Andreas Habertheuer1, Thomas Richards2, Federico Sertic2, Maria Molina2, Prashanth Vallabhajosyula3, Yoshikazu Suzuki2, Dyenaba Diagne2, Edward Cantu2, Ibrahim Sultan4, Maria M Crespo5, Christian A Bermudez6. 1. Department of Surgery and Division of Cardiovascular Surgery, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; Department of Cardiothoracic Surgery, University of Pittsburgh, Pittsburgh, Pennsylvania. 2. Department of Surgery and Division of Cardiovascular Surgery, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. 3. Department of Surgery and Division of Cardiovascular Surgery, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; Department of Cardiothoracic Surgery, Yale University School of Medicine, New Haven, Connecticut. 4. Department of Cardiothoracic Surgery, University of Pittsburgh, Pittsburgh, Pennsylvania. 5. Department of Medicine and Division of Pulmonology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. 6. Department of Surgery and Division of Cardiovascular Surgery, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. Electronic address: christian.bermudez@pennmedicine.upenn.edu.
Abstract
BACKGROUND: No clinically validated tool exists to predict in-hospital mortality in patients requiring extracorporeal membrane oxygenation (ECMO) as a bridge to lung transplantation. We generated a quantitative risk assessment tool for these patients. METHODS: Of 822 patients in the United Network for Organ Sharing (UNOS) database who required ECMO as bridge to lung transplant between 2004-2018, 630 were included in the analysis after exclusion for age <18 years, prior transplant or treatment before 2004. Recipient-specific variables associated with post-transplant in-hospital mortality were incorporated into a multivariable logistic regression model in an automated stepwise fashion. Linear prediction was used to construct the Recipient Stratification Risk Analysis in Bridging Patients to Lung Transplant on ECMO (STABLE) score. K-fold cross-validation provided an unbiased estimate of out-of-sample performance. After further exclusion for University of Pennsylvania patients, remaining cohort was used for external score validation. iOS application was developed to aid clinical use. RESULTS: Six recipient-specific, pretransplant variables were translated into a 24-point score. STABLE scores in the UNOS dataset ranged from 0-21, and each point increased the odds of in-hospital mortality by 22.0% (95% confidence interval, [95%CI]: 1.14 - 1.29, p<0.001). K-fold cross-validation yielded a receiver operating characteristic area under the curve (AUC) of 86.2%. Validation of the STABLE score using our institutional database yielded an AUC of 89%. CONCLUSIONS: The STABLE score is a novel, internally cross-validated tool for risk stratification of patients on ECMO as a bridge to transplant. Its predictive power and accuracy may aid clinical decision-making and improve post-transplant outcomes.
BACKGROUND: No clinically validated tool exists to predict in-hospital mortality in patients requiring extracorporeal membrane oxygenation (ECMO) as a bridge to lung transplantation. We generated a quantitative risk assessment tool for these patients. METHODS: Of 822 patients in the United Network for Organ Sharing (UNOS) database who required ECMO as bridge to lung transplant between 2004-2018, 630 were included in the analysis after exclusion for age <18 years, prior transplant or treatment before 2004. Recipient-specific variables associated with post-transplant in-hospital mortality were incorporated into a multivariable logistic regression model in an automated stepwise fashion. Linear prediction was used to construct the Recipient Stratification Risk Analysis in Bridging Patients to Lung Transplant on ECMO (STABLE) score. K-fold cross-validation provided an unbiased estimate of out-of-sample performance. After further exclusion for University of Pennsylvania patients, remaining cohort was used for external score validation. iOS application was developed to aid clinical use. RESULTS: Six recipient-specific, pretransplant variables were translated into a 24-point score. STABLE scores in the UNOS dataset ranged from 0-21, and each point increased the odds of in-hospital mortality by 22.0% (95% confidence interval, [95%CI]: 1.14 - 1.29, p<0.001). K-fold cross-validation yielded a receiver operating characteristic area under the curve (AUC) of 86.2%. Validation of the STABLE score using our institutional database yielded an AUC of 89%. CONCLUSIONS: The STABLE score is a novel, internally cross-validated tool for risk stratification of patients on ECMO as a bridge to transplant. Its predictive power and accuracy may aid clinical decision-making and improve post-transplant outcomes.