Factors associated with suicidal behaviors in mainland China: a meta-analysis

Electronic database and manual searches were conducted to identify published articles for review. Four databases were used to search for articles published up to January, 2011: Medline/PubMed, EMBASE, CNKI (China National Knowledge Infrastructure) and VIP (Chinese Journal of Science and Technology of VIP). The searches applied a mixture of free text and index terms. The search terms in PubMed were ("Suicide"[Mesh] AND "Risk Factors"[Mesh]) AND "China"[Mesh], ("Suicide"[Mesh] AND "China"[Mesh], (("Risk Factors"[Mesh]) AND "Suicidal Ideation"[Mesh]) AND "China"[Mesh], (("Risk Factors"[Mesh]) AND "Suicide, Attempted"[Mesh]) AND "China"[Mesh], risk factor for suicide in China, suicide in China. The results in Medline/PubMed were limited to full text, human, English and Chinese. The EMTREE search in EMBASE was 'suicide'/exp AND 'risk factor'/exp AND 'china'/exp, 'suicide'/exp AND 'china'/exp, 'risk factor'/exp AND 'suicidal ideation'/exp AND 'China'/exp, 'risk factor'/exp AND 'suicide attempt'/exp AND 'China'/exp. All results in EMBASE were limited to English, human. CNKI and VIP were used to search for articles in Chinese, and the subject words included risk factors for suicide, risk factors for suicide ideation, and risk factors for attempted suicide. We reviewed the references cited in the retrieved articles and then searched the bibliographies of these retrieved papers.

Studied design included cross-sectional, case–control, and cohort studies. Study subjects were from the general population of mainland China. Suicidal behavior outcomes included suicide ideation (thoughts of engaging in behavior intended to end one's life), suicide attempt (engagement in potentially self-injurious behavior in which there is at least some intent to die), and completed suicide (the act of intentionally ending one's life). We also included the studies that were designed with the original purpose of analyzing risk behaviors other than suicide among Chinese people, but the information on the risk factors for suicide ideation, suicide attempt or completed suicide were provided. Studies that focused on subjects from Hong Kong or Taiwan, that investigated psychopathic suicide, that repeated data reported in earlier studies, or that were systematic reviews or meta-analyses were excluded from our analysis.

To determine the validity of the research, the quality of each case–control study was assessed by the second author according to the primary criteria for nonrandomized studies described in the Newcastle-Ottawa Scale (NOS) [14] as follows: A. selection of the study groups; B. comparability of the groups; C. ascertainment of the exposure. For cross-sectional studies, we evaluated four aspects of the quality of the studies: A. representativeness of study participants, B. proper methods to ascertain exposure, C. comparability of the analysis groups, and D. low non-response bias. Each study was scored with respect to meeting (1) or not meeting (0) each of these criteria, and a dash indicated that fulfillment of the criteria could not be determined.

All of the identified articles about suicide among Chinese subjects were examined in detail. Two reviewers independently abstracted the data from potentially relevant articles. Differences were resolved by consensus. For cross-sectional studies, sampling strategy, sample size, characteristics of participants, and information used for comparative analyses were abstracted. For case–control studies, information about the size of the case and control groups; characteristics of participants; and the number of cases/controls exposed and not exposed to the study factors were abstracted from each study. When the results of a study were published more than once, only the most complete data were included in our analysis.

We evaluated heterogeneity between the studies using the Q test [15] and the I-squared statistic (I² = 100% x (Q-df)/Q) [16]. For the Q test, a P value of less than 0.10 was considered statistically significant. If P > 0.10, there was a lack of heterogeneity between the studies. If P ≤ 0.10 (i.e., heterogeneity was significant), we calculated I². If I² ≤ 50%, the heterogeneity might be acceptable; otherwise, there was significant heterogeneity between studies and we conducted a subgroup analysis. For the subgroup analyses, heterogeneity within groups was also tested.

The classification of suicide risk factors is somewhat arbitrary in current reports. To better describe and sort the risk factors, we classified them in the text as follows: demographic factors, socio-family environment, lifestyle/behaviors, negative/stressful life events and psychiatric/psychological factors. To estimate the associations between these factors and suicide, we calculated pooled odds ratios (OR) for suicide ideation, suicide attempt, and completed suicide separately, using RevMan 5.0 software (RevMan software, Version 5.0, Cochrane Collaboration, Oxford, United Kingdom). If there was no significant heterogeneity (P > 0.10 or P ≤ 0.10, but I² < 50%), a fixed effects model was used to pool the data. If significant heterogeneity (P ≤ 0.10, I² >50%) was found between studies, the meta-analysis was conducted using a random effects model. Subgroup analyses were performed to explore possible reasons for the heterogeneity.

A detailed diagram of the full review process is presented in Figure 1. We identified 1101 potentially relevant articles. After reviewing their titles, abstracts, and full text, 43 studies]. (Table 1) with 192,362 subjects had suitable data for meta-analysis and were therefore included in our study (39 cross-sectional studies and 4 case–control studies).

The methodological characteristics of the included studies were then evaluated. In most of the cross-sectional studies, the participants had good representativeness. However, 7 cross-sectional studies [13, 19, 23, 26, 34, 38, 42] had no description of sampling methods or a small sample size. The exposure and control groups in the majority of studies were comparable, except for three studies [21, 34, 35] that did not compare them and seven studies [11, 17, 22, 36‐39] that did not report the non-response rate. Overall, only 3 studies [34, 38, 39] scored below 3 (Additional file 1: Table S1). All case–control studies scored six or above and were judged as being of high methodological quality (Additional file 1: Table S2).

There were 42, 17, and 5 studies with dichotomous data available for a meta-analysis of the factors associated with suicide ideation, suicide attempt and completed suicide, respectively. Heterogeneity tests were carried out on all of these studies and the results are presented in Table 2.

In terms of suicide ideation, significant heterogeneity was found for studies of the association of suicide ideation with gender, education, residence, mood disorders, physical health, suicide of relatives, and the economy. The pooled ORs and 95% CIs indicated that demographic factors (female gender), psychiatric/psychological factors (mood disorders) and negative/stressful life events (poor physical health, suicide of relative) were risk factors for suicide ideation among the Chinese population [pooled OR (95% CI): 1.43 (1.25, 1.64), 5.12 (3.78, 6.94), 4.88 (1.24, 18.68) and 3.02 (1.81, 5.04), respectively]. Lifestyle/behaviors (smoking, alcohol drinking, and substance use), family environment (single or remarried parent), poor academic achievement and study pressure were significant risk factors for suicide ideation among youth (adolescents and college students) [pooled OR (95% CI): 1.67 (1.45, 1.92), 1.96 (1.69, 2.27), 1.55 (1.23, 1.96), 1.97 (1.39, 2.80), 1.25 (1.08, 1.44) and 1.27 (1.17, 1.38), respectively] (Table 2 and Figure 2).

The factors associated with suicide attempt are shown in Table 2 and Figure 3. Heterogeneity tests identified significant heterogeneity among studies of the association of suicide attempt with gender, education, residence and mood disorders. Results of the meta-analysis showed that psychiatric/psychological factors (mood disorders), socio-family environment (poor relationship with family), lifestyle/behaviors (smoking, alcohol use) and negative life events (suicide of relatives) were significantly associated with attempted suicide [pooled OR (95% CI): 3.44 (1.93, 6.14), 3.64 (2.97, 4.47), 1.79 (1.36, 2.35), 2.50 (1.86, 3.36), 4.80 (3.12, 7.39), respectively].

Heterogeneity tests showed that studies of the association between mood disorders, seeking help for mood disorders, previous suicide attempt, suicide of relatives, negative life events, pesticides stored at home and completed suicide lacked significant heterogeneity. According to the pooled ORs and 95% CIs, demographic factors (female gender, residence in rural area, lower education), psychiatric/psychological factors (mood disorder), negative life events and recent change in life predicted completed suicide [pooled OR (95% CI): 2.35 (1.15, 4.78), 2.86 (1.20, 6.84), 2.33 (1.48, 3.66), 30.54 (20.46, 45.58), 10.03 (6.63, 15.19), and 3.79 (2.56, 5.61), respectively](Table 2 and Figure 4).

Given the available collected data, we conducted subgroup analyses of both suicide ideation and completed suicide. Many studies have investigated the factors associated with suicide ideation among youth (adolescents and college students). We carried out subgroup analyses of the association of suicide ideation with gender, mood disorders, physical health, and suicide of relatives by population, namely, youth vs. the rest of the population. The heterogeneity test showed that studies of the effects of physical illness [13, 28, 39] and suicide of relatives [9, 11] on youth suicide ideation lacked heterogeneity, but studies on the effect of gender and mood disorder had significant heterogeneity (P < 0.00001). For factors associated with suicide ideation in the rest of the population, studies lacked heterogeneity, except for those of physical illness (P < 0.00001) [28‐32]. The pooled ORs (and 95% CIs) showed that female gender, mood disorders, physical illness, and suicide of relatives were still risk factors for youth suicide ideation [pooled OR (95% CI): 1.36 (1.18, 1.56), 4.97 (3.37, 7.34), 2.92 (1.39, 6.12) and 1.97 (1.44, 2.68), respectively] (Figure 5).

For the factors associated with completed suicide, the subgroup analyses were performed by year of data collection. Regarding gender and completed suicide, when we excluded study data collected after 2000 [6], the studies using data collected before 2000 still had significant heterogeneity. The pooled results indicated that females had an even higher risk of completed suicide than males [pooled OR (95% CI): 3.03 (1.75, 5.23)]. As for education and completed suicide, when we excluded study data collected after 2000 [7], the studies [7, 8] using data collected before 2000 lacked heterogeneity. The pooled OR still indicated that lower education was a risk factor for completed suicide in China [pooled OR (95% CI): 1.95 (1.57, 2.44)] (Figure 6).