Regarding the Predictors of Recurrence and Progression in Poorly Differentiated Cutaneous Squamous Cell Carcinomas

Dear Editor

I read Roccuzzo et al.’s study [1] with great interest. In this study, the author aimed to use the logistic regression models to identify clinico-pathological predictors of cutaneous recurrence, lymphnode/metastatic progression, and both types of progression for cutaneous squamous cell carcinoma patients. In the univariate logistic regression, tumor clinical diameter, depth of infiltration (DOI), and lymphovascular invasion (LVI) were identified as significant predictors across the various types of progression. Then they concluded that Tumor size, DOI, and LVI were significant predictors of recurrence and metastatic progression. Notably, the size of histologically defined tumor-free margins did not affect the risk of recurrence, while LVI emerged as a key predictor of all forms of progression. Despite definite results, I raise some comments on the statistical methods of this study.

It is commonly accepted that a basic statistical rule demands 1 variable per 10 outcome events for binary logistic regression analysis [2‒4]. However, the univariate analysis (Table 2) in Roccuzzo’s study did obey this basic rule while there are 27 covariates but only 32 progressed patients (outcome) among the 161 cutaneous squamous cell carcinoma patients in this logistic regression models for the predictors of local recurrence and progression. According to this basic statistical rule, 32 progressed patients could only at most analyze 4 covariates in this univariate analysis. In contrast, 27 covariates were analyzed in Table 2 in Roccuzzo’s study; hence, this overfitted analysis model cannot obtain reliable statistical results. Therefore, in this univariate logistic regression, tumor clinical diameter, DOI, and LVI may not be true predictors of recurrence and metastatic progression.

Moreover, 9 significant variables were found in Table 1 between the not recurred/progressed group with the recurred/progressed group. Herein, we are curious why not choose these 9 variables for Table 2 for the logistic regression analysis. Additionally, what are the criteria for choosing these 27 covariates in Table 2 for univariate logistic regression? Lastly, we congratulate Roccuzzo and colleagues for their outstanding work despite these comments.