A systematic literature search of Pubmed and Embase databases was conducted to identify all the studies published before May 1, 2014, which investigated the association of circulating IGF-1 or IGFBP-3 with the ovarian cancer risk. The following keywords were used during the search: ‘Insulin-like growth factor’ and ‘ovarian cancer’. Reference lists of relevant studies and general reviews were also searched. No restriction of languages was imposed. The systematic review was performed following the guideline of the Meta-analysis Of Observational Studies in Epidemiology (13).

The eligibility of studies was assessed by two investigators (Q.W. and H.L.P.) independently. Studies were included into the meta-analysis when they met the following criteria: i) Cohort or case-control studies published in full texts; ii) researching the association of circulating IGF-1/IGFBP-3 with ovarian cancer risk; iii) categorizing circulating IGF-1/IGFBP-3 concentrations into tertile or quartile levels; and iv) reporting odds ratios (ORs) with 95% confidence intervals (CIs), with results from crude and adjusted (adjusted for confounders) models. Study quality was assessed using the nine-star Newcastle-Ottawa scale (14).

The two investigators mentioned performed data extraction independently and discussed the results to make a consensus. The following variables were recorded: First author's name, publication year, geographic region where the study was conducted, study period, study design, sample size, IGFs assay, categorizing and corresponding cut-points of IGF-1/IGFBP-3 levels, ORs with 95% CIs for each category verses reference (the lowest category) of IGF-1/IGFBP-3, the numbers of cases and controls in each exposure category, and confounders considered in the adjustment models. When necessary, the primary authors were contacted for additional information.

Considering that the IGF assays and categorizing cut-points differed across studies, a random effects model (15) was conducted by combining study-specific maximally adjusted ORs for the highest verses lowest exposure levels to evaluate the association of IGF-1/IGFBP-3 with ovarian cancer risk. The pooled ORs and 95% CIs were used to assess the strength of association. Studies reporting the exact numbers of cases and controls in each exposure level were included in the random effects dose-response meta-analysis. The method proposed by previous studies (5,16) was used to calculate the linear trends by relating the log of ORs for different exposure levels.

The degree of heterogeneity among studies was assessed via Q and I² statistics. Subgroup analyses and meta-regression models (17) were conducted to investigate sources of heterogeneity. Sensitivity analyses were performed to evaluate the stability of the results. To estimate the publication bias, funnel plots were used. Funnel plot asymmetry was assessed by Egger's regression test (P<0.05 was considered to indicate a statistically significant publication bias) (18). All the analyses were repeated with unadjusted ORs.

All the statistical tests were performed with the Stata software (version 12.0; StataCorp, College Station, TX, USA). P<0.05 was considered to indicate a statistically significant difference.

A flow diagram of the literature search is shown in Fig. 1. Among the 64 potentially relevant studies, four were finally included in the meta-analysis.

The remaining four studies (10–13,19) were published in English between 2002 and 2007 and involved a total of 1,958 patients (627 cases and 1,358 controls). Overall, the methodology qualities of studies were satisfying, since study designs, sample sizes, exclusion criteria and diagnosis of cases were clearly demonstrated, except for the limited information of follow-up in certain studies. Of these four included case-control studies, three were nested within large prospective cohorts (10–13). The study of Peeters et al (10) was population-based and the others were hospital-based. All three prospective studies reported analyses while excluding cases diagnosed within 1 or 2 years of blood donation. Peeters et al (10) and Lukanova et al (11) reported exclusions of females currently using exogenous hormones at the time of blood donation, which was not mentioned in other studies. Categories of IGF-1/IGFBP-3 levels were calculated on the basis of the distribution of controls (10,19) or the distribution of cases and controls (11,12). Intra-assay coefficients of variation were reported as generally low in all the studies. All the studies adjusted crude ORs with ≥3 confounders. ORs of three studies were further adjusted; IGF-1 models for levels of IGFBP-3, and IGFBP-3 models for levels of IGF-1. One study [Dal Maso et al (19)] reported the results of free and total circulating IGF-1 levels, where the data of total IGF-1 was used. Study characteristics are demonstrated in Table I.

There were no statistically significant associations of circulating IGF-1 or IGFBP-3 with ovarian cancer risk when the maximally adjusted ORs for the highest verses lowest exposure levels in each study were pooled into meta-analysis [OR, 0.85 (95% CI, 0.51–1.40) for IGF-1 and OR, 0.78 (95% CI, 0.43–1.40) for IGFBP-3]. When participants in all the studies were analyzed as a whole, potential heterogeneity among the studies was represented in the analysis model (P=0.103, I²=51.5% in the IGF-1 model; and P=0.042, I²=63.4% in the IGFBP-3 model) (Figs. 2 and 3).

In the subgroup analyses according to the age at diagnosis, the pooled ORs of IGF-1 were 1.89 (95% CI, 0.64–5.59) for the subgroup with cases diagnosed before 55 years (tests of heterogeneity: P=0.063, I²=63.7%) and 0.74 (95% CI, 0.50–1.08) for the subgroup with cases diagnosed at ≥55 years (tests of heterogeneity: P=0.362, I²=0.0%). The pooled ORs of IGFBP3 were 1.08 (95% CI, 0.50–2.32) for the subgroup cases diagnosed before 55 years (tests of heterogeneity: P=0.209, I²=36.1%) and 0.98 (95% CI, 0.73–1.33) for the subgroup with cases diagnosed at ≥55 years (tests of heterogeneity: P=0.617, I²=0.0%) (Figs. 2 and 3).

In the meta-regression models, the residual variation due to heterogeneity was not changed by geographic regions, year of publication, sample type (serum or plasma), assays (immunoradiometric assay or ELISA), confounders (smoking status or ever use of hormones) or range of exposure levels. However, the Knapp-Hartung meta-regression model (17) reported that the heterogeneity in the meta-analysis model of IGF-1 was mainly from studies that did not exclude females using exogenous hormones at the time of blood donation (residual I²=0.0%). However, no significant variation in pooled OR and 95% CI with Knapp-Hartung modification was identified. No evidence of interactions with the above variables was found in the meta-regression analyses of the IGFBP-3 model. The results of the sensitivity analyses indicated that the significance of pooled ORs was not influenced by each individual study. The funnel plot shapes revealed no clear asymmetry. Further Egger's test (18) suggested no evidence of publication bias (P=0.404 in IGF-1 model and P=0.062 in IGFBP-3 model).

Three studies reporting the exact numbers of cases and controls in each exposure category were included in the random effects dose-response meta-analysis (15). No linear association was found between circulating IGF-1/IGFBP-3 levels and ovarian cancer risk [OR, 0.39 (95% CI, 0.13–1.13), P=0.083 for IGF-1 and OR, 0.93 (95% CI, 0.72–1.21), P=0.584 for IGFBP-3]. All the analyses were repeated with unadjusted ORs and similar findings were identified.