Abstract 19199: Does Eligibility for Randomized Controlled Trials Affect Patient Survival With Coronary Revascularization in Observational Data?
Aim: Differences in the coronary revascularization survival benefit when examined in randomized controlled trials (RCT) vs. observational data may be due to selection bias. Our aim was to compare the survival of RCT eligible vs. ineligible patients in an observational dataset.
Methods: Based on selection criteria for Courage, ARTS, ERACI II, MASS II, SoS, and RITA II, we identified contemporaneous Duke Cardiovascular Databank patients meeting RCT eligibility and compared survival curves of eligible vs. ineligible patients for pairwise interventions using propensity and inverse probability weighting (IPW) with imputation for missing data.
Results: Of 23,247 patients, 44% had 1 vessel disease (VD), 26% 2VD and 30% 3VD, and 52% had PCI, 20% CABG and 28% MED. Out of patients receiving PCI, CABG or MED, RCT eligibility ranged from 16% of CABG to 4.7% of PCI patients for CABG/PCI trials, 2.0-3.6% for MED/CABG and 27-22% for MED/PCI. RCT-eligible patients differed significantly from ineligible ones. Pairwise propensity models suggested minimal overlap for both CABG/PCI and MED/CABG. Risk adjustment and IPW yielded similar survival curves with narrowed but relatively stable differences between pairwise treatment survival curves where survival estimates were CABG>PCI, CABG>MED, and PCI>MED. In general, RCT eligible vs. ineligible patients had higher survival estimates, especially for MED/CABG RCTs, but eligibility did not appear to affect pairwise survival differences. For CABG/PCI RCT-eligible patients, however, PCI was superior to CABG with Kaplan-Meier survival curves, but propensity or IPW adjustment for patient characteristics yielded higher CABG survival estimates versus PCI out to 12 years, demonstrating selection bias.
Conclusions: Our results show that RCT-eligible patients represent a minority of patients in a large observational dataset, potentially limiting RCT generalizability. Eligible patients differ significantly from ineligible patients and have higher estimated survival. Adjustments for patient characteristics with propensity and IPW yield similar survival curves and survival differences between interventions. Selection bias is evident for patients undergoing PCI instead of CABG in this observational dataset.
- Myocardial revascularization
- Coronary artery bypass grafting (CABG)
- Percutaneous coronary intervention (PCI)
- Angina pectoris
Author Disclosures: J.B. Wong: None. H. Lee: None. L.K. Shaw: None. E.R. DeLong: None.
- © 2016 by American Heart Association, Inc.