BACKGROUND: Randomized trials evaluating multiarterial grafting (MAG) vs single arterial grafting (SAG) during coronary artery bypass grafting (CABG) have not demonstrated a long-term survival benefit, whereas conventional retrospective studies have consistently reported improved survival with MAG. Whether this discordance reflects true treatment effect heterogeneity or bias from unmeasured confounding in observational analysis remains unclear. OBJECTIVES: Our objective was to evaluate whether the apparent survival advantage associated with MAG in conventional observational analyses persists after accounting for unmeasured confounding using a quasi-experimental instrumental variable (IV) approach and to assess the implications of these findings for long-term survival in an older Medicare population. METHODS: We retrospectively analyzed Medicare beneficiaries who underwent CABG from 2001 to 2019. Surgeon MAG rate during the 12 months preceding each operation was leveraged as an IV. Flexible parametric survival models with time-dependent effects were developed with MAG vs SAG as the exposure variable. The non-IV model adjusted for patient demographics, pre-existing comorbidities, hospital and surgeon characteristics, and procedural details. The IV model incorporated these same covariates plus the IV (surgeon MAG rate) using a 2-stage residual inclusion approach. Regression standardization was used to derive standardized survival probabilities and their differences. RESULTS: Among 1,291,314 beneficiaries, 1,145,760 (88.7%) underwent SAG and 145,554 (12.3%) underwent MAG. In the non-IV model, MAG recipients had improved risk-adjusted median survival as compared with SAG recipients: 10.74 years (95% CI: 10.70-10.79 years) vs 10.33 years (95% CI: 10.31-10.35 years), a difference of 0.41 years. Across 4,164 surgeons, the MAG rate during the 12 months preceding the index CABG was 7.7% ± 9.5% in SAG recipients and 32.9% ± 25.8% in MAG recipients. In the IV model, MAG recipients had similar risk-adjusted median survival compared with SAG recipients: 10.38 years (95% CI: 10.29-10.48 years) vs 10.38 years (95% CI: 10.35-10.40 years), a difference of 0.01 years. CONCLUSIONS: MAG was associated with a modest improvement in long-term survival in a conventional risk-adjusted analysis. However, this association was not robust to a quasi-experimental analysis in which surgeon MAG rate was incorporated as an IV to address unmeasured confounding. The contrast between these models suggests that traditional observational studies may overestimate the survival benefit of MAG because of unmeasured or difficult-to-measure patient characteristics that influence a surgeon's decision to offer MAG.
Schaffer et al. (Fri,) studied this question.