Treatment Delay In Cryptococcal Meningitis And A Diagnostic Model Based On ANN

Part ? Treatment Delay and Acute Stage Prognostic Factors in Cryptococcal MeningitisBackground and objective:Cryptococcal meningitis is a central nervous system mycotic infectious disease caused by the cryptococcus.It is associated with relative high mortality and morbidity.Early diagnosis and prompt effective therapy are crucial for the outcome.Despite the recent advances in diagnosis and treatment for CM,we have observed a lot of treatment delays during clinical work.An earlier research had found significant associations between long-term neurological deficits and treatment delay.However,to the best of our knowledge,the risk factors associated with treatment delay are not yet defined,and it is not known whether the acute stage mortality was related to a delay in treatment.This part aimed to identity the risk factors of treatment delay andassociatedacute stage prognostic factors in CM.Methods:The clinical and laboratory data of adult patients satisfying the diagnostic criteria for cryptococcalmeningitis(n=175)hospitalized at the The First Affiliated Hospital of Zhejiang University during the period 2011-2017 were retrospectively analyzed.Treatment start time was defined as the time interval from onset of symptoms(by patient recall)to initiation of antifungal treatment and was stratified into two categories:early treatment(<20 days)and delay treatment(>20 days)(median delay time is 20 days).Acute stage mortality was defined as death during hospitalization.Univariable and multivariable logistic regression analysis was used to evaluate the determinants of treatment delay and acute stage mortality.Results:The median treatment start time for all CM patients included in the study was 20 days(interquartile range,11-30 days).Multivariate analysis revealed that the first lumbar puncture>14 days(OR = 12.84,95%CI:5.63,31.54),the need for more than one lumbar puncture to obtain a positive smear(OR = 2.67,95%Cl:1.51,5.23)or culture and being misdiagnosed as TBM(OR = 4.38,95%CI:1.06,22.7)had significantly higher risk of the delay.18 patients died in hospital.Multivariate analysis indicated that GCS<15(OR = 7.68;P=0.014;95%CI,1.55 to 43.0),hydrocephalus(OR = 19.6;P==0.005;95%CI,2.42 to 177.0)and plasma albumin<35g/L(OR = 5.2;P=0.035;95%CI,1.16 to 28.1)were independent prognostic factors of acute stage survival.Conclusion:More than half of the CM patients need over 20 days to get appropriate treatment.Our findings indicate that delay of the first lumbar puncture,the need for multiple lumbar puncture to make a definite diagnosisand being misdiagnosed as TBM may lead to delay treatment in CM.Old age,altered mental status,hydrocephalus and hypoalbuminemia predict acute mortality in CM.Part ? Differential Diagnosis of Tuberculous and Cryptococcal Meningitis and a Model Based on ANNBackground and objective:It is difficult to make the differential diagnosis between CM and TBM when the smear and culture are both negative.Currently artificial neural network(ANN)models are apply to different fields in clinical medicine.ANNs are powerful modelling tools,it is used to explore and extract complex non-linear relationships between variables.This part aimed to build an efficient MLP model to help differentiate CM from TBM and compare its performance with a traditional logistic regression model.Methods:In addition to 175 CM patients in the first part,the clinical and laboratory data of another 163 TBM patients were collected.Clinical features were compared between CM and TBM.Multivariable logistic regression analysis was conducted to ascertain the input variables before a logistic regression model and a feed forward MLP model were fitted to the dataset.Dataset randomly divided into two sets;one set of 70%cases for training and an anther set of 30%cases for testing the model.Each of the model was run 10 times to access their efficacy after being built.The accuracy of the models in predicting CM was compared by sensitivity,specificity and classification accuracy.Results:A stepwise logistic regression analysis found nine factors could be used to differentiate CM from TBM:age,fever,CSF glucose,CSF chloride,CSF protein,the first intracranial pressure after admission,erythrocyte sedimentation rate,predisposing factor,a positive T-SPOT and peripheral tuberculous infections.A MLP model with 10 input variables and 1 hidden layers including 45 neurons was built.The sensitivity,specificity and AUC of MLP model was 82.7 ± 5.2%(95%CI,79.0 to 86.6),80.8 ±3.3%(95%CI,78.5 to 83.2)and 0.89 ± 0.02(95%CI,0.87 to 0.90),espectively.Sensitivity,specificity and AUC did not differ significantly between MLP model and logistic regression model.Conclusion:Based on this dataset,the MLP model had a high sensitivity and specificity in diagnosing CM and could be applied to differentiate CM from TBM.The model need further validation before being used in clinics,especially by a prospective research and external data.