Counterfactual Decomposition of Anastomotic Leak Burden Across Gastrointestinal Procedure Classes and Postoperative Care Environments

Abstract

Anastomotic leak is usually discussed through average event rates, even though the clinical burden of leak is rarely distributed evenly across operations or postoperative care settings. That averaging problem matters because hospitals do not manage a single generic anastomosis. They manage a portfolio of procedure types delivered through distinct perioperative pathways, each combining different technical demands, patient severity profiles, and rescue environments. When leak burden is concentrated in a small subset of procedure-setting combinations, pooled estimates can obscure where surveillance, staffing, and escalation design require the most attention. This study examined heterogeneity of leak burden across gastrointestinal procedure classes and postoperative care environments using a retrospective multicenter electronic health record cohort. The primary goal was not to build another individual risk score, but to quantify how much variation in leak burden across settings was attributable to procedure mix, how much remained after standardizing patient severity within procedure class, and how much reflected setting-specific excess within the same procedure class. The analytic cohort included 57,882 restorative gastrointestinal operations performed at 24 hospitals, with 3,873 leaks within 30 days, corresponding to an overall incidence of 6.7%. Seven procedure classes and six postoperative care environments generated 42 procedure-setting cells. Cell-specific rates were estimated after doubly robust standardization for age, frailty, nutritional status, urgency, contamination, operative duration, and early postoperative physiology. Leak burden was highly concentrated: 8 of 42 cells accounted for 57.4% of all events. Across settings, 44.7% of burden dispersion was explained by procedure mix, 13.6% by within-procedure patient severity, and 32.9% by setting-specific excess after standardization, with the remainder arising from interaction. Transfer-complex pathways were dominated by foregut case mix, whereas emergency night pathways showed the largest within-procedure excess. These results indicate that heterogeneity of leak across surgery types and care settings is as much a problem of burden allocation as of individual prediction.