基于可解释性机器学习构建ICU脓毒症相关性脑病患者预后风险预测模型OA

Objective To construct a prognostic risk prediction model for patients with sepsis-associated encephalopathy(SAE)in the intensive care unit(ICU)based on interpretable machine learning.Methods The clinical data of SAE patients admitted to the ICU of the First Affiliated Hospital of Xinjiang Medical University from January 2018 to June 2023 were retrospectively collected.The indicators included gender,age,infection sites(pulmonary,abdominal,urinary tract,thoracic,skin and soft tissue,hematogenous),comorbidities(hypertension,diabetes,cancer,cardiovascular diseases,chronic kidney disease),diagnosis and therapeutic measures[vasoactive drugs,continuous renal replacement therapy(CRRT),surgery,mechanical ventilation],vital signs within 24 hours of ICU admission(body temperature,heart rate,respiratory rate,systolic blood pressure,diastolic blood pressure),laboratory indicators[pH value,fractional inspired oxygen(FiO2),arterial partial pressure of oxygen(PaO2),arterial partial pressure of carbon dioxide(PaCO2),oxygenation index(PaO2/FiO2),blood lactic acid(Lac),serum sodium(Na+),serum potassium(K+),hematocrit(HCT),monocyte count(MON),C-reactive protein(CRP),interleukin-6(IL-6),procalcitonin(PCT),white blood cell count(WBC),neutrophil count(NEU),lymphocyte count(LYM),red blood cell count(RBC),hemoglobin(Hb),platelet count(PLT),blood ammonia(AMM),serum cholinesterase(ChE),alkaline phosphatase(ALP),gamma-glutamyl transpeptidase(GGT),alanine transaminase(ALT),aspartate aminotransferase(AST),AST/ALT ratio,globulin(GLB),albumin(Alb),total protein(TP),indirect bilirubin(IBil),direct bilirubin(DBil),total bilirubin(TBil),creatinine(Cr),blood urea nitrogen(BUN),brain natriuretic peptide(BNP),prothrombin time(PT),activated partial thromboplastin time(APTT),fibrinogen(Fib),international normalized ratio(INR)],as well as critical care scoring systems[acute physiology and chronic health evaluationⅡ(APACHEⅡ),sequential organ failure assessment(SOFA),Glasgow coma scale(GCS)]and ICU length of stay.The 28-day mortality was defined as the primary outcome.Variables with statistically significant differences between the survival group and the death group were screened,and the least absolute shrinkage and selection operator(Lasso)regression was used for further variable selection.Six reliable machine learning algorithms were applied to establish prediction models,and multiple evaluation metrics were adopted to assess model performance.Shapley additive explanation(SHAP)was used to rank the feature importance of the optimal model and realize visualized individual outcome prediction,so as to improve clinical understanding of model mechanism and meet the demand for personalized prognostic assessment.Results A total of 506 SAE patients were enrolled,among whom 243 cases(48.02%)died within 28 days.Using Lasso regression analysis to compress variables and eliminate multicollinearity,11 independent predictive features were identified:sex,age,vasoactive agents,mechanical ventilation,surgery,CRRT,SOFA score,Na+,PCT,NEU,and APTT,respectively.Comprehensive evaluation results showed that the extreme gradient boosting(XGBoost)model presented the best and most stable predictive performance.The areas under the receiver operator characteristic curve(AUC)of the training set and validation set were 0.986 and 0.849,respectively,which were higher than those of other models.SHAP analysis indicated that advanced age,elevated SOFA score and increased NEU were independent risk factors for 28-day mortality in SAE patients.Conclusions The XGBoost model possesses excellent predictive efficacy for short-term prognosis of ICU patients with SAE.It can assist clinicians in early identification of high-risk patients and facilitate the timely implementation of targeted therapeutic strategies.