Interactive training versus self-driven training in the prediction of colorectal polyp histology by trainees using the NICE classification

This research was conducted as a prospective randomized controlled study at five university and academic centers in China from January 1, 2021 to May 31, 2021. Trainees with no previous formal training in the use of Narrow band imaging (NBI) or blue light imaging (BLI) for prediction of colorectal polyp histology using the NICE classification were invited to participate. We enrolled trainees (i) with less than 3 years of colonoscopic experience; (ii) with no previous formal training in the use of NBI or BLI for prediction of colorectal polyp histology; and (iii) with no previous formal training of the NICE classification. Participating trainees provided written informed consent, as they were regarded as the study subjects. They were randomly allocated to either the self-driven training group or the interactive training group using a computerized random number (Fig. 1). The study was preregistered at the Chinese Clinical Trial Registry (Trial registration number: ChiCTR2000031659, Date of registration: 06/04/2020). All methods in this study were carried out in accordance with the Declaration of Helsinki and had been approved by the China Ethics Committee of Registering Clinical Trials approved this study (ChiECRCT20200077).

The classroom interactive training session was conducted as a group for all trainees randomized to this group. The training slides for the NICE classification used in the study were made by Wei-guang Qiao, and the training video was recorded by Yu-tang Ren. Three endoscopists with experience in NBI or BLI for the NICE classification (Si-lin Huang, and Fa-chao Zhi) reviewed the slides and video.

For this training, the trainees could review the training slides and the video once. The training slides contained the key factors for diagnosis via the NICE classification, including the color, the vessel pattern, and the surface pattern (Fig. 2). Type 1 featured a same or lighter color than the background, none or isolated lacy vessels coursing across the lesion with dark or white spots of uniform size, or homogeneous absence of pattern. Type 2 featured browner color relative to the background, brown vessels surrounding white structures, or oval/tubular or branched white structures surrounded by brown vessels. Type 3 featured brown to dark brown relative to the background (sometimes patchy whiter areas), area(s) of disrupted or missing vessels, or amorphous or absent surface pattern. During the training, the trainees were allowed to ask questions. Areas of confusion that could lead to erroneous interpretation were discussed. The training session took a total of 40 min.

The trainees in the self-driven training group were provided with a training video with a voice-over recorded by Yu-tang Ren and training slides provided by Wei-guang Qiao, which were similar to those that were used for teaching in the interactive training group. They reviewed the teaching material individually. The trainees could view the slide presentation and training videos at any time within a 40 min timeframe. To make sure the training materials were actually accessed and reviewed by the trainees, a slide with a QR code used for after-training tests was set as the last page.

Before and after the interactive training or self-driven training, all trainees were given a QR code with 20 selected cases of colorectal polyp under BLI for testing. To decrease investigator bias, the investigators who were in charge of the research design were forbidden to be the test scorers. For control measures, the test scorers were blinded to the randomization. The trainees reviewed these images and made a diagnosis by entering the judgement of the NICE classification, the color, the vessel pattern, the surface pattern, and confidence in diagnosis (low or high). All the images were retrieved during standard care without any interventions or exposure of the patient’ personal information. These images of polyps were obtained using high-definition colonoscopes (EC-L590ZW, EC-L590ZP, EC-760ZP-V/M; Fujifilm Co). The polyps in the collected images in the before-training tests and the after-training tests had been resected and evaluated by pathology for histology confirmation.

We used HyLown Consulting (2019 HyLown Consulting LLC, Atlanta, GA) to calculated the sample size. Under the assumption that the accuracy in the self-driven training group would be 70% compared with 78% in the didactic group, a sample size of 481 observations was needed to detect this 10% difference in accuracy between the two groups with 80% power and alpha of 0.05, given that we had 20 cases for tests and a drop rate of 0.05. A sample size of 26 trainees in each group was needed.

Thirty-one endoscopists were allocated to the interactive training group and self-driven training group in the before-training phase, respectively. There were no statistical differences in sex, age, education, colonoscopic experience, magnifying endoscopic experience, or confidence level. The data of 29 endoscopists in the interactive training group and 27 endoscopists in the self-driven training group were collected after training, respectively. There were four trainees in the self-driven training group, and two trainees in the interactive training group did not complete the training (Fig. 1). The baseline characteristics of all the participants are shown in Table 1.

Overall, the two training groups showed similar accuracy of NICE classification (71.9% vs. 71.6%, P = 0.900 for the before-training phase; 78.1% vs. 79.3%; P = 0.637 for the after-training phase), vessel analysis (69.8% vs. 66.8%, P = 0.246 for the before-training phase; 77.6% vs. 77.8%, P = 0.939 for the after-training phase), and surface pattern analysis (69.8% vs. 67.7%, P = 0.426 for the before-training phase; 76.9% vs. 78.1%, P = 0.616 for the after-training phase). For the after-training phase, the accuracy of color analysis in the interactive training group was better (80.0.1% vs. 74.4%, P = 0.027).

For high-confidence predictions, there was no difference in the accuracy of histology characterization between the two training groups for the before-training phase (77.4% vs. 73.6%, P = 0.270). For the after-training phase, the two training groups showed similar color analysis (81.0% vs. 79.6%, P = 0.601), vessel analysis (78.5% vs. 83.0%, P = 0.091), and surface pattern analysis (78.5% vs. 81.8%, P = 0. 220) but higher accuracy of NICE classification (73.6% vs. 84.4%, P < 0.001) in the interactive training group.

For low-confidence predictions, the interactive training group was better in the accuracy of NICE classification (75.6% vs. 61.8%, P = 0.027), color analysis (75.5% vs. 56.9%, P = 0.003), and vessel analysis (73.6% vs. 60.2%, P = 0.032). The performance characteristics for predicting colorectal polyp histology in both groups are shown in Table 2.

There was no statistical difference for the accuracy of NICE classification between the two groups for all training phases (Table 3). For the after-training phase, the male trainees (79.7% vs. 71.0%, P = 0.014), the trainees with experience less than 500 cases (80.45% vs. 70.0%, P = 0.008), and the trainees with magnifying endoscopic experience less than 100 cases (80.71% vs. 72.08%, P = 0.020) in the interactive training group had better accuracy of color analysis (Table 3). For the before-training phase, the trainees with experience less than 500 cases (71.1% vs. 63.2%, P = 0.047) in the interactive training group showed better accuracy of vessel pattern analysis (Table 3). No other statistical difference was found for the accuracy of vessel pattern between the two groups (Table 3). No statistical difference for the accuracy of surface pattern was found between the two groups regardless of the training phases (Table 3).