We followed the guidelines of Meta-analysis of Observational Studies in Epidemiology (5) to conduct a meta-analysis in order to determine the association between thyroid carcinoma and pregnancy. The search terms used were pregnancy, reproduction and thyroid neoplasms. Three main databases, namely PubMed, OVID and the Cochrane Library, were searched from their inception to April 1st, 2013. The search strategies were medical subject headings combined with key words. To ensure a comprehensive search, we also searched the reference lists of the included studies and previously published reviews. The authors were contacted when data were ambiguous or missing. Subsequently, the abstracts were screened and the full-text articles were accessed. No language restriction was applied. The study selection process is summarized in Fig. 1.

The original studies were prospective or retrospective random case-control trials; except for the investigated factors, there was no other difference between the experimental and control groups. All the diagnoses were pathologically confirmed, pregnancy was diagnosed by ultrasound and the follow-up period in all the cases was > 6 months.

We excluded reviews, editorials, letters, case series, case reports and conference proceedings; studies other that random case-control trials; studies with different diagnostic standards and inappropriate outcome measurements; and studies that provided no original data or only unadjusted analysis. Studies with limited follow-up period were also eliminated.

Two investigators extracted data from the eligible studies. The data included first author, year of publication, study design (case-control or prospective cohort), compared populations, inclusion and exclusion criteria, total sample size, number of patients in the thyroid carcinoma and control groups and number of pregnant individuals in each group. We evaluated and filtered all the eligible clinical studies, grading each study according to the Newcastle-Ottawa scale (NOS) (6), with 0–4 points reflecting low quality and 5–9 high quality. Any disagreements or discrepancies were resolved by consensus.

We classified the selected studies, extracted data and performed a meta-analysis using RevMan 5.1 software (http://tech.cochrane.org/revman/download), which was provided by the Cochrane Library.

Of the 627 retrieved articles, 39 abstracts were selected for full-text screening, including 1 Chinese study written in English. The inclusion criteria were met by 21 of the 39 studies (Fig. 1). According to the exclusion criteria, 18 studies were eliminated (3 studies reported the effect of pregnancy after surgery, 2 focused on the association between thyroid nodules and pregnancy, 1 used fine needle aspiration as the diagnostic method, 1 had a different research object, 2 were meta-analyses, 1 had no control group, 4 focused on surgical treatment, 2 had insufficient samples and 2 did not provide original data). The 21 selected studies included 406,329 cases in total (Table I).

A total of 17 studies, including 402,402 cases, investigated the association between pregnancy and thyroid cancer. Of those 17 studies, 2 were prospective studies of high quality. A proportion of the studies did not include detailed information, such as the number of pregnancies or the age at pregnancy. As there was no heterogeneity (P=0.18), the fixed-effects model was selected. A combined analysis of the 17 studies [odds ratio (OR)=1.00, 95% confidence interval (CI): 0.91-1.11] revealed that pregnancy history is not significantly associated with the risk of thyroid carcinoma (Fig. 2).

Following a precision screening, 8 studies were included. Of those 8 studies, 2 indicated that ≥ 3 pregnancies was a risk factor for thyroid cancer, while the remaining 6 studies reported opposing results. There were 4,553 cases in total, without obvious heterogeneity (P=0.11); therefore, the fixed-effects model was selected. In a combined analysis of all 8 studies (OR=1.39, 95% CI: 1.21-1.59), the results demonstrated that women with ≥ 3 pregnancies exhibited an increased risk for thyroid carcinoma (Fig. 3).

Five studies were included, with a total of 120,278 cases. There was no heterogeneity (P=0.76), so the fixed-effects model was selected. In a combined analysis of all 5 studies (OR=1.53, 95% CI: 1.29-1.81), we observed that, compared to the control group, more women in the thyroid carcinoma group had become pregnant within 5 years since their last pregnancy (Fig. 4). Therefore, an interval of ≤5 years since the last pregnancy is associated with a higher risk for thyroid carcinoma.

A total of 3 studies were included in the analysis of the association between thyroid carcinoma during pregnancy and distant metastasis. There were 671 cases in the experimental and 2,859 cases in the control group. There was no heterogeneity (P=0.27) and the fixed-effects model was used (OR=0.94, 95% CI: 0.53-1.67) (Fig. 5). The results indicated that pregnancy did not increase the distant metastasis rate.

A total of 3 studies were included in the analysis of the association between pregnancy and lymphatic metastasis of thyroid carcinoma (Fig. 6). There were 671 cases in the experimental and 2,859 cases in the control group. There was no significant heterogeneity (P=0.18) and the fixed-effects model was used (OR=1.03, 95% CI: 0.86-1.24). The results indicated that pregnancy did not increase the rate of lymphatic metastasis.

We used a funnel graph to represent publication bias and 17 studies that investigated the association between thyroid carcinoma and pregnancy history did not exhibit publication bias (Fig. 7). There was no significant heterogeneity in the analysis of thyroid carcinoma during pregnancy and lymphatic metastasis, but the heterogeneity was significant, so the sensitivity was analyzed. The effect of each study on the combined estimate was assessed by removing one study at a time, which did not affect the significance of the combined estimate or heterogeneity. It appeared that the heterogeneity originated from the study by Vannucchi et al (3), which included a relatively limited number of cases.