Construction And Application Of Prognostic Scoring Model For Secondary Infection Of Central Nervous System Injury

Objective:Secondary infection follow by central nervous system（CNS）injuries has a high incidence and is difficult to avoid,and its impact on clinical outcomes cannot be accurately estimated.Based on the dynamic changes in GCS and blood indicators,this study is the first attempt to build a scoring model for predicting the prognosis of patients with secondary infection after central nervous system injury（PSICNSI）during their intensive care unit（ICU）stay.Methods:（1）This study was a retrospective case-control study,patients admitted to the ICU of Guangxi Medical University the First Affiliated Hospital from March 2015 to December 2019 that clearly diagnosed with secondary infection after stroke or traumatic brain injury were selected as subjects.With the method of accidental sampling,January 1^st,2019 was taken as the cut-off time point.Inpatients before the time point were included in the model test set whose data were used to construct prediction models,while the rest inpatients were included in the validation set whose data were used to verify the validity of models.（2）Patients’medical records were reviewed retrospectively,information materials such as patients’baseline characteristics,comorbidities,diagnosis treatment measures,clinical disease score,various blood indicators and ICU outcome were recorded at the starting and the end of the study.（3）ICU mortality was used as the primary outcome,based on which patients in the test set were divided into two groups.Univariate analysis and multivariate logistic regression analysis were carried out on each indicator of the test set at the starting point and the end point of the study respectively,and the probability of ICU death at the corresponding time point was calculated according to final regression equations（p-admission and p-discharge）.Predictive abilities of probability and acute physiology and chronic health evaluation scoring system（APACHEⅡ）were shown as the area under the curve（AUC）which analyzed through Receiver Operating Characteristic（ROC）curves.（4）After stratifying and assigning the risk factors in two regression equations,PSICNSI-admission prognostic model for the starting point（representing the initial stage of the disease）and the PSICNSI-discharge for the end point（representing the stage of disease progression）of the study were initially constructed.The PSICNSI prediction models underwent self-verification and cross-verification in the test set and the verification set through ROC analysis compared to corresponding APACHE Ⅱ,respectively.Results:1.A total of 297 patients were selected as the study objects according to the inclusion and exclusion criteria,including 219 patients in test set with 51 ICU deaths（23.3%）and 78 patients in verification set with 13 ICU deaths（16.7%）.2.Univariate analysis of the test set showed that at the start of the study（-ad）,there were statistical differences between the death group and the survival group in APACHE Ⅱ,Glasgow Coma Score（GCS）,blood platelet counts（PLT）,white blood cell counts（WBC）,activated partial thromboplastin time（APTT）,percent of neutrophile granulocyte（%N）,percent of lymphocyte（%L）,serum procalcitonin（PCT）,lactic acid（Lac）（P<0.05）,while at the end of the study（-dis）,there were statistical differences between two groups in APACHE Ⅱ、GCS、C-reactive protein（CRP）、PCT、Lac、oxygenation index（OI）at discharged and whether blood transfusion treatment during hospitalization,types of blood transfusion,volume of blood transfusion,erythrocyte transfusion,plasma transfusion,platelet transfusion,minimum value（-min）of PLT,maximum value（-max）of APTT,prothrombin time（PT）,international standard ratio（INR）,WBC,%N,%L,PDW,CRP and PCT（P<0.05）.3.Multivariate Logistic regression analysis was performed on indicators that different between two groups in univariate analysis at various time points of this study,respectively.It showed that at the starting point,GCS-ad,PLT-ad,WBC-ad and PCT-ad were independent risk factors for ICU death,while at the end point,GCS-dis,CRP-dis,OI-dis and PLT-min were independent risk factors for ICU mortality.ROC analysis was conducted to compare death probability which calculated by regression equations with the corresponding Apache Ⅱ showed that regardless of the starting point（0.868,95%CI 0.806-0.929 vs.0.796,95%CI 0.725-0.866）or the end point（0.980,95%CI 0.963-0.998 vs.0.964,95%CI 0.943-0.986）,the AUC value of the predicted probability was higher than that of Apache Ⅱ,indicating that regression models had good predictive validity.4.The independent risk factors at the starting point and the end point were stratified and assigned,and the PSICNSI-admission and PSICNSI-discharge scoring models along with points-probabilities comparison tables were constructed.The items in the PSICNSI-admission model were assigned as follows:GCS-ad（0～8 points）,PLT-ad（-4～+5 points）,WBC-ad（-1～+6 points）,PCT-ad（-1～+9 points）with a total score ranged from-6 to+28,the corresponding ICU death predictive probability was 0.0001～0.9996;the items in the PSICNSI-discharge model are assigned as follows:GCS-dis（-8～+6points）,PLT-min（-6～+6 points）,OI-dis（-7～+5 points）,CRP-dis（-1～+9 points）,with a total score ranged from-22 to+26,the corresponding ICU death predictive probability was 0.0000～0.9993.5.In the test set,ROC analysis was used to evaluate the predictive validity display of PSICNSI-admission and PSICNSI-discharge models,respectively:at the starting point,the cut-off point value of the PSICNSI-admission model was 8,that is,the risk of death in ICU was high when the predicted total points was≥8 while the risk of death in ICU was low when the total score was less than 8,the sensitivity and specificity of the scoring model was 88.20%and 67.90%,with a AUC value of 0.855;at the end point,the cut-off point value of the PSICNSI-discharge model was 1,that is,the risk of death in ICU was high when the predicted total points was≥1 while the risk of death in ICU was low when the total score was less than 1,the sensitivity and specificity of the scoring model was 98.00%and 85.10%,with a AUC value of 0.975.ROC analysis of scoring models and APACHE Ⅱ at the same time showed that the AUC value of PSICNSI was higher than that of APACHE Ⅱ at the starting point（0.855,95%CI 0.791-0.918 vs.0.796,95%CI0.725-0.866）and the end point（0.975,95%CI 0.956-0.994 vs.0.964,95%CI0.943-0.986）,suggesting that PSICNSI scoring models had good predictive validity.6.In the validation set,ROC analysis was used to evaluate predictive validity of PSICNSI models for ICU prognosis,at the starting of the study,the cut-off point value of the PSICNSI-admission model was 11,means that the predicted total score was≥11 with a high risk of ICU death while the total score was<11 with a low risk of death in ICU,the sensitivity and specificity of the score model were 69.20%and 92.30%,with a AUC value of 0.847;at the end point of the study,the cut-off point value of the PSICNSI-discharge model was 5,means that the predicted total score was≥5 with a high risk of death in ICU while the total score was<5 with a low risk of ICU death,the sensitivity and specificity of the score model were 92.30%and 87.70%,with a AUC value of 0.936.Conclusion:1.The influenced factors of ICU outcomes in patients with secondary infection after CNS injury at different stages（initial and progressive）were different.GCS,PLT,WBC and PCT were excellent prognostic indicators in the early stage of infection,while at the progressive stage the best predictors were GCS,OI,CRP and lowest PLT during ICU stay.2.The PSICNSI scoring models had higher prognostic efficacy on ICU prognosis than that of APACHE Ⅱ,the predictive validity of PSICNSI scoring model in progressive stage was higher than that in the early stage.3.PSICNSI scoring models had good abilities to predict the prognosis of patients with secondary infection after CNS injury,it is simple and easy to operate,and is worthy of clinical application and promotion.