Reproducibility and Consistency of Methods to Define Hospital-Level Procedural Volume Thresholds for Pancreatectomy.

INTRODUCTION

Procedural volume thresholds (VTs) for hospital quality reporting rely on expert consensus or analytic methods that may produce inconsistent VTs (e.g. restricted cubic splines (RCS), optimal cutpoints, classification and regression trees (CART), stratum specific likelihood ratios (SSLR)). The objective of this study was to compare variation in hospital-level VTs for pancreatectomy across multiple methodologies.

METHODS

Patients undergoing pancreatectomy from 2004 to 2021 were identified using the National Cancer Database. RCS, optimal cutpoints, CART, and SSLR were used to compute VTs based on 90-day mortality. From a single clinical data set, VTs were derived multiple times for each method by varying statistical parameters within each model.

RESULTS

Overall, 61,920 patients underwent pancreatectomy at 982 hospitals. VTs associated with reductions in 90-day mortality ranged from 9.2 to 26.1 cases/year (RCS), 15.7-33.8 cases/year (optimal cutpoints), and 11-18 cases/year (CART), all based on modifiable statistical parameters. SSLR analysis yielded a singular VT of 21 cases/year without variability due to lack of statistical input.

CONCLUSION

Among 4 common strategies for identifying VT that we studied, SSLR required the fewest assumptions. This may make it ideal for enhancing transparency and standardization in outcomes reporting.