Aims
Virtual reality (VR) offers an unrivalled immersive education opportunity to provide guaranteed pathology exposure, reproducible instruction and robust pathology recognition key performance indicators. We developed a novel VR platform to address the heterogenous and subjective nature of endoscopy education, which may contribute to the 11.3% published missed cancer rate of upper gastrointestinal endoscopy (EGD). Data demonstrates variability in the quality of endoscopic surveillance of Barrett’s esophagus (BE) but accurate assessment is vital to detect early neoplasia and alter the grave prognosis of esophageal adenocarcinoma. We conducted a pilot validation study for a novel prototype VR endoscopy training environment for Barrett’s esophagus.

Methods
32 endoscopists accredited to perform independent EGDs were recruited. Participants performed 3 virtual reality BE surveillance EGDs using Meta Quest 2 headset and performed targeted biopsies on visible dysplasia. Feedback was collected using the validated Nasa Task Load Index (TLI), and the iGroup Presence Questionnaire (IPQ). Intraclass co-efficients (ICC) assessed inter-observer reliability for gastro-esophageal junction (GEJ) and Prague classification measurements. Visible dysplasia was present in 1/3 cases and was digitally rendered from a real case. Detection was recorded by successful biopsy of abnormal area.

Results
30/32 completed all 3 cases: 22 consultants, 6 trainee gastroenterologists, 4 nurse endoscopists. 95.7% correctly diagnosed BE indicating good content validity. There was good interobserver reliability for GEJ and Prague measurements; ICC 0.84 (95% CI 0.72 – 0.92). 2/32 (6.3%) identified the dysplasia. There was no significant difference between consultant, trainee or nurse endoscopist (X² p = 0.572). The VR cases demonstrated construct validity: 4/32 (12.5%) were tertiary centre Barrett’s experts who were more likely to detect the lesion (likelihood ratio = 9.4, X² p <0.01) and 3/32 (9.4%) performed a weekly Barrett’s surveillance list, which was the strongest predictor of detection (likelihood ratio = 11.1, X² <0.001). The VR experience scored well for immersion in the IPQ for domains of spatial presence (mean = 4.97, SD = 0.86) and involvement (mean = 4.55, SD = 0.86). The mean TLX scores were highest for the mental demand domain (12.5, SD = 4.65). There was no significant difference in scores between age groups (30-40, 40-50, 50-60) calculated with Kruskall Wallis test.

Conclusions:
We have created a VR training environment for BE that demonstrates strong validity evidence and could be used to upskill already clinically independent endoscopists. This could improve early neoplasia detection and improve patient outcomes. Further studies are required with greater numbers of expert and nonexpert endoscopists to assess construct validity and benchmark expert performance.

Alasdair Sinclair

Appearances