LAUSR

It is with great interest that I read the article by Huang et al.1 The effort made by Huang et al to obtain these data is commendable as they present one of the largest studies to date investigating the prognostic role of human papillomavirus (HPV) in nasopharyngeal carcinoma. Using Kaplan-Meier and log-rank test methods, they have found that patients with Epstein-Barr virus–negative/HPVpositive tumors have significantly better disease-free survival and overall survival than another subgroup, namely the Epstein-Barr virus–positive/HPV-negative subgroup. This is valuable research that answers an important question. However, there are certain methodological issues that bear consideration. In their study, Huang et al used Kaplan-Meier curves in conjunction with the log-rank test to analyze survival. These methods are considered univariate analyses because they describe survival with respect to the single variable under consideration (in this case, the HPV status) and do not take into account other prognostic factors. For their results to be more convincing, there is a need to control for other confounding variables that could affect a patient’s prognosis. It is also important to consider that although the log-rank test is able to discern the statistical difference between groups through the respective P values, it does not offer information on the size of the effect between groups. To compensate for the aforementioned shortcomings, multivariate statistical models are frequently reported in survival analyses. The most used of these is the Cox proportional hazards model. The use of such a multivariate model allows the comparison of several variables simultaneously, and it additionally offers estimates of the effect sizes for each constituent variable. Because these models present much more powerful insights into the data, I am wondering whether the authors considered the use of multivariate modeling to strengthen the veracity of their findings.