Richard E. Clark Memorial Paper for Congenital Heart Surgery: Understanding Mortality Following Congenital Heart Surgery: What Do Procedure-Specific Factors Add?

M. Nathan¹, L. Han², K. Zelevinsky³, H. Abing³, J. Mayer⁴, S. T.. Normand³, S. Pasquali^5
1Boston Children's Hospital, Boston, Massachusetts ²North Eastern University, Boston, Massachusetts ³Harvard Medical School, Dept. of Health Care Policy, Boston, Massachusetts ⁴Boston Childrens Hospital, Wellesley, Massachusetts ⁵University of Michigan CS Mott Children's Hospital, Ann Arbor, Michigan

Purpose: The Society of Thoracic Surgeons-Congenital Heart Surgery Database (STS-CHSD) added collection of 82 procedure-specific factors (PSF) for benchmark operations (BMO) in 2013, but their impact has not been studied to date. We sought to assess the contribution of PSFs beyond standard STS risk factors in estimating mortality in this population.1

Methods: BMO (Table 1) within the STS-CHSD across 115 U.S. centers (2017-2022) were included. Coarctation repair was excluded (no PSF collected) along with BMO with missing PSF or operative mortality data. Given low mortality rates and many risk factors, the standard STS-CHSD risk model could not be used and instead we estimated Firth-corrected logistic regression models of operative mortality.2 The baseline model included standard STS-CHSD risk factors.1 We added PSFs to the baseline model through inclusion of interaction terms of BMO with PSF. All models were trained and validated via a random 70-30 data split. The area under the receiver operating characteristic (AUROC) curve (1 is ideal), calibration intercepts (0 is ideal) and slopes (1 is ideal), and Brier scores (lower is better), were compared in the overall sample and by BMO. We also compared variation in estimated mortality (larger is better) and discrimination (larger is better), stratified by observed mortality.

Results: Among 29,266 eligible BMO, median age was 134 (IQR 27, 240) days; overall operative mortality was 2.7%. The proportion of patients with one or more PSF recorded ranged from 3.5% for VSD repair to 99.1% for Norwood. The top five PSFs with the highest associated mortality were: moderate/severe atrioventricular valve regurgitation (Norwood), intact atrial septum/obstructed pulmonary venous return (Norwood); moderate/severe truncal valve stenosis (truncus); presence of sinusoids (Norwood); and moderate/severe truncal valve regurgitation (truncus). The addition of PSFs resulted in a larger range of estimated expected mortality relative to the baseline model (Figure 1) across all operations and some selected BMOs, and the mean (SD) estimated expected operative mortality was higher in the baseline +PSF [mean=0.04, (SD=0.06)] versus baseline model [0.03(0.05)]. For the overall model, the addition of PSF did not meaningfully change the AUROC although some measures of calibration improved (intercept). Within each BMO, the inclusion of PSFs improved discrimination for certain operations (ASO, AVC, Glenn/Hemifontan, and Norwood) but not for others (Figure 1, Table 1).

Conclusion: Addition of PSFs resulted in higher expected mortality estimates and improved discrimination of mortality for some but not all BMO. Ongoing work will identify which risk variables and PSFs best discriminate mortality for individual BMOs, enabling tailored mortality prediction at patient and center level, while reducing data collection burden.

Identify the source of the funding for this research project: This work was funded by a grant from the National Heart, Lung, and Blood Institute: 5R01HL162893 (PI: Normand/Pasquali), titled "Modern Analytics to Improve Quality & Outcome Assessments Following Congenital Heart Surgery."