This is a retrospective cohort study concerning oral squamous cell carcinoma (OSCC) patients that underwent primary curative surgical resection. The aim of the study is to investigate if patients with neither smoking nor drinking history (NSND) differ from those with smoking or drinking history (ND) in patient characteristics, histopathological characteristics, survival and recurrence rates. The authors claim that after adjusting for confounders, NSND patients are at higher risk of second primary tumors (SPT) compared to ND patients.

I have a few comments on the statistical methods used in the study, listed below:

Major:
1. According to supplementary figure 1, out of 261 patients with oral cavity squamous cell carcinoma, 61 (~24%) were excluded due to "primary chemoradiotherapy, palliative or no treatment." How was intoxication status distributed in these excluded patients, and would it impact the representativeness of the remaining sample?

1. Using univariable model p-values and stepwise regression to filter out confounders is generally unsound. In particular, leaving out gender and age as confounder in the multivariable Cox regression models may lead to bias, given that they are strongly associated to intoxication shown in Table 1. The Kaplan-Meier curves show that the SD group generally has favorable outcomes compared to the NSND group. However, after adjusting for age in the analysis for overall survival in Cox regression, the estimated hazard ratio (HR) for SD vs NSND becomes 1.28. This suggests that age is a strong confounder that induces effect reversal for the relationship between intoxication and overall survival. The same phenomenon may well occur in the analysis of other outcomes, including SPT.

2. In the analysis of secondary events, recurrence and SPT, was death treated as censor, event, or competing risk? A competing risk analysis may be more appropriate here since death and these secondary events should be highly dependent.

Minor
1. Please recheck the numbers presented in Table 1 for typos, eg. number of NSND patients with tumor size <= 20 mm should not be 5.

1. Several p-values were presented as "0.00" while others where presented as "<0.001". Please modify them to reflect the number of significant digits.