Pathways to suicide prevention: a population-based cohort study of the mediating role of Bolsa Família in psychosocial stressors in the 100 million Brazilian cohort

Abstract

Introduction

Suicide is a major public health concern in Brazil, and its risk is heightened by psychosocial stressors disproportionately affecting socioeconomically vulnerable groups. This study aimed to investigate how the cash transfer Bolsa Família Programme (BFP) mediates the relationship between psychosocial stressors and suicide risk among low-income individuals in Brazil.

Methods

We analysed data from the 100 Million Brazilian Cohort (2008–2015), which includes individuals registered in the Cadastro Único (national registry of low-income families) linked to hospitalisation and mortality records through deterministic and probabilistic methods. Psychosocial stressors were defined as: (1) violence (hospitalisation for interpersonal violence); (2) impulsivity (unintentional injury hospitalisations); and (3) psychiatric hospitalisation (International Classification of Diseases, 10th Revision (ICD-10):F00–F99). Suicide mortality (ICD-10 X60–X84) was the outcome. Analyses included individuals aged ≥10 years and applied structural equation models adjusted for sociodemographic confounders.

Results

Data from over 22 million participants revealed that 54% were female, 53% self-identified as Pardo/Brown, and 38% were enrolled in BFP. Individuals exposed to violence, impulsivity or mental disorders showed higher risk of suicide (β=0.418, 0.246 and 0.395, respectively; all p<0.001). We confirmed that BFP mediated the association for violence and mental disorders (indirect effects: β=−0.005 and β =−0.006; p<0.001), but not for impulsivity. All models showed good fit.

Conclusions

BFP was directly and indirectly associated with a reduction in suicide mortality among individuals exposed to violence and mental disorders. These analyses underscore the complex relationship between poverty alleviation and decreased suicide risk within a large administrative data set, highlighting the need to elucidate pathways to enhance suicide prevention strategies.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Conditional cash transfer programmes, such as Bolsa Família in Brazil, have been shown to reduce suicide risk and improve mental health outcomes in vulnerable populations. Previous studies have demonstrated that poverty, exposure to violence, impulsivity and mental disorders are significant psychosocial stressors associated with increased suicide risk.

WHAT THIS STUDY ADDS

• This study provides new evidence on the mediating role of Bolsa Família in reducing the probability of suicide among individuals exposed to violence and mental disorders. The findings show that Bolsa Família is not only associated with a direct reduction in suicide risk but also mediates the association between psychosocial stressors (such as violence and mental disorders) and suicide risk.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study emphasises the importance of integrating mental health considerations into social protection policies like Bolsa Família. Policy-makers should consider expanding the scope of such programmes to address mental health factors and the effects of violence, as this could further reduce suicide risk in vulnerable populations. These findings could influence future research on the intersection of social welfare and mental health, leading to more comprehensive interventions.

Introduction

Suicide is a significant global public health challenge, affecting approximately 703 000 individuals each year.¹ Most suicides occur in low-income and middle-income countries^{1 2} and disproportionately affect socioeconomically vulnerable groups.¹ Suicide patterns also vary by sex^{1 3 4} and across countries and regions, reflecting sociocultural and economic contexts.¹ Vulnerability to suicide is further reinforced by psychosocial stressors such as violence, impulsivity and mental disorders.5,7 Poverty alleviation programmes, such as Brazil’s Bolsa Família Programme (BFP), may help counteract these risks by reducing hardship^{8 9} and facilitating access to healthcare and social support.¹⁰ Understanding the complex nature of the issue is imperative for targeted and effective prevention efforts to alleviate its burden.

Recent conceptual frameworks emphasise the bidirectional relationship between poverty and mental disorders,¹¹ given that poverty is consistently associated with an increased risk of suicide.12,14 This could be attributed to increased stress, family hardship, alcoholism, violence and mental illnesses among economically vulnerable individuals.14,16 Conversely, mental illnesses can worsen socioeconomic vulnerabilities, creating a cyclical pattern.¹¹

One of the world’s largest conditional cash transfer programmes (CCTP) was implemented in 2003 in Brazil aiming at alleviating poverty—the BFP.¹⁷ BFP is targeted to families registered in Cadastro Único (CadÚnico), Brazil’s national social programme registry and eligibility requires a monthly per capita income below half the minimum wage (eg, BRL778.00 in 2015, ≈US$155) or a total family income of up to three minimum wages.¹⁸ Conditionalities include a minimum school attendance rate of 85% for children and participation in routine healthcare visits, such as prenatal care and vaccination schedules.¹⁰ By linking cash transfers to education and health, the programme also strengthens access to basic healthcare and social support networks.¹⁰

Studies suggest that the BFP and other CCTP can reduce suicide rates,^{13 14} violence exposure^{19 20} and mental disorders.²¹ These programmes address key social determinants of well-being, mitigating risks associated with poor mental health by targeting vulnerable populations. In Brazil, a country with regional economic and social disparities, evidence indicates that the BFP protective effect on mental health outcomes, such as suicide, persists across geographic areas and is strengthened with higher coverage and longer duration of benefit receipt.¹³

Psychosocial stressors—such as violence, impulsivity and psychiatric hospitalisations—are indicators of severe social or clinical vulnerability and are strongly associated with adverse mental health outcomes, including suicide.5,7 Individuals exposed to these stressors often have fewer opportunities and resources to improve their socioeconomic status, limiting their ability to cope effectively with these adverse conditions.^{11 22} Cash transfer programmes, like BFP, may offer protective benefits by helping to mitigate these challenges through poverty alleviation, improved access to healthcare services,²³ enhanced food security⁹ and strengthened social support networks.²⁴ These pathways provide a foundation for building resilience against adverse psychological outcomes.

The extent to which such programmes can mediate the effects of psychosocial stressors on suicide risk remains underexplored. Although some studies have examined the effects of CCTP on suicide^{8 13} mediation analyses in this field have generally focused on socioeconomic factors such as income and economic empowerment.^{25 26}

In addition, the existing studies have primarily focused on exposures to adverse conditions, such as violence,^{25 26} and generally addressed suicidal behaviours,^{25 27 28} particularly ideation and attempts. Many of these investigations were conducted in specific populations or clinical settings—often with relatively small samples—for example, in South Africa²⁵ and Brazil.²⁶ Our research concentrates on suicide deaths only, whereas prior work has predominantly examined suicidal ideation and attempts. This reflects a methodological choice, as attempts and deaths differ in important ways; psychosocial stressors may precipitate suicidal behaviour, with individuals often developing thoughts of self-harm when exposed to adverse conditions.²⁹ However, those who die by suicide represent a distinct group, with demographic, clinical and behavioural characteristics that may differ from those who attempt.^{30 31}

For example, although women attempt suicide more often, men account for most suicide deaths due to their greater use of highly lethal methods.^{4 30} Suicide deaths are more strongly associated with severe mood disorders, psychosis and comorbid substance use, whereas other psychiatric comorbidities, such as personality disorders, are more frequently observed among attempters.³¹ Studying those who die by suicide not only fills an important research gap but also prevents combining distinct groups in ways that could obscure critical nuances and risk patterns unique to suicide deaths.

To address the research gaps and advance understanding of the potential role of CCTP, we use a robust methodological approach with a large cohort derived from administrative-linked data consisting of individuals registered in the BFP. This study aimed to assess whether participation in a poverty alleviation programme mediates the relationship between psychosocial stressors—such as violence exposure, impulsivity and mental disorders—and suicide. We hypothesise that individuals exposed to these psychosocial stressors are more likely to qualify for and receive BFP, thereby potentially experiencing benefits associated with programme participation. We propose that BFP participation is indirectly associated with lower suicide risk by partially mediating the relationship between these stressors and suicide. Exploring its role as a mediator allows us to better understand the pathways through which cash transfer programmes impact health outcomes, offering insights into the multifaceted ways these programmes can mitigate the consequences of adverse conditions.

Materials and methods

Study design and participants

This study used longitudinal data extracted from the 100 Million Brazilian Cohort, an open cohort that links data from some large national social and health databases.³² We included individuals aged >10 years, either male or female, who enrolled in the Brazilian national social programme register (CadÚnico, Cadastro Único) between 2008 and 2015. CadÚnico is used in Brazil to identify and target social programmes (eg, Bolsa Família). Enrolment occurs at the household (family) level at municipal Social Assistance Reference Centres or via home visits; a responsible adult registers the household and lists all co-residents who share address and income. Each listed person is assigned a Social Identification Number. In this study, the analytic unit was the individual. We set the age threshold and study period given the rarity of suicide cases under the age of 10 (only 57 registered from 2004 to 2015) and the availability of variables related to the study purpose, derived from the databases included in the linkage process (see online supplemental data sources S1).

Eligibility criteria and follow-up

Follow-up began at cohort entry (the date of the person’s first CadÚnico record) and ended at the earliest of suicide, death from other causes, or 31 December 2015. To ensure the temporal relationships between variables in the conceptual models, we applied the following exclusions: (1) BFP start and end on the same date; (2) any exposure recorded after BFP start or end; (3) death from other causes during the BFP period; (4) suicide before or on the BFP start date; (5) suicide before or on the exposure date; (6) exposure dated before follow-up start and (7) follow-up of only 1 day. We apply these eligibility criteria and establish three distinct databases, one for each exposure: violence (n=22 180 620), impulsivity (n=21 831 440), and mental disorders hospitalisations (n=22 118 469). The final samples from these datasets represented 16.82%, 16.57% and 16.79% of the original data, respectively. Online supplemental figure S1 illustrates the eligibility application process (see online supplemental figure S1).

The temporal flow of exposure, mediation and outcome with non-eligibility criteria for temporal consistency is shown in Figure 1. We adhered to the methods and analyses outlined in the previously published research protocol.¹⁴

Figure 1. Temporal flow of exposure, mediation and outcome with non-eligibility. Diagram illustrating the temporal framework used to define exposure, mediation and outcome periods in the cohort study. The figure details the temporal consistency criteria applied to ensure correct sequencing between hospitalisation records, Bolsa Família Programme (BFP) participation and suicide.

Procedures

The final dataset was provided to the authors, deidentified, covering the period from 2008 to 2015. The original data collection varies across administrative databases. Trained employees from the Social Assistance Reference Centre gather data for CadÚnico. From the Hospitalisation Information System of the National Unified Health System (Sistema de informações Hospitalares/Sistema Único de Saúde, SIH/SUS), medical professionals, maxillofacial surgeons and obstetric nurses complete the Hospital Admission Authorisation (Autorização de Internação Hospitalar) forms based on their expertise.³³ Within the Mortality Information System (Sistema de Informação sobre Mortalidade, SIM), death certificates are the primary data source, completed at the location of death. In cases of natural deaths without medical intervention, the responsibility falls on the Death Verification Services (Serviço de Verificação de Óbito, SVO), where a doctor completes the Death Certificate. If an SVO is unavailable, any doctor must complete the document.³⁴

Exposures

We assessed data about violence, impulsivity and hospitalisation for mental disorders from SIH/SUS database. The variable ‘hospitalisation for violence’ was created following the mandatory approach for recording hospitalisations established by the Brazilian Ministry of Health,³⁵ which recommends recording injuries as the primary cause (ICDs: S00–T88) and aggression as the secondary cause (ICDs: X85–Y09). This coding structure is required to mitigate potential systematic errors to ensure comparability within national health statistics. In the context of data linkage, strict adherence to this standard is particularly important to maintain the reliability of merged datasets. For this reason, individuals with any classification of hospitalisation for violence that did not adhere to the standardised recommendation were excluded (ie, hospitalisations where violence was recorded as the primary cause or as the secondary cause without an injury as the primary cause, n=508). This variable was included in the models as binary (absence=0, presence=1), and potential duplicates were removed based on the event date.

Impulsivity was specified from the SIH/SUS database with the records of hospitalisation due to injuries (ICDs: S00–T88) and included as binary in the models (absence=0, presence=1). Injuries spanned various anatomical regions, including the head (S00–S09) and upper limbs (S40–S69). We used unintentional injury hospitalisations as a proxy for impulsivity, conceptually grounded in definitions of impulsivity as disinhibition and risk-taking³⁶ and empirically supported by evidence linking unintentional injuries to impulsive traits in both clinical³⁷ and non-clinical populations.³⁸ We excluded hospitalisations for injuries related to interpersonal violence (ICDs: X85–Y09), as this exposure is analysed separately as a mediator in our models, and self-inflicted injuries (ICDs: X60–X84), given their conceptual proximity to the study outcome. These exclusions ensured that the impulsivity measure remained conceptually and operationally distinct from the variable capturing violence-related hospitalisations.

Hospitalisations due to mental disorders were specified using SIH/SUS records under the ICD codes for: organic disorders (F00–F09), substance use-related disorders (F10–F19), schizophrenia and mood disorders (F20–F39), neurotic and stress-related disorders (F40–F48), behavioural syndromes (F50–F59), personality disorders (F60–F69), intellectual disability (F70–F79), developmental disorders (F80–F89) and behaviour/emotional disorders in childhood/adolescence (F90–F98), plus unspecified mental disorders (F99–F99). This variable was included as a binary in the models (absence=0, presence=1).

Mediator

The BFP, Brazil’s largest cash transfer initiative, aims to combat poverty and promote social equity. It provides monthly transfers to low-income households, with amounts adjusted for household income, size and the presence of children or adolescents, with an average monthly benefit of approximately BRL 170–190 per household during the study period. Continuation of benefit receipt was conditional on school attendance and health check-ups, and vaccination of children, as well as prenatal care for pregnant women, and duration of receipt varied according to continued eligibility.^{10 18}

To ensure a balanced comparison group (ie, individuals with similar profiles but differing in terms of programme exposure), BFP participants were identified as those who began receiving benefits within 6 months of CadÚnico registration.⁸ The unexposed group consisted of individuals who applied in the same year but did not receive benefits within this period, with follow-up ending if they later joined the programme. Individuals who were never enrolled in BFP were also categorised as unexposed.

Outcome

Suicide was obtained from the SIM database when the primary cause of death was a self-inflicted injury (ICD codes X60–X84) and encoded as a binary variable (absence=0, presence=1).

Covariates

We used the directed acyclic graphs (DAG) criteria (see online supplemental figure S2) for selecting the covariates for adjustment to define the minimum set of covariates to adjust for confounding and to reduce variable selection bias.^{39 40} We were guided by existing literature^{5 6 14 21} to represent the theoretical model and elucidate the involvement of covariates in the three mediation models. The DAGs were built using DAGitty software (V.3.1; Nijmegen, GE, The Netherlands). The covariates suggested by the backdoor criteria to adjust for confounding included sex, age, race/skin colour, location of residence, educational level and socioeconomic status. In addition, for the model with impulsivity as the exposure, mental disorders and violence were identified as confounders, while for the model with hospitalisation for mental disorders as the exposure, violence was identified as a confounder.

Demographic, economic and social information was collected during participants’ CadÚnico registration. Sex assigned at birth (female/male) and birth date were sourced from official documents or self-reported if unavailable. Age was calculated by subtracting the birth date from the enrolment date in the follow-up; we considered as missing the outliers those aged >120 years; for the models, we categorised as ‘youth (10–24 years)’, ‘adults (25–59 years)’ and ‘older adults (60+ years)’. Race/skin colour is self-declared and was classified as White, Black, Pardo, Indigenous or Asian, following the categories used by the Brazilian Institute of Geography and Statistics. In Brazil, Pardo refers to individuals with predominantly Black and mixed ancestry, including European, African and Indigenous backgrounds.⁴¹ Pardos and Black individuals often share similar social determinants and experiences of structural disadvantage and exhibit comparable sociodemographic and health characteristics.⁴² The educational levels included ‘have never been to school’, ‘preschool’, ‘primary school or less’, ‘junior high school’, ‘high school’ and ‘college/university’. The location residence variable categorised the place of residence as ‘urban’ or ‘rural’, with urban areas classified as 0 and rural areas as 1. The ‘living conditions’ variable was determined by aggregating household characteristics such as access to water, household construction material, garbage disposal, sewage disposal and electricity.⁴³ Each question in the ‘living conditions’ variable was assigned 1 point for the best conditions and 0 for the worst.

Statistical analysis

Descriptive analyses were conducted on all variables of interest. Differences in continuous variables among participants were assessed using the Mann-Whitney U test, given the non-normal distribution of the data. The χ² test was applied to categorical variables. As detailed in Procedures, three analytical datasets (violence, impulsivity, psychiatric hospitalisation) were used. Because they were derived from the same linked sources (CadÚnico, SIH/SUS, SIM) with similar eligibility criteria, sociodemographic distributions were equivalent. We, therefore, present the violence dataset in the main text and provide the others in the online supplemental tables S1 and S2.

In the inferential analysis, three models were specified to explore the relationship between the exposures and suicide, with the BFP as a mediator. Model A examined violence, model B analysed impulsivity and model C hospitalisation for mental disorders as exposures. In all models, BFP served as the mediator and suicide as the outcome.

We specifically applied path analysis within the structural equation modelling (SEM) framework to test mediation, assessing whether BFP participation influenced the association between psychosocial stressors (exposures) and suicide (outcome). This method estimates direct, indirect and total effects simultaneously. In our models, path a represents the direct association between psychosocial stressors and BFP, path b between BFP and suicide and path c the direct link between psychosocial stressors and suicide. The indirect effect shows how much of the association between psychosocial stressors and suicide can be explained by BFP participation. The total effect is the sum of this indirect effect and the direct effect.

All models were fitted using data from participants with complete information on BFP participation, mediators, confounders and outcomes. Sex, race/skin colour and location of residence were recorded as nominal variables. Educational level and living conditions were treated as ordinal variables. The variables for notification of violence, impulsivity, hospitalisation for mental disorders and BFP enrolment were treated as categorical variables in the analysis. Sociodemographic data were available only at baseline and were included in the models as covariates. The final set of confounders was included in the models: sex, race/skin colour, age, location of residence, education level and living conditions.

Model fit was assessed based on established criteria, where they were deemed good fit if a root mean square error of approximation (RMSEA) ≤0.08, a Comparative Fit Index (CFI) >0.95⁴⁴, Tucker-Lewis Index (TLI) >0.90⁴⁵ and standardised root mean square residual <0.068.⁴⁶ The robust method with the Weighted Least Squares Mean and Variance adjusted estimator was chosen to calculate SEs of indirect effects, and the theta parameterisation was applied, which fixes the variance of observed variables at 1. This method is suitable for handling non-normal variables and categorical or ordinal data.⁴⁷ Additionally, the models were run under the Bayesian framework (see online supplemental file) as sensitivity analyses. Point estimates and directions of the associations were close to those reported in the manuscript.

Data preparation and descriptive analyses were performed using R software (V.3.6.3, The R Foundation, Vienna, Austria, 2020). To perform SEM we used the MPLUS V.8.4 software (Muthen & Muthen, Los Angeles, California, USA) and an alpha of 0.05. By default, MPLUS uses listwise deletion to estimate the model and manage missing data, meaning that cases with missing values on x-variables are excluded from the analysis.⁴⁸ Sensitivity tests were conducted using the SEM Bayesian approach in MPLUS, detailed in online supplemental sensitive analyses S1 and table S3.

Results

From a cohort of over 130 million participants, we assessed over 22 million eligible individuals. Among these, 54.8% were female and 38.9% were enrolled in the BFP. The majority, 54.1%, were aged between 25–59 years. Most of the population self-identified as pardo and were adults, with many reporting access to adequate living conditions: running water (76.1%), houses constructed from bricks or cement (81.4%), a public garbage collection system (81.3%) and sewage disposal facilities (94.5%). Regarding education, a higher percentage of women reached high school (32.3%), while men were more likely to have only primary education or less (32.4%). The median time of enrolment in the BFP was 0.58 (IQR: 0.00, 3.72) years. The overall cumulative incidence of violence, impulsivity, psychiatric hospitalisation and suicide during follow-up was found to be 0.1, 2.9, 0.6 and 0.04%, respectively, with males exhibiting higher frequencies of impulsivity, psychiatric hospitalisations and suicide occurrence compared with females (table 1). Only a small proportion of data had missing values, specifically in race/skin colour (5.8%), education (7.0%), location residence (0.2%), access to water (3.0%), house material construction (3.0%), access to electricity (3.0%), means of garbage disposal (3.0%) and means of sewage disposal (5.5%) (table 1).

Table 1. Sample descriptive analysis stratified by sex.

	Overall	Male	Female	P value*
n	22 180 620	10 021 906	12 158 714
Age (mean (SD))	33.64 (16.95)	34.32 (16.78)	33.07 (17.07)	<0.001
Age group (%)				<0.001
Youth (10–24)	8 000 026 (36.1)	3 268 336 (32.6)	4 731 690 (38.9)
Adults (25–59)	12 000 088 (54.1)	5 753 615 (57.4)	6 246 473 (51.4)
Older adults (60+)	2 179 600 (9.8)	999 489 (10.0)	1 180 111 (9.7)
NA	906 (0.0)	466 (0.0)	440 (0.0)
Race/skin colour (%)				<0.001
White	7 277 270 (32.8)	3 098 900 (30.9)	4 178 370 (34.4)
Black	1 599 754 (7.2)	713 716 (7.1)	886 038 (7.3)
Asian descendants	109 120 (0.5)	43 775 (0.4)	65 345 (0.5)
Brown	11 785 245 (53.1)	5 408 625 (54.0)	6 376 620 (52.4)
Indigenous	116 309 (0.5)	56 106 (0.6)	60 203 (0.5)
NA	1 292 922 (5.8)	700 784 (7.0)	592 138 (4.9)
Education level (%)				<0.001
Have never been to school	2 018 276 (9.1)	1 044 982 (10.4)	973 294 (8.0)
Preschool	212 597 (1.0)	105 774 (1.1)	106 823 (0.9)
Primary school or less (5 years)	6 348 357 (28.6)	3 244 540 (32.4)	3 103 817 (25.5)
Junior high school (6–10 years)	5 180 722 (23.4)	2 355 424 (23.5)	2 825 298 (23.2)
High school (10–12 years)	6 327 219 (28.5)	2 404 194 (24.0)	3 923 025 (32.3)
College/university (13 years)	547 445 (2.5)	175 577 (1.8)	371 868 (3.1)
NA	1 546 004 (7.0)	691 415 (6.9)	854 589 (7.0)
Location residence (%)				<0.001
Urban	18 323 404 (82.6)	7 945 723 (79.3)	10 377 681 (85.4)
Rural	3 820 156 (17.2)	2 043 445 (20.4)	1 776 711 (14.6)
NA	37 060 (0.2)	32 738 (0.3)	4322 (0.0)
Access to water (%)				<0.001
Public network (running water)	16 884 251 (76.1)	7 375 363 (73.6)	9 508 888 (78.2)
Well, natural sources or other	4 631 901 (20.9)	2 330 673 (23.3)	2 301 228 (18.9)
NA	664 468 (3.0)	315 870 (3.2)	348 598 (2.9)
Means of garbage disposal (%)				<0.001
Burned, buried, outdoor disposal or other	3 488 742 (15.7)	1 870 370 (18.7)	1 618 372 (13.3)
Public collection system	18 027 322 (81.3)	7 835 596 (78.2)	10 191 726 (83.8)
NA	664 556 (3.0)	315 940 (3.2)	348 616 (2.9)
House material construction (%)				<0.001
Bricks/cement	18 046 591 (81.4)	7 966 671 (79.5)	10 079 920 (82.9)
Wood, vegetal materials and other	3 469 556 (15.6)	1 739 327 (17.4)	1 730 229 (14.2)
NA	664 473 (3.0)	315 908 (3.2)	348 565 (2.9)
Access to electricity (%)				<0.001
Community electricity metre	1 279 678 (5.8)	552 081 (5.5)	727 597 (6.0)
Own electricity metre	18 118 551 (81.7)	8 180 922 (81.6)	9 937 629 (81.7)
Use of lamp, candle or other type of lighting	2 117 950 (9.5)	973 041 (9.7)	1 144 909 (9.4)
NA	664 441 (3.0)	315 862 (3.2)	348 579 (2.9)
Means of sewage disposal (%)				<0.001
Ditch or other	1 340 522 (6.0)	671 235 (6.7)	669 287 (5.5)
Homemade septic tank	4 971 430 (22.4)	2 423 369 (24.2)	2 548 061 (21.0)
Public network	11 678 313 (52.7)	4 910 907 (49.0)	6 767 406 (55.7)
Septic tank	2 961 990 (13.4)	1 385 898 (13.8)	1 576 092 (13.0)
NA	1 228 365 (5.5)	630 497 (6.3)	597 868 (4.9)
BFP yes=yes (%)	8 622 770 (38.9)	4 070 164 (40.6)	4 552 606 (37.4)	<0.001
Time receiving BFP (median (IQR))	0.58 (0.00, 3.72)	0.50 (0.00, 3.59)	0.67 (0.00, 3.83)	<0.001
Hospitalisation for impulsivity=yes (%)	645 735 (2.9)	438 541 (4.4)	207 194 (1.7)	<0.001
Exposure to violence=yes (%)	12 435 (0.1)	9927 (0.1)	2508 (0.0)	<0.001
Hospitalisation for mental disorders=yes (%)	136 475 (0.6)	82 823 (0.8)	53 652 (0.4)	<0.001
suicide=yes (%)	9771 (0.0)	7620 (0.1)	2151 (0.0)	<0.001

Note: This table displays the sample distribution of the violence dataset. Descriptive analyses for the impulsivity and mental disorders hospitalisations datasets can be found in the online supplemental tables S1 and S2. As expected, the distributions of sociodemographic covariates were very similar across the three analytic datasets, given that they were derived from the same population with different eligibility criteria to ensure temporal consistency.

^*χ² test for categorical variables and or t-test for continuous variables.

BFP, Bolsa Família Programme; NA, no answer.

The initial models, which examined direct pathways from the exposures—violence (model A), impulsivity (model B) and mental disorders (model C)—to suicide, as well as indirect pathways through BFP, demonstrated suboptimal fit, with a CFI and TLI below 0.80. Alternative models, excluding education and living conditions as cofactors, subsequently resulted in improved fit (table 2) and were chosen as final models. Details of the direct and indirect effects are provided in table 3 and figure 2.

Table 2. Models examining pathways to suicide.

	Fit indices
	x2	df	P value	CFI	TLI	RMSEA
Model A: Hospitalisation for violence → BFP participation → suicide	18 741.128	4	<0.001	0.90	0.62	0.015; 0.015; p<0.05
Model B: Hospitalisation for impulsivity → BFP participation → suicide	36 796.723	6	<0.001	0.87	0.54	0.017; 0.017; p<0.05
Model C: Hospitalisation for mental disorders → BFP participation → suicide	30 266.750	5	<0.001	0.84	0.44	0.017; 0.017; p<0.05

Models were adjusted for sex, race/skin colour, age and location of residence. Model B was additionally adjusted for violence and hospitalisation for mental disorders. Model C was additionally adjusted for violence.

BFP, Bolsa Família Programme; CFI, Comparative Fit Index; RMSEA, root mean square error approximation; SRMR, standardised root mean square residual; TLI, Tucker-Lewis Index.

Table 3. Standardised direct, indirect and total effects exposures on suicide in final models (SEs shown in parentheses).

	Effects
	Direct†	Indirect‡	Total
Model A: Hospitalisation for violence → BFP participation → suicide	0.418 (0.012)*	−0.005 (0.001)*	0.413 (0.011)*
Model B: Hospitalisation for impulsivity → BFP participation → suicide	0.246 (0.007)*	0.000 (0.000)	0.247 (0.007)*
Model C: Hospitalisation for mental disorders → BFP participation → suicide	0.395 (0.008)*	−0.006 (0.001)*	0.389 (0.008)*

Models were adjusted for sex, race/skin colour, age and location of residence. Model B was additionally adjusted for violence and hospitalisation for mental disorders. Model C was additionally adjusted for violence.

^*Statistically significant effects (p<0.001).

^†Direct effect of exposure on suicide risk.

^‡Indirect effects of exposures through BFP participation (mediation).

BFP, Bolsa Família Programme.

Figure 2. Simplified diagram from structural equation modelling analysis. (A–C) The structural paths between hospitalisation for (A) violence, (B) impulsivity and (C) mental disorders, participation in the Bolsa Família programme (BFP) and suicide outcomes. Each model displays standardised coefficients (β) and SEs for direct, indirect and total effects. Model fit indices are reported below each diagram, including the root mean square error of approximation (RMSEA), Comparative Fit Index (CFI), Tucker-Lewis Index (TLI) and standardised root mean square residual (SRMR). VAR indicates residual variance, fixed at 1 under theta parameterisation for model identification. Significance levels are indicated as p<0.001 (*).

In model A (see figure 2A), individuals exposed to violence were more likely to receive Bolsa Família benefits, as indicated by the positive association between violence exposure and BFP enrolment (path a, β=0.146, SE=0.002; p<0.001). BFP itself had a direct effect that was associated with a decrease in suicide rates (path b, β= –0.035, SE=0.005; p<0.001). Violence exposure was directly associated with an increase in suicide (path c, β=0.418, SE=0.012; p<0.001). The indirect effect of violence on suicide, mediated by BFP, was associated with a reduction in suicide (β= –0.005, SE=0.001; p<0.001), with a total effect estimate of β=0.413, SE=0.011 (p<0.001). In model B (figure 2B), impulsivity (measured by injury) was positively associated with BFP enrolment, indicating that individuals who suffered injury were more likely to receive Bolsa Família benefits (path a, β=0.057, SE=0.001; p<0.001). In this model, BFP enrolment did not show a significant direct effect on suicide (path b, β=0.000; SE=0.000; p>0.05). Impulsivity was directly associated with an increase in suicide (path c, β=0.246, SE=0.007; p<0.001). No significant indirect effect was found through BFP in this model, and the total effect estimate for impulsivity on suicide remained positive (β=0.247, SE=0.007; p<0.001).

In model C (figure 2C), hospitalisation for mental disorders was positively associated with BFP enrolment, indicating that individuals with a history of psychiatric hospitalisations were more likely to receive Bolsa Família benefits (path a, β=0.150, SE=0.004; p<0.001). BFP itself had a direct effect that was associated with a decrease in suicide rates (path b, β=–0.042, SE=0.004; p<0.001). Hospitalisation for mental disorders was directly associated with an increase in suicide (path c, β=0.395, SE=0.008; p<0.001). The indirect effect of psychiatric hospitalisation on suicide, mediated by BFP, was associated with a reduction in suicide (β= –0.006, SE=0.001; p<0.001), with a total effect estimate of β=0.389, SE=0.008; (p<0.001).

Discussion

It is well established that poverty negatively impacts mental health, and prior studies have reported that cash transfer programmes are associated with lower suicide risk.^{8 13 19 21 49} However, our results are the first to demonstrate how cash transfers mediate key risk factors for suicide. Our analysis of longitudinal administrative linked data confirmed that it was statistically associated with lower suicide risk and with a partial mediation of the relationship between psychosocial stressors and suicide. These associations suggest that social protection programmes may play a role in mitigating suicide risk for beneficiaries facing specific adversities, such as exposure to violence and hospitalisation for mental disorders, with no associations for individuals hospitalised by impulsivity.

These results shed light on the potential of using administrative databases to enhance our understanding of interventions associated with reductions in suicide. In the model with violence as exposure, we found an expected positive relationship between violence victimisation and BFP participation, as individuals who entered the health system in Brazil due to violence victimisation should be referred to other types of assistance. We confirmed a direct association between BFP and reduced suicide risk, corroborating previous studies.^{13 14} Our mediation analysis is consistent with previous reports,¹⁹ showing that higher Bolsa Família coverage was associated with lower rates of violence-related hospitalisations. By demonstrating that BFP also mediates the relationship between violence exposure and suicide, our study highlights BFP’s broader role in mitigating both immediate and long-term consequences of violence. Additionally, our analysis showed that violence correlates with increased suicide rates, aligning with stress theories,^{50 51} which suggest that adverse living conditions, such as poverty, escalate secondary stressors like violence, thereby negatively impacting mental health.^{52 53}

In the model with impulsivity as exposure, we also identified an expected positive relationship between impulsivity, measured by injuries and hospitalisations, and BFP participation, possibly because both low-income status and financial stress have been associated with greater levels of impulsive behaviour.⁵⁴ We did not find a direct or mediated association of BFP on suicide in this model, which may be related to the use of impulsivity as a proxy variable. Alternatively, it may indicate that impulsivity elevates suicide risk primarily through proximal mechanisms, such as disinhibition and poor emotional regulation.⁵⁵ Nevertheless, a direct relationship between impulsivity and suicide was observed. This aligns with the Developmental Psychopathology perspective, which suggests that adverse experiences can disrupt emotional regulation and increase impulsivity, ultimately elevating suicide risk,⁵⁵ and with evidence showing that impulsivity is a robust correlate of suicidal behaviour.⁵⁶

In the model with hospitalisation for mental disorders as the exposure, a positive association with BFP participation was found, likely due to the increased risk of mental health issues among individuals facing socioeconomic disadvantages. The significant mediation in this model corroborates findings from Bonfim et al,⁵⁷ who showed that BFP reduced mortality among individuals hospitalised for psychiatric disorders. Our study adds to this by incorporating a counterfactual element by including non-hospitalised individuals in the data, and by evaluating the mediating role of BFP in further reducing suicide risk among those registered in social programmes. Finally, this model is consistent with previous work linking mental disorders as major risk factors for suicide.^{58 59}

Our findings must be considered within the context of certain limitations. Proxies can be unreliable measures of the target behavioural observed variables. In particular, impulsivity was operationalised through hospitalisations due to unintentional injury, with self-inflicted cases excluded to avoid overlap with the suicide outcome, and violence was captured only through hospitalisations. These operational choices may underestimate the broader spectrum of impulsive behaviours and exposure to violence. Although all exposures were derived from the same hospitalisation system (SIH/SUS), potential variability in reporting practices could affect the robustness of the estimates. Nevertheless, linkage through the 100 Million Brazilian Cohort ensured consistency in data integration.³² Future studies should consider more granular indicators beyond hospitalisation and explore integrated data collection across social protection and health services to strengthen the robustness of mediation analyses. Moreover, due to convergence issues, adjustment variables such as living conditions and education could not be incorporated into the models.

Also, due to the pattern of missingness and the method of estimation, stratification and multiple groups models did not converge. While CFI and TLI did not meet conventional thresholds, RMSEA, recognised as more robust for administrative data,⁶⁰ indicated adequate fit, supporting model validity. Although this study captures a significant sample of the Brazilian population, its focus on socioeconomically vulnerable groups limits generalisability to higher-income populations or those outside Brazil. Furthermore, despite longitudinal data capturing every pathway being unavailable, evidence supports that BFP acts on critical determinants of well-being, which in turn help reduce suicide risk.^{9 23 24} Mediation analysis remains a robust method for capturing indirect effects within this framework, even without explicitly measuring all pathways.

Finally, although BFP addresses socioeconomic challenges, it was not originally designed to mitigate risk factors associated with poor mental health. However, our mediation analysis revealed that BFP indirectly contributes to reducing suicide risk by mediating the effects of psychosocial stressors, such as violence and mental disorders. This highlights the potential of a comprehensive approach, suggesting that integrating mental health policies into cash transfer frameworks could be a promising strategy to improve mental health outcomes among vulnerable populations.

For instance, Brazil has an established network of services that address violence and mental health, including including Social Assistance Reference Centres (Centros de Referência de Assistência Social, CRAS), Specialised Social Assistance Reference Centres (Centros de Referência Especializados de Assistência Social, CREAS) and Psychosocial Care Centres (Centros de Atenção Psicossocial, CAPS), but there is no formal articulation between these services and the BFP. Referrals for beneficiary families are mostly local and unsystematic. Our findings point to an opportunity to create an integrated family support network that combines violence-prevention education, psychological assessment and structured referral pathways for mental healthcare. This approach aligns with WHO recommendations for multisectoral strategies linking social protection, mental health and violence–prevention interventions.61,63

This study contributes to understanding the factors influencing suicide among Brazil’s vulnerable populations, emphasising the need for interventions targeting violence, impulsivity and mental disorders. If the programme did not exist, individuals would likely have greater barriers to services, as Bolsa Família helps connect families to local assistance and health networks.²⁴ The recognition of demographic, economic and environmental factors implicated in the models also reinforces the role of social determinants in mental health, as endorsed by the United Nations Sustainable Development Goals.⁶⁴ By identifying potential pathways that help mitigate the link between these psychosocial stressors and suicide risk, the findings presented here are critical for improving the identification of clinical and social avenues for intervention.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability statement

Data are available on reasonable request.