Survival Odds to Minimize Risk Heterogeneity Bias in Heart Failure Trials: Application to Dapagliflozin

Patients with cardiovascular conditions like heart failure (HF) often exhibit significant heterogeneity of the risk of clinical events. In clinical trials, large risk heterogeneity can result in an underestimation of treatment effects derived from Cox proportional hazards models. This occurs due to selection bias when estimating the hazard ratio, stemming from a disproportionate reduction of event-free patients in the control group compared with an effective active group over time, ultimately reducing the statistical power. Therefore, it is important to explore alternative analysis methods for outcome trials that are robust with respect to risk heterogeneity.

We used clinical data from 2 dapagliflozin HF trials-DAPA-HF (Dapagliflozin in Patients with Heart Failure and Reduced Ejection Fraction) and DELIVER (Dapagliflozin in Heart Failure with Mildly Reduced or Preserved Ejection Fraction) to characterize the extent of risk heterogeneity and nonproportionality of hazards in HF. We then evaluated a candidate method for estimating treatment effects in HF outcome trials, namely the survival proportional odds model, and compared this to traditional Cox regression in a simulation study.

In the dapagliflozin trials, nonproportional hazards were a larger issue in the HFpEF population of the DELIVER trial compared with the more homogeneous heart failure with reduced ejection fraction population of the DAPA-HF trial. In simulations of populations with varying degrees of heterogeneity, the survival proportional odds model was more robust to heterogeneity and demonstrated higher power compared with traditional Cox regression in high heterogeneity populations, while performing similarly or slightly worse in more or less heterogeneous populations. Reanalyses of the dapagliflozin trials confirmed these findings, with the survival proportional odds model providing consistently higher power in the DELIVER trial and similar power in the DAPA-HF trial.

In HF trials, the survival proportional odds model is a viable and more robust alternative for analyzing time to event outcomes, also providing an intuitive interpretation of the treatment effect directly linked to survival probability: improved odds of being event-free in the active group compared with the control group.

URL: https://www.clinicaltrials.gov; Unique identifiers: NCT03036124 and NCT03619213.