A total of 226 articles were collected using the search strategy on electronic database PubMed, SinoMed and EMBASE with exception of 35 duplicated articles. Moreover, we excluded 21 case reports, 7 meeting abstracts, 70 reviews and meta-analysis, and some other irrelevant articles. These irrelevant articles included patients with bipolar disease, male patients, epilepsy management or VPA as comparator and so on. In this way, we got 21 full-text articles and assessed the eligibility. After screening each individual article, we extracted the key information and found there were two articles[13, 21] enrolling prepubertal girls, which did not meet our requirements. Besides, we also excluded four articles[36, 39, 44, 47] not providing enough useful data information in each group. Finally, fifteen articles[3, 5, 6, 16, 22, 23, 26, 27, 30-32, 34, 35, 46, 56] were included and their details were given in the table 1. Although two studies had some of the same authors published in the same year, their subjects were recruited from different clinical centers. Hence, we considered that they were not repetitive. Figure 1 showed the flow diagram of the identification and screening of studies.
The fifteen studies included in this paper were prospective, involving 698 women with epilepsy treated with VPA, 927 women with epilepsy treated with other AEDs, 121 women with epilepsy untreated and 423 healthy women as controls. However, we emphasized that all end points were observed not only after menarche but also before menopause. Anti-epilepsy duration was no less than 6 months in order to explore the VPA effects on female reproductive endocrine system, including monotherapy and polytherapy. Except for VPA, carbamazepine (CBZ), lamotrigine (LTG), clonazepam (CNZ), primidone (PMD), phenytoin (PHT) and oxcarbazepine (OXP) were used as comparators, among which CBA and LTG were most common used. The patients involved in each individual study were not all available for PCOS and its components. There were 15 studies showing data in other AEDs treated women[5, 6, 16, 22, 23, 26, 27, 30-32, 34, 35, 46, 56], 4 studies showing data in untreated women[5, 13, 23, 34] and 5 studies showing data in healthy controls[3, 23, 26, 27, 30].
At first, we compared the incidence of PCOS between VPA and other AEDs therapy table 2. Premature withdrawal and follow-up loss led to a major not all proportion of participants completing the study on PCOS. The PCOS incidence was 24.32%, 9.09%, 10.59% and 11.96% respectively in these four groups, VPA treated group, other AEDs group, untreated group and healthy controls. The statistic significant difference in PCOS was present between VPA treated and other AEDs treated group (p<0.001, OR 4.08, 95% CI 2.77-6.01; in Fig 2). The symmetrical Begg’s funnel plot indicated the almost absence of publication bias (Begg’s Test, p=1) while Egger’s publication bias plot showed the minor publication bias (Egger’s Test, p=0.65). Although the diagnostic criteria of PCOS have changed over years, I2 was 8.0%, meaning that the effect of heterogeneity could be neglected.
Then we compared the three critical features of PCOS in VPA treated group and other AEDs group. PCO was considered with 8-10 or more follicles and diameter 2-8mm in 6 studies[6, 22, 30-32, 34], or with transvaginal or transabdominal (in the case with an intact hymen) each ovary existing with 12 or more follicles with size 2-9 mm in diameter and/or volume of ovary >10 mL[3, 13, 46, 56] as well a total of at least 10 cysts 2 to 8 mm in diameter[26, 27]. One study defined ovaries as polycystic when ≥10–15 cysts were seen in a single plane[23]. There is a significant difference of PCO between women treated with VPA and those without VPA (p<0.001, OR 2.58, 95% CI 1.90-3.51; in Fig 3). Hyperandrogenism was a clinical and/or biochemical medical condition characterized by excessive production and/or secretion of androgens with symptoms including hirsutism, acne, alopecia and seborrhea. Women using VPA medication were more susceptible to hyperandrogenism and/or hyperandrogenemia than women using other AEDs (p<0.001, OR 2.31, 95% CI 1.66-3.23; in Fig 4). Menstrual disorders meant the presence of any of the following for more than 6 months: amenorrhea (absence of menstruation), oligomenorrhea (cycle length longer than 35 days), irregular menstrual cycles (cycle length varying more than 4 days, between 21 and 35 days), and polymenorrhea (menstrual cycles lasting less than 21-25 days). VPA treated women were more susceptible to menstrual disorders than non-VPA treated women (p<0.001, OR 1.71, 95% CI 1.32-2.22; in Fig 5).
Funnel plot and Begg’s as well as Egger’s test demonstrated that this meta-analysis would not be subject to publication bias. Cochrane Q test and I2 test did not show heterogeneity in PCOS in total (Fig 2). However, heterogeneity was presence in PCO (Fig 3), hyperandrogenism (Fig 4) and menstrual disorders (Fig 5) of VPA vs other AEDs according to Q test (p<0.1) and I2 test (41.8%-73.0%). For sensitivity analysis of PCO, hyperandrogenism and menstrual disorders, the summary results would not change despite reduced heterogeneity. Finally, we also used a measurement tool AMSTAR 2 (Assess Systematic Reviews, http://www.amstar.ca/) to assess the quality of our meta-analysis and improved the content of our study according to AMSTR checklist. Based on the comprehensive literature search, we listed the excluded articles and justified the exclusions.