Artificial Intelligence Versus Surgeon Decision-Making for Operative Intervention in Small Bowel Obstruction

Document Type

Conference Proceeding - Restricted Access

Publication Date

5-8-2026

Abstract

Small bowel obstruction (SBO) is one of the most common pathologies addressed by general surgeons. Computed tomography (CT) remains the most common method of diagnosis, with recent data citing a sensitivity and specificity of 95% for high grade SBO1,2. Despite this, CT's predictive value for patients requiring operative intervention in patients with SBO remains uncertain. CT findings with phrases such as "high grade," "closed loop," and "transition point" often play a role in a surgeon's decision to pursue operative intervention. As the use of artificial intelligence (AI) has increased, it has not been studied as a tool to aid surgeon decision-making for operative versus non-operative management of SBO. The primary objective of this study is to compare surgeon and AI decision-making with CT scans demonstrating SBO with high-risk features.

This study is a retrospective chart review of patients admitted at Corewell Health Blodgett Hospital with SBO between 2017 and 2022. Patients with high-risk CT findings meeting inclusion criteria were identified via chart review and enrolled into two arms, operative and non-operative. A total of 50 consecutive patients were enrolled in each arm. De-identified CT scan impressions were entered into the AI interpreter, ChatGPT 4.0, and AI decision-making was compared to surgeon decision-making. The correctness of AI and surgeon decision-making were then compared. A correct decision was defined as operative intervention with positive findings or non-operative intervention without progression to surgery. Incorrect decisions were defined as the opposite. Summary statistics were calculated utilizing the chi-square test and Fisher's exact test, as appropriate. Significance was assessed at p< 0.05.

Between the two arms, AI and surgeons pursued surgery for 77.8% and 50.5% of patients, respectively. Surgeons had a higher rate of incorrect non-surgical decisions (10.1%) compared to incorrect surgical decisions (3.0%). The Fisher's exact test p-value for this was 0.041, suggesting that surgeon decision-making was differentially impactful in correctness between non-operative and operative management. AI's tendency towards surgery was associated with a high number of incorrect surgical decisions (23.2%). When AI opted for non-operative management, it was incorrect 3% of the time. The Fisher's exact test p value for this was not significant (0.172), suggesting that AI's incorrectness was not significantly different when choosing operative versus non-operative management. A follow-up analysis is underway to compare the correctness of AI and surgeon decision-making within the non-operative arm using an additional consecutive 200 patients from the original data set with results currently pending.

This study provides unique perspective into surgeon and AI decision-making with CT findings of high-risk SBO. Thus far, the data demonstrate that AI is more likely than surgeons to pursue surgery with an associated high rate of incorrect operative decisions. Further analysis of the non-operative group will continue to identify more patterns in decision-making between surgeons and AI. Further research comparing AI and physician assessment is essential and may lead to future AI incorporation in clinical decision-making tools.

Comments

2026 Research Day Corewell Health West, Grand Rapids, MI, May 8, 2026. Abstract 2014

This document is currently not available here.

Share

COinS