The present study systematically searched the following databases; Cochrane Library (http://www.cochranelibrary.com/), PubMed (https://www.ncbi.nlm.nih.gov/pubmed) and EmBase (https://www.elsevier.com/solutions/embase-biomedical-research) accessed on and prior to 31 July 2016. Articles investigating s-p53 antibodies being used for the detection of CRC were taken into account in the present study. The following combined search terms were used: ‘Colorectal cancer’, ‘colorectal carcinoma’, ‘blood OR serum’, ‘seropositive OR serum antibody’ and ‘p53 OR tp53’. PubMed was used to identify associated articles, which were searched manually for associated functions and references (11). The study was conducted according to the meta-analysis of observational studies in epidemiology (MOOSE) guidelines (12,13) for systemic reviews and meta-analysis was conducted using PRISMA.

All articles were evaluated carefully and eligible articles were included if they met the following criteria: i) Studies investigating the diagnostic value of serum (s)-p53 antibody in CRC; ii) articles that diagnosed CRC using the established gold standard (pathological examination of biopsies) and also tested patient serum for the detection of anti-p53 prior to any treatment and used controls without any other cancer; iii) the results of the included articles on diagnostic accuracy were able to be summarized in a 2×2 table.

The exclusion criteria were as follows: i) If the same author reported results of patients in several publications, only the most relevant one was selected; ii) if available data for analysis was not complete, the study was excluded; iii) all systematic reviews, meta-analyses, case reports and editorials were excluded.

All articles that were included in the present study were assessed by two different authors, any disagreements between the two authors were resolved by discussion. All the included studies were assessed using the quality assessment for studies of diagnostic accuracy (QUADAS) guidelines (14). The Cochrane Collaboration Methods group on screening and diagnostic tests recommended the 11 items of QUADAS. A QUADAS score of ≥7 was considered as good quality and score of <7 was considered as suboptimal quality. The following data were extracted from the included studies: First author's name, year of publication, country of publication, golden criteria of diagnosis, threshold value(s), number of TP (true positive), FP (false positive), FN (false negative), TN (true negative) diagnoses, diagnosis time (the period of time between admission to hospital and diagnosis), stage of tumor (TNM; Japanese classification of colorectal carcinoma, 8th Edition) (15) and research methods. To avoid transcriptional errors all data were reviewed twice.

Meta DiSc statistical software (version 1.4; Universidad Complutense, Madrid, Spain) and STATA (version 14.0, Stata Corporation, College Station, TX, USA) were used to analyze the extracted data. The present study adopted standard methods previously recommended for the meta-analysis of diagnostic test evaluations (16). The sensitivity, specificity, positive and negative likelihood ratio (PLR and NLR), and 95% confidence intervals (CIs) were calculated with a random effects model according to the Mantel-Haensed method, as previously described (17). The diagnostic odds ratio (DOR) was determined using Moses' constant of linear method and was used to measure the accuracy, which indicates the change in diagnostic performance of the test under study per unit increase in the covariant (18). Summary receiver operating characteristic curves (SROC) were used to summarize overall test performance, and the area under the curve (AUC) was calculated. Additionally, the issue associated of sensitivities and specificities of 100% were solved by the default method of adding 0.5 to all cells within the diagnostic 2×2 table (16,19).

The χ² test was applied to detect heterogeneity in the included studies. Inter-study heterogeneity was assessed using the I² test according to the following formula: I²=100% × (Cochran Q-degrees of freedom)/Cochran Q (20). Meta regression was conducted to detect the heterogeneity between studies. In order to detect cut-off threshold effects, Spearman's correlation coefficient was used to evaluate the association between sensitivity and specificity. A scatter plot of the inverse square root of the effective sample size (1/ESS1/2) vs. the diagnostic log odds ratio (lnDOR) was used to visually assess the publication bias is, and should demonstrate a symmetrical funnel shape (21).

A total of 206 abstracts and titles were identified following all search strategies as described in Fig. 1. A total of 136 records were retained in the present study following rejection of replication errors in the included records. In total, 94 articles were excluded, as they did not meet the aforementioned inclusion criteria on the basis of the titles and abstracts. Of the remaining 42 studies, 18 additional articles were excluded due to the lack of a study control, 13 studies were excluded from the meta-analysis due to the lack of s-p53 detection in serum. The final 11 articles included in the present study met all inclusion criteria and were of good quality (Tables I and II).

A total of 11 clinical trials presenting the diagnostic value of s-p53 antibody for CRC diagnosis were investigated in the present study, demonstrating different characteristics. For example, studies included in the meta-analyses were conducted in different countries; 6 studies were conducted in Japan (22–27), 1 was conducted in the USA (28), 1 was conducted in Germany (29), 1 was conducted in France (30) and 2 were conducted in China (31,32). The year of publication ranged between 1997 and 2011. All 11 studies were retrospective; however, 4 studies did not provide TNM stage data (22,23,25,27). In total, 10 studies included health volunteers as a control and 1 used patients with benign disease as the control (32).

Computation of the Spearman correction coefficient for the logit of sensitivity and logit of 1-specificity of s-p53 antibody calculated by metadisc was 0.518 (P=0.102) (data not shown), indicating that there was no threshold effect and positive correlations were not statistically significant.

The random-effects model was applied to pooled data as there were distinct definitions of outcome and differing patient baseline characteristics between trials. Results presented in Fig. 2 demonstrate that the pooled DOR was 6.70 (95% CI, 4.59–9.76), heterogeneity χ² =11.99 (P=0.286) and I²=16.60%. In the present study, the symmetrical SROC of P-53 was 0.73 (Fig. 3). Thus, according to results of the present study, s-p53 antibody exhibited reasonable accuracy in terms of differential diagnosis in cases of CRC. The range of the sensitivity was between 13 and 63%, pooled sensitivity was 0.19 (95% CI, 0.18–0.21; Fig. 5), specificity was between 87 and 100% and pooled specificity was 0.93 (95% CI, 0.92–0.94) (Figs. 4 and 5). In the present study, results demonstrated a pooled PLR of 4.56 (95% CI, 3.27–6.34), suggesting that patients with CRC exhibited 4.56-fold increased chance of testing positive for s-p53 antibody compared with patients without CRC (Fig. 6). Furthermore, the NLR was 0.78 (95% CI, 0.71–0.85; Fig. 7). Significant heterogeneity was identified for all eligible studies. Heterogeneity χ² =84.27 (P<0.001) and I²=88.10%.

The meta-regression was adopted to explore the possible sources of heterogeneity, which included: Variation in the quality of methodology (QUADAS), study design, sample size, assay method, staging system (stage I %) (TNM) and the time taken to collect and fix the sample. Meta-regression indicated that study design (pre- or post-treatment) [relative diagnostic odds ratio, 1.68; 95% (0.65–4.36)] was the probable source of heterogeneity.

Despite the studies included in the meta-analysis exhibiting some heterogeneity, the results of the present study demonstrated that no publication bias was detected by using Egger's test (P=0.72). Furthermore, the funnel plots (Fig. 6) used to detect publication bias also demonstrated no asymmetry upon visual inspection. Sensitivity analysis was conducted in terms of statistical analysis methods, study design and sample size. The results of sensitivity analysis produced no obvious changes. When the studies without matched cases and control sample size were excluded, results were not affected.