Improving Risk Adjustment in the Assessment of Congenital Heart Center Surgical Quality.

Abstract

Understanding center performance in congenital heart surgery remains challenging, with limitations to standard regression approaches. Modern causal inference methods may improve estimation of expected mortality through better balancing case mix but have not been studied. Benchmark operations across 115 centers (2016-2022) from The Society of Thoracic Surgeons Congenital Database were included. The standard approach uses mixed-effects logistic regression to estimate centers' expected operative mortality inclusive of all patients, even those treated at the center of interest ("target") and those from other centers regardless of how closely aligned with the target they are. The causal approach creates tailored comparison groups, including only similar patients from other centers, by using weights. The study used 3 different weights (entropy balancing weight [EBW], stable balancing weight [SBW], and covariate balancing propensity score [CBPS]) and compared them with standard regression. Among 42,579 benchmark operations, overall operative mortality was 2.37%. Centers' mean (25th-75th percentiles [interquartile range; IQR]) regression-based expected mortality was lower (2.03% [IQR, 1.42%-2.58%]) than weighted estimates (EBW, 2.10% [IQR, 1.64%-2.67%]; SBW, 2.09% [IQR, 1.57%-2.69%]; CBPS, 2.10% [IQR, 1.63%-2.66%]) and less variable. The mean ratio of weighted to regression mortality estimate was 1.03 for all estimates, with IQRs: EBW [IQR, 0.89-1.16], SBW [IQR, 0.87-1.15], and CBPS [IQR, 0.92-1.17]. The distributions of operation type and other risk factors were better aligned with the target center in weighted estimates. The largest differences among methods were observed at smaller-volume centers. Causal inference methods constructed more tailored comparison groups for estimating centers' expected mortality, with better alignment of case mix. Adoption of newer approaches may be warranted.