated 1 in 132 patients, gamma-aminobutyric acid type B receptor in 46, contactin-associated protein-like 2 in 20, and metabotropic glutamate receptor 5 in 11 patients. Overall, 64 patients, or 12.9 percent of the entire cohort, experienced an unfavorable functional outcome at 12 months, formally defined as a modified Rankin Scale score greater than 2. Among the antibody-positive individuals, 46 of 382 experienced poor outcomes, including 21 deaths, whereas 18 of 116 antibody-negative patients had poor outcomes, including 10 recorded deaths.

Methodologically, the investigators implemented a two-stage analytical model designed to avoid collinearity and adjust for institutional variance across the 11 contributing hospitals. Features were first screened using a Least Absolute Shrinkage and Selection Operator regression with ten-fold cross-validation to select parameters yielding the minimum cross-validation error. Variables identified in this step were subsequently evaluated using multivariable fixed-effects logistic regression as well as generalized linear mixed-effects models where the treating hospital was treated as a random intercept. This mixed-effects framework allowed the team to measure center-level variance through the intraclass correlation coefficient, thereby accounting for potential variations in critical care protocols, specialist availability, and diagnostic timing between tertiary care centers.

Within the overall cohort of 498 patients, the mixed-effects regression analysis demonstrated that discharge Clinical Assessment Scale in Autoimmune Encephalitis scores were significantly associated with an unfavorable 12-month outcome, showing an adjusted odds ratio of 1.60 with a 95 percent confidence interval ranging from 1.25 to 2.06. Similarly, discharge modified Rankin Scale scores exhibited an adjusted odds ratio of 2.55 with a confidence interval between 1.57 and 4.15. Patient age at disease onset of 65 years or older demonstrated an odds ratio of 6.42 with a confidence interval of 1.53 to 27.02, while initial presentation with atypical features—specifically predominant impairment of consciousness, language deficits, movement disorders, or ataxia—yielded an odds ratio of 3.28 with a confidence interval of 1.02 to 10.52. The overall intraclass correlation coefficient was calculated at 0.055 with a p-value of 0.23, demonstrating that only 5.5 percent of total outcome variance was attributable to institutional differences across hospitals.

When examining patients strictly by serostatus, striking divergences emerged between the two cohorts. Among the 382 antibody-positive patients, multivariable mixed-effects models confirmed that clinical impairment persisting at discharge remained the primary driver of poor 12-month recovery: discharge Clinical Assessment Scale scores yielded an odds ratio of 1.56, while discharge modified Rankin Scale scores produced an odds ratio of 3.74. Advanced age at onset of 65 years or older was also an independent predictor, associated with an odds ratio of 5.78. Center-level variance yielded an intraclass correlation coefficient of 0.088. Although baseline severe features such as intensive care unit admission and tumor comorbidity—present in 28.3 percent of poor outcome cases versus 6.5 percent of favorable cases—were frequent, the failure to achieve prompt functional improvement prior to discharge remained the decisive independent metric of long-term disability.

Conversely, in the subgroup of 116 antibody-negative patients, standard discharge metrics failed to retain predictive significance in either fixed-effects or mixed-effects multivariable regression models. Neither discharge Clinical Assessment Scale scores nor modified Rankin Scale scores demonstrated a statistically significant association with one-year functional status. Instead, poor functional outcomes were predominantly determined by initial atypical clinical presentation, which demonstrated an adjusted odds ratio of 26.77 in the generalized linear mixed-effects model with a wide confidence interval ranging from 1.76 to 406.72, and an age at onset of 65 years or older, which yielded an odds ratio of 29.59 with a confidence interval from 1.16 to 755.94. The intraclass correlation coefficient in this subgroup was estimated at approximately 0.000, indicating negligible institutional variance across centers.

The study also evaluated whether clinical phenotype altered the therapeutic window for initiating immunotherapy. Delayed immunotherapy, defined as beginning treatment 30 days or more after the initial onset of symptoms, was evaluated using interaction terms in multivariable models. A statistically significant interaction was detected specifically between delayed treatment and atypical clinical presentations, showing an interaction p-value of 0.024. Among individuals presenting with atypical manifestations, receiving immunotherapy after 30 days was associated with an adjusted odds ratio of 9.16 for an unfavorable outcome in the mixed-effects analysis, with a confidence interval between 1.31 and 63.93. Conversely, no statistically significant association with delayed intervention was observed among patients presenting with canonical manifestations, such as predominant seizures, memory loss, or neuropsychiatric disturbances.

Drawing upon these multivariable prognostic determinants, the researchers constructed and internally validated a bedside prognostic nomogram designed to calculate an individual patient’s probability of poor functional recovery at 12 months. The model assigned weighted point values to discharge Clinical Assessment Scale scores, discharge modified Rankin Scale scores, initial clinical phenotype, categorical age brackets, and tumor comorbidity. Internal validation conducted through 1000 bootstrap resamples demonstrated high discriminative performance, achieving an area under the receiver operating characteristic curve of 0.909 with a 95 percent confidence interval of 0.86 to 0.96. Calibration assessment revealed close agreement between predicted and observed probabilities, documenting a mean absolute error of 0.022 across the validation iterations.

The authors outlined several critical methodological constraints inherent to the research design. Due to its retrospective nature, the investigation carried potential risks of unmeasured confounding and selection bias. Furthermore, the antibody-negative subcohort comprised a modest sample of 116 patients with 18 poor outcome events, resulting in wide confidence intervals for key predictor estimates that require cautious clinical interpretation and independent validation in larger prospective cohorts. Additionally, because the study population was derived exclusively from tertiary referral institutions managing complex caseloads, the findings may carry referral bias and may not generalize completely to secondary or community hospital environments, even though low intraclass correlation coefficients indicated consistent management standards across the tertiary network.

Subject of Research: Prognostic determinants and therapeutic windows in autoimmune encephalitis

Article Title: Serostatus‐Specific Prognostic Profiles and Phenotype‐Dependent Therapeutic Windows in Autoimmune Encephalitis: A Multicenter Cohort Study

Article References: Tang, X., Wang, J., Zhu, K., Zhang, Z., Wu, L., Wang, W., Li, C., Li, Q., Li, H., Chen, J., Lian, Z., Liu, B., Liu, T., Wen, C., Wang, N., Qun, S., Xiao, L., & Lai, Q. (2026). Serostatus‐Specific Prognostic Profiles and Phenotype‐Dependent Therapeutic Windows in Autoimmune Encephalitis: A Multicenter Cohort Study. Annals of Clinical and Translational Neurology, Article acn3.70571. https://doi.org/10.1002/acn3.70571

Image Credits: AI Generated

DOI: 10.1002/acn3.70571

Keywords: autoimmune encephalitis, neuroimmunology, prognostic models, delayed immunotherapy, mixed-effects modeling, neurology research, antibody-negative encephalitis, clinical neurology, Serostatus, Specific, Prognostic, Profiles