LAUSR

We were interested to read the paper by Kozaki and colleagues that was published in Archives of Gynecology and Obstetrics [1]. The authors aimed to examine the efficacy of the chemotherapy, predictive factors for the effects of chemotherapy, and prognostic factors in these patients. They concluded that therapy-free interval (TFI) has predictive value for response to chemotherapy and prognosis of patients with recurrent cervical cancer after definitive concurrent chemoradiotherapy (CCRT). Although the useful study has been conducted, some methodological issues need to be noted. First, the Cox proportional hazard (PH) model was used to find predictors of overall survival in uniand multi-variable analyses which should be more checked. The Cox PH model has an important assumption that must be evaluated before applying to the data. Otherwise, biased results may be reported. The Extended Cox can be used when this assumption is not established [2]. So, Kozaki and colleagues should consider this point and use appropriate model. Second, the author developed multivariable regression model but no information has been provided on how the independent variables were imported into the model. In fact, it is unclear based on which criteria the variables were considered for multivariable model. In addition, no multicollinearity is assessed between included variables. Scientists suggest that the Testimation bias and multicollinearity can appear in the multivariable models which have not followed standard model building process [3, 4]. Finally, Kozaki and colleagues stated that TFI has predictive value for response to chemotherapy and prognosis of patients which is questionable. The prediction performance of the developed model should be validated in their study using the Bootstrapping method or in an independent cohort through cross-validation; otherwise, their interpretation is optimistic [4]. The take-home message is that the assumption of regression models and standard process of model building should be considered in epidemiological studies. Also, the validation of prediction model should be evaluated before applying them into clinical practice.