Abstract Body (Do not enter title and authors here): Background: Traditional congenital heart surgery quality assessments rely on indirect standardization via regression, which can be complicated by heterogeneity in case-mix, surgical volume, and low mortality rates. Our prior work revealed causal inference methods can better balance such differences across centers, yet these methods have not been widely studied to date.

Aim: To compare quality assessments using traditional regression vs. causal inference weighting approaches.

Methods: Using data from the Society of Thoracic Surgeons Congenital Heart Surgery Database (2016-22) for 10 Benchmark Operations (BMO), center observed:expected (O:E) operative mortality was calculated. Regression pooled all patients, adjusting for 21 covariates and 38 procedure-age random effects to estimate expected mortality. The causal approach divided observed mortality at each center by a weighted average of observed mortality at other centers using two weighting approaches (stable balancing [SBW] and covariate balancing propensity scores [CBPS]) and up to 204 covariates for balancing. O:E ratios (95% bootstrap intervals) were compared across methods.

Results: Across 42,579 BMOs from 115 US centers (median age 462 days, 43% female), overall operative mortality was 2.37% (range: 0.38%-6.8% in centers with ≥100 operations). Regression yielded lower, less variable O:E ratios (mean=1.37 [SD=1.21]) than causal methods (SBW = 1.76 [2.30]; CBPS = 1.79 [2.47]; Panel A). Lower volume centers had greater variability across estimates (Panel B). Regression classified fewer centers as having higher-than-expected mortality (9/115; 7.8%) compared to SBW (19/115; 17%) and CBPS (18/115; 15%); and more centers as lower-than-expected (Panel C). Compared to centers classified as higher-than-expected by all approaches, discordant centers had smaller volumes, treated more adults, and performed fewer higher complexity operations.

Conclusions: Congenital heart surgery quality assessments are sensitive to the approach for case-mix adjustment. As causal inference approaches better account for such differences across centers and rely on fewer statistical assumptions, their adoption is encouraged.