Comparison Of Clinical Features And Development And Validation Of Differential Diagnostic Models For Tuberculous And Cryptococcal Meningitis In HIV-negative Patients

【Background & Objectives】Meningitis is a commonly encounterly central nervous system infectious disease with a high incidence in developing countries and with high mortality and disability.A prompt diagnosis and timely treatment are essential for reducing both their mortality and neurological sequelae of meningitis patients.Tuberculous and cryptococcal meningitis are two of the most common types of infectious meningitis.Owing to the overlapped clinical features,neuroimaging and cerebrospinal fluid indices between the two types of meningitis,frequent misdiagnosis and confusion of treatment occurs.It is difficult for clinicians to achieve a rapid and accurate differential diagnosis between the two types of meningitis at the early disease stages.TBM and CM often occur in HIV-positive people.Many studies have compared the clinical characteristics of two types of the meningitis in HIV-positive people and established differential diagnostic model.Early and rapid diagnosis of TBM and CM can be achieved in HIV population by differential diagnostic model.However,there have been relatively few clinical data on TBM and CM in HIV-negative people.Therefore,the goals of this research were:1)To compare the clinical features and investigate the prognostic risk factors of tuberculous meningitis and cryptococcal meningitis in HIV-negative people.2)To establish a nomogram model for the differential diagnosis of tuberculous meningitis and cryptococcal meningitis in HIV-negative people,which can be used by clinicians to achieve rapid and accurate diagnosis of meningitis in patients.3)Logistic regression and machine learning(ML)algorithms were used to identify tuberculous meningitis and cryptococcal meningitis respectively,build a predictive model and evaluate its predictive performance and select the optimal algorithm.【Methods】We retrospectively collected the clinical medical records of patients with TBM and CM admitted to the department of neurology,Xijing Hospital Affiliated to Air Force Military Medical University from June 2008 to December 2019.General information(sex,age),clinical symptoms and signs(fever,headache,seizures,cough,night sweats,blurred vision,hearing impairment,neck stiffness,GCS score),blood routine,the first cerebrospinal fluid examination(CSF cells count,CSF protein,CSF sugar)after admission,neuroimaging(CT or MRI)within one week after admission and Glasgow Coma Scale score within 24 hours after admission.The final outcome of patients was analyzed with Glasgow Outcome Score(GOS)at discharge from the hospital.(1)To compare the clinical features,neuroimaging,cerebrospinal fluid,blood laboratory indexes of CM and TBM patients,and analyze the prognostic risk factors of CM and TBM patients by univariate and multivariate logistic regression,respectively.(2)The study population was divided into training and validation set according to the admission time.The prediction model of nomogram was developed with independent differential factors in the training sets and was validated in the validation sets.(3)In the training set,logistic regression and ML algorithms,including artificial neural network(ANN),Naive Bayes(NB),bagged trees(BT)and random forest(RF)to establish a model.The performance of the models was evaluated by a series of evaluation indicators,including the accuracy,AUC,Hosmer-Lemeshow test,sensitivity,specificity,PPV and NPV in this study.【Results】A total of 292 people were included in this study,including TBM patients(n=227),CM patients(n=65),training set(n=240),and validation set(n=52).(1)The clinical features of the two groups were compared.It was found that compared with TBM patients,CM patients were older(P<0.001),more likely to have autoimmune diseases(P<0.001);TBM patients were more likely to have fever(P= 0.020);CM patients had higher CSF opening pressure(P=0.005);TBM patients had higher CSF white blood cell count(P<0.001),CSF protein(P=0.004),CSF glucose(P=0.014),platelets(P=0.031) and Albumin(P=0.01).Demyelination(P=0.03)was more likely in CM patients.(2)In terms of disease prognosis,GCS score <15 points,cerebrospinal fluid protein increased and hydrocephalus at admission were independent risk factors for poor prognosis of TBM patients.Among CM patients,advanced age is an independent risk factor for poor prognosis of CM patients.(3)Six independent differential factors including age,fever,autoimmune disease,cerebrospinal fluid opening pressure,white blood cell count in cerebrospinal fluid,and glucose in cerebrospinal fluid were selected to build a nomogram model for the differential diagnosis of TBM and CM.The accuracy of the nomogram model was 78.85%,the sensitivity was 80%,the specificity was 76.47%,the PPV was 87.5%,and the NPV was 61.9%,respectively.(4)Based on the ML algorithm,10 variables including cerebrospinal fluid white blood cell count,cerebrospinal fluid glucose,cerebrospinal fluid protein,hydrocephalus,cerebrospinal fluid opening pressure,age,blood sodium,fever,immune abnormalities,and albumin were selected to build ML models.The ML models were compared with the conventional logistic regression model.The study found that the ML model based on the BT algorithm had the best performance,with an accuracy of 0.846,a sensitivity of 0.971,a specificity of 0.588,a positive predictive value of 0.829,a negative predictive value of0.909,and AUC is 0.872.【Conclusions】In non-HIV patients,patients with CM are older and more likely to have autoimmune diseases than patients with TBM.Compared with CM patients,TBM has aggravated inflammatory response(fever symptoms,increase in CSF protein,CSF white cell count,blood albumin and other indicators).In terms of prognosis,among TBM patients,GCS score <15 points,cerebrospinal fluid protein increased,and hydrocephalus at admission were independent risk factors for poor prognosis of TBM patients.Among CM patients,advanced age is an independent risk factor for poor prognosis of CM patients.(2)A nomogram model for differential diagnosis of TBM and CM was successfully constructed based on 6 independent differential factors including age,fever,autoimmune disease,cerebrospinal fluid opening pressure,cerebrospinal fluid white blood cell count,and cerebrospinal fluid glucose.The prediction model obtained good discrimination and calibration ability by verification in the validation set as well as the training set.(3)The differential diagnosis model based on the BT algorithm has the best performance in the validation set and is superior to the traditional logistic regression model.In the future,multi-center and large-sample prospective studies are needed to improve the model performance.The diagnosis level of infectious meningitis in primary hospitals will be improved by the BT model.