Re: The Predictive Value of the Neutrophil-to-Lymphocyte and Platelet-to-Lymphocyte Ratio in Patients with Recurrent Idiopathic Granulomatous Mastitis

Dear Editor,

We read the comment on our article (1), “The Predictive Value of the Neutrophil-to-Lymphocyte and Platelet-to-Lymphocyte Ratio in Patients with Recurrent Idiopathic Granulomatous Mastitis” with great interest.

We agree with Akbulut and Sahin. Granulomatous mastitis (GM) is a heterogeneous group of diseases of unknown etiology; however, idiopathic granulomatous mastitis (IGM) is a diagnosis of exclusion.

With a cut-off value of 5.02, the preoperative neutrophil-to-lymphocyte ratio had a sensitivity of 62.5% and a specificity of 84.8%, as we reported, as well as a positive predictive value (PPV) of 50.0% and a negative predictive value (NPV) of 90.3% in predicting recurrent IGM. Sensitivity and specificity are independent of the population of interest subjected to the test. They are dependent on the cut-off value above or below which the test is positive. In other words, these diagnostic performance parameters are threshold dependent. However, PPV and NPV are dependent on the prevalence of the disease in the population of interest and is known as changeable parameters according to prevalence (2). In rare cases such as IGM, the recurrence rate may change from region to region as well as according to treatment strategies. Thus, if the prevalence of the disease in a 2 × 2 table is not the same as in the population, interpreting the results according to PPV or NPV may not be the right approach (3).

Multivariate analysis is used to describe analyses of data where there are multiple variables or observations for each unit or individual and is used to further examine the variables indicated by the univariate analysis. Theoretically, every variable collected in the study could be a candidate predictor. However, to reduce the risk of false-positive findings and improve model performance, the events per variable rule of thumb is commonly applied and at a minimum set to 10 for multivariate logistic regression. This rule of thumb recommends that at least 10 individuals need to have developed the outcome of interest for every predictor variable included in the model (4). For logistic regression, the number of events is given by the size of the smallest of the outcome categories. In our study, because of relatively low number of patients and the low recurrence rate (8/33), we could not run multivariate analysis. Interestingly, we are not able to understand how Akbulut and Sahin run a multivariate analysis, without any dataset in their hands. Of course, this is impossible. Because selecting variables for regression analysis using univariate analysis is not the only way; there are multiple methods used for choosing the variables to be included in the final model without introducing bias into the analysis. These variables can be determined by the literature review, the experience in the field, correlation, or maybe risk factors for the disease. We think that to perform a multivariate analysis using the data in table 2 (1) will give a wrong direction.