Introduction

In China, around 130 000 people died from suicide in 2016, accounting for approximately one-sixth of suicides worldwide, at a rate similar to the global average [1]. Previous studies in China have reported major differences in the distribution of suicides from that in high-income countries [2,3], including higher rates in women and rural areas [4–6]. More recent investigations, however, suggest an epidemiological transition to a distribution seen in many high-income countries including higher risk in men [7], but these reports have been limited by small sample size and selected samples.

Although the effect of some risk factors has remained constant over time [2,3,7], this research is uncertain on the magnitude of effects and the effect of a range of modifiable risk factors, such as psychiatric disorders. Furthermore, most studies on suicide risk factors in China have been limited by cross-sectional and case-control designs, which likely overestimate the association as risk factors and suicide outcomes are typically measured at the same time. In addition, evidence on risk factors is mostly based on psychological autopsy studies, relying on family members and friends for background information that might be prone to measurement error and recall bias [8,9]. Prospective research has been very limited and mostly from selected samples based on patient characteristics, specific age groups, and particular regions [10,11]. Studies have also mainly focused on sociodemographic and lifestyle correlates, rather than longitudinal associations between psychiatric factors and suicide outcomes [12]. Despite these limitations in the research to date, sociodemographic factors such as rural residence and poverty and stressful life events have been linked to an increased risk of suicide.

To address these uncertainties, we have examined a broad range of risk factors for suicide and possible suicide from a large prospective study, the China Kadoorie Biobank (CKB), of 0.5 million adults across China.

Methods

Results

Among the 512,715 participants at baseline, the mean age at baseline was 52 (SD = 10.7), 59% were female, 55.9% lived in rural areas, and 9.4% were single. Around 15% reported having physical illnesses and prevalence of mental disorders ranged from 0.2% to 16.8% (Tables 1 and 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic and lifestyle characteristics and stressful life events of participants by suicide status during follow-up.

https://doi.org/10.1371/journal.pmed.1003545.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Physical and mental health status of participants by suicide status during follow-up.

https://doi.org/10.1371/journal.pmed.1003545.t002

During the average follow-up of 9.9 years, there were 520 deaths by suicide (at a rate of 102 suicides per 100,000 persons or 10.2 per 100,000 annually), and 51.3% were in men. Around 70% of suicide cases in individuals with lower education and 78% from rural residential areas. The suicide mortality rates increased with age and were higher in rural than urban areas (145 versus 46 per 100,000) and in men than women (127 versus 84 per 100,000) (Fig 1). The rate of possible suicide was 108 per 100,000 (n = 554). Similar demographic patterns were found for possible suicide by areas of residence (148 versus 58 per 100,000, in rural and urban, respectively) and between males (163 per 100,000) and females (70 per 100,000) (S1 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Mortality rates of suicide by gender, residence, and age group.

Fig 1a. Suicide per 100,000 population by gender; Fig 1b. Suicide per 100,000 population by residence; Fig 1c. Suicide per 100,000 population by gender and residence.

https://doi.org/10.1371/journal.pmed.1003545.g001

All risk factors in the domain of physical health status, all but 1 risk factor in each of 2 domains of sociodemographic factors and mental health status, and 1 risk factor in the other 2 domains, lifestyle factors and stressful life events, were associated with an increased risk of suicide (Figs 2 and 3). After adjustment for sociodemographic factors, HRs for suicide ranged from 1.3 (95% CI: 1.1 to 1.6, p = 0.004) for low-income status to 2.6 (2.1 to 3.3, p < 0.001) for rural residence. In addition, subgroup analyses indicated that there were no clear differences in the link between income and suicide risk by residential areas, which was 1.5 (0.9 to 2.5, p = 0.091) in urban areas versus 1.3 (1.0 to 1.5, p = 0.016) in rural areas. Higher age band was associated with a higher risk of suicide (1.3 [1.2 to 1.5], p < 0.001). Age was associated with suicide risk when it was entered to the regression model as a continuous variable (B = 0.03, SE = 0.01, p < 0.001, HR = 1.03 [1.02, 1.04]), indicating that 1 year increase in age was associated with a 3% higher risk of suicide. Sensitivity test of nonlinearity of the age–suicide link did not indicate a better model than the linear model (df = 1, χ² = 0.18, p = 0.67). Adjusted hazard ratios (aHR) for suicide ranged from 0.8 (0.6 to 1.0, p = 0.115) for ever having been regular smoker to 1.4 (1.2 to 1.7, p < 0.001) for physical inactivity among lifestyle factors. In addition, when BMI was entered into the model as a continuous variable, 1 unit of increase in BMI score was associated with a 7% lower risk of suicide (B = −0.07, SE = 0.01, p < 0.001, HR = 0.93 [0.91, 0.96]). A lower BMI was associated with a higher risk of suicide (1.6 [1.3 to 1.8, p < 0.001]) when BMI was categorized according to WHO cut-off scores. Hazards were increased for family-related stressful life events (aHR = 1.8 [1.2 to 2.9, p = 0.004]). However, there were no clear links between suicide and other type of stressful life events. In relation to physical health, aHR varied from 1.5 (1.3 to 1.9, p < 0.001) for current major illnesses to 2.1 (1.7 to 2.6, p < 0.001) for self-reported poor physical health. For individual mental disorders, aHR ranged from 1.4 (1.2 to 1.7, p = 0.001) for sleep disorders to 11.0 (7.1 to 17.0, p < 0.001) for schizophrenia-spectrum disorders.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Hazard ratios for suicide by sociodemographic and lifestyle factors and stressful life events.

HRs adjusted for sex, age, rural residence, education, income, and single status. HR, hazard ratio; MET, Metabolic Equivalent Task.

https://doi.org/10.1371/journal.pmed.1003545.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Hazard ratios for suicide by physical and mental health status.

HRs adjusted for sex, age, rural residence, education, income, and single status. HR, hazard ratio; BMI, body mass index.

https://doi.org/10.1371/journal.pmed.1003545.g003

We additionally controlled for risk factors in other domains in addition to sociodemographic factors. After additionally adjusting for lifestyle factors including weekly drinking, regular smoking, and physical inactivity, aHR was 1.5 (1.2 to 1.9, p < 0.001) for major physical illnesses and 2.0 (1.6 to 2.5, p < 0.001) for poor self-reported physical health. aHR varied from 1.2 (1.0 to 1.4, p = 0.020) for sleep disorders to 10.6 (6.8 to 16.5, p < 0.001) for schizophrenia-spectrum disorders. There were no material changes between HRs with and without additional adjustments. In addition, after adjusting for current psychiatric disorders, the association between sleep disorders and increased suicide risk remained (aHR = 1.2 [1.0 to 1.4, p = 0.021]). Finally, we compared the associations between risk factors and suicide by gender and residence. There were no differences by gender or residence in the magnitude of the link between risk factors and later suicides (S1 Table).

In a further sensitivity analysis, we investigated risk factors for possible suicide (n = 554). Overall, the effect was similar for sociodemographic, physical health, and lifestyle domains but not for all mental health variables (S2 and S3 Figs). Several specific psychiatric disorders were not associated with possible suicide, including depression (aHR = 0.5 [0.1, 2.0], p = 0.324), generalized anxiety disorders (aHR = 0.7 [0.1, 4.7, p = 0.681]), and sleep disorders (aHR = 1.1 [0.9, 1.3, p = 0.547]) (S3 Fig).

Discussion

In this prospective study of 512,715 individuals from 10 geographical regions in China, we examined suicide rates and risk factors. In total, 520 individuals died by suicide during an average follow-up period of 9.9 years (10.1 per 100,000 persons annually), comparable to the global average rate (10.5 per 100,000 persons annually) [28]. We found that the risk of suicide was higher in men than women, and those from rural areas had a 3 times higher suicide risk than urban dwellers. Among examined risk factors, mental health factors were associated with the highest suicide risks.

The new work adds to the evidence base in the following ways. First, we have shown that diagnosed mental disorders were associated with a higher risk of suicide. Overall, though, the relative risks of mental disorders were lower than that reported in high-income countries, consistent with the finding of lower absolute prevalence of mental disorders in suicide [29]. For example, we estimated an 11-fold increased risk of suicide in those with a schizophrenia-spectrum disorder when compared to those without this disorder, while in a population-based cohort study in Sweden, the reported risk of suicide in the same diagnostic group was 20 times higher [30], similar to a meta-review of risk factors for suicide [31]. Previous work has reported a HR of 9 for individuals with depressive disorders in a meta-analysis of 35 studies [32], higher than the increased hazard of 2.6 in the current study.

The underdiagnosis of mental disorders due to stigma in China might explain this discrepancy [33,34]. For instance, the rate of depression was 0.7% in this cohort, lower than a rate of 1.7% in a WHO survey using CIDI in Beijing and Shanghai [35]. In this cohort study, individuals with and without mental disorders were followed up for 10 years for their suicide outcomes. If mental disorders were underdiagnosed, some of the individuals who died by suicide would be categorized as without mental disorders when they did have mental disorders. In this scenario, hazards of suicide in individuals without mental disorders would be overestimated, and the ratio of the hazard risk of suicide among individuals with a mental disorder to the risk of suicide among individuals without a mental disorder would be underestimated. This suggests that our risk estimates are likely conservative, and future work is required to replicate these findings. Another explanation for this discrepancy could lie in the study design. Most studies examining the link between mental disorders and suicide risk have been case-control psychological autopsy studies in which typically a psychiatric diagnosis directly prior to suicide death was examined [32]. Longer follow-ups have been associated with lower hazards for mental disorders. Our findings are more consistent with a cohort study which had a 10-year follow-up and used DSM-criteria for depressive disorders (HR = 1.8 [1.7 to 2.0]) [36]. Nevertheless, the lower relative risk for mental disorders in China compared to other countries might be a real finding. At the same time, mental disorders still had the highest risk estimates among the examined risk factors. Thus, improving access to mental health services and providing evidence-based assessment and treatment is one clear implication of these findings.

A second novel finding is the contribution of sleep disorder. We found that sleep disorders were modestly associated with an elevated risk of suicide. This is consistent with the findings of prospective investigations in other Asian countries [37,38], and one Norwegian study [39]. Sleep problems are highly prevalent, affecting 10% to 40% of the general population [40,41] and consistent with this the current study reported a rate of 26%. The increased risk of suicide in this group suggests that clinical assessment of suicide risk among individuals with persistent sleep problems might be beneficial for suicide prevention in China.

Third, we found that suicide rates were higher in rural areas, which is consistent with previous research in China [42,43]. The higher rate might be due to lower availability of effective interventions after a suicide attempt and more ready access to lethal means of suicide such as hazardous pesticides [44]. Thus, even if one assumes the same rate of suicide attempts, suicide death might be more likely to occur in the rural areas as medical assistance is less accessible in emergencies, and the overall quality of medical services is poorer [45]. In support, a psychological autopsy study of 895 suicide victims in China reported that 61% of suicide deaths were associated with unsuccessful medical resuscitation [46]. Other explanations for the higher suicide rate in the rural areas could be selection effects as individuals with a higher education, less underlying psychiatric morbidity, and more economic resources move to cities. The higher risk reported in rural areas is also reported in some high-income countries [47–49]. These findings suggest targeted approaches in rural populations could be considered in national suicide prevention efforts. Further research into risk factors of rural suicide is required [50].

Although suicide risk was higher in rural than urban areas, the groups contributing to these differences have changed over the last 20 to 30 years. Previously, young women in rural areas were at high risk [6]. However, economic and social changes have provided women with increased opportunities in education and work and provided more independence including financially [51]. The change to higher risk in men has been shown in previous studies [42,43], and the earlier excess in women in the 1990s [12] has not been found since. In the current study, rural men between 70 and 79 years old had the highest suicide risks, with a rate of 413 per 100,000 population during the follow-up of 9.9 years. Suicide risk in elderly might be linked to loneliness and lack of social support network [52,53], and we found that 20% of suicide decedents were living alone and that 46% were single in those aged 70 to 79 years, compared to less than 5% living alone and 15% single in those aged 30 to 59 years old. Limited data has suggested some differences in suicide method by residence. For instance, in this study, pesticide was used more often in rural areas as a suicide method (9/415, 2%) than in urban areas (1/145, 1%). Future research is needed to examine this.

A final finding of the current study is that individuals with a major physical illness had a 50% increased risk of suicide. This estimate is lower than case-control studies in China which reported a 3-fold higher risk among individuals with physical illness than matched controls [54]. Chronic physical illnesses were more prevalent in the older people. In this study, prevalence was 3.9% in those aged 30 to 39 years but 32.0% in the 70 to 79 age group. However, the effects of physical illness on suicide risk were similar between age groups and residential areas. The effects of specific physical illnesses needs further research [55], a finding consistent with research in high-income settings [56].

Family-related events, such as separation or divorce, family conflict, and death of a spouse or illness were linked to a higher risk of suicide. However, financial stressors, including loss of income, bankruptcy, and loss of job or retirement, were not clearly associated. This suggests that family-related events had stronger associations with suicide than financial concerns. This finding is similar to research on suicidal behaviors in Chinese adolescents, which suggests that problems in family relationships are important to consider in risk assessment and management [16]. As a consequence, the role of family-based interventions on reducing suicide thoughts and behaviors [57] needs examination in China. Furthermore, data from the same cohort also reported that, compared to finance-related stressful life events, family-related stressful life events had a higher correlation with other negative outcomes such as depression [58]. Nevertheless, in this study, financial stress was defined as experiencing loss of job or income or bankruptcy over the past 2 years. Longer-term erosion of economic opportunity, including the perceived prospects of poor occupation status [59], unemployment [60], and low salary [61], have been linked to increased suicide risk. These findings are consistent with the elevated risks observed for rural residents and men who were unsatisfied with life.

Many risk factors associated with suicide were also linked to possible suicide, although in general with lower effect sizes. However, there were large differences in mental health factors, where common disorders including depression and generalized anxiety disorders were not associated with possible suicide. The effects of any psychiatric disorder and schizophrenia-spectrum disorders on possible suicide were also considerably smaller. These differences in the effect of psychiatric risk factors suggests that possible suicide and suicide (that includes undetermined deaths) as defined in this study might belong to different phenotypes. This underscores that our decision to classify possible suicide separately was justified, and further work will need to clarify whether these deaths are more accurately classified as accidents. It is also possible that evidence of mental health is one of the factors which contribute to a death being determined as suicide, especially as there is limited information from coroners in China in determining suicide death [62]. If there is a background of mental illness, then it is possible that in uncertain cases, these are more likely to be classified as suicide rather than possible suicide in China. This may explain the lack of an association between some mental disorders and possible suicide, and the overall weaker associations between them than the link between mental health and suicide death.

Our study has several strengths. First, we conducted the largest cohort study to date of suicide using data from more than half a million individuals in China with a broad geographic and social diversity. Therefore, the findings are generalizable to the general Chinese population. Second, by using a prospective design, we compared individuals with and without the examined risk factors on later suicide outcomes during follow up. This helps to mitigate recall bias which is often present in retrospective studies [9] and provide a more precise estimation of the associations between risk factors and suicide than cross-sectional and psychological autopsy studies. Third, we used structured clinical assessments and self-report data in the baseline measurement, unlike many earlier psychological autopsy studies in which information on risk factors were from other informants who might not have reliable or sufficiently detailed knowledge of the individuals who died by suicide. In addition, the proxy informants and the interviewers in previous autopsy studies were also aware of the reported cause of death being suicide, which could introduce report bias. Thus, our study design allowed for a more reliable measurement of risk factors.

Some limitations have to be noted. First, we cannot infer a causal relationship between identified risk factors and suicide. However, the studied variables were independently linked to the suicide risk, as we adjusted for a wide range of potential confounders. Second, we only examined the effects of a limited number of mental disorders. Other conditions, such as bipolar disorder, personality disorders, eating disorders, and substance use disorders, have been associated with increased suicide risk [31,63]. Future research on the link between these mental disorders and suicide is needed. Third, due to lack of data, we were not able to examine attempted suicide, as it appeared to be a strong risk factor for suicide death [64]; future study as a risk factor as well as an outcome is recommended. Fourth, the studied cohort were aged from 30 to 79 years at baseline measurement. Thus, findings may not be applicable to younger populations, such as adolescents, where suicide rates are relatively low. In addition, 33% of invited individuals visited the local assessment center to participate in this study. As sociodemographic data of individuals who did not participate in the study were not collected, it is not possible for us to know how well the cohort could be generalized to the general population. However, it is possible to compare the included sample with general population based on information from other official sources. This suggests differences in sex distribution but similar in terms of urban versus rural residence. Compared to the general population during the baseline measurement years (2004 to 2008) in China [65,66], the Kadoorie cohort had a slightly higher percentage of female (59.0% versus 48.6%) but a similar residence distribution (55.9% versus 54.0% rural). Furthermore, income at separate time periods might play a different role on suicide risk, and future studies could examine this. Finally, although CIDI is a validated diagnostic instrument, the face-to-face interview format and stigma toward mental illnesses in China might be associated with underreporting of mental disorders in this population. If this is the case, then our reported links with mental disorders are conservative estimates.

In summary, we followed up over half a million individuals for suicide for an average of 10 years. We prospectively examined the relationships between a range of risk factors, including sociodemographic, lifestyle, physical and mental health, and stressful life events, and suicide outcomes. Almost all of the studied risk factors were associated with increased hazards of suicide death. Men and individuals from rural areas had higher suicide risks, and we identified rural elderly men as a high-risk group. Mental disorders as a category were associated with the highest suicide risk. This pattern and magnitude of risk factors should be considered as part of suicide prevention strategies in China.

Supporting information