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

Area of Science:

• Surgical Quality Measurement
• Health Services Research
• Biostatistics

Background:

• Hospital quality reporting often uses procedural volume thresholds (VTs) derived from expert consensus or various analytic methods.
• Methods like restricted cubic splines (RCS), optimal cutpoints, classification and regression trees (CART), and stratum-specific likelihood ratios (SSLR) can yield inconsistent VTs.
• This inconsistency poses challenges for standardized hospital quality reporting.

Purpose of the Study:

• To compare the variability of hospital-level VTs for pancreatectomy across multiple statistical methodologies.
• To evaluate which method provides the most consistent and transparent VTs for quality reporting.

Main Methods:

• Utilized the National Cancer Database to identify patients undergoing pancreatectomy from 2004 to 2021.
• Applied RCS, optimal cutpoints, CART, and SSLR to calculate VTs based on 90-day mortality.
• Derived VTs multiple times for each method by adjusting statistical parameters within a single dataset.

Main Results:

• Analyzed data from 61,920 pancreatectomy patients across 982 hospitals.
• VTs varied significantly based on methodology: RCS (9.2-26.1 cases/year), optimal cutpoints (15.7-33.8 cases/year), and CART (11-18 cases/year).
• SSLR produced a singular, consistent VT of 21 cases/year, unaffected by statistical parameter variations.

Conclusions:

• SSLR demonstrated the least variability and required the fewest assumptions among the studied methods for VT identification.
• The consistency of SSLR makes it a potentially superior method for enhancing transparency and standardization in surgical outcomes reporting.