Comparative Validation of Abdominal CT Models that Predict Need for Surgery in Adhesion-Related Small-Bowel Obstruction

Abstract

Introduction

Adhesion-related small-bowel obstruction (ASBO) can be managed without surgery in selected patients. The aim of this study was to validate three previously published computed tomography (CT) models that predict need for surgery.

Methods

A retrospective study of patients with ASBO admitted to a tertiary referral hospital between November 2009 and April 2015 was conducted. Data on clinical variables were extracted from medical records. CT signs were assessed by a radiologist who was blinded to whether or not the patients required surgery. Three previously published models were validated by testing their ability to predict need for surgery.

Results

The cohort comprised 233 patients with ASBO (mean age 69.7 years, 47.6% male), of whom 73 (31.3%) required surgery. A predictive model using a combination of mesenteric oedema, free intraperitoneal fluid and absence of small-bowel faecalisation had a sensitivity of 38% [95% CI 27–50%], specificity of 88% [81–92%], positive likelihood ratio (LR+) of 3.1 [1.6–5.1] and negative likelihood ratio (LR−) of 0.7 [0.6–0.8]. Only the results of one previously published model (which used a combination of obstipation, free intraperitoneal fluid and high-grade or complete obstruction) could be reproduced. This model had a potentially clinically useful LR+ of 2.9 [1.1–7.4] and LR− of 0.9 [0.8–1.0]. The poor performances of the other two models may be partially explained by measurement bias.

Conclusion

The performances of the previously published predictive models in this validation study were varied. Future attempts to develop models should use clearly defined, standardised and reproducible predictors wherever possible.