The influence of decreasing variable collection burden on hospital-level risk-adjustment

Area of Science:

• Surgical Quality Improvement
• Pediatric Surgery
• Health Services Research

Background:

• Risk-adjustment is crucial for the American College of Surgeons National Surgical Quality Improvement Program-Pediatric (NSQIP-Ped).
• Accurate risk-adjusted model variables require meticulous data collection and regular evaluation.
• This study examines the impact of removing congenital malformation (CM) predictors from NSQIP-Ped models.

Purpose of the Study:

• To evaluate the effect of excluding congenital malformation (CM) predictors on risk-adjusted outcome models within NSQIP-Ped.
• To determine if CM variables are superfluous and can be removed without compromising model performance.
• To assess the implications for data collection burden and risk-adjustment accuracy.

Main Methods:

• Retrospective cohort study utilizing NSQIP-Ped data from 2019 (141 hospitals).
• Compared six risk-adjusted mortality and morbidity models with and without CM predictors.
• Assessed model performance using C-index and Hosmer-Lemeshow (HL) statistics; evaluated hospital-level performance changes and performed correlation analysis on odds ratios (ORs).

Main Results:

• Model performance remained similar after removing CM predictors, with minimal differences in C-index (≤0.002) and HL statistics.
• Hospital-level performance metrics (outlier status, quartile ranks) showed negligible changes.
• Strong correlation (r ≥ 0.993) observed between log-transformed ORs in models with and without CM.

Conclusions:

• Excluding CM predictors from NSQIP-Ped risk-adjusted models has a minimal impact on overall performance.
• Efforts to streamline data collection by removing non-essential variables can be explored without compromising risk-adjustment quality.
• This approach may help balance data collection efficiency with the need for robust risk adjustment in pediatric surgical quality initiatives.