LAUSR

We took an interest to read the article “Relationship between LAT1 expression and resistance to chemotherapy in pancreatic ductal adenocarcinoma” conducted by Altan and colleagues that was published in Cancer Chemotherapy And Pharmacology recently [1]. This was a well-written article and meaningful in this area. However, some statistical methods should be reconsidered seriously because of the rigor of science. The univariate and multivariate survival analyses were performed using Cox’s proportional hazards model for survival in this study. There are two small problems in this statistical method. The first one, the predictors are assumed to have proportional effects during follow-up in the Cox’s proportional hazards model. However, the proportional hazards’ assumption was not examined by the authors. If the proportional hazards’ assumption is violated, the invalid conclusion may be made [2]. The extended Cox model is more appropriate if the proportional hazards assumption is violated. The second one, the clinicopathological factors with p value less than 0.05 in the univariate analysis were proceeded to the multivariate analysis by the authors. Although this strategy can reduce problems of overfitting, the prediction effect of these factors will be overestimated and the Testimation bias will be induced in the outcome [3, 4]. The factors with p values less than 0.2 were imported into the multivariable model which may be more appropriate. The authors indicated that the L-type amino acid transporter 1 as an independent prognostic marker is a potential molecular targeting gene to reduce chemoresistance and tumor growth in patients with pancreatic ductal adenocarcinoma. To draw a more scientific conclusion, especially in small sample size study, the prediction models need to be tested by some appropriate methods such as cross-validation and bootstrapping. Otherwise, these conclusions would be overoptimistic expectations [5].