A 4-tier Protocolized Radiological Classification System for Leaks Following Sleeve Gastrectomy.

ABSTRACT

OBJECTIVES: To develop and validate a classification of sleeve gastrectomy leaks able to reliably predict outcomes, from protocolized computed tomography (CT) findings and readily available variables. SUMMARY OF BACKGROUND DATA Leaks post sleeve gastrectomy remain morbid and resource-consuming. Incidence, treatments, and outcomes are variable, representing heterogeneity of the problem. A predictive tool available at presentation would aid management and predict outcomes. METHODS: From a prospective database (2009-2018) we reviewed patients with staple line leaks. A Delphi process was undertaken on candidate variables (80-20). Correlations were performed to stratify 4 groupings based on outcomes (salvage resection, length of stay, and complications) and predictor variables. Training and validation cohorts were established by block randomization. RESULTS: A 4-tiered classification was developed based on CT appearance and duration postsurgery. Interobserver agreement was high (κ = 0.85, P < 0.001). There were 59 patients, (training 30, validation 29). Age 42.5 ± 10.8 versus 38.9 ± 10.0 years (P = 0.187); female 65.5% versus 80.0% (P = 0.211), weight 127.4 ± 31.3 versus 141.0 ± 47.9 kg, (P = 0.203). In the training group, there was a trend toward longer hospital stays as grading increased (I = 10.5 d; II = 24 d; III = 66.5 d; IV = 72 d; P = 0.005). Risk of salvage resection increased (risk ratio grade 4 = 9; P = 0.043) as did complication severity (P = 0.027).Findings were reproduced in the validation group risk of salvage resection (P = 0.007), hospital stay (P = 0.001), complications (P = 0.016). CONCLUSION: We have developed and validated a classification system, based on protocolized CT imaging that predicts a step-wise increased risk of salvage resection, complication severity, and increased hospital stay. The system should aid patient management and facilitate comparisons of outcomes and efficacy of interventions.