External validation of predictive models for prolonged postoperative ileus following colorectal resection

INTRODUCTION: Prolonged postoperative ileus (PPOI) is a frequent complication after colorectal surgery. Several prediction models have been proposed to estimate PPOI risk, but few have undergone external validation, limiting their generalizability and clinical usefulness. The primary aim of this study was to externally validate two predictive models for PPOI after colorectal resection; a secondary aim was to explore factors associated with PPOI in our cohort. METHODS: Data from all consecutive patients who underwent elective colorectal resection in our department between 2019 and 2022 were retrospectively analyzed from a prospective database. Eligible criteria were age ≥ 18 years, elective colorectal resection, and ASA score I-III. Based on a recent systematic review, we selected the models by Hain et al. and Wolthuis et al. for external validation, as all required variables were available in our dataset. Model performance was assessed in terms of discrimination, calibration, and overall accuracy. An exploratory multivariable regression analysis was also performed to assess factors associated with PPOI. RESULTS: Among 200 patients undergoing colorectal resection, 43 (21.5%) developed PPOI. Both prediction models showed poor external performance. The Hain model had a C-statistic of 0.597 (95% CI 0.514-0.681) and the Wolthuis model a C-statistic of 0.589 (95% CI 0.501-0.677), with suboptimal calibration and limited overall accuracy. In secondary exploratory multivariable analyses, postoperative opioid use was associated with PPOI in both models. In the model excluding postoperative oral intake initiation and autonomous postoperative mobilization, splenic flexure mobilization was associated with lower odds of PPOI; in the fully adjusted model, delayed postoperative oral intake initiation was associated with PPOI. These exploratory local findings should be interpreted as hypothesis-generating. CONCLUSIONS: External validation of two previously published PPOI prediction models demonstrated poor performance in this cohort, limiting their transportability to our setting. The secondary exploratory analysis identified potentially relevant postoperative factors, but these findings should be considered hypothesis-generating. Further research should prioritize harmonized PPOI definitions and robust multicentre validation of prediction models.