| Objective: Cardiogenic shock is characterized by acute onset,critical condition,and poor prognosis,and the incidence of cardiovascular diseases in Xinjiang is higher than that in other regions of China.Therefore,it is necessary to find out the relevant risk factors reflecting the prognosis of patients with cardiogenic shock and build prediction models,as well as verify and optimize them based on the local population of Xinjiang.In this study,the MIMIC III database(Medical Information Mart for Intensive Care III)was used to screen out the relevant risk factors reflecting the prognosis of cardiogenic shock.Relevant risk factors were obtained to construct a predictive model that reflects the prognosis of cardiogenic shock,and the predictive power of the obtained predictive model was compared with that of the existing critical illness prognosis score system.The obtained prediction model was used to verify the prediction ability of the population in Xinjiang,and to find new biomarkers that reflected the prognosis and optimized the model accordingly.Methods: 1)We applied for access to MIMIC database and used structured query language to extract data.According to the inclusion and exclusion criteria,a total of 1,131 patients with cardiogenic shock were extracted,including 662 males and 469 females.2)Admission data were collected for patients with cardiogenic shock,such as demographic parameters,vital signs,laboratory parameters,comorbidities,and critical illness scores.Extract indicators include whether combination of coronary artery disease,congestive heart failure,atrial fibrillation,and other diseases.Red blood cell distribution width,bicarbonate,anion gap and other laboratory indicators were also extracted,as well as sequential organ failure assessment(SOFA)and simplified acute physiology score ?(SAPS ?).In this study,the 30-day all-cause mortality was the primary clinical endpoint,while the 90-day and 365-day all-cause mortality were thesecondary endpoints.3)We used methods such as univariate analysis and Cox proportional hazard regression model to screen out independent risk factors for poor prognosis of cardiogenic shock,and produced receiver operating characteristic curve(ROC)for each independent risk factor.By calculating the area under the curve,the predictive ability of each risk factor for the prognosis of cardiogenic shock was tested.In addition,a generalized additive model was used to identify the relationship between various risk factors and poor prognosis.4)Independent risk factors with significant influence on the prognosis of patients with cardiogenic shock in MIMIC database were used to construct nomograms of 30,90 and 365-day all-cause mortality according to Akaike information criterion(AIC).The area under the curve was used to evaluate the advantage of the model,and the calibration curve was used to judge its prediction compliance,Delong method was used to compare the prediction ability of different models.5)The new model constructed in MIMIC III was further validated in Xinjiang patients,and the area under the ROC curve was used to evaluate the prediction ability of the model.The new model was optimized by using AIC value for variable selection.Similarly,area under ROC curve was used to evaluate the merits of the optimal model for the prognosis of cardiogenic shock(OMPCS).The calibration curve was used to judge its prediction compliance,Delong method was used to compare the prediction abilities of different models(OMPCS,SOFA,and APACHE ?).Results: 1)In multivariate analysis,adjusted for ethnicity,gender,and other confounding factors,we found age,blood urea nitrogen,red blood cell distribution width,lactate,blood glucose,and anion gap were significant predictors of risk of all-cause mortality at 30 days,90 days,365 days.Age,blood urea nitrogen,red blood cell distribution width,lactate,blood glucose,and anion gap had poor predictive power for the prognosis of patients with cardiogenic shock,the corresponding AUC values were 0.639,0.634,0.600,0.686,0.588,and 0.675.The curve fitting results showed that age,blood glucose and anion gap values had an approximately U-shaped relationship with the 30-day all-cause mortality,while blood urea nitrogen,red blood cell distribution width,and lactate were nearly positive linear correlation with 30-day all-cause mortality.2)The selection of modeling variables was based on the AIC value.The new model=-6.27856 + 0.00492 * blood glucose + 0.14570 * lactate + 0.01282 * blood urea nitrogen + 0.03167 * age + 0.09829 * red blood cell distribution width,and the corresponding nomogram was constructed,with the nomogram AUC=0.7302.The calibration curve showed that the predicted probability comparison line was close to the ideal completely consistent line,with the average absolute error=0.038.A similar approach yielded nomograms of 90-day,365-day all-cause mortality.The new model,SOFA score,and SAPS ? score were used to draw ROC curves to assess the prognosis of patients with cardiogenic shock,the corresponding AUC values were new model: 0.7302(95%confidence interval: 0.6936~0.7688),SOFA score: 0.6773(95%confidence interval: 0.6379~0.7166),SAPS ? score: 0.7226(95%confidence interval: 0.6852~0.7600).The AUC of the new model was greater than the SOFA score and the comparison between the two models was statistically significant(P=0.0193).The AUC of the new model was also greater than the SAPS ? score,but the comparison was not statistically significant(P=0.6753).3)The new model from MIMIC III,which reflected the 30-day mortality risk of cardiogenic shock,was brought to the Xinjiang patients with cardiogenic shock for verification.The verified AUC=0.7515(95%confidence interval: 0.6878~0.8152).The risk factors of age,blood urea nitrogen,red blood cell distribution width,lactate,and blood glucose,NT-proBNP and cTnI were selected according to AIC value,and the model was optimized.OMPCS=-5.27820 +0.39690* lactate +0.20704* red blood cell distribution width +0.10486* cTnI +0.00014* NT-proBNP,and the AUC of OMPCS=0.8525.The calibration curve showed that the predicted probability comparison line was close to the ideal completely consistent line,with an average absolute error=0.045.The OMPCS,new model,SOFA score,and APACHE ? score were used to draw ROC curves to assess the prognosis of patients with cardiogenic shock,corresponding AUC values were OMPCS: 0.8525(95%confidence interval: 0.7943~0.9116),and the new model: 0.7515(95%confidence interval: 0.6878~0.8152),SOFA score: 0.7308(95%confidence interval: 0.6511~0.8104),APACHE ? score: 0.8187(95%confidence interval: 0.7540~0.8834).It could be obtained that the AUC of OMPCS was greater than the new model(P=0.0144)and the SOFA score(P=0.0164),and also greater than the APACHE ? score,but there was no statistical significance between these two models(P=0.4170).Conclusions: 1)Based on MIMIC database analysis showed that age,blood urea nitrogen,red blood cell distribution width,lactate,blood glucose,and anion gap were significant predictors of risk of all-cause mortality at 30 days,90 days,365 days.2)The new model prognostic nomogram constructed in MIMIC database could better predict the risk of all-cause mortality at 30 days,90 days,365 days in patients with cardiogenic shock.The predictive ability of the new model for short-term prognosis was better than that of SOFA score and similar to that of SAPS ? score.3)The new model from MIMIC database could also better predict the short-term poor prognosis of patients with cardiogenic shock in Xinjiang,revealing that the new model had good predictive power in both the United States and Xinjiang.The predictive ability of the OMPCS based on Xinjiang population was stronger than the new model based on MIMIC database,and the OMPCS was superior to SOFA in predicting short-term prognosis of patients with cardiogenic shock in Xinjiang,similar to APACHE ?.The above conclusions are still needed to verify by multicenter,large-sample prospective studies. |
PDF Full Text Request