Predictive Value Of Clinical Characteristics In Anti-N-Methyl-D-Aspartate Receptor Encephalitis

Background and Objective:Anti-N-methyl-D-aspartate(anti-NMDA)receptor encephalitis,was first described in 2007 by Dalmau et al.As the most common type of autoimmune encephalitis,anti-NMDA receptor encephalitis has a high diagnostic accuracy rate within neurology,psychiatry,pediatrics,and critical care medicine due to increasing insights into its pathogenesis.This disease presents with a range of clinical symptoms,and has an acute onset with rapid progression,including a variety of clinical manifestations.Seventy to seventy-five percent of patients with anti-NMDA receptor encephalitis are admitted to the intensive care unit(ICU)to receive monitoring and respiratory and circulatory support due to severe disorders of consciousness and central hypoventilation.Antibody diagnosis and immunotherapy are well established,with early diagnoses and aggressive treatment leading to good prognoses.Prognoses are good in 75-81%of patients,with a reported mortality rate of approximately 4%and a recurrence rate of approximately 20-25%.Currently,there are various conclusions about prognostic factors and is a lack of studies on relapse-related predictors.Magnetic resonance imaging(MRI)of brain is the most common auxiliary examination for evaluating the state of the nervous system and is easy to perform;the rate of brain MRI abnormalities at the onset of anti-NMDA receptor encephalitis ranges from 11 to 83%.However,there is no solid evidence that proves and validates the application and importance of relevant imaging findings.It is mostly accepted within the medical literature that the brain MRI findings of anti-NMDA receptor encephalitis are nonspecific and rarely have obvious associations with clinical characteristics,prognoses,and disease recurrence.At present,machine learning prediction model is relatively new and popular,and has good accuracy.This study aimed to comprehensively describe the clinical characteristics and magnetic resonance imaging features of patients with anti-N-methyl-D-aspartate receptor encephalitis,examine their associations,and evaluate their predictive power for long-term disease recurrence and prognosis.The risk model for recurrence and poor prognosis of anti-NMDA receptor encephalitis was constructed using machine learning prediction model,and its predictive value was evaluated.Methods:1.Clinical characteristics and prognostic factors of anti-NMDA receptor encephalitis:This retrospective study included patients diagnosed with anti-NMDA receptor encephalitis admitted to the First Affiliated Hospital of Zhengzhou University from January 2013 to October 2019.The clinical features,auxiliary examinations,treatment regimens,and follow-up were recorded.The follow-up time was 3 months,6 months,1 year,2 years and 3 years since the onset of the disease.Patients were followed up until October 31,2020.The outcomes were 2-year disease relapse,modified Rankin Scale(mRS)scores and death.The factors influencing recurrence were used as covariates in Cox regression analyses,and the prognostic factors were identified using binary logistic regression.2.MRI features and prognostic factors of anti-NMDA receptor encephalitis:We retrospectively extracted the clinical data of anti-NMDA receptor encephalitis patients diagnosed and treated at our tertiary medical center,including brain magnetic resonance imaging findings(site,number,symmetry,outcomes).Patients were subjected to a regular two-year follow-up to assess the disease outcomes,recurrence,and mRS scores.We evaluated the associations of brain magnetic resonance imaging findings at the onset with clinical symptomology,recurrence,and prognosis.The associations between different imaging features and clinical characteristics were assessed via binary logistic regression analyses,the brain magnetic resonance imaging factors influencing recurrence were used as covariates in Cox regression analyses,and the brain magnetic resonance imaging predictor of prognosis were identified using binary logistic regression.3.Construction and evaluation of clinical outcome prediction model for anti-NMDA receptor encephalitis:The population was randomly divided into training set and test set,and statistical test was conducted for each independent variable that might affect clinical outcome.Chi-square test was used for categorical variables,and rank-sum test was used for continuous variables.Screening indicators with significant difference between the disease recurrence group/non-recurrence group and the poor prognosis group/good prognosis group;In the training set,RandomForest model(R3.6.0,RandomForest package)is used for cross validation to select the indicators used for prediction,and then the Random Forest model is established;Calculate the probabilities of various clinical outcomes of the training set and test set through the model,draw the receiver operating characteristic(ROC)Curve,and calculate the Area Under the ROC Curve(AUC).Evaluate model differentiation.Based on the random forest prediction results,the Accuracy(ACC)of the training set and test set was calculated using the confusion matrix method to evaluate the Accuracy of the model.Results:1.Clinical characteristics and prognostic factors of anti-NMDA receptor encephalitis:A total of 160 patients were included.The median age was 26 years(17,41.75),with 67 males(41.9%)and 93 females(58.1%).Tumors were present in 47%of patients,and 95%of patients received first-line immunotherapy(glucocorticoids,intravenous immunoglobulin,and plasmapheresis).Follow-up was 1-7 years(median 2 years).Consequently,6(5%)deaths,34(25.4%)relapses,and 19(15.2%)patients had a poor prognosis(modified Rankin score(mRS)≥3)were recorded.The multivariable analyses showed that age(Hazard Ratio[HR]=1.04,95%Confidence Interval[CI]:1.01-1.07,P=0.011),abnormal magnetic resonance imaging(MRI)(HR=3.03,95%CI:1.20-7.62,P=0.019),glucocorticoid pulse(HR=0.31,95%CI:0.13-0.74,P=0.009),and intracranial pressure(HR=1.01,95%CI:1.001-1.01,P=0.023)were independently associated with the relapse,while age(Odds Ratio[OR]=1.04,95%CI:1.004-1.08,P=0.030)and pulmonary ventilation dysfunction(OR=4.78,95%CI:1.28-17.89,P=0.020)were independently associated with a poor prognosis at 2 years.2.MRI features and prognostic factors of anti-NMDA receptor encephalitis:A total of 144 patients were enrolled(52[36.1%]with normal MRI findings and 92[63.9%]with abnormal/atypical MRI findings);65(45.1%)presented with typical abnormalities while 27(18.8%)presented with atypical abnormalities(white matter lesions[WML],ventriculomegaly,cerebral atrophy);34(29.3%)developed recurrence and 10(9.4%)had poor prognoses(mRS scores>3),including four(3.7%)deaths.Binary logistic regression analyses revealed the associations between MRI features and initial clinical symptoms:insula abnormalities were associated with acute seizure(odds ratio[OR]=3.048,95%confidence interval[CI]:1.026-9.060),and WML was associated with cognitive impairment(OR=2.730,95%CI:1.096-6.799).Risk factors for a poor two-year prognosis included the number of brain MRI abnormalities(OR=1.573,95%CI:1.129-2.192,P=0.007)and intensive care unit(ICU)admissions(OR=15.312,95%CI:1.684-139.198,P=0.015).Risk factors for two-year recurrence were age(hazard ratio[HR]=1.031,95%CI:1.005-1.058,P=0.018)and thalamus abnormalities(HR=2.896,95%CI:1.159-7.238,P=0.023).3.Construction and evaluation of clinical outcome prediction model for anti-NMDA receptor encephalitis:Relapse prediction model(Model 1)included CRP,hospital stay,age,fibrinogen,diastolic blood pressure;The AUC predicted by the Model 1 of training sets was 0.6632(95%CI:0.5342-0.7922),and the AUC predicted by the Model 1 of test sets was 0.6644(95%CI:0.4498-0.879),indicating low differentiation of the model.ACC of the training set was 0.784,and ACC of the test set was 0.810,indicating that the prediction results were in good agreement with the actual results and had good accuracy.The predictive model of poor prognosis(Model 2)included:ICU admission,autonomic nervous dysfunction,pulmonary ventilation dysfunction,urinary protein and cerebrospinal fluid white blood cell count;The AUC predicted by the Model 2 of training sets was 0.7839(95%CI:0.6062-0.9615),and the model differentiation was medium.The AUC predicted by the Model 2 of training sets was 0.678(95%CI:0.1177-0.9444),and the model differentiation was low.The ACC of training set was 0.902,and that of test set was 0.892.The prediction results were consistent with the actual results and had good accuracy.Conclusions:1.Glucocorticoid pulse therapy reduces the relapse of anti-NMDA receptor encephalitis.Age,abnormal MRI,and intracranial pressure are risk factors for relapse,while age and pulmonary ventilation dysfunction are independently associated with poor prognosis.2.The features of MRI abnormalities were closely associated with clinical manifestations,prognoses,and recurrence.insula abnormalities were associated with acute seizure,and white matter lesion was associated with cognitive impairment.More number of brain MRI abnormalities and ICU admissions were predictive of poor prognoses.Age and thalamus abnormalities were risk factors for recurrence.3.Machine learning prediction models have some value in predicting the risk of recurrence and poor prognosis of anti-NMDA receptor encephalitis.