Introduction

Suicide is a major global public health challenge [1, 2]. Approximately, 800,000 individuals (annual rate of 10.6 per 100,000 population) die from suicide annually, with a suicide rate of 13.5 per 100,000 population in men and 7.7 per 100,000 population in women [3]. Persons with a past suicide attempt (SA) [4] usually have 20–30 times higher risk of future completed suicide than those without [5]. The global annual prevalence of self-reported SA is approximately 0.3% [6, 7]. Apart from increased likelihood of completed suicide, SA is highly associated with other negative outcomes, such as physical injuries, hospitalization, and increased treatment burden [8, 9]. For instance, a study found that global rate of years of life lost (YLL) associated with suicidality was 458.4 (438.5–506.1) per 100 000, accounting for 2.18% (1.9–2.2%) of total YLL [10]. Another study in Switzerland found that the extrapolated direct medical cost for medical treatment of SA per year amounted to 191 million Swiss Francs (CHF) [11].

Pregnant and postpartum women with SA also have various negative outcomes; for example, suicidality was associated with 2.2–13% of maternal deaths [12,13,14,15], greater risk of premature labor and caesarean delivery, and increased need for blood transfusion [16, 17]. To allocate health resources, develop relevant policies, implement effective preventive measures and treatment, and reduce poor SA-related health outcomes of in pregnant and postpartum women, better understanding of SA patterns is important. In the past years, a range of studies have examined the prevalence of SA and its correlates among pregnant and postpartum women, with mixed findings. For instance, one study found that the prevalence of SA was 40 per 100,000 pregnant women and risk factors included premature labor, cesarean delivery, and need for blood transfusion [18], while another study found that the prevalence of SA was 5190 per 100,000 pregnant women and risk factors included anxiety/depression, and experience of verbal or physical/sexual abuse [19]. In contrast, a study found that prevalence of SA was 10 per 100,000 postpartum women and risk factors included single, widowed or divorced marital status, history of a caesarean delivery or suicidality, and postpartum depression [20].

To date, no meta-analysis or systematic review synthesized the prevalence and correlates of SA; thus, we conducted this meta-analysis to examine the prevalence and moderating factors (e.g., study design and sites, sample size and mean age) of SA in pregnant and postpartum women. Based on previous findings that pregnant and postpartum depression was common [21,22,23,24] and the association between depression and suicidality [25, 26], we hypothesized that SA in pregnant and postpartum women would be common.

Methods

Data sources and search strategies

The study protocol has been reviewed in the International Prospective Register of Systematic Reviews (PROSPERO: CRD42020188798). This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. A systematic review of relevant publications in PubMed, EMBASE, Web of Science, PsycINFO, Medline complete, Chinese National Knowledge Infrastructure database (CNKI), Chinese Wanfang and Chongqing VIP database was independently searched by three researchers (WWR, YY and TJM) from the inception dates of the target databases up to March 28, 2019. The following search terms were used: attempted suicide, suicide attempt, suicide attempt*, parasuicide*, suicide*, self-injurious behavior, postpartum, perinatal, antenatal, mother, mom, maternal, wife, pregnant women.

Study eligibility

Titles and abstracts of relevant publications were screened, and then the full texts were read independently by the same three researchers. Any disagreements were resolved by a discussion and consensus, with final arbitration by with a senior researcher (YTX) if necessary. Studies were included if they met the following inclusion criteria: (a) SA occurred during pregnancy or the first-year postpartum; (b) cross-sectional or cohort studies; (c) studies reporting prevalence of SA or providing relevant data that enabled the calculation of prevalence of SA. Case reports, study protocols, editorials, systematic reviews and those conducted in special populations [such as those with mental disorders or major medical conditions, or health professionals] were excluded.

Data extraction

Data of study and participant characteristics were independently extracted by two researchers (YY and TJM). Any disagreements were resolved by a discussion, consensus, or consulting another researcher (WWR). Study (publication year and language, country, continent, study design, initial sample size, actual sample size, attrition rate, study site, source of data, and year of survey) and participant characteristics (period, time, mean age and standard deviation, age range and primipara), and suicide attempt-related data (i.e., frequency) were extracted.

Quality assessment

Study quality was assessed using the Parker’s quality evaluation tool for prevalence studies [27], which has been widely used [28,29,30]. The tool contains six items, including definition and representativeness of targeted population, sampling methods, response rate, definition of the target symptom or diagnosis and validation of the assessment instrument. Each item was rated as “1 (yes)” or “0 (no or unclear)”. The total score ranges from 1 to 6 with a higher score indicating better quality [31]. All studies were independently assessed by two researchers (YY and TJM).

Statistical analyses

Prevalence of SA with 95% CI was calculated using the DerSimonian and Laird random-effects model with Natural Log transformation. Heterogeneity was examined using the Q and I2 statistic, and significant heterogeneity was defined as p value ≤ 0.05 in the Q test and I2 > 50%. Publication bias was assessed by visual inspection of the funnel plots, Begg’s and Egger’s tests. The “metatrim” command with the linear trimming estimator was performed in trim and fill adjusted analysis [32]. Subgroup analyses were performed to examine the moderating associations between SA prevalence and the following categorical variables: publication language (English or Chinese), type of study site (hospital, community, or mixed), study design (cross-sectional or longitudinal), and continent (i.e., Asia, America, Europe or Africa). The associations between prevalence of SA and continuous variables including year of publication, survey year based on end year, sample size, age and study quality were analyzed by meta-regression analyses [33]. Sensitivity analyses were used to examine the robustness of the primary results by excluding the included studies one by one. STATA, Version 12.0 for Windows (Stata Corporation, College Station, Texas, USA) and CMA, Version 2 (Biostat Inc., Englewood, New Jersey, USA). Significance level was set at 0.05 (two-sided).

Results

Study selection and characteristics

In the literature search, 1464 relevant studies were identified; of them, 420 were excluded due to duplicates, 955 were excluded by screening titles and abstracts, then 75 were excluded by reviewing the full texts. Finally, 14 studies with 6,406,245 participants were included. The procedure of study selection is summarized in Fig. 1. One study provided data of SA in both antenatal and postnatal women; therefore, the data were extracted and analyzed in two separate groups [34]. Study and participant characteristics are shown in Table 1. The included studies (antenatal: n = 8; postnatal: n = 5; antenatal and postnatal: n = 1) were published between 2006 and 2019, with the sample size ranging from 32 to 4,833,286. Five studies were conducted in Asia, two in Europe, two in Africa, four in North America, and one in South America. Two studies were longitudinal [34, 35], while the remaining were cross-sectional.

Fig. 1
figure 1

Flowchart of literature selection

Full size image
Table 1 Characteristics of the studies included in the meta-analysis
Full size table

Quality assessment, publication bias and sensitivity analysis

Table S1 shows that quality assessment scores of the 14 studies, ranging from 3 to 6, with the median of 4. The funnel plot was symmetrical, and Egger’s and Begg’s tests did not suggest publication bias [antenatal: (Egger: t = 2.21, p = 0.063; Begg: Z = − 0.63, p = 0.532); postnatal: (Egger: t = 0.87, p = 0.435; Begg: Z = − 0.19, P = 0.851); Fig. 2]. The Duval and Tweedies’s trim and fill analysis did not reveal any missing studies, which indicates that no missing effect size qualitatively affected the pooled results. Sensitivity analyses did not find individual study which could significantly change the robustness of the primary results.

Fig. 2
figure 2

Funnel plot of publication bias

Full size image

Prevalence of suicide attempt, subgroup and meta-regression analyses

During pregnancy, the pooled prevalence of SA was 680 per 100,000 (n = 4,883,231; 95% CI 0.10–4.69%; I2 = 99.6%; Fig. 3). During the first year of postpartum, the pooled prevalence of SA was 210 per 100,000 (n = 1,568,376; 95% CI 0.01–3.21%; I2 = 99.5%; Fig. 3). Subgroup analyses found that data collected in field surveys were associated with higher prevalence of SA in both pregnancy and the first year of postpartum (both p < 0.05), while other study characteristics were not associated with prevalence of SA (Table 2). The prevalence estimates of SA by study site and continent are presented in Figures S1 and S2. Meta-regression analysis did not find any study characteristics significantly associated with prevalence of SA (p values > 0.05, Table 2).

Fig. 3
figure 3

Forest plot of the prevalence of suicide attempts in pregnant and postpartum women

Full size image
Table 2 Subgroup and meta-regression analyses
Full size table

Discussion

To the best of our knowledge, this was the first meta-analysis to examine the worldwide prevalence of SA in pregnant and postpartum women. The prevalence of SA was 680 per 100,000 (95% CI 0.10–4.69%) within the pregnancy period and 210 per 100,000 (95% CI 0.01–3.21%) within the first postpartum year. The SA during prenatal and postpartum period could be due to several reasons. First, prenatal and postpartum depression is common [36, 37], which is a major risk factor for suicide-related behaviours including SA [38]. Second, prenatal and postpartum women often experience psychological symptoms related to suicide-related behaviours, such as premenstrual irritability, perceived pregnancy complications, negative attitude toward pregnancy, anxiety about birth, and social withdrawal [35, 39, 40]. Third, some studies [41, 42] found that certain delivery methods, such as caesarean delivery, were associated with increased SA risk [20] among women who are introverted and have high emotionality [43] and lower self-efficacy [44]. Fourth, intimate partner violence, including verbal abuse, physical and/or sexual violence, is associated with increased risk of SA during pregnancy [19].

To date, no meta-analysis on SA among pregnant and postpartum women was published; therefore, we cannot directly compare our findings with estimates from previous studies. We chose to examine differences between our study estimates and those derived from studies of one-year prevalence of SA in other populations. Prevalence of SA in this meta-analysis is similar to the one-year prevalence of SA in the female general population in some studies, e.g., the figure was 300 per 100,000 in high-income countries and 500 per 100,000 in low- and middle-income countries [7]. However, our result is lower than the figures in college female students worldwide (1260 per 100,000, 95% CI 0.82–1.8%) [45] and in Korean female adolescents (6300 per 100,000) [46]. It should be noted that different population characteristics and prevalence timeframe (e.g., during pregnancy and the postpartum period vs. 1 year) hinder direct comparisons between studies. The prevalence of SA in this study was not as high as we hypothesized, which probably may be due to the following reason. Depression is a major contributor of SA [47, 48]. However, most previous studies reporting common pregnant and postpartum depression only focused on mild-moderate depressive symptoms as measured by screening scales [49, 50], rather than the clinical diagnosis of major depressive disorder (MDD). Mild-moderate depressive symptoms may be more likely to increase the risk of suicidal ideation and suicide plan, rather than severe suicidality, such as SA and suicide in clinical practice. As expected, a positive association between data collected in field surveys and the prevalence of SA was found. Data scrutinized from databases were often underestimated because some useful information could not be recorded due to lack of face-to-face interviews and use of standardized instruments which were often used in field surveys.

Strengths of this study include the large sample size and use of sophisticated data analyses (e.g., subgroup and meta-regression analyses, and sensitivity analysis). However, some limitations should be noted in this meta-analysis. First, only articles in English and the Chinese languages were included, thus, those published in other languages may be missed. Second, similar to other meta-analyses of epidemiological studies on suicidality [4, 51], high heterogeneity still remained, even though subgroup analyses were performed. The source of heterogeneity may be related to certain unreported factors, such as pregnancy care, way of delivery, and family relationships. Third, some variables related to SA, such as economic status, information of prior SA in people who died of suicide, history of psychiatric disorders, substance use [8, 52], were not analyzed due to insufficient data reported by included studies. Fourth, the large difference in sample size between studies was mainly due to different source of data (e.g., secondary analyses of databases vs. survey-based samples). However, no moderating effect of the sample size on the results was found in subgroup analyses. In addition, sensitivity analyses did not find outlying studies with large or small sample size that could significantly affect the primary results. Fifth, some studies included in this meta-analysis had lower scores in quality assessment, which might bias the results. Sixth, fatal SAs were not recorded in included studies, which might bias the results. Seventh, due to the relatively low prevalence of SA and/or small sample size in some studies, the confidence intervals were large in the forest plot. Finally, most studies were conducted in Asia and North America, which limits the generalizability of the findings.

In summary, this meta-analysis found that SA is not high in pregnant and postpartum women compared to 1-year prevalence of SA in other populations. Considering the negative impact of SA on health outcomes and potential loss of life, hence clinicians should routinely screen SA in pregnant and postpartum women and undertake effective treatments if necessary, such as public education on suicide prevention and providing hotline services. Appropriate psychotropic treatments should be prescribed for those with severe psychiatric symptoms. Prospective studies on the association between SA and other demographic and clinical variables in pregnant and postpartum women are warranted in the future. In addition, international studies could be conducted to explore the impact of different sociocultural and economic factors on suicide attempt in this population.