Development and Validation of a Simple-to-use Nomogram for Predicting In-hospital Mortality in Acute Heart Failure Patients Undergoing Continuous Renal Replacement Therapy

Overview

Acute heart failure (AHF) is one of the most common causes of hospitalization and life-threatening medical condition around worldwide. The AHF patients admitted to the intensive care unit (ICU) usually be critically ill with multiorgan failure, in which the kidneys are most frequently involved. The goals of treatment of AHF in ICU were to improve hemodynamic stability and organ perfusion, alleviate symptoms, and limit cardiac and renal damage, which can be achieved by continuous renal replacement therapy (CRRT), a continuous extracorporeal blood purification. CRRT can mimic urine output to slowly and continuously remove patient's plasma water, providing accurate volume control and hemodynamic stability. Acute Heart Failure Global Survey of Standard Treatment (ALARM-HF) study showed that hospital mortality of AHF patients was about 17.8% in the intensive care unit (ICU). But the patients undergoing CRRT, the mortality up to 45%-62.1%. For this reason, an early model or score to a screening of AHF patients undergoing CRRT who at high mortality risk is crucial, which can help clinicians to rapidly intervene and ameliorate disease outcomes. 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) , 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 model to predict the in-hospital mortality of AHF patient undergoing CRRT. Based on the acute heart failure unit (AHFU) of Qilu Hospital and the medical information mart for intensive care III (MIMIC III) database, the investigators collected the data of AHF adults undergoing CRRT. The present study aimed to develop and validate a simple-to-use nomogram model comprised of independent prognostic variables for predicting in-hospital mortality in AHF adults undergoing CRRT by using multivariate logistic regression analysis. With this model, the investigators can guide the early screening of high-risk patients in in-hospital mortality.

The eligible patients randomly into training cohort and validation cohort. The univariate logistics regression analyses were performed to determine the independent risk characteristics in the training cohort of the presence of in-hospital all-cause death. Odds Ratios (ORs) and 95% confidence intervals (CIs) of these variables were estimated to quantify the strength of these associations. All 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. A nomogram model, producing by using the rms package, was formulated based on the results of independent risk factors in multivariate logistic regression. Based on the nomogram model, the total scores and prediction of in-hospital mortality risk of each patient were added by each eligible variable and then were converted to predicted probabilities both in the training cohort and validation cohort. To evaluated the model for the prediction value of in-hospital mortality, firstly, the investigators calculated the calibration of the model was measured by calibration with 1000 bootstrap samples to decrease the overfit bias. Model fitting was assessed using the Hosmer-Lemeshow test to evaluate the goodness of fit. Secondly, the Harrell concordance index (C index) and receiver operating characteristic curve (ROC curve) to evaluate the predictive performance and discrimination of the nomogram. The ROC curve analysis was used to calculate the optimal cutoff values that were determined by maximizing the Youden index. Third, the clinical effectiveness of the resulting model was evaluated by decision curve analysis (DCA), which was a method for evaluating alternative diagnostic or prognostic tools that had advantages over others\[16\]. The increase in the discriminative value of MEWS and the resulting model for mortality was assessed by the net reclassification index (NRI).