Validation of colonoscopic findings from a structured endoscopic documentation database against manually collected medical records data.

BACKGROUND: Endoscopic documentation software can generate research data on large numbers of subjects automatically. There are increasing numbers of published studies based on endoscopic databases such as the Clinical Outcomes Research Initiative. However, no study has yet validated such data. We compared colonoscopic findings reported by an endoscopic documentation software (Provation) against manually collected medical records data from two similar patient cohorts in the same endoscopy unit.

METHODS: In November 2011, our unit switched from dictation-based text documentation to the Provation system. As a quality control initiative, we collected data on 9614 patients who had undergone colonoscopies from January 2010 to November 2011, using manual electronic chart review. We compared these data against those generated by Provation on 7091 similar patients who underwent colonoscopy from November 2011 to March 2013.

RESULTS: Age, sex and procedural indication distribution were similar between the Manual and Provation cohorts, as were the large (≥1 cm) polyp (7.6 vs. 8.1%; p = 0.25) and advanced neoplasia (8.3 vs. 8.2%; p = 0.80) prevalences. However, there were significant differences in the polyp (46.9 vs. 49.8%) and adenoma prevalences (31.3 vs. 26.8%; p < 0.001). Furthermore, the Manual cohort had a higher prevalence of diverticulosis and hemorrhoids, and a lower colonoscopy completion rate. Stratification by indication resulted in additional discrepancies between the two cohorts for screening and surveillance patients. There were also differences in the anatomic (right vs. left colon) distribution of large polyps.

CONCLUSIONS: There were significant discrepancies between data from Provation and manually collected medical records data. Although the two cohorts were enrolled during slightly different time periods, they came from the same endoscopy unit, had the same endoscopists and indications, and demonstrated similar demographics, making it unlikely for there to be true differences between the cohorts independent of documentation method. Thus, caution is advised when using endoscopic data for research.