Development Of A D-GLAD Model For Predicting In-hospital Mortality In Acute Heart Failure Patients Undergoing Continuous Renal Replacement Therapy

BackgroundAcute heart failure(AHF)is one of the most common causes of hospitalization and life-threatening medical condition around worldwide,characterized by high in-hospital mortality and high risk of re-hospitalization.AHF patients often amalgamate a variety of viscera failures including acute and chronic kidney injury.The treatment principle of AHF was to maintain hemodynamic stability and organ perfusion,alleviate symptoms,limit cardiac and renal damage and deal with protopathy.The guideline recommends that renal replacement therapy should be considered in patients with refractory volume overload and acute kidney injury Continuous renal replacement therapy(CRRT),a continuous extracorporeal blood purification,can mimic urine output to slowly and continuously remove a patient’s plasma water,providing accurate volume control and hemodynamic stability.As the most commonly used model of renal replacement therapy in intensive care units,CRRT is increasingly used in patients with AHF in recent years.The mortality of critically ill patients who confirmed undergoing CRRT up to 45%-62.1%,2-folds than AHF patients without CRRT.How to screen and intervene in patients at high risk of mortality is important to clinical.The most popular tools,especially that can predict mortality for critically ill patients,are the Acute Physiology Assessment and Chronic Health Evaluation Ⅱ(APACHE Ⅱ)scoring systems,and Simplified Acute Physiologic Score Ⅱ(SAPS Ⅱ).But variables in these scoring systems are complex,which was not convenient to assess at any time.Modified Early Warning Score(MEWS)and SUPER score,much more concise than APACHE Ⅱ and SAPS Ⅱ not only can be used for early warning of the onset of AHF in patients with the risk of heart failure but also has a positive correlation with mortality in these patients.However,up to our knowledge,there was no scores or models specially for predicting the in-hospital mortality of AHF patient undergoing CRRT.Therefore,there is an urgent need for a simple and effective clinical prediction model that can early screen AHF patients who undergoing CRRT with a high risk of in-hospital mortalityObjectiveTo explore the independent related factors that affect the in-hospital mortality of AHF patients who undergoing CRRT,and construct a clinical prediction nomogram model.Meanwhile,the effectiveness of APACHE II,SAPS II,and MEWS in predicting the in-hospital mortality of AHF patients receiving CRRT was verified,and the most suitable model was selected to widely popularize and apply and to provide an individualized and accurate theoretical basis for screening patients with a high risk of in-hospital mortality.MethodWe retrospectively enrolled a total of 121 patients diagnosed with acute heart failure(AHF)undergoing CRRT from 2011.09 to 2020.08 in the acute heart failure unit(AHFU)and 105 patients in MIMIC III databases.We collected the demographic information,vital signs information,laboratory data,and hospital outcomes of eligible patients.Patients were divided into death cohort and survival cohort according to whether in-hospital mortality occurred.Eligible patients were randomly(7:3)classified as training cohort(159)and validation cohort(67).The univariate logistics regression analyses and multivariate logistics regression analysis were performed to determine the independent risk characteristics in the training cohort of the presence of in-hospital all-cause death.A nomogram model was obtained using the visual processing logistic regression model by R language.The receiver operating characteristic curve(ROC)and concordance index(C index),calibration curve and Hosmer-Lemeshow test,and decision curve analysis were used to verify the discrimination,calibration,and clinical effectiveness of the resulting model respectively.The net reclassification index(NRI)was used to compare the ability of the new model and MEWS to correctly classify research objects.ResultA total of 226 AHF patients who undergoing CRRT were enrolled.The eligible patient can be categorized into 98 non-survivors(43.4%)and 128 survivors(56.6%),the mortality was 46.3%in the validation cohort and 42.1%in the training cohort.Age,cardiopulmonary resuscitation,systolic blood pressure,diastolic blood pressure(DBP),mean arterial pressure,oxygen saturation,aspartate aminotransferase,urine volume,creatinine,glucose,lactic acid,days after admission before CRRT,mechanical ventilation,potassium,and SUPER score about 15 variables that showed a univariate relationship with in-hospital mortality or that were considered clinically relevant were candidates for stepwise multivariate analysis in the training cohort.Age,days after admission before CRRT,lactic acid,glucose,and.DBP were shown as significant prognostic factors in logistic regression analyses and included in the nomogram,also called the D-GLAD model,as predictors.An online webserver of D-GLAD model can be accessed at https://ahfcrrt--d-glad.shinyapps.io/DynNomapp/.The discrimination ability of the D-GLAD model constituted by the above 5 factors was both good in training groups(C-index 0.829,95%CI 0.767-0.891)and validation group(C-index 0.740,95%CI 0.620-0.860).The calibration and Hosmer-Lemeshow test showed good accuracy of the nomogram for in-hospital mortality prediction both in the training cohort(P=0.868)and validation cohort(P=0.104).When compared with MEWS,APACHEII and SAPSII,the clinical efficacy of D-GLAD model was much better.The ROC curve showed that the optimal cutoff values of prediction of in-hospital mortality were 0.266,with a corresponding total nomogram score was about 125.NRI showed that the resulting model improved prognostic discrimination for in-hospital mortality in the training cohort(NRI=59.14%,P<0.001),and validation cohort(NRI=23.84%,P=0.187)by contrast with MEWS.When choosing 0.266 as the cutoff values,the sensitivity,specificity,positive predictive value,and negative predictive value when used in screening high-risk patients were 73.5%,76.6%,57.2%,and 87.2%in the training cohort,and 83.87%,44.44%,56.52%,and 76.19%in the validation cohort,respectively.The in-hospital mortality in low-,moderate-,and high-risk group was 14.46%,40.74%,and 71.91%.ConclusionBased on AHFU and MIMIC Ⅲ,this study developed and validated a D-GLAD model associated with the prognosis of in-hospital mortality in AHF patients who undergoing CRRT,which were formed by independent predictors,such as age,days after admission before CRRT,lactic acid,glucose,and DBP.The D-GLAD model has good discrimination,calibration,and clinical effectiveness when compared with APACHE Ⅱ、MEWS、SAPS Ⅱ,which achieved an optimal prognosis prediction of in-hospital mortality in AHF adults undergoing CRRT.Using the Simple-to-use model,the risk for an individual patient of in-hospital mortality can be determined,which can be useful to guide early screening of high-risk patients.The online web server makes the clinicians and researchers apply the model more conveniently and simply.Combined with the SUPER score,the D-GLAD model can more accurately assess the risk of in-hospital mortality in the AHF patients receiving CRRT.