Roel Bogie

Development and validation of an educational web-based system for endoscopic classification of laterally spreading tumors

Endoscopic classification of LSTs before and after training in fellows Before training, the IOA for the endoscopic Kudo LST classification among endoscopy fellows was fair to moderate (Gwet’s AC1 coefficient: 0.43, 95% CI: 0.37 – 0.50, Table 2.3 ). Full agreement was present in 4 cases, while in 4 cases only 1 rater disagreed. The overall mean pairwise agreement was 57.4%, with the highest proportion discordant pairs between LST-NG-FE and LST-NG-PD (12.5%) and between LST-G-H and LST-G-NM (11.7%, Table 2.4a ). After recoding the endoscopic Kudo classification into 2 categories (granular vs non-granular subtype), the IOA was also moderate to substantial (Gwet’s AC1 coefficient: 0.63, 95% CI: 0.55 – 0.71; Fleiss kappa: 0.63, 95% CI: 0.62 – 0.65) with a pairwise agreement of 81.5%. After training, the IOA for the endoscopic Kudo LST classification among endoscopy fellows significantly increased (Gwet’s AC1 coefficient: 0.59, 95% CI: 0.53 – 0.65, P <0.001). Full agreement was now present in 6 cases, while in 14 cases only 1 rater disagreed. The Fleiss kappa of all four LST subtypes increased after training ( Table 2.3 ). The overall mean pairwise agreement increased to 68.9%, with the highest proportion discordant pairs between LST-NG-FE and LST-NG-PD (13.4%) ( Table 2.4b ). After recoding endoscopic Kudo classification into granular vs non-granular, the IOA increased to substantial/almost perfect agreement (Gwet’s AC1 coefficient: 0.81, 95% CI: 0.75 – 0.88) with a pairwise agreement of 90.6%. IOA for the Paris classification also statistically improved after training ( P <0.001). Before training, a small, non-significant difference in agreement between fellows of the MUMC+ and NTUH existed: Gwet’s AC1 0.41 (95% CI: 0.34 – 0.48) vs 0.47 (95% CI: 0.40 – 0.54) respectively ( P =0.223). After training, the Gwet’s AC1 value was similar: 0.58 for the MUMC+ (95% CI: 0.51 – 0.65) vs 0.60 for the NTUH (95% CI: 0.53 – 0.66) ( P =0.754). Mean age was slightly higher among the NTUH fellows: 31.4 years (MUMC+, SD: 1.5) vs 33.3 years (NTUH, SD: 1.8) ( P =0.019). Furthermore, the fellows in the NTUH had performed more colonoscopies under supervision (Modus 1-500 in MUMC+ and >2000 in NTUH, P =0.004) and unattended (Modus 0-500 inMUMC+ and 501-1000 in NTUH, P <0.001) than the fellows of the MUMC+. Finally, NTUH fellows were also more experienced with endoscopic resections (Modus 0 in MUMC+ and 1-20 in NTUH, P =0.004). Initially, endoscopy fellows had 68.1% concordance with the overall expert opinion in the endoscopic Kudo classification (see Table 2.5 ). For determining granularity vs non-granularity, the concordance was 87.3%. After the training, the concordance with the experts on endoscopic Kudo classification became 76.5% and for granular vs non-granular status 93.1%. Both before and after training, the endoscopy fellows had most concordant pairs with the cases that the experts classified as LST-G-NM. The largest improvement after training was for the LST-NG-FE cases: from 64.4% to 77.3% accordance. Regarding endoscopy fellows, the concordance with experts increased for 17 raters, remained stable for one rater and decreased for three raters after training (see Figure 2.4 ).

2

29