Introduction

Suicide is a serious global public health problem. It is among the top 20 leading causes of death worldwide, with approximately 800,000 deaths per year globally. Suicide causes more deaths than malaria, breast cancer, war, or homicide [1]. The age-standardised suicide rate worldwide is 10.5 per 100,000 inhabitants [1], with a wide variety around the world, and is 6.1 in Brazil [2]. The World Health Organization (WHO) recommends that suicide prevention should become a higher priority on the global public health agenda [3]. However, the best interventions to prevent suicide at the population level remain unclear.

There is growing evidence that economic recessions lead to a rise in suicide rates in various countries [4–6], including Brazil [7].

The Coronavirus Disease 2019 (COVID-19) pandemic is expected to lead to a severe global recession, increasing poverty, and resulting in massive unemployment worldwide [8], with possible global increases in suicide rates in the coming years [9,10], which is already occurring in Japan [11]. Suicide prevention interventions may help to mitigate this increase, if proven effective.

Social and economic factors, such as poverty, economic uncertainty, unemployment, and income inequality, have been consistently associated with increased suicide rates [12–17], including the young population [18]. Therefore, socioeconomic interventions, such as cash transfer programmes, could potentially decrease suicide rates by improving beneficiaries’ welfare, and reducing known risk factors for suicide, such as financial problems, family instability, and alcohol consumption [3,16,19,20]. However, evidence on the impact of socioeconomic interventions to reduce suicide rates is limited, due to small sample sizes, difficulties in conducting randomised trials, due to difficulties recruiting people, high costs, ethical issues, and data availability [19,21–23].

In 2004, Brazil implemented one of the largest poverty alleviation programmes in the world—the “Bolsa Família” programme (BFP), a conditional cash transfer (CCT). By 2015, approximately 46 million people had benefited from the programme [24]. It has 3 main aims: an income supplement guarantee for the immediate relief of poverty; access to public services (improving families’ education, health, and civic participation); and productive inclusion, which involves job skills training, to increase individual capacity to seek jobs and job opportunities [25].

Brazil’s continental size, and the availability of data, create a unique opportunity to investigate the association of a CCT programme with suicide, and other relevant health outcomes [26,27]. Based on current knowledge, we hypothesise that BFP, and similar programmes, could protect against suicide. Therefore, this study aims to investigate the association of a large CCT programme with the reduced occurrence of suicide.

Methods

This study had a quasi-experimental design, comparing a group who participated in a cash transfer intervention with those who did not. We utilised linked Brazilian administrative data from the 100 Million Brazilian Cohort, covering a 12-year period (2004 to 2015); this data is stored at the Centre for Data and Knowledge Integration for Health (CIDACS/FIOCRUZ) [28]. Data after this period was not included, since it had not been provided by the government at the time of submission. This study is presented according to the reporting of research conducted using observational, routinely collected health data (RECORD) guidelines (S1 RECORD Checklist).

Statistical analysis

Data linkage.

Linkage was performed by a team of statisticians and information technology professionals in a security-controlled environment at CIDACS/FIOCRUZ. The analysis of linkage accuracy included manual verification and assessing the receiver operating characteristic (ROC) curve (S1 Table and S1 Fig). Details are reported elsewhere [34], and a summary is included in the Supporting information (S2 Text). After the dataset has undergone all of the accuracy tests, it is made available to the researchers through a secure environment, which they access via a VPN [28].

For the current study, and other similar studies [26,27,35], BFP data was deterministically linked to the 100 Million Brazilian Cohort, using a social identification number (NIS). The linkage algorithm used 5 variables to identify matching records from 2 databases (SIM and CadÚnico), and they were recorded in both of the datasets that were being linked: the beneficiary’s name, mother’s name, sex, municipality of residence code, and date of birth. Individuals who had died, and their cause of death, were identified in the cohort, by linking mortality data with the cohort baseline. Further details are provided in the Supporting information (S2 Text, S1 Table and S1 Fig) and are fully described elsewhere [29,29,34,36,37].

Follow-up.

Since BFP was only implemented in 2004, individuals registered on CadÚnico before 2004 were considered as first registered to benefit from social programmes on 1 January 2004. For those not benefiting from BFP, follow-up began on: the CadÚnico registration date (A), or 1 January 2004, if registered before that time (B), and follow-up ended either on the individual’s death by any cause (C), when they became a BFP beneficiary (D), or on 31 December 2015 (E), if C and D were not satisfied.

For beneficiary individuals, follow-up began when they started to receive BFP benefits (within 6 months) and ended on: the individual’s death by any cause (C), when the beneficiary stopped receiving BFP benefits (F), or on 31 December 2015 (E), if C and F were not satisfied. The contribution of each individual’s person-years was calculated by subtracting the start date from the end of follow-up expressed in years.

Statistical modelling.

All analyses used an alpha of 0.05, and 2-sided p-values, and were performed using STATA version 15.0. We fitted Poisson models to estimate IRRs for suicide associated with BFP experience, unadjusted and adjusted for age (as a continuous variable), sex, education level (never attended school, pre-school, primary school or less, junior high school, high school, and college/university), unemployment, living alone, location of residence (urban or rural resident), and household characteristics, as a proxy for socioeconomic status (water supply, waste, construction materials, sanitation, and crowding) and year of registration on the cohort baseline [23,38–41]. Crowding was measured by dividing the number of individuals living in the same household by the number of rooms available [26,27]. We included follow-up (person-years) as an offset, to allow the risk of mortality to be adjusted by the observation period. Logistic models (which are usually used for binary outcomes) were not adequate for our analysis, since they do not consider individuals’ follow-up time. In addition, suicide was measured in the models as a rate instead of a binary outcome.

We estimated the association of BFP using inverse probability of treatment weighting (IPTW) [42]. IPTW uses the propensity score (PS) to balance baseline characteristics in the beneficiary and nonbeneficiary groups by weighting each individual by the inverse probability of receiving treatment. We estimated the weights for BFP beneficiaries (weight = 1) and nonbeneficiaries (weight = E(ps)/(1-E(ps))), where “E” is the beneficiary group, and “PS” is the propensity score that indicates the probability of receiving treatment (further details in S3 Text). To estimate these associations, we used an average treatment effect on the treated (ATT) estimator and fitted Poisson models to estimate the marginal IRR for BFP among recipients. Participants with missing covariate data were excluded from the main analysis. We considered the same covariates used in Poisson models to estimate the PS.

Sensitivity analyses.

The following analyses were performed to assess the robustness of the results: (A) We performed propensity score matching (PSM) analyses to investigate possible biases due to differences between the intervention and control group (further details in S4 Text); (B) we repeated the analysis, to test if there were potential biases due to differential start follow-up between BFP beneficiaries and nonbeneficiaries, excluding the initial 6-month follow-up in both groups; (C) we repeated the analysis, to investigate possible biases due to the exclusion of missing data, including missing covariate values as missing categories (S2 and S3 Tables in the Supporting information); (D) to test if there were potential biases, due to differential loss of follow-up between BFP beneficiaries and nonbeneficiaries, we censored each matched pair by the smallest contribution of person-years at risk (PYR), so that each matched pair of beneficiaries contribute to the same number of PYR. Finally (E), we ran kernel matching to check the robustness of the results. The method estimates the ATT after matching by weights, based on the same covariates, as used previously when calculating the PS, but this time using a kernel, nonparametric function [43]. Pairs were exactly matched by year of registration on the cohort baseline, and the PS matched with kernel weights [43] for age, sex, education level, unemployment, living alone, location of residence, and household characteristics, as a proxy for socioeconomic status (water supply, waste, construction materials, sewage, and crowding). We also performed subgroup analyses by sex and age (10 to 24 years old, 25 to 59 years old, and age 60 or older).

The PS was defined as the probability of receiving BFP, conditional upon the confounders listed above, and estimated using multiple logistic regression. We performed 1:1 nearest-neighbour matching with a 0.05 calliper, allowing the same nonbeneficiary to match with more than 1 beneficiary (i.e., matching with replacement) [44]. See Supporting information for matching details (S2 Text). The confounder balance was assessed using the standardised mean difference (SMD), taking absolute values of 0.1, or greater, to indicate potential for confounding by that characteristic [45] (Table 1 and S2 Fig in the Supporting information).

Download:

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

Table 1. Description of nonbeneficiaries (non-BFP) and beneficiaries of the BFP in the original and matched cohorts from 2004 to 2015.

https://doi.org/10.1371/journal.pmed.1004000.t001

Results

In the cohort, 76,532,158 individuals were aged 10 or older and, among these, 36,742 had suicide as the cause of death. After excluding individuals with missing data in the baseline characteristics, 62,766,964 remained for the main analysis (Fig 1); 27,913,305 (44%) and 34,853,659 (56%) were BFP beneficiaries and nonbeneficiaries, respectively.

Download:

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

Fig 1. Flowchart of selection of study population.

BFP, Bolsa Família programme; SIM, Mortality Information System.

https://doi.org/10.1371/journal.pmed.1004000.g001

At the baseline, there were sociodemographic differences between BFP beneficiaries and nonbeneficiaries. BFP beneficiaries were younger (median age 27.4 versus 35.4), had higher unemployment rates (56% versus 32%), a lower level of education, resided in rural areas, and experienced worse household conditions (Table 1). In the sensitivity analysis, PSM successfully matched all but 2 BFP beneficiaries with nonbeneficiaries with a similar estimated PS, and generated a matched cohort of 69,707,312 individuals with similar baseline characteristics. Following PSM, the beneficiary and nonbeneficiary group had very similar baseline characteristics; the SMD between both groups was less than 0.07 in all covariates (Table 1). There were 489,500,000 person-years in the original cohort and 305,229,883 person-years in the matched cohort, with 33,281 suicide cases (Table 2).

Download:

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

Table 2. Incidence of suicide rate for BFP participation, overall, by sex and age group, in the original and matched cohorts from 2004–2015.

https://doi.org/10.1371/journal.pmed.1004000.t002

Lower suicide rates were observed among BFP beneficiaries in all of the models analysed. Suicide rates among beneficiaries and nonbeneficiaries were 5.4 overall (95% CI = 5.32, 5.47, p < 0.001) and 10.7 (95% CI = 10.51, 10.87, p < 0.001) per 100,000 individuals in the original cohorts, and 5.5 (95% CI = 5.44, 5.61, p < 0.001) and 11.1 (95% CI = 10.41, 11.81, p < 0.001) in the matched cohort (Table 2). BFP beneficiaries had a 56% lower risk of suicide than nonbeneficiaries in the final adjusted model with IPTW (IRR = 0.44, 95% CI = 0.42, 0.45, p < 0.001) (Table 3).

Download:

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

Table 3. Suicide IRR for BFP participation in the matched and original cohorts from 2004–2015.

https://doi.org/10.1371/journal.pmed.1004000.t003

Analysing by subgroups, the difference between beneficiaries and nonbeneficiaries was more prominent among women, with a suicide rate of 2.1 (95% CI = 2.08, 2.21, p < 0.001) and 5.1 (95% CI = 4.90, 5.22, p < 0.001) per 100,000 individuals in the original cohorts, and 2.1 (95% CI = 2.09, 2.24, p < 0.001) and 5.6 (95% CI = 5.04, 6.11, p < 0.001) in the matched cohort. It was also more prominent among individuals aged 25 to 59, with a suicide rate of 6.6 (95% CI = 6.45, 6.70, p < 0.001) among beneficiaries, and 12.34 (95% CI = 12.08, 12.61, p < 0.001) among nonbeneficiaries per 100,000 individuals in the original cohorts, and 6.8 (95% CI = 6.66, 6.94, p < 0.001) and 13.42 (95% CI = 12.61, 14.28, p < 0.001) in the matched cohort (Table 2). BFP association with decreased suicide was also more prominent among women (IRR = 0.36, 95% CI = 0.33, 0.38, p < 0.001), and individuals aged 25 to 59 (IRR = 0.41, 95% CI = 0.40, 0.43, p < 0.001), followed by younger individuals, aged 10 to 24 (IRR = 0.45, 95% CI = 0.43, 0.48, p < 0.001) (Table 4).

Download:

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

Table 4. Suicide IRR for BFP participation in the matched and original cohorts by sex and age groups from 2004–2015.

https://doi.org/10.1371/journal.pmed.1004000.t004

We obtained similar point estimates for the association between receiving BFP and lower suicide rates in all of the sensitivity analyses performed, also after repeating the analyses excluding the initial 6-month follow-up in both groups, and including missing covariate values as missing categories (S2 and S3 Tables in the Supporting information). In the unmatched cohort, unadjusted Poisson models estimated a 50% lower suicide rate (IRR = 0.50, 95% CI = 0.49, 0.52, p < 0.001) and adjusted Poisson models estimated a 56% lower suicide rate (IRR = 0.44, 95% CI = 0.42, 0.45, p < 0.001). PS matching generated similar estimates to the primary analysis, with rates 61% lower (IRR = 0.39, 95% CI = 0.37, 0.41, p < 0.001) among beneficiaries (Table 3). The ATT measured using a kernel matching model, indicated 3 fewer suicide cases per 100,000 individuals among BFP beneficiaries, which is approximately a 50% decrease in the overall suicide rate (ATT = ‒0.00003; 95% CI = ‒0.00004, ‒0.00001, p < 0.001) (Table 5).

Download:

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

Table 5. ATT of suicide for BFP participation in the original cohort from 2004–2015.

https://doi.org/10.1371/journal.pmed.1004000.t005

Discussion

The study results consistently found that lower suicide rates occurred in the group of BFP beneficiaries, when compared to nonbeneficiaries. The association remained strong following adjustment for other measured variables, and after generating a propensity score-matched cohort. BFP was associated with 56% lower suicide rates, and all of the sensitivity tests showed similar results. To the best of our knowledge, our study is the first using individual-level data to show that a large-scale cash transfer programme is associated with lower suicide rates.

There has been a growing awareness that social and economic factors play a role in determining suicide [23,38–41]. A recent systematic review of low- and middle-income countries showed that approximately 62% of all studies identified the association of suicide with general poverty measures [41]. An ecological study conducted in Indonesia reported a reduction in suicide associated with a CCT programme [19], with a decrease of approximately 18% in suicide rates in the subdistricts where it was implemented [19]. In Brazil [21], an ecological study found a 6% decrease in municipalities with higher BFP coverage. However, both studies were ecological, while our research evaluated the association of the Brazilian BFP with suicide in a large, individual dataset, which allowed an evaluation of the association of the programme with reduced suicide among those who received the benefit.

It is plausible that BFP could help to prevent suicide, since poverty is directly related to factors that can lead to suicide, such as unemployment, financial strain, family instability, or violence, as well as a greater predisposition to mental disorders, such as alcoholism and depression [7,11,14,21,23,39,46–50]. A recent systematic review summarised the causal evidence and mechanisms for the relationship between poverty and common mental illnesses [49]. It explained that poverty increases worry, early-life conditions, violence, and crime, and these affect mood and anxiety disorders, while mood and anxiety disorders impact productivity, economic decision-making, female empowerment, and child development, which increase the risk of poverty, as a consequence [49].

Poverty may also be a barrier to accessing goods, resources, and services (including mental health services) contributing to the feeling of social injustice generated from inequities [20,51]. Cash transfers could increase beneficiaries’ welfare by providing greater financial stability [19,21] and by immediate poverty alleviation through the transfer of benefits to poor and extremely poor families, as well as improving access to health and social care services [27]. The interconnected nature of these determinants not only with poverty as the target, but also job skills training, access to health services and education (and the conditionality linked to continued access to these services/resources) may have a buffering impact beyond the direct implications of the cash transfer. Through conditionalities, cash transfers play a role in additional access to resources [27], and convey some hopefulness towards future prospects [39,40], which may be of particular importance in suicide prevention. See Supporting information for further details on the potential causal mechanisms (S1 Text and S3 Fig).

We have demonstrated the potential public impact of targeting a social determinant, such as poverty, to prevent a phenomenon that has been studied as an exclusively psychiatric matter. When interpersonal problems, psychological, or psychiatric factors are added to socioeconomic stressors, it can make life much harder, especially in low- and middle-income countries, where a large proportion of the population struggles to have their basic needs met. It could also indicate that in these settings, targeting social determinants through a cash transfer programme may have more potential to prevent suicide at the population level than an exclusively clinical approach. Furthermore, our results indicate the need to perform further investigation of the potential of such programmes to help prevent suicide, for example, studies trying to understand the mechanisms and investigating the potential of using a combination of programmes to try to prevent suicide.

Priorities in the suicide prevention field should be urgently established, since the current pandemic has increased economic instability, making many people more vulnerable to mental health problems [52–54], including suicidal behaviour [10]. The mental health consequences of this unprecedented situation are likely to affect societies for a considerable time [10]. This may be especially important for hard-hit countries, including Brazil, and those which already have high suicide rates. In Brazil, no consistent evidence of pandemic-related worsening psychopathology has been found, but socioeconomic disadvantages have been associated with increased odds of psychiatric disorders during the COVID-19 pandemic [55].

The overall response to the COVID-19 crisis should consider suicide prevention measures [9]. Programmes targeting poverty during the pandemic may have an impact on reducing suicide in the coming years. Evidence of country-level strategy efficacy is more critical than ever [10].

Supporting information

Acknowledgments

The authors would like to thank the data production team at CIDACS/FIOCRUZ, and all of the collaborators for their work on compiling the 100 Million Brazilian Cohort, and providing appreciated input during the study. DBM and JMP had full access to all of the study data and take responsibility for data integrity and accuracy of the analyses.