Information bias in endoscopic assessment.

Abstract

Endoscopic assessment is observer dependent and patient history may influence the interpretation of images. This study assessed the impact of knowledge bias in a large sample of endoscopists. The same 2-minute video sequence of a normal gastroscopy was shown twice as part of a diagnostic quiz. The 129 endoscopists were unaware that they participated in a validation experiment. Two fictional patient histories accompanied the videos. The first case suggested reflux-like symptoms. The second case described epigastric pain in an elderly woman. Endoscopists gave diagnoses in free writing. The diagnostic suggestions were categorized into five classes: (a) normal findings; (b) reflux-related findings (e.g., erosions, hiatal hernia, Barrett's metaplasia); (c) peptic ulcer or atrophic gastritis; (d) other specified pathology; (e) no diagnostic suggestion. Only 30/129 (23%) endoscopists gave the same diagnosis for the two identical video cases. The patient history had a major impact on the diagnostic suggestions: In case 1 (reflux symptoms) 54/129 (42%) of the diagnoses were related to reflux disease compared with only six diagnoses of ulcer or atrophic gastritis (P < 0.0001). In case 2 (epigastric pain) 56 endoscopists (43%) diagnosed ulcer or atrophic gastritis whereas only 16 (12%) suggested reflux-related findings (P < 0.0001). Endoscopists are biased by the patient history when they interpret endoscopic images.