VALIDATION OF PREDICTIVE MODELS FOR INADEQUATE BOWEL PREPARATION IN COLONOSCOPIES IN A TERTIARY HOSPITAL POPULATION

Aims According to the ESGE, bowel preparation is a major quality criteria for colonoscopies. Several models were developed to identify patients at risk for inadequate preparation but none is yet validated in external cohorts.

Our aim was to validate inadequate bowel preparation predictive models in our population.

Methods All colonoscopies performed between April-November/2019 were retrospectively included. Bowel preparation was considered adequate if Boston Bowel Preparation Scale (BBPS)≥6. Patients with insufficient data, incomplete colonoscopies and colectomized patients were excluded.

Two predictive models were tested: model-1 by Dik V. et al, 2015 (tricyclic antidepressants, opioids, diabetes, constipation, previous abdominal surgery, previous inadequate preparation, admitted patient and ASA score ≥3); Model-2 by Gimeno-Garcia A. et al, 2016 (co-morbidities, tricyclic antidepressants, constipation and previous abdominal surgery).

Statistical analysis: chi-square test; α=0.05.

Results We included 514 patients (63% males; average age 61.7±15.6), 448 with BBPS ≥6. The main indications for colonoscopy were endoscopic treatment (24.9%), inflammatory bowel disease (26.1%) and colorectal cancer surveillance (12.3%). Most patients used a polyethyleneglycol solution (72.2%). Previous abdominal surgery (36.2%), ASA score ≥3 (23.7%) and diabetes (21.4%) were the most commonly identified risk factors.

Model-1 identified 202 patients with inadequate cleansing while model-2 identified 186 and both were associated with accurate predictions (p< 0.01 for both). Model-1 had a sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of 64.5%, 65.5%, 92.6% and 21.3%, respectively. Model-2 had sensitivity, specificity, PPV and NPV of 67.6%, 62.1%, 92.4% and 22%, respectively.

Conclusions Both models had a similar performance to the original papers (66% and 50% sensitivity for model-1 and model-2; 79% and 80% specificity for model-1 and model-2) and were validated in our population. Both models identified the majority of patients with low risk of inadequate bowel preparation (92.6% and 92.4%, respectively) but also identified more than 60% of patients with inadequate bowel preparation.