The Colorado Heart Failure Acuity Risk Model (CHARM) Score: A Mortality Risk Model for Waitlisted Cardiac Transplant Patients

Abstract

ABSTRACTImportanceAlthough the Organ Procurement and Transplantation network provides structured policies and guidance for waitlisted cardiac transplant patients, the heart transplantation community lacks a mathematical model that can accurately estimate the short-term risk of death associated with being waitlisted. Importantly, the CHARM score provides a risk management and ranking system for patients based on a well-defined and sensitive medical urgency metric.ObjectiveWe had three primary objectives in completing this study. First, to increase relevance and applicability, we selected patient attributes that were clinically justified and readily available. Second, we designed and implemented an intuitive, formal system that accurately defined the relative risk of death while being waitlisted at 30-day, 90-day, and 1-year censoring periods. Third, we developed and validated a medical urgency metric that is intuitive, practical, and can be implemented nationally.DesignWe present a multivariable, prognostic model and risk management strategy for adult waitlisted heart transplant patients (N=1,965) from the Scientific Registry of Transplant Recipients (SRTR) database that were waitlisted from January 1, 2008, to September 2, 2022. To independently validate each model, we randomly split this cohort into a discovery set (N=1,174) and validation set (N=784). Twelve independent patient attributes were selected, and three linear regression formulas were derived to estimate and rank the relative risk of dying while waitlisted. Four independent validation methods were used to measure each model’s performance as a classifier and ranking system.SettingThe United StatesParticipantsThis cohort (N=1,965) consisted of adult heart transplant candidates without missing laboratory data who were placed on the waitlist from January 1, 2008, to September 2, 2022. Patients listed for multi-organ transplantation were excluded.Patients with missing laboratory data were analyzed independently.ExposuresThe short-term risk of death remaining on the heart transplant waitlist.Main Outcomes and MeasuresThe primary outcome of this study was the design, development, and validation of a formal risk management system for waitlisted heart transplant candidates experiencing end-stage heart failure. We derived three linear regression formulas and calibrated a seven-tiered risk index to accurately rank patients who were more likely to die on the waitlist at 30-day (30D), 90-day (90D), and 1-year (1Y) censoring periods. Four independent validation methods were used to measure each model’s classification and ranking performance.ResultsUsing six interaction terms, we applied the 5-fold cross-validation procedure to the CHARM to discover an area under the ROC curve of 96.4%, 90.4.%, and 78% for the 30D, 90D, and 1Y models, respectively. The mean positive predictive values of the tiered risk system were 99.2% (30D), 94.1% (90D), and 88% (1Y). Risk indices for all three models were >99% correlated to the observed mortality rate across the seven tiers for the 30D, 90D, and 1Y models.Conclusions and RelevanceWe designed, implemented, and validated an intuitive and formal risk scoring and ranking system which is ideal for prioritizing waitlisted heart failure patients based on a well-defined medical urgency metric. The CHARM score provides extreme sensitivity in predicting short-term mortality outcomes. The CHARM score is extensible to larger patient populations experiencing end-stage heart failure.KEY POINTSQuestionCan pre-operative patient characteristics be used to develop a formal system to accurately estimate, rank, and predict the relative short-term mortality of waitlisted heart transplant patients?FindingsUsing twelve patient attributes, we derived three linear regression equations to accurately predict the 30-day, 90-day, and 1-year mortality of waitlisted heart transplant patients. We developed and calibrated a seven-tiered risk index for each model that was 99% correlated to the observed mortality rate. Using several independent validation methods, we achieved extreme sensitivity (>98%) in ordinally ranking patient groups who were more likely to survive 30 days on the waitlist. Model performance was measured using the area under the receiver operating characteristic (ROC) curve. Using six interaction terms, the area under the ROC curve was 96.4% (30-day), 90.4% (90-day), and 78% (1-year).MeaningOur models accurately discriminate among patient subgroups who are more likely to die while waitlisted. Because our tiered ranking system is simple, extremely sensitive, and well calibrated, it is ideal for prioritizing waitlisted heart transplant patients based on a well-defined medical urgency score. These models are generalized and therefore extensible to defining medical urgency in larger patient populations experiencing end-stage heart failure.