AN ELECTRONIC MEDICAL RECORD PREDICTION MODEL TO IDENTIFY INADEQUATE BOWEL PREPARATION IN PATIENTS AT OUTPATIENT COLONOSCOPY

Background and aims: Inadequate bowel preparation during colonoscopy is associated with decreased adenoma detection, increased costs, and patient procedural risks. This study aimed to develop a prediction model for identifying patients at high risk of inadequate bowel preparation for potential clinical integration into the EMR.

Methods: A retrospective study was conducted using outpatient screening/surveillance colonoscopies at the University of North Carolina (UNC) from 2017 to 2022. Data were extracted from the EMRs of Epic and ProVation, including demographic, socioeconomic, and clinical variables. Logistic regression, LASSO regression, and gradient boosting machine (GBM) models were evaluated and validated in a held-out testing set.

Results: The dataset included 23,456 colonoscopies, of which 6.25% had inadequate bowel preparation. The reduced LASSO regression model demonstrated an area under the curve (AUC) of 0.65 [95% CI 0.63-0.67] in the held-out testing set. The relative risk of inadequate bowel prep in the high-risk group determined by the model was 2.42 (95% CI 2.07-2.82), compared to patients identified as low risk. The model calibration in the testing set revealed that among patients categorized as having 0-11%, 11-22%, and 22-33% predicted risk of inadequate prep, the respective proportions of patients with inadequate prep were 5.5%, 19.3%, and 33.3%. Using the reduced LASSO model, a rudimentary code for a potential Epic FHIR application called PrepPredict was developed.

Conclusions: This study developed a prediction model for inadequate bowel preparation with the potential to integrate into the EMR for clinical use and optimize bowel preparation to improve patient care.