Comment on “An Analysis of the Timeline to Diagnosis and Treatment in Oral Cavity and Oropharynx Cancer”

We read with great interest the article recently published in this Journal entitled “An Analysis of the Timeline to Diagnosis and Treatment in Oral Cavity and Oropharynx Cancer”(Martínez-Ramírez et al. 2025). At an initial reading, the use of Cox proportional hazards models in a study with a cross-sectional design drew our attention (“As an exploratory step, univariate Cox proportional hazards models were fitted…”). Therefore, we consulted the Supporting Information of the original article by Martínez-Ramírez et al. and found that the authors did indeed present the results of this exploratory stage in the form of hazard ratios (HR). In this context, considering that the outcome analyzed was the time interval and that the predictive factors corresponded to clinicopathological characteristics, the exploratory logic adopted by the authors in this initial stage of the analysis is understandable. Another aspect that initially appeared unusual was the use of multivariate negative binomial regression, with the effects presented as incidence rate ratios (IRR). In the Statistical Analysis section, the authors state: “Given the skewness and overdispersion observed in the interval data (variance exceeding the mean), negative binomial regression (NB) was selected as the primary multivariable modeling strategy. In this approach, time (in months) was modeled as count data, and results were expressed as incidence rate ratios (IRR).” Indeed, the study is cross-sectional in its epidemiological design; however, the outcome analyzed is not prevalence, as is the case in most cross-sectional studies, but rather a count variable (time intervals) measured retrospectively in months. Thus, it is understood that the authors are not estimating risk or incidence, but rather the ratio between expected times. The IRR presented, therefore, does not correspond to an epidemiological incidence, but to a mathematical ratio derived from a count model. In this sense, the use of IRR is statistically appropriate to the context, particularly for overdispersed count data, as widely described in the medical literature (Du et al. 2012; Schober and Vetter 2021), although the terminology may potentially generate confusion among clinical readers who are less familiar with this type of modeling. It should be noted, however, that at no point did the authors interpret the IRR as incidence or risk. On the contrary, the IRR was consistently used as a ratio between expected times, derived from negative binomial regression applied to time intervals treated as count data. This conceptual clarity is reflected in the way the results are discussed throughout the manuscript, as exemplified in the statement that “patients with p16-positive OPSCC were slightly more likely to experience a longer PI”. In conclusion, we emphasize that these observations are essentially didactic and terminological, not methodological. We also highlight the relevance of the study and the high quality and transparency of the analysis presented by the authors, which contributes consistently to the understanding of delays in the diagnosis and treatment of oral cavity and oropharyngeal cancer. João Alberto de Almeida Gonçalves: writing – original draft. Alan Carvalho Darós: writing – original draft. Kívia Linhares Ferrazzo: conceptualization, writing – review and editing, supervision. The authors declare no conflicts of interest.