Abstract 4363040: Re-Evaluating the Unassuming Assumption: Lessons from Hemodynamic Predictors in Heart Transplant Outcomes

Background: Cox proportional hazards models are common statistical methods widely used in cardiovascular research that depend on several key assumptions of the data to ensure validity of results. The assumption of linearity (that is, proportional), or the linear relationship between predictor variables and the log of the hazard, is rarely tested but recently observed in our group’s research to be unreliable. This discovery may prevent biased risk estimates if true relationships are nonlinear. Research Question: To interrogate the assumed linearity between key hemodynamic variables used in heart transplant risk stratification and patient outcomes, understanding the methodological reliance on standard Cox modeling in research design. Methods: Using a population subgroup defined for another study of adult, heart-only transplant candidates (Status 1-3) without mechanical circulatory support (MCS) listed in the UNOS Registry from 10/18/2018 to 09/30/2024 with follow-up through 10/04/2024. Hemodynamic values at listing were assessed for association with the outcome of death/deterioration on the waitlist. To evaluate the potentially nonlinear relationship between hemodynamic variables and the log cause-specific hazard, we applied restricted cubic splines to the above hemodynamic variables prior to inclusion into the cause-specific Cox proportional hazards model. Results: In total, 2,718 non-MCS patients were included. There was a significant non-linear association between left ventricular cardiac power output (CPOLV) (p-nonlinear = 0.012) and pulmonary artery pulsatility index (PAPI) (p-nonlinear = 0.033) with the outcome of death/deterioration on the waitlist. Additionally, these findings display threshold effects, where these variables can be treated as continuous below certain cutoff values but plateau beyond them. For instance, risk of death/deterioration increased sharply below PAPI = 2.42 but plateaued beyond that, violating linearity assumptions. Conclusion: Our findings demonstrate that commonly used hemodynamic markers may not exhibit linear risk relationships. This implies that Cox regression may misestimate true risk at certain hemodynamic values when analyzing not only CPOLV and PAPI, but perhaps other physiologic estimates in cardiac research. Thus, our results highlight the need for more routine testing for nonlinearity in Cox models when employed in cardiovascular outcomes research.