Bulletin of the World Health Organization

Pesticide exposure and suicidal ideation in rural communities in Zhejiang province, China

Jianmin Zhang a, Robert Stewart b, Michael Phillips c, Qichang Shi a & Martin Prince b

a. Office of Mental Health, Zhejiang Provincial Tongde Hospital, Hangzhou, China.
b. Institute of Psychiatry, King’s College London (Institute of Psychiatry), De Crespigny Park, London, SE5 8AF, England.
c. WHO Collaborating Center for Research and Training in Suicide Prevention, Beijing Hui Long Guan Hospital, Beijing, China.

Correspondence to Robert Stewart (e-mail: r.stewart@iop.kcl.ac.uk).

(Submitted: 18 April 2008 – Revised version received: 10 December 2008 – Accepted: 23 January 2009 – Published online: 28 July 2009.)

Bulletin of the World Health Organization 2009;87:745-753. doi: 10.2471/BLT.08.054122


The use and availability of pesticides are significant concerns in the field of mental health, not only because these chemicals are used in suicide attempts1 but also because their possession may be directly associated with mental disorder. Suicide rates are reported to be higher in areas where organophosphates are used2 and exposure is a possible risk factor for Parkinson disease3,4 and Alzheimer disease,3 for depressive and anxiety disorders5,6 and for mortality ascribed to mental disorder.7 However, research in this area remains controversial.8,9 Pesticides have been widely used in agriculture since the 1950s10 but, despite precautionary measures, intentional and occupational poisoning remain major concerns. For example, pesticide ingestion was implicated in 62% of suicides in China between 1996 and 2000, which corresponds to around 175 000 cases per year.1,11

Suicide in China accounts for 44% of all suicides worldwide.12 Moreover, suicide is the fifth leading cause of death in China13 overall and the leading cause in 15–34-year-olds.14 In rural areas, suicide rates are 2–5 times those in urban areas13,15 and, in contrast to Western populations, the rate is higher in women than men.13 Although mental disorder, especially depression, is associated with suicide,16 the link between suicide and mental disorder may be relatively weak in China.13 The reasons for the high rate and unique pattern of suicide in China have yet to be established.

Organophosphate pesticides are used widely in China and the potential for exposure is high. Despite this, the mental health risks of prolonged exposure remain unclear. We investigated the association between the storage of pesticides at home and recent suicidal ideation in rural China by analysing data from a province-wide survey of mental disorders that was carried out in 2001 as part of a WHO and Chinese Ministry of Health project.


The study was performed in Zhejiang province, a densely populated coastal province of China that in 2000 had a total population of 45 million, 70% of which lived in rural areas. It is one of the most developed provinces in China and in 2000 had an average per capita gross domestic product (GDP) of 13 461 renminbi (US$ 1660), the fourth highest among China’s 31 provinces and independent municipalities.

Sample size

The survey was designed to ascertain a prevalence of mental disorder of 15% with 1% precision. In addition, the sample size was increased to 15 000 to enable comparisons to be made according to gender and urban or rural residence. The ratio of urban to rural residence was estimated to be 1:2.17

Sampling procedure

The survey employed a multistage sampling process that has previously been described in detail by Shi et al.18 The sampling frame was the province-wide computerized household registry, which was updated during the 2000 national census. Seven indicators (i.e. population density, birth rate, the proportion of non-agricultural labourers, per capita GDP, illiteracy rate, crude death rate and proportion aged > 65 years) for 11 cities and 63 rural counties were subjected to a principal components factor analysis. Two factors accounted for 65.2% of the variance. The factor scores were then used to construct five strata: (i) 5 cities (comprising 45.5% of the total urban population); (ii) 6 cities (comprising 54.5% of the total urban population); (iii) 19 counties (comprising 30.2% of the total rural population); (iv) 13 counties (comprising 20.6% of the total rural population); and (v) 31 counties (comprising 49.2% of the total rural population). This analysis considered only the three rural strata, which broadly corresponded to mountain, plains and coastal regions.

Groups of three, two and five counties were selected from the three rural strata (mountain, plains and coastal regions, respectively) by sampling in proportion to the population aged 15 years or over. Thereafter, five districts within each county and, subsequently, two villages within each district were sampled in proportion to their populations. Systematic random sampling was used in each stratum to select the counties, districts, and villages. A total of 100 villages were selected from the 50 rural districts sampled from the 10 counties. The final stage of sampling, which was again carried out in proportion to population size, involved the identification of 1000 residents aged 15 years or more in each of the 10 counties. At each village, simple random selection was used to identify target subjects from the computerized registry of community residents. The demographic characteristics of the selected population sample were similar to those of the province as a whole.17 At each village, a 20% supplemental sample was also selected using the same method.

During the survey at least three attempts were made to contact selected individuals but 3372 (22.5%) could not be contacted, primarily because of non-residence. A similar proportion of men and women (i.e. 22.6% and 22.4%, respectively) could not be contacted, and these individuals were, on average, 5 years younger than those who could be contacted: mean age 39.3 years and 44.3 years, respectively. People of a similar age and gender to replace the 3372 individuals who could not be located were randomly selected from three sources: the supplemental sample at that location (n = 567), neighbouring households of the target subjects (n = 2133) and, for two sites at which most residents had relocated, neighbouring communities (n = 672).


Interviews were carried out in study participants’ homes by 34 trained psychiatric nurses between September and December 2001. All participants provided verbal informed consent after the study had been fully explained. Local ethics review boards were not in place at the time of the survey. According to standard practice at that time, the survey was initially considered and approved by WHO and further consideration and approval were then provided by the Department of Disease Control and the Department of International Cooperation at the Chinese Ministry of Health, as well as by the Mental Health Leading Group of Zhejiang province. The principal independent variable considered in this analysis was the storage of pesticides at home. In households where informants reported storing pesticides (excluding rodenticides), the type of pesticide and method of storage were recorded.

All participants were asked the following questions: “Have you ever considered suicide or deliberate self-harm in your lifetime?”, “Have you ever planned suicide or deliberate self-harm in your lifetime?”, and “Have you ever attempted suicide or deliberate self-harm in your lifetime?”. If any of the above were answered positively, further questions were asked about the respondent’s age at the first and most recent episode or occurrence. The primary dependent variable was whether or not the respondent reported suicidal ideation in the 2 years before the interview. No participant reported attempting or planning suicide without previous suicidal ideation.

The Chinese version of the 12-item General Health Questionnaire (GHQ) was administered to all participants and a cut-off score of 3/12 was used to define GHQ caseness,19 which was treated as a covariate. This instrument was translated into Chinese and validated in the 1980s20,21 and has been widely used. Recent specific ethnographic evaluation has supported its applicability in a rural Chinese context.22 Other covariates were age, sex, years of formal education, marital status, annual per capita household income (in three strata), family history of suicidal behaviour, and subjective global physical health status in the prior month, which was categorized on a 5-point scale and then dichotomized to “poor health” versus “others”. The rural stratum of the participant was also considered as a covariate and as an effect modifier. For unadjusted analyses, the three rural strata were ordered according to the frequency of suicidal ideation, although this ordering differed from that of the socioeconomic indices (i.e. the order of areas with ascending socioeconomic indices were mountain followed by coastal followed by plains regions).

Statistical analysis

Statistical weighting was applied so that the participants selected accurately represented the total population of Zhejiang province. Within each cluster (i.e. each village), the initial weight was the number of individuals in the population represented by the cluster divided by the number of completed interviews in that cluster. Thus, the weight incorporated an adjustment for individuals included in a sample but not interviewed. In addition, weights were further adjusted to reduce the effect of extreme weights: in clusters in which the weight was greater than two standard deviations above the average weight, the value was reduced to equal two standard deviations. Finally, post-stratification weights were obtained to compensate for any mismatch between the age and gender distributions of each stratum and the age and gender distributions of individuals aged 15 years or more in Zhejiang province. Post-stratification weights that took design effects and clustering into account were used in the analysis. Standard errors were adjusted for unequal sampling fractions within each stratum and for possible homogeneity within each cluster. The analysis employed the complex survey procedures in SPSS 15.0 statistical software (SPSS Inc., Chicago, Illinois, United States of America). Logistic regression analysis was used to examine the relationships between suicidal ideation in the prior 2 years, reported storage of pesticides in the home, and other covariates.


Of the 10 035 individuals in the selected sample from the 10 rural counties, 9811 (97.8%) completed the interview. The 224 who did not complete the interview included 202 who refused to participate, 17 who were unable to participate because of severe illness or intervening mortality and 5 who completed only part of the interview. Among those who did complete it, 7627 (77.7%) were from the original sample and 2184 (22.3%) were replacements from the sources described above.

Of the 9811 respondents, 5116 (52.1%) stored pesticides in the home. Among the 5088 who reported the method of storage, 249 (4.9%) locked up the pesticides, 1588 (31.2%) stored them in a high location not easily accessible by children, and 3251 (63.9%) stored them where they were easily accessible by all household residents. Of the pesticides stored at home, 86.9% comprised or included organophosphates. The most commonly stored pesticide was methamidophos, which was present in 63.4% of households that stored pesticides.

Self-report lifetime prevalence estimates of suicidal ideation, planning suicide and attempting suicide in the 9811 respondents were 4.8%, 0.9% and 0.4%, respectively. Prevalence estimates for the most recent 2 years were 2.0%, 0.4% and 0.2%, respectively. To avoid potential misclassification of cases, the 270 individuals who only reported suicidal ideation more than 2 years previously were excluded. This left 9541 participants. Moreover, the analysis was further restricted to 9159 participants for whom complete data on all covariates were available.

Associations between pesticide storage at home and demographic, socioeconomic and clinical covariates, adjusted for design effects and clustering only, are summarized in Table 1. The presence of pesticides at home was significantly associated with male gender, older age, fewer years of education, lower income, current or previous marriage, a family history of suicidal behaviour and GHQ caseness. Significant differences were also found between the three regions.

Associations between suicidal ideation in the 2 years prior to the study and pesticide storage and demographic, socioeconomic and clinical covariates, adjusted for design effects and clustering only, are summarized in Table 2. The odds ratio (OR) for the association between pesticides stored at home and suicidal ideation over the prior 2 years was 2.40 (95% confidence interval, CI: 1.68–3.43). Suicidal ideation was also significantly associated with pesticide storage method, female gender, older age, fewer years of education, lower income, poor physical health, a family history of suicidal behaviour and GHQ caseness, and was significantly different between the three regions.

In the logistic regression models, suicidal ideation in the prior 2 years remained significantly associated with pesticide storage at home. As shown in Table 3, the unadjusted OR for the association between pesticides stored at home and suicidal ideation over the prior 2 years was 2.12 (95% CI: 1.54–2.93). Adjustment for only design effects and clustering had a minor effect on the estimated OR, which became 2.40 (95% CI: 1.68–3.43). The only substantial change in the OR of interest occurred following adjustment for the rural stratum. After full adjustment (i.e. model 11 in Table 3), the ORs for suicidal ideation for different storage methods compared to no storage at home were as follows: for locked storage, 1.13 (95% CI: 0.32–4.00); for unlocked high-up storage, 1.40 (95% CI: 0.86–2.25); and for easy access, 1.76 (95% CI: 1.18–2.59).

Further analyses were carried out separately for the three rural strata (Table 4). After full adjustment, a significant association between pesticide storage at home and suicidal ideation was found in the plains region but not in the mountain or coastal region. Finally, analyses were repeated using all (i.e. not just recent) suicidal ideation as an outcome and home storage of organophosphates as the only exposure. Overall, there was no meaningful change in the associations of interest (data not shown).


We found a positive, independent association between the presence of pesticides in the home and suicidal ideation in the prior 2 years in a large representative sample of the rural population of Zhejiang province, China. We also found that reported suicidal ideation became more common with increasing ease of access to pesticides and, furthermore, that the highest prevalence of reported suicidal ideation was in the two rural strata (i.e. the plains and mountain regions) with the highest prevalence of pesticide storage at home.

The strong points of this study are that it involved a large, representative sample and that the response rate was high, both of which indicate that the results can be readily applied to the source population.

An important limitation is that individual pesticide exposure was not directly measured. Chronic low level exposure to pesticides is difficult to ascertain directly, particularly in a large-scale community study where biological assays are not feasible. Occupational exposure is also difficult to ascertain because of the near universal use of pesticides in agriculture. We therefore had to use self-report data on home storage as a proxy measure of exposure. Pesticide storage is not a sensitive issue in these communities so it is unlikely that respondents intentionally underreported home storage or that their reports were influenced by the presence or absence of previous suicidality. Moreover, measurement inaccuracy would have the effect of diluting the association of interest. Similarly, although prior suicidal thoughts may be underreported, underreporting is unlikely to vary with respect to pesticide storage. If the associations described here do indeed relate to an individual’s level of pesticide exposure rather than stem from reverse causality or unmeasured confounding, the true association between suicidality and underlying exposure is likely to be substantially stronger than that estimated here using the proxy measure.

With respect to confounding, the logistic regression analysis showed that none of the covariates accounted substantially for the association identified between exposure to pesticides and suicidal ideation. However, it should be borne in mind that physical and mental health were assessed relatively briefly and additional research is required to clarify potential mediating and confounding effects further. Although the GHQ has been translated into Chinese and validated in Chinese populations, it may nonetheless underestimate mental distress since it makes use of a 12-point scale that cannot encompass all forms of psychopathology, such as the expression of emotional distress through somatic complaints, which has been observed in Chinese populations.23

The cross-sectional nature of the study does not enable the direction of cause and effect to be definitively inferred. It is possible that the possession of pesticides was a consequence of suicidal ideation, although we think that this is unlikely for the following reasons: (i) the hypothesized outcome was rare compared with pesticide exposure; (ii) the association between pesticide storage at home and suicidal ideation was independent of GHQ status; (iii) the association of interest was little affected by whether suicidal ideation throughout life or in the previous 2 years was defined as the outcome; and (iv) the association increased with ease of pesticide access.

Pesticide possession was frequent in this population and many of the commonly stored pesticides contained methamidophos, an organophosphate insecticide that persists in water and soil following contamination and that has been classified by WHO as a class‑I pesticide (i.e. extremely or highly toxic). In addition, the liquid preparation used in agriculture is volatile, making it particularly hazardous.24 The assumption underlying our analysis is that participants who keep pesticides at home will have experienced higher chronic low-level exposure, principally through transdermal or respiratory absorption, possibly during storage but more likely due to increased use or spillage. While biologically plausible, the association between home storage and the level of individual exposure has not, to our knowledge, been directly demonstrated, although home storage has been cited as an undesirable potential source of exposure.25 Most studies have focused on occupational use as a proxy for chronic low-level exposure and have compared high- and low-exposure workers. This is not feasible in rural China because of the homogeneity of occupation. One study found higher levels of urine metabolites in children from communities where pesticides were used and stored,26 suggesting that exposure may take a wide variety of routes. However, home storage cannot definitively be implicated and it is possible that exposure could occur through contamination of clothing or other fabrics or deficient pesticide preparation practices.

Several studies have indicated that pesticide use is associated with neurological disorders and worse mental health36 as well as with suicidal thoughts and behaviour.2 There are a variety of potential causal mechanisms. Imbalance in cholinergic pathways are implicated in depressive disorder27 and organophosphates, including methamidophos, inhibit cholinesterase activity.28 Paraoxonase activity may be a modifying factor since paraoxonase is known to hydrolyse a wide variety of organophosphates and plays a role in their detoxification.29,30

Although chronic exposure to organophosphates may increase suicide through increasing depressive symptoms,2 a large number of people in China who develop suicidal thoughts or who complete suicide do not have an apparent mental illness.13,31 There may, therefore, be different aetiological mechanisms. The ready availability of highly lethal means could, for example, convert impulsive, low-intent suicide “attempts” among persons without a mental illness into completed suicides.32 Suicidality may be a separate symptom domain independent of depression,33 and it has been suggested that some risk factors for suicidality do not act through increasing depression.3437 Frontal lobe syndromes and increased impulsivity would be one example of an alternative causal pathway. The observation that adjustment for GHQ score did not affect the association of interest may reflect this; however, as highlighted earlier, some aspects of depressive symptomatology may not be adequately identified by this instrument in the Chinese population.

The findings of this study might partially account for the much higher incidence of suicide in rural than urban areas of China. The observed association in our study between pesticide storage in different regions and suicidal ideation in those regions is interesting. Less easy to explain are the apparent differences in the association of interest between regions. They require further replication and evaluation. The direction of cause and effect also requires further research that makes use of more accurate markers of chronic exposure. Regardless of the direction of causation, the association between pesticide exposure and suicidality demonstrates that potentially vulnerable groups with previous suicidal thoughts have increased access to highly toxic chemicals. This is of particular concern in China and other agriculture-based Asian and Pacific nations where suicidal behaviour is often related to impulsivity and the ready availability of a lethal method.12,13

The influence of pesticide exposure on mental disorder, suicide and overall health is a very important public health issue. To address this, in 2006 three WHO departments announced a global public health initiative and released a report on community interventions for safer access to pesticides.38 It is vital that these messages are disseminated as widely as possible to the appropriate policy-makers. ■


We thank the staff of the Zhejiang Office of Mental Health, participating nurses and doctors from psychiatric hospitals in the province for their assistance in data collection and entry, and Mark Davies for his assistance in the statistical analysis.

Funding: The survey was part of a WHO and Chinese Ministry of Health Mental Health Project and was supported by a WHO grant (GL/GLO/MNH/343/XE/00.J.999.00). Dr Zhang prepared this paper at King’s College London while supported by a Chevening Scholarship from the British Council and by a grant from the Institute of Social Psychiatry, London, United Kingdom. Robert Stewart is funded by a National Institute for Health Research Specialist Biomedical Research Centre for Mental Health award to the South London and Maudsley NHS Foundation Trust, and the Institute of Psychiatry, King’s College London, United Kingdom. Michael Phillips is also affiliated with the Department of Psychiatry and Department of Epidemiology, Columbia University, United States of America.

Competing interests: None declared.