This study was registered at PROSPERO (International Prospective Register of Systematic Reviews, www.crd.york.ac.uk/prospero). Number CRD42020198548.

Referring to the international consensus on colorectal liver metastases (16), synchronous CPM could be defined as peritoneal metastases detected at or before diagnosis or at the time of surgery for the primary CRC.

Comparative studies of primary colorectal tumours (with or without synchronous PM data reported about their clinicopathological and molecular characteristics) were eligible for inclusion. The included studies met the recognized diagnostic criteria as follows: primary colorectal tumours; the primary tumour’s pathological diagnosis was confirmed; and the patient’s synchronous PM was confirmed by an imaging diagnosis before surgery, an intraoperative exploration or by a histopathological examination.

The exclusion criteria were as follows: (1) case reports, review articles and animal studies; (2) non-English publications; (3) studies that were not related to CRC or PM; (4) metachronous PM; (5) no analysis of the risk factors; (6) no comparator group; (7) no relevant data, including articles published only in abstract form as well as studies without complete data or an inability to construct a 2×2 contingency table from the present data; (8) mixed primary tumours; (9) a nonstandardized histological type; and (10) synchronous CPM was not clearly or correctly defined.

We selected relevant studies by searching PubMed, Embase and the Cochrane CENTRAL Register of Controlled Trials. The following combined terms were used in the search: (peritoneal metastasis OR peritoneal metastases OR peritoneal carcinomatosis) AND (colorectal OR colon OR rectal). The latest search was implemented on July 14, 2020, and there was not limit to the earliest date of publication.

Two independent authors (Y Zhang and X Qin) checked the title and abstract of each study, and the studies that satisfied the potential eligibility were obtained for further full-text assessment. Disagreements were resolved by discussion with the senior authors (Huaiming Wang or Hui Wang) until a consensus was achieved.

By using standardized forms, two authors (Y Zhang and X Qin) independently extracted the data from each eligible study. The authors resolved any disagreements by discussion with the senior authors (Huaiming Wang or Hui Wang). The following data were extracted from each eligible study: author, year of publication, country where the study was conducted, setting of the centre, type of study, enrolment interval, number of primary CRC patients with or without synchronous PM and the clinicopathological and molecular characteristics. In addition, the Newcastle–Ottawa Scale score (N-O score) was also calculated and extracted for all of the eligible studies.

We used Comprehensive Meta-Analysis (version 2.0) and Stata (version 12.0) for all statistical analyses. All of the pooled outcomes were determined using a random effects model (DerSimonian–Laird method). In the pooled analyses of the associations between the clinicopathological-molecular characteristics and synchronous PM, the effect sizes were calculated as the odds ratios (ORs) with 95% confidence intervals (CIs). The χ²-based Cochran Q test was used to assess heterogeneity among the studies, in which P < 0.1 indicates the presence of heterogeneity (17). We also performed I² inconsistency testing to assess the extent of the heterogeneity among studies, with values greater than 50% regarded as a moderate-to-high heterogeneity (18). For significant heterogeneity, a sensitivity analysis or a subgroup analysis was performed to find the potential source of the heterogeneity. The sensitivity analysis was performed by omitting each study sequentially in order to test the influence of each individual study on the pooled result. The evidence for a publication bias was evaluated by the visual inspection of the funnel plot for symmetry (an asymmetric plot suggested possible publication bias) and was quantified by means of the Begg’s test, with a P value < 0.05 regarded as a significant publication bias (19).

The qualities of the included studies were assessed using the Newcastle–Ottawa Scale (20), in which a score ≥ 6 indicates a high-quality study. The qualities of the studies were evaluated by examining 3 categories: patient selection, comparability of the 2 study groups, and the assessment of exposure (maximum score 9), as shown in the Newcastle–Ottawa Scale.

The initial search yielded a total of 9470 studies. After removal of duplicates, a total of 7659 studies were screened by analysing their titles and abstracts, and 7435 studies were removed because they met one or more of the exclusion criteria. The remaining 224 studies were then assessed for eligibility by full-text examination, and a further 199 were excluded due to ineligibility. The reasons for exclusion were recorded. Finally, 25 studies were included in the final analysis ( Figure 1 ) (2, 10–13, 21–40).

Among the 25 included studies, 7 had a multicentre setting, and 18 had a single centre design. Five of the included studies were prospectively performed; the remaining twenty were retrospective. All included studies were considered high quality (N-O score ≥ 6). Complete characteristics of the included studies are available in Table 1 .

Six clinicopathological and molecular factors could not be included in the quantitative synthesis because they had only a single study in their subgroup, or their methodology did not permit for the pooling of the data. The six factors were serum CEA (21), serum CA125 (29), connective tissue growth factor (CTGF) (37), discoidindomain receptor 2 (DDR2) (27), vimentin (VIM) (39), and TP53 (25). We included these factors in Table 1 for completeness, but they were not included in the final quantitative synthesis through the meta-analysis.

Finally, 21 studies on 13 factors were included in the quantitative synthesis through the meta-analysis: 7 studies on sex, 4 studies on the tumour invasion depth, 3 studies on lymph node metastasis, 5 studies on the differentiation, 6 studies on the primary tumour site, 7 studies on the histological findings, 2 studies on serum CA19-9, 2 studies on PROK1/PROKR2, 9 studies on BRAF, 6 studies on KRAS, 2 studies on NRAS, 2 studies on PIK3CA and 4 studies on the MSI-H/dMMR status.

Seven studies (2, 10, 12, 13, 21, 30, 38) that included 160679 patients (30366 synchronous pmCRC, 130313 nonpmCRC) and that evaluated the patients’ sex were included in the meta-analysis. The pooled analysis indicated that females were positively associated with synchronous PM compared to males (OR 1.299; 95% CI, 1.118 to 1.509; P = 0.001) ( Figure 2A ). There was significant heterogeneity (Cochran Q, P < 0.001; I² = 76.9%). To explore the possible sources of the heterogeneity, a sensitivity analysis was performed by omitting each study sequentially to test the influence of each individual study on the pooled result. When one study was omitted (12), there was no significant heterogeneity (Cochran Q, P = 0.099; I² = 46.0%), with no noticeable influence on the pooled OR confidence interval (OR 1.233; 95% CI, 1.051 to 1.445; P = 0.010). It is noteworthy that the proportion of females in the PM group was > 50% in that one study, but in the others, the proportions were < 50%.

Four studies (10, 21, 22, 38) that included data from 19432 patients (809 synchronous pmCRC, 18623 nonpmCRC) regarding the tumour invasion depth were included in the meta-analysis. The pooled analysis indicated that T4 was positively associated with synchronous PM compared with T1-3 (OR 12.331; 95% CI, 7.734 to 19.660; P < 0.001) ( Figure 2B ). There was significant heterogeneity (Cochran Q, P = 0.009; I² = 74.2%). When one study was omitted (38), there was no significant heterogeneity (Cochran Q, P = 0.593; I² = 0%), with no noticeable influence on the pooled result (OR 16.028; 95% CI, 11.439 to 22.457; P < 0.001).

Three studies (10, 22, 38) that included data from 16097 patients (702 synchronous pmCRC, 15395 nonpmCRC) and that compared lymph node metastasis were included in the meta-analysis. The pooled analysis indicated that N1-2 was positively associated with synchronous PM compared with N0 (OR 5.665; 95% CI, 3.628 to 8.848; P < 0.001) ( Figure 2C ). There was significant heterogeneity (Cochran Q, P = 0.068; I² = 62.7%). The heterogeneity disappeared if the study was omitted (Cochran Q, P = 0.765; I² = 0%) (22), with no noticeable influence on the pooled result (OR 4.558; 95% CI, 3.755 to 5.533; P < 0.001).

Five studies (10, 12, 21, 38, 40) that included data on 108360 patients (21986 synchronous pmCRC, 86374 nonpmCRC) and that compared the differentiation, were included in the meta-analysis. The pooled analysis indicated that a poorly differentiated grade was positively associated with synchronous PM compared with a well/moderately differentiated grade (OR 2.560; 95% CI, 1.537 to 4.265; P < 0.001) ( Figure 2D ). There was significant heterogeneity (Cochran Q, P < 0.001; I² = 94.5%). The heterogeneity disappeared when one of the studies was omitted (Cochran Q, P = 0.636; I² = 0%) (12), with no noticeable influence on the pooled result (OR 3.352; 95% CI, 2.875 to 3.909; P < 0.001).

Six studies (2, 10, 13, 21, 22, 38) regarding colon cancer were included in the meta-analysis. The PM status of right and left colon cancer patients were listed as follows, respectively: Right colon (720 synchronous pmCRC, 6158 nonpmCRC) and left colon cancer (568 synchronous pmCRC, 7822 nonpmCRC). Quantitative synthesis showed that synchronous PM was positively associated with right colon cancer (OR 2.468; 95% CI, 2.050 to 2.970; P < 0.001) ( Figure 3A ). There was no significant heterogeneity (Cochran Q, P = 0.119; I² = 42.9%). Besides, synchronous PM was not associated with left colon cancer (OR 1.000; 95% CI, 0.761 to 1.314; P = 0.998) ( Figure 3B ). There was significant heterogeneity (Cochran Q, P = 0.004; I² = 71.4%). When one study was omitted through the sensitivity analysis (10), the heterogeneity was less significant (Cochran Q, P = 0.049; I² = 58.0%), with no noticeable influence on the pooled result.

Five studies (2, 13, 21, 22, 38) that included data on 23278 patients (1519 synchronous pmCRC, 21759 nonpmCRC) and that evaluated rectal cancer, were included in the meta-analysis. The pooled analysis indicated that rectal cancer was negatively associated with synchronous PM compared with colon cancer (OR 0.323; 95% CI, 0.284 to 0.368; P < 0.001) ( Figure 3C ). No significant heterogeneity existed (Cochran Q, P = 0.969; I² = 0%).

Six studies (22, 23, 33, 35, 38, 40), which included data on 24252 patients (1600 synchronous pmCRC, 22652 nonpmCRC) regarding nonmucinous adenocarcinoma (NMC), were included in the meta-analysis. Synchronous PM was negatively associated with NMC (OR 0.319; 95% CI, 0.237 to 0.429; P < 0.001) ( Figure 4A ). There was significant heterogeneity (Cochran Q, P = 0.005; I² = 70.4%). The heterogeneity disappeared if one of the studies was omitted (Cochran Q, P = 0.106; I² = 47.5%) (33), with no noticeable influence on the pooled OR and confidence interval (OR 0.353; 95% CI, 0.285 to 0.437; P < 0.001).

Seven studies (12, 22, 23, 33, 35, 38, 40), which included data on 154377 patients (29448 synchronous pmCRC, 124929 nonpmCRC) regarding mucinous adenocarcinoma (MC), were included in the meta-analysis. Synchronous PM was positively associated with MC (OR 3.565; 95% CI, 2.095 to 6.064; P < 0.001) ( Figure 4B ). There was significant heterogeneity (Cochran Q, P < 0.001; I² = 97.1%). To explore the possible sources of the heterogeneity, a subgroup analysis was performed. According to the rate of PM, two of the studies were divided into subgroup one, and there was no significant heterogeneity (Cochran Q, P = 0.228; I² = 31.2%) (12, 33), with no noticeable influence on the pooled result (OR 7.518; 95% CI, 4.952 to 11.412; P < 0.001). The other studies were divided into subgroup two that also had no significant heterogeneity (Cochran Q, P = 0.174; I² = 37.0%) (22, 23, 35, 38, 40), with no noticeable influence on the pooled result (OR 2.645; 95% CI, 2.169 to 3.226; P < 0.001).

Three studies (22, 23, 35), which included data on 5741 patients (673 synchronous pmCRC, 5068 nonpmCRC) regarding signet-ring cell carcinoma (SRCC), were included in the meta-analysis. Synchronous PM was positively associated with SRCC (OR 4.480; 95% CI, 1.836 to 10.933; P = 0.001) ( Figure 4C ). There was significant heterogeneity (Cochran Q, P = 0.036; I² = 69.7%). When one study was omitted (22), there was no significant heterogeneity (Cochran Q, P = 0.656; I² = 0%) and no noticeable influence on the pooled result (OR 2.986; 95% CI, 1.741 to 5.123; P < 0.001). It is noteworthy that the omitted study had a much higher OR value.

Levels of up to 37.0 µ/ml were taken as the upper cut-off values for the Serum CA19-9 reference ranges. Two studies (21, 36), which included data on 728 patients (24 synchronous pmCRC, 704 nonpmCRC) regarding serum CA19-9, were included in the meta-analysis. Synchronous PM was positively associated with elevated serum CA19-9 (OR 12.868; 95% CI, 5.196 to 31.867; P < 0.001) ( Figure 5A ). No significant heterogeneity existed (Cochran Q, P = 0.710; I² = 0%).

Two studies (31, 34), which included data on 944 patients (29 synchronous pmCRC, 915 nonpmCRC) regarding PROK1/PROKR2, were included in the meta-analysis. Synchronous PM was positively associated with PROK1/PROKR2 positivity (OR 2.244; 95% CI, 1.031 to 4.884; P = 0.042) ( Figure 5B ). There was no significant heterogeneity (Cochran Q, P = 0.344; I² = 0%).

Nine studies (2, 11, 13, 24–26, 28, 30, 32) that included data on 4979 patients (704 synchronous pmCRC, 4275 nonpmCRC) regarding the patients’ BRAF statuses, were included in the meta-analysis. Synchronous PM was positively associated with BRAF mutations (OR 2.586; 95% CI, 1.674 to 3.994; P < 0.001) ( Figure 5C ). There was significant heterogeneity (Cochran Q, P = 0.019; I² = 56.3%). When one study was omitted (25), there was no significant heterogeneity (Cochran Q, P = 0.073; I² = 45.9%) and no noticeable influence on the pooled result (OR 2.305; 95% CI, 1.569 to 3.385; P < 0.001). It is clear that the study had a smaller sample size.

Six studies (2, 11–13, 25, 30), which included data on 134197 patients (28362 synchronous pmCRC, 105835 nonpmCRC) regarding the KRAS status, were included in the meta-analysis. Synchronous PM was not associated with KRAS mutations (OR 0.972; 95% CI, 0.576 to 1.638; P = 0.914) ( Figure 6A ). There was significant heterogeneity (Cochran Q, P < 0.001; I² = 92.4%). The heterogeneity disappeared if one of the studies was omitted (Cochran Q, P = 0.774; I² = 0%) [12], with no noticeable influence on the pooled result (OR 1.202; 95% CI, 0.994 to 1.453; P = 0.057). It was found that the rate of KRAS mutations in the synchronous PM group was much lower in the omitted study.

Two studies (11, 25), which included data on 731 patients (43 synchronous pmCRC, 688 nonpmCRC) regarding the patients’ NRAS status, were included in the meta-analysis. Synchronous PM was not associated with NRAS mutations (OR 1.140; 95% CI, 0.133 to 9.748; P = 0.905) ( Figure 6B ). No significant heterogeneity existed (Cochran Q, P = 0.373; I² = 0%).

Two studies (11, 30), which included data on 897 patients (93 synchronous pmCRC, 804 nonpmCRC) regarding the PIK3CA status, were included for eligibility in the meta-analysis. Synchronous CPM was not associated with PIK3CA mutations (OR 0.667; 95% CI, 0.289 to 1.540; P = 0.343) ( Figure 6C ). There was no significant heterogeneity (Cochran Q, P = 0.415; I² = 0%).

Four studies (11–13, 28), which included data on 131015 patients (27922 synchronous pmCRC, 103093 nonpmCRC) regarding their MSI-H/dMMR status, were included in the meta-analysis. Synchronous CPM was not associated with MSI-H/dMMR (OR 1.087; 95% CI, 0.351 to 3.367; P = 0.885) ( Figure 6D ). There was significant heterogeneity (Cochran Q, P = 0.097; I² = 52.5%). When one study was omitted (13), there was no significant heterogeneity (Cochran Q, P = 0.153; I² = 46.6%), with no noticeable influence on the pooled result (OR 1.481; 95% CI, 0.536 to 4.087; P = 0.449).

No significant publication bias was found according to the visual inspection of the funnel plot and Begg’s test ( Supplementary Figures S1 – S6 ).