The prognostic value of CDKN2A hypermethylation in colorectal cancer: a meta-analysis

Search strategy and selection criteria

We conducted and reported this systematic review and meta-analysis following the PRISMA statement (Moher et al, 2009). The following were the criteria for the eligibility of included studies: (1) the study assessed the association between CDKN2A methylation and overall survival (OS) of patients with colorectal cancer; (2) the study evaluated the methylation status of CDKN2A promoter in primary tumour tissues after surgical resection; (3) the study provided a hazard ratio (HR) and 95% confidence interval (CI) directly or gave the data, which allow for the estimation of the HR and 95% CI; (4) the sample size of the study is not <40 patients.

Pubmed, Embase and ISI web of knowledge were searched by using the combinations of following terms: ‘CDKN2A', ‘p16', ‘methylation', ‘colon', 'rectum', 'colorectal', ‘cancer', ‘carcinoma', ‘prognosis', ‘prognostic' and ‘survival'. The search concluded in February 2013, and no lower date limit was used. We only included the studies that were published in the English language, given the fact that other languages were often not available for both authors and readers.

The bibliographies in selected articles were also examined to identify other relevant studies. Conference abstracts were not in the scope of this analysis owing to the insufficient data provided by the authors. All studies were carefully examined to avoid the inclusion of the duplicate data. Two reviewers (XX and CW) assessed the eligibility of the screened studies independently. Agreement was reached for the discrepancies by discussion.

Data extraction and management

Data were extracted independently by two authors (XX and CW) from each eligible study. The predefined form used for data extraction documented the most relevant items including author's name, year of publication, study location, number of patients, methylation detection method, methylation rate and follow-up.

Methodological assessment

Methodological assessment for each of the included studies was performed by three investigators (XX, CW and CM) and scored them using REMARK guidelines and ELCWP quality scale (Steels et al, 2001; McShane et al, 2005). The three investigators reported the quality scores of reviewed studies independently, and reach a consensus value for each item.

Statistical analysis

For the quantitative aggregation of the survival data, HRs and their 95% CIs were used to assess the impact of CDKN2A hypermethylation on survival of colorectal cancer patients. Studies reporting results of multivariate or univariate analysis for survival were used for the aggregation of the survival data. If HRs and their corresponding SEs were not directly reported in the included studies, they were estimated according to the available survival data by using a method reported by Parmar et al (1998). The individual HR estimates were pooled into a summary HR using the approach provided in a previous published study (Yusuf et al, 1985). For the analysis of the association of CDKN2A hypermethylation and clinicopathological features, odds ratios (ORs) and their 95% CIs were applied to estimate the effect. An observed HR or OR >1 implied a worse survival for the group with CDKN2A hypermethylation or the significant association between CDKN2A hypermethylation and clinicopathological characteristics, respectively. We pooled HRs and ORs of the studies by using Review Manager Version 5.1 (The Nordic Cochrane Centre, Copenhagen, Denmark). Publication bias was assessed by using a method reported by Egger et al (1997a). We also explored reasons for inter-study heterogeneity using meta-regression analysis and subgroup analysis by study location, publication year, number of patients, methylation rate and quality score. The analysis of publication bias and meta-regression was performed by using STATA version 10.0 (StataCorp, College Station, TX, USA). Sensitivity analysis was conducted by omission of each single study to investigate the stability of the results.

Study selection and characteristics

Through the database search, a total of 135 articles were identified for initial evaluation (Figure 1). After excluding the studies out of the scope of our meta-analysis, 46 studies evaluating the prognostic effect of CDKN2A hypermethylation in patients with colorectal cancer were considered for further assessment in detail. By further review, 28 were excluded because the estimation of HRs in these studies was not allowed because of the insufficient data provided by the authors, three were excluded because the authors determined the methylation status of CDKN2A promoter by using DNA from other than tumour tissues (e.g., serum), one was excluded because it had overlapped data with another study, two were excluded because the studies were not publications in English, and one was excluded because of the small sample size in this study. Therefore, 11 studies meeting the inclusion criteria were finally enrolled in this systematic review and meta-analysis (Liang et al, 1999; Esteller et al, 2001; Maeda et al, 2003; Ward et al, 2003; Ishiguro et al, 2006; Shen et al, 2007; Barault et al, 2008; Kim et al, 2010; Shima et al, 2011a; Bihl et al, 2012; Veganzones-de-Castro et al, 2012; Figure 1).

The characteristics of the 11 eligible studies are extracted and summarised in Table 1. One study assessed the patients from China, one from Korea, two from Japan, two from Spain, one from France, one from Switzerland, two from USA and one from Australia. The total study sample size was 3440 with a mean of 313 (range, 84–902 patients). Four of these studies included <100 patients and 5 studies enrolled >200 patients (Table 1). These eligible studies were published from 1999 to 2012.

Study results report and meta-analysis

CDKN2A gene was found to be methylated in 23% of total colorectal cancer patients. The survival data by multivariate analysis can be obtained from three studies (27.3% Table 1). Five of 11 studies (45.5%) revealed that CDKN2A hypermethylation was a poor prognostic factor for survival of patients, and 6 (54.5%) reported that CDKN2A hypermethylation did not have an unfavourable impact on survival.

The individual HRs of the 11 included studies were estimated using the methods reported by Parmar et al (1998). Five of these 11 studies directly provided their HRs. Two studies reported the total number of events and the log-rank statistic or its P-value according to which HRs can be approximated. In the four remaining studies, HRs had to be extrapolated from the graphical representations of the survival distributions.

Figure 2 demonstrates a forest plot of the individual HRs and results from the meta-analysis. Overall, CDKN2A hypermethylation in the primary tumour had significant association with an enhanced mortality risk of colorectal cancer patients in the random-effects model (combined HR 1.65, 95% CI 1.29–2.11), despite the exhibition of heterogeneity among studies (I² 69%, P=0.0004). For the exploration of the source of heterogeneity, meta-regression and subgroup analysis were conducted by study location, publication year, number of patients and methylation rate, REMARK score and ELCWP score. Study location was found to be significantly correlated with the inter-study heterogeneity (P=0.008) while other covariates was not (Table 2). Furthermore, subgroup analysis revealed that the significant correlation between CDKN2A hypermethylation and OS was present in Europe (HR 1.49; 95% CI 1.28–1.74) and Asia (HR 3.30; 95% CI 1.68–6.46). We also observed the significant association of CDKN2A hypermethylation and OS of patients in studies with REMARK score >12 (HR 1.75, 1.22–2.51), further confirming the prognostic role of CDKN2A hypermethylation in colorectal cancer. Of note, the pooled HR estimate by multivariate analysis was 1.38 (95% CI: 1.02–1.87), suggesting that CDKN2A hypermethylation might be an independent prognostic factor for patients with colorectal cancer.

In addition, we assessed the association between CDKN2A hypermethylation and clinicopathological features of colorectal cancer patients. As indicated in Table 3, CDKN2A hypermetylation had a significant association with lymphovascular invasion (positive vs negative: OR 1.68, 95% CI 1.15–2.47), lymph node metastasis (positive vs negative: OR 1.68, 95% CI 1.09–2.59) and tumour location (proximal vs distal: OR 2.09, 95% CI 1.34–3.26). The meta-analysis of the correlation between CDKN2A hypermethylation and lymphovascular invasion, lymph node metastasis, Duke's stage and tumour size did not exhibit inter-study heterogeneity (I² 0%), whereas the analysis of other clinicopathological parameters exhibited heterogeneity (I² 51–92%).

Sensitivity analysis revealed that omitting any single study did not affect the pooled HR significantly. The evaluation of publication bias indicated that the Egger test reached the significance (P=0.063<0.1) for studies involved in the analysis of OS. The funnel plots for publication bias also demonstrated a certain degree of asymmetry (Figure 3).