A comprehensive validation of the accuracy of two predictive models for inadequate bowel preparation

Introduction

Colonoscopy is the gold standard investigation of the bowel. Adequate cleansing is a prerequisite of a good quality examination. However, inadequate bowel preparation occurs in 20% of procedures, limiting the accuracy of the test. Predictive models devised by Dik et al and Gimeno-Garcia et al, stratified the risk of poor bowel preparation using a cumulative score. This study aims to assess the validity of these models using an independent cohort

Methods

Data generated from the EBOPS study was utilised. Adequacy was defined using the Boston Bowel Preparation Scale (BBPS), with a score of ≥ 2 in each segment. Each participant had a score derived from each predictive score, with high risk defined as a score ≥ 2 in the Dik et al score and ≥ 1.225 in the Gimeno Garcia score. Descriptive analyses, AUROC and an assessment of calibration were conducted between the models and the EBOPS independent cohort.

Results

507 participants were included. 89 (17.5%) had inadequate preparation. 31% of the cohort had a Dik et al score ≥ 2 with a sensitivity of 46%, specificity of 71%, PPV of 29.7% and NPV 84% for poor preparation. 48% of the cohort had a Gimeno-Garcia score of ≥1.225 with a sensitivity for inadequate bowel preparation of 60%, specificity of 55%, PPV of 26% and NPV of 84%. AUROC was 0.71 for the Dik et al model and 0.7 for the Gimeno-Garcia model.

Conclusions

A similar performance of the predictive models is observed with this independent cohort, compared with the initial internal validation. The models do identify a cohort of patients at high risk of poor bowel preparation with a modest degree of discrimination, allowing a stratification of those at an elevated risk.