The success of colonoscopy in colorectal cancer (CRC) prevention is dependent on operator competence in the detection and removal of adenomatous polyps. However, studies have shown substantial variability in polyp detection rates. Changes in clinical practice have led to a reduction in training opportunities; thus alternative methods of teaching have been developed. The advert of virtual-reality (VR) colonoscopy training has become a viable method of teaching the psychomotor skills required for the intubation of the colon. However, simulated-training of pathology detection is yet to be extensively studied. To address this gap we have developed 20 novel VR polyp detection tasks, we aim to assess their construct-validity, face-validity, novice learning-curves and define benchmark criteria.
25 participants were recruited; 15 novices (training-grade doctors), 2 intermediates and 8 experienced colonoscopists. Participants were required to navigate around the colon, detect abnormalities and target them. All participants performed 4 repetitions of two types of VR-simulated tasks on the Olympus ENDO-TS1 VR-colonscopy simulator: clean colon model and dirty colon model. In addition, 10 novices performed 12 more repetitions in a randomized order for learning-curve analysis. On completion of the study, intermediate and experienced groups completed a face-validity questionnaire.
The intermediate cohort was too small and thus excluded from analysis. Both clean and dirty colon models distinguished successfully between initial skill-levels of the novice and experienced colonoscopists in pathology detection metrics including: number of abnormalities located, number of polyps located and number of angiodysplasias (p <0.05). Both cohorts performed worse in the dirty colon tasks compared to the clean colon tasks in all validated metrics. In both models, individual performance of experienced colonoscopists remained stable throughout task repetition, thus median scores of the experienced cohort in validated metrics determined the benchmark criteria for novices.
Both the clean and dirty colon tasks showed statistically learning curves for all pathology detection metrics (p<0.05). By the end of the training programme, novices had progressed towards benchmarks achieved by the experienced cohort. Plateaus of learning curves most occurred at the third or fourth attempt. Although, the dirty colon ‘number of angioplasia located’ plateau occurred much later at the sixth attempt (p= 0.821). Results from the face-validity questionnaires showed that the models scored highly in realism (mean=4.2/5 p= 0.0005) and that the tasks would be significantly useful for the inexperienced population.
This study represents the first colonoscopy simulator study focused on VR teaching of pathology detection and removal; an integral skill for successful CRC prevention. The novel simulator tasks and metrics possess excellent construct-validity; hence can act as a form of objective assessment. In addition, learning-curve analysis found that repetitive practice did improve performance in the novice group. This is the first study to demonstrate that specific targeted VR simulator-training can improve the proficiency of the colonoscopist in mucosal inspection. It is hoped that this targeted intervention will instil the principles of vigilance and meticulous inspection of the colon in competent polyp detection early in an endoscopist career.