Specificity of Metabolic Colorectal Cancer Biomarkers in Serum Through Effect Size.

INTRODUCTION___ Colorectal cancer is one of the most diagnosed cancers, leading to a large number of deaths. In addition to existing screening methods, metabolic profiling can help both to diagnose and to understand the various states of the disease. Objectives We aimed to find candidate biomarkers (CB) of the active state of colorectal cancer that are specific to it, in comparison to the remission state of the disease evaluated on distinct patients (cross-sectional study).___ METHODS___ All serum samples were analyzed using comprehensive two-dimensional gas chromatography (GC×GC) coupled to high resolution time of flight mass spectrometry through an optimized and validated untargeted analytical method regulated by a quality control (QC) system. First, we used a specific multi-approaches data (pre)processing workflow to highlight, annotate and assess the performances of the most altered metabolites between the active state (CRC, n=18) and healthy control samples (HC, n=19) specifically matched for age and gender, two of the most influential bias factors. On the contrary, due to the difficulty to control and to match for all clinical and demographic traits when sampling small cohorts, we used samples from patients in remission (R-CRC, n=17) that were not matched for age and gender. Because of the consequent risk of bias, the usual null hypothesis significance tests (NHST) could not be applied reliably. Therefore, we compared the R-CRC samples to another specifically matched group of healthy controls (R-HC, n=17), and used this comparison to indirectly compare the two states of the disease (CRC vs R-CRC) through a measure called effect size (ES) whose methodological aspects were investigated.___ RESULTS___ We obtained 24 candidate biomarkers able to discriminate the CRC and HC samples efficiently (ROC AUC of 0.86, sensitivity and specificity of 0.72 and 0.78). 10 of those were found to have signals significantly returning to normal, near healthy levels in the R-CRC samples and were therefore specific to the active state of colorectal cancer. Regarding the effect size, we observed that, in a point biserial case, many statistical values led to identical ES, and r-like and d–like ES were directly convertible. Only linear and rank-based ES were different. Hedges’ g, Spearman’s rho and Kendall’s tau are therefore recommended, along with an unstandardized ES such as the percentage of variation of the raw analytical signals. The confidence intervals were of particular interest to consider the uncertainty of the measure and were well represented through scatterplots and distribution curves.___ CONCLUSION___ The candidate biomarkers found, along with their specificity, could, after proper validation on independent cohorts, complement the detection of colorectal cancer and help understand its occurrence and remission. The effect size, used and studied on a MS global profiling data set, is an ideal complement to NHST and a useful tool to compare and combine distinct cohorts, here within a study but also between studies (meta-analysis).