The Influence of Decreasing Variable Collection Burden on Hospital-Level Risk-Adjustment
jpedsurg.org shows its articles on its own site.
Read the article on jpedsurg.org ↗Article · Oct 2021 · 1 min read
In brief
In brief
Analysis of 141 hospitals demonstrates that removing congenital malformation as a predictor variable from NSQIP-Ped risk-adjustment models maintains model performance while reducing data collection burden. Hospital outlier status and performance rankings remained stable, suggesting opportunities to streamline registry data collection without compromising quality benchmarking.
Written by the GCMD Library team from the article.
Background
Risk-adjustment is a key feature of the American College of Surgeons National Surgical Quality Improvement Program-Pediatric (NSQIP-Ped). Risk-adjusted model variables require meticulous collection and periodic assessment. This study presents a method for eliminating superfluous variables using the congenital malformation (CM) predictor variable as an example.
Methods
This retrospective cohort study used NSQIP-Ped data from January 1st to December 31st, 2019 from 141 hospitals to compare six risk-adjusted mortality and morbidity outcome models with and without CM as a predictor. Model performance was compared using C-index and Hosmer-Lemeshow (HL) statistics. Hospital-level performance was assessed by comparing changes in outlier statuses, adjusted quartile ranks, and overall hospital performance statuses between models with and without CM inclusion. Lastly, Pearson correlation analysis was performed on log-transformed ORs between models.
Results
Model performance was similar with removal of CM as a predictor. The difference between C-index statistics was minimal (≤ 0.002). Graphical representations of model HL-statistics with and without CM showed considerable overlap and only one model attained significance, indicating minimally decreased performance (P = 0.058 with CM; P = 0.044 without CM). Regarding hospital-level performance, minimal changes in the number and list of hospitals assigned to each outlier status, adjusted quartile rank, and overall hospital performance status were observed when CM was removed. Strong correlation between log-transformed ORs was observed (r ≥ 0.993).
Conclusions
Removal of CM from NSQIP-Ped has minimal effect on risk-adjusted outcome modelling. Similar efforts may help balance optimal data collection burdens without sacrificing highly valued risk-adjustment in the future.