Letter to the editor on ‘‘Traction-assisted endoscopic submucosal dissection for resection of ileocecal valve neoplasia: a French retrospective multicenter case series’’

The ileocecal valve (ICV) is considered to be one of the most difficult locations for endoscopic submucosal dissection (ESD). In a recent issue of Endoscopy, Yzet et al. [1] evaluated the efficacy and safety of traction-assisted ESD in this situation, and en bloc and R0 rates were determined. Based on their univariable and multivariable analysis, the authors found that factors predicting non-R0 resection were lesions covering ≥75% of the ICV, and those involving the anal lip or more than two sites on the ICV. Finally, large lesions and anal lip involvement appeared to be factors predictive of difficulty. Despite definite results, we raise some statistical concerns about this study, which could obtain more accurate predictive factor results if the following fundamental statistical rules were considered.

According to the fundamental statistical rule of 10 outcomes for 1 variable for the predictor logistic regression analysis [2] [3] [4], 25 non-R0 resection patients in this study could, at most, analyze 2 predictive variables in Table 3, which shows the factors predicting non-R0 resection in traction-assisted ESD for ICV neoplasia. In contrast, there were 8 variables in Table 3. Thus, 4-fold variables in this overfitted logistic regression analysis model could not obtain reliable results. In other words, at least 80 (8×10) non-R0 resection patients would be needed for predictor analysis in this study. Thus, lesions covering ≥75% of the ICV, and involving the anal lip or more than two sites on the ICV may not be accurate predictors of non-R0 resection in traction-assisted ESD for ICV neoplasia.

Moreover, we wonder about the rationale for selecting these eight variables for the non-R0 resection predictor analysis in Table 3: were they selected according to their clinical endoscopy experience or at random? According to the commonly accepted statistical rule, the author could make the comparison between the R0 resection group and non-R0 resection group by screening the significant variables between these two groups and subsequently using these variables to conduct a predictor logistic regression analysis for non-R0 resection, which would obtain more reliable and accurate predictive results.

Finally, it is our great honor to comment on the outstanding work of Yzet et al. despite these concerns.

Publication History

Article published online:
28 January 2025

• References

• 1 Yzet C, Wallenhorst T, Jacques J. et al. Traction-assisted endoscopic submucosal dissection for resection of ileocecal valve neoplasia: a French retrospective multicenter case series. Endoscopy 2024;
• 2 Pate A, Riley RD, Collins GS. et al. Minimum sample size for developing a multivariable prediction model using multinomial logistic regression. Stat Methods Med Res 2023; 32: 555-571
• 3 van Smeden M, Moons KG, de Groot JA. et al. Sample size for binary logistic prediction models: beyond events per variable criteria. Stat Methods Med Res 2019; 28: 2455-2474
• 4 Riley RD, Ensor J, Snell KIE. et al. Calculating the sample size required for developing a clinical prediction model. BMJ 2020; 368: m441