Age, choline-to-N-acetyl aspartate, and lipids-lactate-to-creatine ratios assemble a significant Cox's proportional-hazards regression model for survival prediction in patients with high-grade gliomas

To the Editor,

With great interest, we read the article¹ Age, choline-to-N-acetyl aspartate, and lipids-lactate-to-creatine ratios assemble a significant Cox's proportional-hazards regression model for survival prediction in patients with high-grade gliomas (by Roldan-Valadez et al¹ 2016). In this work, the authors developed a Cox's models to predict survival of patients with high-grade gliomas.

In this article, Cox's model¹ (N = 28) predictors included 20 clinical and MR variables. When fitting a prediction model, it is well known that we pick up part of the idiosyncratic characteristics of the data along with the systematic relationship between dependent and explanatory variables. This phenomenon is known as overfitting and generally occurs when a model is excessively complex relative to the amount of data available.² This is commonly referred to as the high-dimensional, low sample size (i.e. p ≫ n) setting.³ Overfitting, which is characterized by high accuracy for a classifier when evaluated on the training data set but by low accuracy when assessed on a separate test set, has been recognized as a problem in p ≫ n settings.³ Overfitted models make the results of a research project appear better than they are, thus (falsely) impressing colleagues and referees.⁴ As a result, the authors are recommended to use variable selection to represent more precise fit of their raw data. Further study with a large cohort is required.