Validation of Automated Colonoscopic Findings from an Endoscopic Documentation Database (Provation) against Manually Collected Data

Abstract

Purpose: Many endoscopy units are now using computer-based endoscopic documentation software such as Provation (Wolters Kluwer, Minneapolis). Such software can generate sophisticated outcomes data on large numbers of patients automatically, resulting in a growing number of published studies based on endoscopic databases. However, no published study has yet validated endoscopic software data against manually collected data. To this end, we compared the colonoscopic findings reported by Provation against manually collected findings from two similar cohorts of patients in the same endoscopy unit. Methods: In November 2011, our endoscopy unit switched from a dictation-based documentation system to the Provation system. As part of a long-term quality control initiative, we had collected data on 9,714 patients who had undergone elective colonoscopies from January 2010 to November 2011, using manual chart review involving a trained data abstractor. We compared these findings against those generated by Provation on 7,091 patients who underwent colonoscopy from November 2011 to March 2013. Results: The Provation cohort was similar to the Manual cohort with regard to mean age (59.2 vs 58.8 years; p=0.31), gender distribution (49.5% male vs 48.3%; p=0.13), and indications (48.9% screening, 30.6% surveillance and 20.5% diagnostic vs 46.4%, 30.8% and 22.9% respectively; p=0.10). The incidence of poor bowel preparation was similar between the two groups, as were the percentages of patients with one or more large polyp(s) (≥1 cm) or mass lesion(s). However, the Manual cohort had a higher prevalence of diverticulosis and hemorrhoids, while the Provation cohort had a lower colonoscopy completion rate, and a higher prevalence of polyps of any size (Table). If we restrict our analysis to screening patients, the Manual cohort showed a polyp prevalence of 45.2%, large polyp prevalence of 7.1% and mass prevalence of 0.2%, while the Provation cohort had prevalences of 45.7%, 7.6% and 0.3% respectively (all p-values >0.05).TableConclusion: Data generated by Provation are slightly different from data from manual chart review, although the large polyp and mass prevalence are similar. If we consider manual collected chart review to be the “gold standard”, this study partially validates the Provation data, but caution is advised when using endoscopic data for outcomes research.