Prognostic DNA methylation markers for sporadic colorectal cancer: a systematic review

Abstract

Background: Biomarkers that can predict the prognosis of colorectal cancer (CRC) patients and that can stratify high-risk early stage patients from low-risk early stage patients are urgently needed for better management of CRC. During the last decades, a large variety of prognostic DNA methylation markers has been published in the literature. However, to date, none of these markers are used in clinical practice.

Methods: To obtain an overview of the number of published prognostic methylation markers for CRC, the number of markers that was validated independently, and the current level of evidence (LoE), we conducted a systematic review of PubMed, EMBASE, and MEDLINE. In addition, we scored studies based on the REMARK guidelines that were established in order to attain more transparency and complete reporting of prognostic biomarker studies. Eighty-three studies reporting on 123 methylation markers fulfilled the study entry criteria and were scored according to REMARK.

Results: Sixty-three studies investigated single methylation markers, whereas 20 studies reported combinations of methylation markers. We observed substantial variation regarding the reporting of sample sizes and patient characteristics, statistical analyses, and methodology. The median (range) REMARK score for the studies was 10.7 points (4.5 to 17.5) out of a maximum of 20 possible points. The median REMARK score was lower in studies, which reported a p value below 0.05 versus those, which did not (p = 0.005). A borderline statistically significant association was observed between the reported p value of the survival analysis and the size of the study population (p = 0.051). Only 23 out of 123 markers (17%) were investigated in two or more study series. For 12 markers, and two multimarker panels, consistent results were reported in two or more study series. For four markers, the current LoE is level II, for all other markers, the LoE is lower.

Conclusion: This systematic review reflects that adequate reporting according to REMARK and validation of prognostic methylation markers is absent in the majority of CRC methylation marker studies. However, this systematic review provides a comprehensive overview of published prognostic methylation markers for CRC and highlights the most promising markers that have been published in the last two decades.

Keywords: Biomarker; Colon cancer; Colorectal cancer; DNA methylation; Methylation marker; Patient outcome; Prognosis; REMARK; Survival.

Fig. 1

Flowchart of the study identification process. A total of 83 studies were selected for qualitative assessment

Fig. 2

Quality assessment of methylation marker studies. a Histogram depicting the REMARK score distribution for all studies included in the analysis (mean REMARK score = 10.711, standard deviation = 2.820). b Histogram showing the completeness of reported REMARK items

Fig. 3

Association between REMARK score and significance level of studies. a Box plot comparing the REMARK score between studies that reported statistically significant findings versus studies that did not report statistically significant findings (Mann-Whitney test, p = 0.005). b Dot plot showing that there was, however no significant correlation between the REMARK score and the reported p values (Pearson’s correlation coefficient = 0.0543, p value = 0.499). c Dot plot depicting that for the reported survival analyses, we found a stronger, but still statistically not significant correlation between the number of patients used and the reported p values (Pearson’s correlation coefficient = 0.1814, p value = 0.051)

Fig. 4

Forest plots of reported methylation markers in colorectal cancer studies. Forest plots were prepared for methylation markers that were reported in two or more publications or study populations. The hazard ratios (HR) are sorted according to the REMARK score. HRs with a statistically significant association are depicted with a solid line; HRs of reported markers with no significant association are depicted with a dotted line; HRs of subgroup analyses are depicted in blue. Univariate HRs and confidence intervals (CI) are reported unless multivariate HRs were available (a). As for IGFBP3 and TFAP2E the HRs of the study of Perez-Carbonell et al. [90] and Zhang et al. [117], respectively, were both associated with worse survival. For this figure, the HR was reversed for visualization purposes (b). A multivariate HR for BNIP3 methylation was available in the study of Shimizu et al., however was not statistically significant (c)