Conditional cash transfer program and child mortality: A cross-sectional analysis nested within the 100 Million Brazilian Cohort.

BACKGROUND: Brazil has made great progress in reducing child mortality over the past decades, and a parcel of this achievement has been credited to the Bolsa Família program (BFP). We examined the association between being a BFP beneficiary and child mortality (1-4 years of age), also examining how this association differs by maternal race/skin color, gestational age at birth (term versus preterm), municipality income level, and index of quality of BFP management. METHODS AND
FINDINGS: This is a cross-sectional analysis nested within the 100 Million Brazilian Cohort, a population-based cohort primarily built from Brazil's Unified Registry for Social Programs (Cadastro Único). We analyzed data from 6,309,366 children under 5 years of age whose families enrolled between 2006 and 2015. Through deterministic linkage with the BFP payroll datasets, and similarity linkage with the Brazilian Mortality Information System, 4,858,253 children were identified as beneficiaries (77%) and 1,451,113 (23%) were not. Our analysis consisted of a combination of kernel matching and weighted logistic regressions. After kernel matching, 5,308,989 (84.1%) children were included in the final weighted logistic analysis, with 4,107,920 (77.4%) of those being beneficiaries and 1,201,069 (22.6%) not, with a total of 14,897 linked deaths. Overall, BFP participation was associated with a reduction in child mortality (weighted odds ratio [OR] = 0.83; 95% CI: 0.79 to 0.88; p < 0.001). This association was stronger for preterm children (weighted OR = 0.78; 95% CI: 0.68 to 0.90; p < 0.001), children of Black mothers (weighted OR = 0.74; 95% CI: 0.57 to 0.97; p < 0.001), children living in municipalities in the lowest income quintile (first quintile of municipal income: weighted OR = 0.72; 95% CI: 0.62 to 0.82; p < 0.001), and municipalities with better index of BFP management (5th quintile of the Decentralized Management Index: weighted OR = 0.76; 95% CI: 0.66 to 0.88; p < 0.001). The main limitation of our methodology is that our propensity score approach does not account for possible unmeasured confounders. Furthermore, sensitivity analysis showed that loss of nameless death records before linkage may have resulted in overestimation of the associations between BFP participation and mortality, with loss of statistical significance in municipalities with greater losses of data and change in the direction of the association in municipalities with no losses.
CONCLUSIONS: In this study, we observed a significant association between BFP participation and child mortality in children aged 1-4 years and found that this association was stronger for children living in municipalities in the lowest quintile of wealth, in municipalities with better index of program management, and also in preterm children and children of Black mothers. These findings reinforce the evidence that programs like BFP, already proven effective in poverty reduction, have a great potential to improve child health and survival. Subgroup analysis revealed heterogeneous results, useful for policy improvement and better targeting of BFP.

Introduction

Worldwide, remarkable progress in reducing child mortality over the past 3 decades has been observed. From 1990 to 2018, the global under-5 mortality rate declined 58%, from 93 to 39 deaths per 1,000 live births. Still, the burden of child deaths remains significant: In 2019 alone, 5.2 million children died before their fifth birthday [1]. With an impressive 67% reduction in under-5 mortality from 1990 to 2015, Brazil met Millennium Development Goal 4 ahead of schedule [2], as rates dropped from 52 to 14 per 1,000 live births during this period, with an average annual decrease rate of 4.41%. However, if disaggregated from national levels, significant disparities can be found at both the beginning and end time points of this period, reflecting the still persistent inequalities in the country. In 1990, under-5 mortality varied from 38 to 114 per 1,000 live births in states in the North Region and Northeast Region, the poorest regions in the country, but ranged from 23 to 41 per 1,000 live births in the rest of Brazil. In 2015, these disparities lessened in magnitude but were still present, with rates ranging from 14 to 23 in states in the North Region and Northeast Region and 13 to 15 in the rest of the country [3].

Given that a considerable proportion of child deaths are related to poverty and result from health issues that can be treated and prevented easily and economically [1,4], income redistribution initiatives have been a successful strategy for the improvement of child survival, especially in low- and middle-income countries [5-9]. Among such strategies are conditional cash transfer (CCT) programs, created with the purpose of breaking the intergenerational cycle of poverty by transferring cash to low-income families as long as they comply with the conditions of investing in their children’s health and education [10]. Almost all countries in Latin America have CCT programs, and they are also present in countries including Bangladesh, Indonesia, Nigeria, Malawi, and Turkey [11,12], sparking interest even in high-income countries, given their impact on social development [13]. In Brazil, the CCT program, branded as the Bolsa Família program (BFP), was initiated in 2003 and rapidly implemented throughout the country, covering over 13 million families in 2015 and becoming the world’s largest CCT.

Because of its extensive coverage and effectiveness in alleviating poverty, BFP has been pointed out as one of the driving forces behind the successful story of childhood mortality reduction in Brazil up to the year 2015 [14-16]. BFP’s conditionality involves mainly health and education activities, as children must complete vaccine schedules and attend classes for at least 85% of the school year, and pregnant women must complete prenatal visits [17]. This is important for our understanding of the mechanisms through which the program can affect child mortality. As proposed by Rasella et al.’s model, BFP, like other CCTs, can affect child health through 2 pathways, one based on income improvement and the other on health-related requirements. Income improvement can objectively increase access to food and other living necessities, and health-related requirements can improve access to crucial services such as immunization, growth monitoring, and emergency care [15].

To date, there are 4 studies on BFP and child mortality [14-16,18], all pointing to a positive impact of the program on child survival. However, all of them are ecological and derived from aggregate municipal-level data. Here we tested the hypothesis that receiving a BFP stipend is associated with lower risk of child mortality, and we believe this study can contribute to the literature by using large-scale individual-level data from families enrolled in Brazil’s Unified Registry for Social Programs (Cadastro Único [CadÚnico]), the BFP payroll database, and the Brazilian Mortality Information System (SIM), all linked as part of the 100 Million Brazilian Cohort [19,20].

Our main goals are (a) to analyze the association between receiving a BFP stipend and the risk of childhood mortality (1–4 years of age) and (b) to use subgroup analysis to explore whether the observed association varies according to indicators of poverty and quality of BFP management at the municipality level and according to maternal and perinatal characteristics, in particular, maternal race/skin color and gestational age at birth.

The rationale for pursuing the subgroup analysis by municipal-level poverty is based on previous research pointing to an inverse relationship between program coverage and municipal level of social and economic development [21,22]. Therefore, we expect the association between BFP participation and child mortality to be stronger in the poorest municipalities, since those have better targeting and coverage of the program. The basis for exploring stratification by municipal-level indicators of quality of BFP management is that there is evidence of considerable heterogeneity in these indicators among Brazilian municipalities [23-26], and that municipalities with better management and coverage of the program show stronger indicators of poverty reduction [27] and monitoring of program health conditionality [28]. We expect the association between BFP participation and child mortality to be stronger in municipalities with indicators of better CCT management, considering that there is greater monitoring of child health, immunization, and development in these contexts.

Methods

Data description

This is a cross-sectional analysis nested within the 100 Million Brazilian Cohort, a population-based cohort primarily built from CadÚnico, a shared registry for more than 20 social programs, which covers the poorest half of the Brazilian population (families with monthly income equal to or below 3 times minimum wage, approximately US$578). So far, the cohort includes data of approximately 114 million people (nearly half of the country’s population) from 2001 to 2015, being continuously linked to other public health databases to generate data for population-health-relevant questions and epidemiological studies [19,20]. We restricted our study window to the period 2006–2015 based on data linkage performance quality (S2 Text).

For the purposes of this study, the 100 Million Brazilian Cohort baseline dataset (consisting of data from individuals at their first registration in CadÚnico) was linked to 3 different databases: (a) the BFP payroll dataset, in order to identify the CCT’s beneficiaries; (b) the Brazilian SIM, in order to identify the death records of children aged 1–4 in the 100 Million Brazilian Cohort baseline dataset; and (c) the Brazilian Live Birth Information System (SINASC), to assess relevant perinatal data such as birth weight and gestational age at birth.

Data linkage

All the linkage steps were done at the individual level. The linkage process between the 100 Million Brazilian Cohort baseline dataset and BFP payroll dataset was deterministic, based on the NIS (Social Identification Number or “Número de Identificação Socia”) number—a unique identifier similar to a social security number. The linkage between the 100 Million Brazilian Cohort baseline dataset, SIM, and SINASC was performed by similarity matching using CIDACS-RL, an open-source linkage algorithm from the Center for Data and Knowledge Integration for Health (CIDACS) that generates a similarity score on the basis of several identifiers [29]; the linked records were verified through manual analysis of a sample of 2,000 randomly selected pairs from all possible paired records (Table A in S2 Text).

The process to identify deaths of under-5 children linked to the 100 Million Brazilian Cohort was done in 3 stages, and a detailed report of the linkage methodology, including sensitivity and specificity indexes, can be found in S2 Text. Authors had access to a pseudonymized version of the linked database to create the study population, without the personally identifiable information fields.

Statistical analyses

Population definition

We included all children under 5 years of age whose families enrolled in CadÚnico between 1 January 2006 and 31 December 2015. To allow for the analysis of mortality between the ages 1 and 4 years, we included those who survived beyond the first year of life and up until age 5 until 31 December 2015, or died between the ages 1 and 4 years and 11 months during the same period of time. Other exclusions were related to inconsistencies between dates (e.g., date of registry in CadÚnico later than date of death) and unmeasured outcomes (Fig 1).

Fig 1

Flowchart of selection of study population.

Flow diagram of selection and exclusion criteria for the population eligible for this study after linkage of the 100 Million Brazilian Cohort baseline dataset with Bolsa Família and mortality data.

Exposure definition

The beneficiary group (exposed) was defined as children whose family received a BFP stipend, uninterruptedly, from the first to the fifth year of the child’s life. The non-beneficiary group (unexposed) was composed of children in families that did not receive a BFP stipend prior to the child reaching 5 years of age or death, receiving a stipend only after that time point or never receiving a stipend.

If BFP stipends were randomly assigned to families, one could assess the effect of the CCT on childhood mortality by simply comparing the difference in mortality between beneficiaries and non-beneficiaries. In fact, BFP is not randomly assigned. Instead, according to the program’s eligibility criteria [17], it is a process of families’ “self-selection” since whether a family receives the stipend or not is determined by per capita income and a set of family socioeconomic characteristics. Regarding per capita income, concerns about its reliability have been raised, given it is a self-reported measure [30]. To account for the issue of self-selection into the exposure group, we followed a kernel matching approach for the choice of a set of BFP non-beneficiary observations inside the CIDACS 100 Million Brazilian Cohort, allowing us to balance the 2 groups on their observable characteristics and control for possible bias arising from socioeconomic factors.

The analysis plan originally outlined by the study protocol (S1 Text) anticipated the use of propensity score (PS) methods and regression discontinuity. Due to limitations of the income data in CadÚnico, we were not able to implement the latter.

Analytical approach

Our analysis consists of a combination of kernel matching and weighted logistic regressions. In the first part of our analysis, we used a logit model to estimate the probability of receiving a BFP stipend (PS) based on baseline characteristics of the child’s mother and living conditions.

We include baseline characteristics of the child’s mother and living conditions as regressors in the PS model using the following variables: household density (≤2 versus >2 people per room), maternal education (3 years of education or less, 4 to 7 years of education, or 8 years of education or more), self-reported maternal race/skin color (Black, white, Asian, mixed/brown [pardo], or indigenous), maternal marital status (lives with a partner [married or in a relationship] or no partner [single, widowed, or divorced]), maternal parity (0, 1, 2, 3, or 4 or more children), maternal age (≤19, 20–34, or ≥35 years), region (North, Northeast, Central-West, Southeast, or South), and year of first registration in CadÚnico (ranging from 2006 to 2015). We chose not to include income as a PS predictor. Although BFP eligibility is, by regulation, defined by per capita income, it is a self-reported variable highly subject to manipulation [30]. Variables with more than 10% missing values ​​were not included in the analysis.

After we estimated the PSs, we performed a kernel matching procedure, as proposed by Heckman et al. [31]. Kernel-based matching estimation is a non-parametric approach that uses weighted averages of all individuals in the control group to construct the counterfactual outcome, where more weight is given to units close to the one that needs to be matched. Intuitively, the procedure selects individuals who were not BFP beneficiaries but “look like” the set of beneficiary observations, in terms of their PSs, and gives them a greater weighting. Weights in kernel-based matching depend on the distance between each individual from the control group and the participant observation for which the counterfactual is estimated [32]. When applying this approach, one has to choose a kernel function and the bandwidth parameter, which determines how narrow a band of values around the participants’ PSs receive high weights. In this work, we use an Epanechnikov kernel function with a bandwidth h, chosen through the pair-matching algorithm [33]. This yields a set of matching weights for the control group, which allows us to obtain an appropriate set of counterfactual observations. We impose a common support condition that drops beneficiary observations whose PS is higher than the maximum or less than the minimum of the non-beneficiaries.

To reduce bias, standard error estimates were obtained by bootstrap methodology, based on estimating from resampling with replacement from the original sample; this methodology has been widely applied to calculate standard error estimates in this setting [34-36].

In the final logistic models, weighted by the kernel weights from the previous stage, we adjust for relevant perinatal conditions (number of prenatal visits, birth weight, gestational age at birth, and type of delivery). For this method to provide unbiased estimates, we need to believe that—conditional on the sample restriction, common support condition, and kernel weights—the regressors are not correlated with the error term. If this assumption is satisfied, then our estimate represents an unbiased estimate of the “average treatment effect on the treated” (ATT). Besides this, for final logistic models, we also considered cluster effects on the estimation of standard errors, because the observations are clustered within a household, and adjusting for this cluster effect returns more robust estimates.

Subgroup analysis

The subgroup analysis was established by our research protocol (see S1 Text, objective 3). We aimed to explore BFP’s association with child mortality across subgroups. To analyze the association between receiving a BFP stipend and child mortality according to individual characteristics, we looked at subgroups according to maternal race/skin color and gestational age at birth. To conduct this analysis, all kernel-weighted logistic models were calculated separately within each subgroup of gestational age at birth (preterm [<37 weeks] versus term [≥37 weeks]) and maternal race/skin color (Black, white, indigenous, mixed/brown). This implies that our subgroup analysis should be interpreted as “the estimates of x on y for group 1 and the estimates of x on y for group 2,” as is appropriate with separate regressions.

We also analyzed heterogeneities in BFP’s association with child mortality by presenting estimates of this relationship for the municipalities in the highest and lowest income quintiles and also those with different indexes of quality in BFP’s management. To achieve this goal, we ranked municipalities into quintiles of per capita income (Municipal Human Development Index–Renda [MHDI-R]) and CadÚnico’s Decentralized Management Index (DMI) and separately conducted all the analysis steps within each quintile of these indicators, including the estimation of PSs, kernel matching, and final weighted logistic models to provide separate estimates.

The DMI is an indicator of the quality of the management of BFP at the municipality level. It varies from 0 (worst) to 1 (best), and it is calculated based mainly on the timeliness of CadÚnico updates by the municipalities, their success in capturing families in extreme poverty, and their performance in monitoring the upholding of health and education conditions by the program’s beneficiaries [37]. Given that in municipalities with better DMI beneficiaries are probably more compliant with the conditions and families in extreme poverty are more likely to be included in the program, we expect BFP to be associated with a greater reduction of child mortality in these contexts. Our choice to quintilize these municipal indicators represents an exploratory approach. Although there is theoretical support for selecting these variables [21-28], we did not identify any studies in the literature that have worked with these specific variables when analyzing BFP association with child health outcomes.

To formally assess whether the association between receiving a BFP stipend and child mortality varies across subgroups, we performed a statistical test for interaction by including BFP status × subgroup indicator terms in our subgroup models. Analysis of the presence of heterogeneity in BFP association with mortality across subgroups was based on the statistical significance (p < 0.05) of the interaction terms [38] and on applying the likelihood ratio test to evaluate the difference between nested models, considering that the model without the interaction term is nested within the model with the interaction term.

The use of interaction terms results in numerous hypothesis tests, especially in the context of categorical covariates. To account for that, we applied the Bonferroni correction and adjusted p-values for multiple comparisons within factor variable terms.

Robustness checks

We checked the robustness of our results by using inverse probability of treatment weighting (IPTW) as an alternative approach. For this analysis, the association between BFP participation and child mortality was estimated through logistic models with weights equal to PS/(1 − PS) for the non-beneficiaries and weights equal to 1 for the beneficiaries. Literature on the comparison between the properties of kernel matching and IPTW estimates suggests that IPTW surpasses kernel matching in terms of precision and, since it does not require the selection of a bandwidth parameter, is quicker to compute than kernel matching, reducing the time to bootstrap the standard errors [39].

Finally, to assess the impact of linkage bias in our estimates, we break down the analysis into 7 subgroups of municipalities according to their percentage of nameless records not submitted to linkage due to missing information (see S2 Text for more details). All analyses were done using Stata/MP version 15.0. This study is reported as per the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guideline (S1 Checklist).

The study protocol was reviewed and approved by the Institute of Public Health Ethics Committee at the Federal University of Bahia (CAAE registration number: 56003716.0.3001.5030), and patient consent was not required as the study used only de-identified registry-based secondary data.

Results

Of the 12,510,126 under-5 children registered in the 100 Million Brazilian Cohort between 2006 and 2015, 6,309,366 children from 4,627,984 families were initially included in this study (Fig 1). Among this sample, 4,858,253 were BFP beneficiaries (77%) and 1,451,113 (23%) were not. After the kernel matching procedure, 5,308,989 (84.1%) children were included in the final weighted logistic analysis, comprising 14,897 linked deaths. Regarding BFP status, 4,107,920 (77.4%) were beneficiaries and 1,201,069 (22.6%) were not. Table 1 shows the distribution of baseline characteristics between beneficiaries and non-beneficiaries, indicating that the 2 groups became more balanced in such covariates after the kernel procedure (S3 Text).

Table 1

Baseline characteristics of Bolsa Família program (BFP) beneficiaries and non-beneficiaries before and after the kernel matching procedure—100 Million Brazilian Cohort, 2006 to 2015.

Characteristic	Absolute frequency and unweighted (crude) proportion					Kernel-weighted proportion
	BFP beneficiary(N = 4,858,253)		Non-beneficiary(N = 1,451,113)		Difference1	BFP beneficiary(N = 4,107,920)	Non-beneficiary(N = 1,201,069)	Difference1
	N	Percent	N	Percent
Region
North	654,492	13.5	166,399	11.5	2.0	13.1	13.2	−0.1
Northeast	1,970,529	40.6	413,395	28.5	12.1	38.2	37.7	0.6
Southeast	1,538,119	31.7	523,548	36.1	−4.4	33.1	33.7	−0.6
South	404,724	8.3	203,779	14.0	−5.7	9.3	9.2	0.1
Central-West	290,389	6.0	143,992	9.9	−3.9	6.2	6.3	0.0
Missing	0	0.0	0	0.0
Maternal education
≤3 years	2,065,350	42.5	845,553	58.3	−15.8	43.1	42.9	0.2
4–7 years	1,924,239	39.6	446,234	30.7	8.9	40.9	41.2	−0.3
≥8 years	746,711	15.4	126,652	8.8	6.6	16.0	15.9	0.1
Missing	121,953	0.3	32,674	2.2	−1.9
Household density
≤2 people per room	4,497,436	92.6	1,382,043	95.2	−2.6	94.7	94.7	0.0
>2 people per room	247,166	5.1	23,589	1.6	3.5	5.3	5.3	0.0
Missing	113,651	0.2	45,481	3.1	−2.9
Maternal race/skin color
White	1,380,976	28.4	550,947	38.0	−9.6	29.7	29.7	0.0
Black	236,240	4.9	56,641	3.9	1.0	4.7	4.9	−0.2
Asian descent2	16,800	0.4	5,980	0.4	0.0	0.3	0.3	0.0
Mixed/brown	3,179,209	65.4	834,486	57.5	7.9	64.4	64.4	0.1
Indigenous	44,856	0.9	2,948	0.2	0.7	0.8	0.7	0.1
Missing	172	0.0	111	0.0
Maternal marital status
Has a partner	1,617,927	33.3	603,500	41.6	−8.3	34.3	34.0	0.3
No partner (single, divorced, widowed)	3,149,217	64.8	822,818	56.7	8.1	65.7	66.0	−0.3
Missing	91,109	0.2	24,795	1.7	−1.5
Maternal parity
0 children	1,309,968	29.6	623,445	42.9	−13.3	29.6	29.5	0.1
1 child	1,332,313	27.4	412,560	28.4	−1.0	30.1	30.7	−0.6
2 children	836,486	17.2	162,515	11.2	6.0	18.9	19.1	−0.2
3 children	438,084	0.9	56,445	3.9	−3.0	9.9	9.8	0.1
4 children or more	521,636	10.8	58,148	4.0	6.8	11.6	10.9	0.7
Missing	419,766	0.8	138,000	9.5	−8.7
Maternal age
≤19 years	1,183,187	24.4	405,803	28.0	−3.6	22.3	23.2	−0.8
20–34 years	3,355,218	69.1	962,022	66.3	2.8	70.8	70.4	0.4
≥35 years	318,198	6.6	82,769	5.7	0.8	6.9	6.5	0.4
Missing	1,650	0.0	519	0.0

1The difference in proportions of each category between BFP beneficiaries and non-beneficiaries (BFP beneficiary proportion minus non-beneficiary proportion).

2The Asian group could not be included in the final models due to small sample size and number of linked deaths (n = 12).

Regarding the association of BFP with child mortality, the weighted logistic regression analysis results are presented in Table 2. In the model adjusted for the number of prenatal visits, birth weight, gestational age at birth, and type of delivery, being in a family that received a BFP stipend was associated with lower mortality (weighted odds ratio [OR] = 0.83; 95% CI: 0.79 to 0.88; p < 0.001). This finding was consistent in the subgroup analyses for quintiles of municipal income, quintiles of DMI, maternal race/skin color, and gestational age at birth, with noticeable differences across levels of these indicators (Tables 3–6).

Table 2

Regression results: Coefficients of unadjusted and adjusted kernel-weighted logistic regressions of Bolsa Família Program (BFP) participation on mortality between ages 1 and 4 years.

Coefficient	Unadjusted model			Adjusted model1
	Weighted odds ratio (95% CI)	Robust standard error	p-Value	Weighted odds ratio (95% CI)	Robust standard error	p-Value
Beneficiary status (BFP participation = 1)	0.84 (0.79 to 0.88)	0.0232	<0.001	0.83 (0.79 to 0.88)	0.0231	<0.001
Constant	0.0029	0.0001		0.0024	0.0001

Sample size after kernel matching = 5,308,989.

1Model adjusted for number of prenatal visits, birth weight, gestational age at birth, and type of delivery.

Table 3

Regression results: Coefficients of adjusted kernel-weighted logistic regressions within subgroups of municipal quintiles of per capita income (Municipal Human Development Index–Renda [MHDI-R]).

MHDI-R	Weighted odds ratio1 (95% CI)	Robust standard error	p-Value	N
Model 3a				713,577
1st quintile (lowest income)	0.72 (0.62 to 0.82)	0.050	<0.001
Constant	0.004 (0.004 to 0.005)	0.0004	<0.001
Model 3b				802,524
2nd quintile	0.75 (0.66 to 0.85)	0.047	<0.001
Constant	0.003 (0.003 to 0.004)	0.0003	<0.001
Model 3c				722,243
3rd quintile	0.84 (0.73 to 0.97)	0.062	0.020
Constant	0.002 (0.002 to 0.003)	0.0002	<0.001
Model 3d				856,961
4th quintile	0.87 (0.76 to 0.98)	0.057	0.027
Constant	0.002 (0.002 to 0.003)	0.0002	<0.001
Model 3e				2,210,567
5th quintile (highest income)	0.92 (0.84 to 1.01)	0.043	0.086
Constant	0.002 (0.002 to 0.003)	0.0002	<0.001

All the analytical steps (propensity score estimation, kernel matching, and weighted logistic regression) were conducted separately within each level of MHDI-R. All models (3a to 3e) were done separately, within each of the MHDI-R quintiles, and adjusted for prenatal visits, birth weight, gestational age at birth, and type of delivery. Unadjusted estimates are available in S3 Text.

1Beneficiary status (Bolsa Família participation = 1).

Table 6

Regression results: Coefficients of adjusted kernel-weighted logistic regressions within subgroups of gestational age at birth.

Gestational age at birth	Weighted odds ratio1 (95% CI)	Robust standard error	p-Value	N
Model 6a				4,960,905
≥37 weeks	0.84 (0.79 to 0.89)	0.025	<0.001
Constant	0.002 (0.002 to 0.003)	0.001	<0.001
Model 6b				345,266
<37 weeks	0.78 (0.68 to 0.90)	0.057	<0.001
Constant	0.004 (0.003 to 0.005)	0.004	<0.001

All the analytical steps (propensity score estimation, kernel matching, and weighted logistic regression) were conducted separately within each level of gestational age at birth. All models (6a and 6b) were done separately, within each of these levels, and adjusted for prenatal visits, birth weight, and type of delivery. Unadjusted estimates are available in S3 Text.

1Beneficiary status (Bolsa Família participation = 1).

Across quintiles of municipal income, the association between BFP participation and mortality for children aged 1–4 years ranged from a 28% reduction in the odds of mortality in the poorest municipalities (weighted OR = 0.72; 95% CI: 0.62 to 0.82; p < 0.001) to a 13% reduction in the fourth quintile (weighted OR = 0.87; 95% CI: 0.76 to 0.98; p < 0.001) and no association in the fifth quintile, indicating a stronger association between BFP and lower risk of mortality for children living in the poorest 20% of municipalities of Brazil (Table 3).

The association between BFP participation and child mortality was also stronger in municipalities in the highest DMI quintile, ranging from a 12% reduction in the odds of mortality in the lowest quintile (weighted OR = 0.88; 95% CI: 0.81 to 0.96; p < 0.001) to a 24% reduction in the highest quintile (weighted OR = 0.76; 95% CI: 0.66 to 0.88; p < 0.001), suggesting a stronger association of BFP with lower risk of child mortality in municipalities in which the program is best administered (Table 4).

Table 4

Regression results: Coefficients of adjusted kernel-weighted logistic regressions within subgroups of Cadastro Único’s Decentralized Management Index (DMI).

DMI	Weighted odds ratio1 (95% CI)	Robust standard error	p-Value	N
Model 4a				2,309,348
1st quintile (worst)	0.88 (0.81 to 0.96)	0.040	0.005
Constant	0.002 (0.001 to 0.002)	0.0001	<0.001
Model 4b				1,008,954
2nd quintile	0.88 (0.78 to 1.00)	0.055	0.047
Constant	0.002 (0.002 to 0.003)	0.0002	<0.001
Model 4c				689,239
3rd quintile	0.83 (0.72 to 0.96)	0.061	0.010
Constant	0.003 (0.002 to 0.003)	0.0002	<0.001
Model 4d				675,491
4th quintile	0.79 (0.69 to 0.91)	0.056	0.001
Constant	0.002 (0.001 to 0.002)	0.0001	<0.001
Model 4e				622,812
5th quintile (best)	0.76 (0.66 to 0.88)	0.055	<0.001
Constant	0.003 (0.002 to 0.003)	0.0003	<0.001

All the analytical steps (propensity score estimation, kernel matching, and weighted logistic regression) were conducted separately within each level of DMI. All models (4a to 4e) were done separately, within each of the DMI quintiles, and adjusted for prenatal visits, birth weight, gestational age at birth, and type of delivery. Unadjusted estimates are available in S3 Text.

1Beneficiary status (Bolsa Família participation = 1).

Considering differences across individual characteristics, an association between BFP participation and child mortality was found for all maternal race/skin color groups except for children of indigenous mothers (weighted OR = 0.99; 95% CI: 0.51 to 1.96; p < 0.001), possibly due to the relatively small sample size of this group. Estimates varied from a small 10% decrease in the odds of mortality for BFP beneficiary children of white mothers to a 19% decrease for BFP beneficiary children of mixed/brown (pardo) mothers (weighted OR = 0.81; 95% CI: 0.75 to 0.86; p < 0.001) and a 26% decrease for beneficiary children of Black mothers (Table 5). Considering gestational age at birth, an association between BFP participation and child mortality was found for both term (weighted OR = 0.84; 95% CI: 0.79 to 0.89; p < 0.001) and preterm (weighted OR = 0.78; 95% CI: 0.68 to 0.90; p < 0.001) children, with a stronger association for the latter (Table 6).

Table 5

Regression results: Coefficients of adjusted kernel-weighted logistic regressions within subgroups of maternal race/skin color.

Maternal race/skin color	Weighted odds ratio1 (95% CI)	Robust standard error	p-Value	N
Model 5a				1,701,111
White	0.90 (0.83 to 0.99)	0.041	0.019
Constant	0.002 (0.002 to 0.002)	0.0001	<0.001
Model 5b				3,311,091
Mixed/brown (pardo)	0.81 (0.75 to 0.86)	0.028	<0.001
Constant	0.003 (0.002 to 0.003)	0.0001	<0.001
Model 5c				239,587
Black	0.74 (0.57 to 0.97)	0.101	0.029
Constant	0.002 (0.002 to 0.003)	0.0004	<0.001
Model 5d				35,690
Indigenous	0.99 (0.51 to 1.96)	0.345	0.993
Constant	0.008 (0.004 to 0.021)	0.004	<0.001

All the analytical steps (propensity score estimation, kernel matching, and weighted logistic regression) were conducted separately within each level of maternal race/skin color. All models (5a to 5d) were done separately, within each of these levels, and adjusted for prenatal visits, birth weight, gestational age at birth, and type of delivery. Unadjusted estimates are available in S3 Text.

1Beneficiary status (Bolsa Família participation = 1).

As a formal test of our subgroup effects, we conducted an interaction analysis (Tables F and G in S3 Text). The results of the likelihood ratio test indicated our interaction terms to be statistically significant for BFP status × MHDI-R quintile (p < 0.001), BFP status × DMI quintile (p = 0.001), BFP status × maternal race/skin color (p = 0.002), and BFP status × gestational age at birth (p < 0.001).

As illustrated in Fig 2, the predictive margins for probability of child mortality indicate noticeable heterogeneity across subgroups (all other variables held constant). The likelihood ratio test comparing the models with versus without the interaction terms was also significant in all 4 subgroup models (Tables F and G in S3 Text).

Fig 2

Predictive margins for probability of child mortality, with 95% confidence intervals, by subgroup.

Beneficiary status: 0 = non-beneficiary; 1 = Bolsa Família program beneficiary. (a) Predictive margins (95% CI) by quintile of municipal per capita income (Municipal Human Development Index–Renda [MHDI-R]). Model adjusted for number of prenatal visits, birth weight, gestational age at birth, and type of delivery. (b) Predictive margins (95% CI) by quintile of Cadastro Único Decentralized Management Index (DMI). Model adjusted for number of prenatal visits, birth weight, gestational age at birth, and type of delivery. (c) Predictive margins (95% CI) by maternal race/skin color. Model adjusted for number of prenatal visits, birth weight, gestational age at birth, and type of delivery. (d) Predictive margins (95% CI) by gestational age at birth. Model adjusted for number of prenatal visits, birth weight, and type of delivery.

The robustness analysis with IPTW yielded similar results to the kernel-weighted analysis, without changes in the overall association between BFP participation and child mortality (Table J in S3 Text) or within subgroups (Tables K to N in S3 Text).

Although the linkage process yielded good overall indexes of sensitivity and specificity (Table A in S2 Text), the percentage of nameless death records not submitted to linkage at the municipal level varied considerably across Brazil. Supposing this loss was concentrated among BFP beneficiaries, this could lead to overestimation of our calculated association, and if concentrated among non-beneficiaries, this could lead to an underestimated association, as a parcel of what we considered non-cases in the non-beneficiary group could, in fact, be among the death records initially left out of linkage. To account for this possible bias and as a form of sensitivity analysis, we calculated the percentage of nameless death records submitted to linkage in each municipality, and categorized municipalities into 8 subgroups based on this percentage: 0% (all nameless death records excluded from linkage), 0.1% to 20.0%, 20.1% to 40.0%, 40.1% to 60.0%, 60.1% to 80.0%, 80.1% to 90.0%, 90.1% to 99.9%, and 100.0% (all nameless death records submitted to linkage) (Table C in S2 Text).

Analyzing the association between BFP participation and mortality in children aged 1–4 years in each of these subgroups, we found that in the group of municipalities without any loss of nameless records from linkage, the measured association changed direction, although the lower bound of the confidence interval was close to the null value. For the subgroups of municipalities with 40.0% or less of their nameless records submitted, and those with 90.1%–99.9% submitted, the null value is well within their confidence intervals, indicating no significant association between BFP participation and mortality in these subgroups (Table C in S2 Text).

Discussion

We found that receiving a BFP stipend is significantly associated with a reduced probability of death between the ages 1 and 4 years. This association was stronger for preterm children, children of Black mothers, children living in poorer municipalities, and children living in municipalities with better indexes of BFP management. These findings are consistent with previous studies reporting significant effects of BFP [14-16,18] and other CCTs on childhood mortality [7,40,41]. However, while these previous studies used aggregate municipal data, the present one used individual data from the 100 Million Brazilian Cohort [19,20].

Issues regarding inference based on ecological versus individual-level studies have been discussed for decades in various fields of research. We believe that interpreting individual-level findings is an advantage and a contribution of the present study to the knowledge on BFP, since we were able to deal with potential confounders through adjusting for individual variables, and using weighting techniques to address selection bias and covariates that are known predictors of child mortality.

Because of our large sample, it was also possible to explore heterogeneity in the association between BFP participation and mortality across multiple subgroups, findings that, when taken as a whole, are aligned with a recent review on CCTs and child health in low- and middle-income countries showing that such programs show substantial heterogeneity across subgroups defined not only geographically but also by indicators of socioeconomic status and program implementation intensity [42].

Considering the subgroups of maternal race/skin color and gestational age at birth—2 important upstream determinants of child health—our results went in a similar direction. The strongest association of the cash transfer with reduced child mortality was found among Black individuals, a group that is historically underprivileged socially and economically and in access to health in Brazil [43,44]. Studies have suggested that women exposed to a CCT program engage in higher maternal and child health service utilization and show substantial schooling accumulation. Both healthcare utilization and maternal education are significant predictors of child mortality, especially among preterm babies, which could explain the stronger association between the CCT program and child mortality among preterm babies. As proposed by Cooper et al.’s recent review of the heterogeneous effects of CCTs, this suggests that cash transfers might work to augment or attenuate upstream determinants of child health [42].

In addition, we found evidence of heterogeneous results regarding BFP and mortality between the ages 1–4 years across broader contextual indicators, such as quintiles of municipal per capita income and DMI, an index that measures the quality of BFP and CadÚnico administration. In accordance with our hypothesis, the association between BFP participation and child mortality was stronger for beneficiaries living in poorer municipalities, meaning these cities’ beneficiary children had lower odds of mortality than their non-beneficiary counterparts.

The relationship between socioeconomic status and health is one of the most robust and well-documented findings in social and health sciences. However, findings regarding the differential impacts of cash transfers across contexts of different poverty levels are not conclusive, with effect gradients going in opposite directions in studies of CCTs and child health conducted in Mexico, Ecuador, Niger, and India [45-49].

Regarding heterogeneous results across DMI levels, our results align with previous studies finding CCT effects to be dependent on implementation quality and management indicators [50-52]. As the 100 Million Brazilian Cohort comprises the poorest half of the country, this finding is of great importance, showing that not only the presence but also the quality of poverty-alleviating policies matters when targeting health promotion among vulnerable populations. This finding is aligned with our stated hypothesis that BFP’s association with child mortality would be stronger in contexts where the program was best administered.

Strengths and weaknesses of the study

Our study has notable strengths. We used individual data from a population-level database of more than 6 million children, making this the largest study on CCTs and child mortality to date, to the best of our knowledge. We applied a robust analytical approach, using kernel-based PS weighting to account for measured socioeconomic confounders and verified our results through an alternative approach, using an inverse probability weighted analysis. The beneficiary and non-beneficiary groups were well balanced for distributions of covariates, and weighted and stratified analyses confirmed our main findings to be consistent across several subgroups. We have also accounted for heterogeneities not previously assessed regarding the association of BFP participation with child health outcomes, providing specific results by maternal race/skin color group, gestational age at birth, and municipal level of poverty and program administration quality.

Through the kernel-matching procedure, we limited the potential for measured confounders to influence the result. Nevertheless, important unmeasured factors need to be considered, especially family income, a variable that could not be included in this study given its possibility for manipulation because it is a self-reported measure that can influence eligibility. Therefore, our methodology’s main limitation is that our PS approach does not account for this and other possible unmeasured confounders. This could be dealt with by an instrumental variable approach, but we were unable to identify a suitable instrument in the data that is currently available within the 100 Million Brazilian Cohort. Future studies should revisit our hypothesis through a more rigorous quasi-experimental approach in order to support causal inference.

Since our evidence is associational, we recognize that causal claims should not be made solely based on our findings. Approaches such as the ones applied by Okeke and Abubakar to a Nigerian CCT [12] and Barham to Mexico’s Progresa CCT [9] through randomized experiments are closer to providing all the elements expected for causal inference and impact evaluation. In the context of BFP, however, such experimental designs are not viable due to the program’s characteristics and data availability. Nonetheless, we believe our contribution is still valuable to the broader literature on CCTs and child health by providing evidence based on a large population and focusing on several subgroups.

Limitations also arise from the linkage process, as loss of nameless records before linkage may have resulted in a possible overestimation of the association between program participation and child mortality, since we observed that in the municipalities without such losses, the direction of the measured association changed, and the lower bound of the confidence interval was very close to the null value of 1.00. Furthermore, in the very few municipalities (n = 22) where losses were greater than 60%, the association between BFP participation and mortality lost statistical significance.

Considering, however, the distribution of our study population across these municipal strata of nameless records lost before linkage, over 70% of our observations came from municipalities where the measured association was consistent with the one reported for the whole sample, with kernel-weighted odds ratios varying between 0.77 and 0.85 (Table C in S2 Text). It is also important to consider that nameless death records correspond to only 3.99% of the total death records for children aged 1–4 years in Brazil in our analysis period (5,218 nameless records out of a total of 130,858 deaths; Table B in S2 Text), and that less than 1% (22 municipalities) of the Brazilian municipalities had losses of linkage records greater than 40%.

Lastly, we were not able to isolate the effect of other interventions that could also be targeting poor families with children inside CadÚnico, such as the Child Labor Eradication Program (PETI) and the housing program (Minha Casa Minha Vida).

Conclusions

CCTs like BFP have a great potential to improve child health in vulnerable populations. Using data from the 100 Million Brazilian Cohort, we conclude that BFP participation had a significant association with lower risk of child mortality, especially in municipalities with low per capita income. Moreover, as the magnitude of the association was greater in municipalities with better indicators of BFP targeting and management, we can also conclude that when properly administered, these programs can reinforce their relevance in improving child health outcomes.

Our findings were mostly consistent with other studies that have reported an important positive impact of BFP and other similar CCTs. Furthermore, this was, to our knowledge, the first evaluation of the association between BFP participation and child mortality based on individual-level data, and the first to account for heterogeneity, providing evidence necessary for better targeting and for generating more precise knowledge in this regard.

RECORD Checklist.

REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guideline.

(DOCX)

Click here for additional data file.

Linkage process between the CIDACS 100 Million Brazilian Cohort, mortality, and live birth records.

Flowchart of the linkage process between the CIDACS 100 Million Brazilian Cohort baseline dataset, mortality data (Brazilian Mortality Information System [SIM]), and live birth records (Brazilian Live Birth Information System [SINASC]).

(TIF)

Click here for additional data file.

Receiver operating characteristic (ROC) curve for the linkage between deaths of children aged 1–4 years (with name on the death certificate) and the CIDACS 100 Million Brazilian Cohort baseline dataset.

(TIF)

Click here for additional data file.

Receiver operating characteristic (ROC) curve for the linkage between deaths of children under 5 years (without name on the death certificate—using the name of the mother) and the CIDACS 100 Million Brazilian Cohort baseline dataset.

(TIF)

Click here for additional data file.

Propensity scores common support area.

Distribution of propensity scores across beneficiaries and non-beneficiaries.

(TIF)

Click here for additional data file.

Research protocol.

(PDF)

Click here for additional data file.

Data linkage information.

(DOCX)

Click here for additional data file.

Kernel-matching procedure and robustness check.

Information on propensity score estimation, kernel matching, subgroup analysis, and robustness check.

(DOCX)

Click here for additional data file.

9 Dec 2020

Dear Dr Ramos,

Thank you for submitting your manuscript entitled "Conditional cash transfer program and child mortality: a study of the 100 Million Brazilian Cohort" for consideration by PLOS Medicine.

Your manuscript has now been evaluated by the PLOS Medicine editorial staff, as well as by the special issue guest editors, and I am writing to let you know that we would like to send your submission out for external peer review.

However, before we can send your manuscript to reviewers, we need you to complete your submission by providing the metadata that is required for full assessment. To this end, please login to Editorial Manager where you will find the paper in the 'Submissions Needing Revisions' folder on your homepage. Please click 'Revise Submission' from the Action Links and complete all additional questions in the submission questionnaire.

Please re-submit your manuscript within two working days, i.e. by .

Login to Editorial Manager here: https://www.editorialmanager.com/pmedicine

Once your full submission is complete, your paper will undergo a series of checks in preparation for peer review. Once your manuscript has passed all checks it will be sent out for review.

Feel free to email us at plosmedicine@plos.org if you have any queries relating to your submission.

Kind regards,

Artur A. Arikainen,

Associate Editor

PLOS Medicine

22 Feb 2021

Dear Dr. Ramos,

Thank you very much for submitting your manuscript "Conditional cash transfer program and child mortality: a study of the 100 Million Brazilian Cohort" (PMEDICINE-D-20-05781R1) for consideration in PLOS Medicine’s Special Issue on Global Child Health.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with the Special Issue Guest Editors, and sent to independent reviewers, including a statistical reviewer. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Mar 15 2021 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

1.Title: Please revise your title according to PLOS Medicine's style. Your title must be nondeclarative and not a question. It should begin with main concept if possible. "Effect of" should be used only if causality can be inferred, i.e., for an RCT. Please place the study design ("A randomized controlled trial," "A retrospective study," "A modelling study," etc.) in the subtitle (ie, after a colon).

2.Data availability statement: The Data Availability Statement (DAS) requires revision. Please include the relevant information for requesting access to the data from CIDACS (weblink and/or email contact). The contact cannot be one of the authors of the study. It is not clear if the relevant data are available by accessing the provided link.

3.Throughout: Please include line numbers with the revised version of the manuscript.

4.Throughout: Please use square brackets for in-text citations, like this [1].

5.Throughout: Please use “White” rather than “Caucasian”

6.Abstract: Please structure your abstract using the PLOS Medicine headings (Background, Methods and Findings, Conclusions).

7.Abstract Background: Provide the context of why the study is important. The final sentence should clearly state the study question.

8.Abstract: Methods and Findings: Please be sure the abstract describes the population and setting, number of participants, years during which the study took place and main outcome measures.

9. Abstract: Methods and Findings: Please quantify the main results (with 95% CIs and p values).

10. Abstract: Methods and Findings: In the last sentence of the Abstract Methods and Findings section, please describe the main limitation(s) of the study's methodology.

11. Abstract: Conclusions: Please address the study implications without overreaching what can be concluded from the data; the phrase "In this study, we observed ..." may be useful.

12. Author Summary: At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary

13. Introduction: Please temper claims of primacy: e.g in second last paragraph of introduction: “To the best of our knowledge, this is the first study using nationwide individual level data to analyze the association of BFP with child mortality.”

14. Methods: Thank you for including the RECORD checklist. Please add the following statement, or similar, to the Methods: "This study is reported as per the REporting of studies Conducted using Observational Routinely-collected Data (RECORD) guideline (S1 Checklist)."

15. Methods: Did your study have a prospective protocol or analysis plan? Please state this (either way) early in the Methods section.

a) If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant prospectively written document with your revised manuscript as a Supporting Information file to be published alongside your study, and cite it in the Methods section. A legend for this file should be included at the end of your manuscript.

b) If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and when/why any data-driven changes to analyses took place.

c) In either case, changes in the analysis-- including those made in response to peer review comments-- should be identified as such in the Methods section of the paper, with rationale.

16. Methods: Please move the information pertaining to the study ethical approval and participant consent to the Methods section.

17. Throughout: Please clarify the language used- perhaps “highest income” or “greatest wealth” rather than “richest”

18. Results: For all findings presented in the text, please provide both the 95% CIs and p values along with the odds ratios. For the adjusted ORs, please also provide results from unadjusted analyses, this can be in a supporting information table.

19. Results: Please avoid the use of causal language (e.g. “effect”) by revising to the following: “Across quintiles of municipal income, associations with BFP ranged from a 28% reduction in the odds of mortality in the poorest municipalities (weighted OR= 0.72; CI 95% [0.62-0.82]) to 13% (weighted OR= 0.87; CI 95% [0.76-0.98]) in the fourth quintile, with no statistically significant association identified with the fifth quintile, indicating a stronger association between BFP and lower risk of mortality for children living in the poorest 20% municipalities of Brazil compared to those in the highest quintile (richest).” Please clarify whether comparisons were done between quintiles, to support the second half of the statement.

20. Results: Please clarify whether tests were done comparing DMI quintiles: “The association between BFP and mortality between ages 1-4 was also stronger in municipalities at the highest DMI quintile, ranging from a 12% reduction in the odds of mortality in the lowest quintile (weighted OR=0.88; CI 95% [0.81-0.96]) to 24% in the highest (weighted OR=0.76; CI 95% [0.66-0.88]), suggesting a stronger association of BFP with lower risk of child mortality in municipalities in which the program is best administered.”

21. Results: Please revise to avoid causal language: “Considering differences in association across individual characteristics, associations between BFP and child mortality were found in all race groups…” or similar.

22. Results: Please provide the analyses supporting this statement, or clarify if statistical significance is meant here: “The percentage of nameless death records not submitted to linkage at the municipal

level were significantly different across Brazil.”

23. Results: Please clarify this sentence, revising to avoid causal language: “The null value for the 95%

and 90% subgroups of municipalities are well within their confidence intervals, indicating no significant effects of the program in these subgroups (see Table 1c - supplementary material 1).”

24. Discussion: Please avoid the use of causal language throughout, e.g.”effect on” in the second paragraph: “Because of our large sample, it was also possible to explore heterogeneity in effect of BFP on 1-4 years old mortality across multiple subgroups…”

25. Conclusions: Please revise to avoid the use of causal language “Using data from the 100 Million Brazilian Cohort, we conclude that the programme had a significant effect on reducing child mortality, especially in municipalities with low per capita income. Moreover, as the magnitude of the effect grew in municipalities

with better indicators of BFP targeting and management, we can also conclude that when properly administered these programmes can increase their already significant impact on child health outcomes.”

26. References: Please use the "Vancouver" style for reference formatting, and see our website for other reference guidelines https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references

27. Checklist: When completing the RECORD/STROBE items on the checklist, please use section and paragraph numbers, rather than page numbers.

28. Figures/Tables: Please provide titles and legends for all figures and tables.

29. Figure 1: Please provide a descriptive legend, and please define the abbreviation for PBF.

30. Figure 2: Please indicate in the legend if these are adjusted or unadjusted OR- and if adjusted please list the variables included in the adjustment. Please provide the unadjusted analyses.

31. Table 1: Please define abbreviations in the legend (BFP) and please describe the BFP1-BFP0 in the legend.

32. Table 2 and Table 3: Please provide the p values associated with these OR

33. Table 3: Please indicate in the legend if these are adjusted OR, and indicate the variables included in the adjustment in the legend.

34. Supporting information files: Please include a descriptive title and legend for each figure/table in the Supporting Information files.

35. Supporting information Table 1C: Please revise the title to avoid causal implications: “Table 1c - Estimates of BFP impact on child mortality…” Please also provide p values, and note factors included for adjustment.

36. Supporting information Table 2A: Please also provide p values for these associations.

Comments from the reviewers:

Reviewer #1: It is a relevant manuscript to the public health area, since it addresses the issue related to the access to cash transfer program for low-income Brazilian population and infant mortality. Therefore, it is aligned with the scope of the journal Plos Medicine.

The manuscript aimed to investigate the association between receiving Programa Bolsa Familia and the risk of mortality before 5 years old, through a 100 Million Brazilian Cohort whose data were obtained from Cadastro Único (CadÚnico).

The manuscript is well written and structured.

In the Introduction section, the authors described the present state of cientific literature related to the central theme, pointing out how far the studies have come and the gap to be filled to advance knowledge about the effects of income transfer programs in reducing poverty and mortality of children under 05 years. In this sense, the hypothesis was highlighted.

In the Methods section, procedures and analyzes for achieving the objectives were consistently listed. Results and Discussion were equally clear and concise, pointing out the reach and the limits of the study. In the Conclusion section, authors summarized th fundamental findings.

Due to its unprecedented nature, as it is the first national study with individual data investigating the effects of a cash transfer program on infant mortality, I recommend the manuscript to be eligible for publication in Plos Medicine Journal, after minor changes listed below:

Abstract:

"Conclusions: These findings reinforce the evidence that CCTs like BFP..."

- I suggest the authors to write the full name for CCTs, once it is the first time it appears in the text.

Introduction:

- Fourth line: authors might write the full name of MGD 4

Tables and Figures:

- It is necessary to review the titles of Tables and Figures, as well as to include captions for the acronyms. Tables and figures should be self-explanatory, without requiring the readers to use the text for their understanding.

References:

- It would be interesting to update the references, especially those below the year of 2014.

Reviewer #2: This cross sectional study aims to test the hypothesis that poverty-alleviating policies can reduce child mortality, by examining the association between the Brazilian Bolsa Familia conditional cash transfer program and child mortality (1-4 years) as well as exploring heterogeneous effects by causes of death, maternal race, gestational age, municipality income level and indexes of quality of BFP management.

Comments:

The RECORD statement checklist of items, extended from the STROBE statement, that should be reported in observational studies using routinely collected health data is provided within the supplementary documentation.

"In fact, Bolsa Família is not randomly assigned. Instead, it is a process of families' 'self-selection' since whether a family receives the stipend or not is determined by a set of family socioeconomic characteristics. To account for the issue of self-selection bias into the exposure group, we follow a kernel matching approach for the choice of a set of PBF non-beneficiary observations inside the Cidacs' 100 Million Cohort that can allow us to compare the two groups. "

The authors have undertaken a rigorous analytical approach which supports the minimisation of potential biases in the data.

See acronym typo in quote - which happens more than once throughout the manuscript and requires remedying.

Overall, this is a well written paper with a clearly explained, detailed, thorough, and technically appropriate methodology.

The authors have undertaken suitable subgroup analyses, accounted for bias appropriately, and acknowledged the main limitations in the discussion section for accurate and transparent interpretation of the study outcomes.

Reviewer #3: Referee report for "Conditional cash transfer program and child mortality: a study of the 100 Million Brazilian Cohort"

Main comments

This paper uses an impressive sample bringing data together from different administrative sources to analyze the relationship between exposure to Brazil's conditional cash transfer program and child mortality. As such it revisits a question on which there is existing evidence, but with a larger sample (gathering data from 2006 to 2015) covering 5 million children. While the scale is impressive, the exercise unfortunately cannot say much about the causality of the relationship that is being analyzed. And as there is existing rigorously identified evidence from large scale programs (both from quasi-experimental designs, e.g. Barham 2011 in Mexico, covering more than a 1 million children; and from large-scale experimental designs (Okeke and Abubakar 2020 in Nigeria), it is not clear the contribution is sufficient to warrant publication in PLOS Medicine. These papers also need to be recognized, as well as the wider literature in economics on the impact of CCTs on health investments.

The reasons the evidence cannot be interpreted as causal is because the assumptions underlying the Kernel matching estimations are most likely violated in this setting. Matching estimators allow for causal identification only if it is reasonable to assume that selection into the program is based on variables that are observable to the analyst. But we know from a relatively large economics literature on Bolsa Familia that exposure to the program is in part driven by political criteria/processes (e.g. de Janvry et al 2012; Brollo et al 2020; among many others), and in part related to self-reported (and therefore likely strategically biased) income. This is indeed why Bolsa is one of the few CCT programs in Latin America without a straightforward set of impact evaluation results.

Was the sub-group analysis pre-specified? If not, why these cuts of the data and not others? Should we not be concerned with multiple hypotheses testing?

Other comments

The author reference other work on the relationship between Bolsa and child mortality, but claim they do better because they use individual data. It is never particularly clear which key questions they think they can answer better with individual data, that cannot be answered with aggregate data. It would be good to highlight this more specifically.

There is a relatively long literature on using bootstrap methods for matching estimators in economics that needs to be recognized. See Abadie and Imbens (2008, 2016) and many related papers, including on bias-corrected matching estimators (Abadie and Imbens, 2011). Also, given the sensitivity of matching estimators to strong assumptions, it is good practice to show robustness with various alternative estimators.

References

Abadie, Alberto and Guido Imbens, 2011. Bias-Corrected Matching Estimators for Average Treatment Effects

Journal of Business and Economic Statistics, January 2011, 29(1), 1-11.

Abadie, Alberto and Guido Imbens, 2008. On the Failure of the Bootstrap for Matching Estimators Econometrica 76(6), 1537-1557.

Abadie, Alberto and Guido Imbens, 2016. Matching on the Estimated Propensity Score, Econometrica, 84(2), 781-807.

Barham, Tania, A Healthier Start: The Effect of Conditional Cash Transfers on Neonatal and Infant Mortality in Rural Mexico, Journal of Development Economics, 2011, 94(1), 74-85

Brollo, F K Kaufmann, E La Ferrara, 2020. The political economy of program enforcement: Evidence from Brazil Journal of the European Economic Association 18 (2), 750-791

de Janvry, Alain, Frederico Finan and Elisabeth Sadoulet, 2012. Local Electoral Accountability and Decentralized Program Performance, with Review of Economics and Statistics, 94(3): 672-685.

Okeke Edward and Isa S. Abubakar, 2020. "Healthcare at the beginning of life and child survival: Evidence from a cash transfer experiment in Nigeria" Journal of Development Economics, 143: 102426

Any attachments provided with reviews can be seen via the following link:

[LINK]

16 Mar 2021

Submitted filename: Requests from the editors_DR_NB_RF_MB_March13.docx

Click here for additional data file.

13 Apr 2021

Dear Dr. Ramos,

Thank you very much for submitting your manuscript "Conditional cash transfer program and child mortality, a cross-sectional analysis nested within the 100 Million Brazilian Cohort" (PMEDICINE-D-20-05781R2) for consideration in PLOS Medicine’s Special Issue: Global Child Health: From Birth to Adolescence and Beyond.

Your revised paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with the Special Issue guest editors, and re-reviewed by one of the reviewers. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a further revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by May 04 2021 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

1. Reviewer Comments: Please fully address the comments of Reviewer 3.

2. Abstract: Line 39, and throughout: Please capitalize Black.

3. Author summary: Line 62 (and throughout): Please use “low-middle income countries” or similar rather than “developing” countries.

4. Introduction: Line 127-128: Please revise to avoid causal implications: “To address this gap, we tested the hypothesis that receiving BFP is associated with reduced child mortality…”

5. Methods: Please explicitly state in the Methods whether or not your study had a prospectively developed analysis plan. At line 276-278, please refer to the supporting information file containing your protocol (S1_Protocol). Please clarify if the analyses/outcomes described here are prospectively described in the included protocol. In either case, any changes in the analysis-- including those made in response to peer review comments-- should be identified as such in the Methods section of the paper, with rationale.

6. Results: Please provide statistics/results to accompany the statements comparing the strength of associations between subgroup associations, for example where associations between BFP and mortality are reported to be stronger for preterm compared to associations for term children: “Considering gestational age, an association between BFP was found for both term (Weighted OR=0.84; CI 95%: 0.79 to 0.89, p<0.0001) and preterm (weighted OR=0.78; CI 95%: 0.68 to 0.90, p<0.0001) children, with stronger association for the latter.” Please also complete this for the DMI quintile and income associations in the previous paragraphs.

7. Discussion: Lines 425-429: Would you please explain further regarding the potential for applying a regression discontinuity design in your study? It would be ideal to conduct this analysis in the current study, if feasible.

8. Discussion: Between the Limitations and Conclusions sections, please include a short section summarizing some implications and next steps for research, clinical practice, and/or public policy;

9. Discussion: Lines 441-443: Please temper causal language throughout, for example, in the following sentence: “...we can also conclude that when properly administered these programmes can increase their already significant impact on child health outcomes.”

10. Funding, Competing Interests, Data Sharing: Page 14: Please remove these sections from the main text. This information should be accurately and completely entered in the manuscript submission form.

11. Patient consent, Page 14: Please remove this section and ensure the relevant participant consent information is noted in the Methods section.

12. Table 1: Please present numbers in addition to percentages.

13. Table 3: Please define all abbreviations used in the table.

14. Figure 1: Please define all abbreviations in a figure legend (PBF under “exclusions”)

15. RECORD Checklist: Thank you for including the checklist. Please do remove references to page numbers. Locations can be referred to by section, and paragraph within section, such as: “Title Page” or “Methods, 2nd paragraph” for example.

16. Supporting Information Figures and Tables: Please provide a title and legend for each Table and Figure, and please define all abbreviations used in the figures and tables. Please report p values as p<0.001 rather than p<0.000 where applicable. Please define the asterisks (*) in Table S2a.

Comments from the reviewers:

Reviewer #3: I appreciate the efforts of the authors to revise the manuscript, and in particular to tone down the causal language throughout the manuscript. This indeed is appropriate. The revision however points to one new first order concern.

In particular, the results in Table S1c clearly show that for municipalities for which 90% or more of records are submitted, there is no significant relationship between Bolsa and child mortality. This seems to imply that possibly all the results in the paper are driven by sample selection related to missing death records, rather than any real associations between the program and child mortality (as for the subsample for which selection is small, no significant association is found. Most notably, no significative associations are found in municipalities with the full (100%) or almost full (95%) records, and indeed the results there have the opposite sign). On line 431 the authors suggest that loss of records could lead to a small overestimate. This is however not a small overestimate - as the entire relationship becomes insignificant or indeed reversed! This is particularly a concern as a very large share of the observations (3.8 million) come from municipalities with 50% of records missing. At the very least, a bounding exercise based on reasonable assumptions about the missing records would seem important, though given the magnitude of the selection, this is likely to result in very wide confidence intervals, which would unfortunately confirm that little can be said from this incomplete data about the true relationship.

In addition, the revision I continue to have many concerns with this revised manuscript that are similar to the ones raised earlier, but also with the responses to the authors to my earlier comments, which I do not believe fully address the central points I made.

In particular

1) In order to demonstrate the contribution of this paper to the literature, it is crucial not only to review what it is known from the international literature about CCTs and health investments, and what is known about Bolsa Familia in particular, but also specifically to refer to studies in the literature that causally identify the impact of this type of program on child mortality. The current approach in this paper, which refers to the international literature regarding broader health findings, but specifically excludes the papers focused on child mortality (including Barham, 2011 and Okeke and Abubakar, 2020 that I cited earlier), is highly misleading, and frankly hard to understand. I strongly suggest to acknowledge these studies, and to discuss the contribution of the current paper in comparison with them (which appears to be bringing evidence on brazil, at scale and for different subgroups, even if it is not causal).

Indeed, the concern I raised earlier is about the contribution of this paper - not about the plausibility. Given that what we already know from more rigorous designs on how these type of programs affect child mortality, why is showing associational evidence for Bolsa a sufficient contribution for publication in a high impact journal such as Plos Medecine?

2) My earlier comments on the drivers of selection into the program are NOT about suggesting a different research question - rather it points out that any study that wants to say something about "effects" (something the authors continue to claim they want to do in their response to my comments) needs to start from a good understanding of the selection criteria, given that the matching estimator proposed by the authors specifically relies on the assumption that it is observable characteristics that drive selection into the program (and that the variables used for that selection are observable to the analysist too).

The references I provided from the economics literature are just some of many that explain (and show empirical support) that in the case of Bolsa Familia, selection is not based on the type of observables the authors account for in the propensity score estimation (which means the assumptions underlying the propensity score estimates are invalid). The concern is in part with manipulation of eligibility - either because of issues with self-reported income, or local political processes. All these capture unobserved factors driving eligibility, and therefore raise serious doubts about the assumptions underlying the matching estimations proposed by the authors (to the extent that it makes it very, very likely that the assumptions are indeed violated.). Here too, ignoring the large literature that has considered this question and instead claiming starting on page 196 that "whether a family receives the program or not is determined by a set of family socio-economic characteristics" is misleading. In this paragraph, a correct discussion on what actually drives selection into the program is important. The statement needs to be revised to give a more comprehensive review of factors affecting receipt of the program.

Given the above, the matching estimator here merely is used as a tool to compare treated and non-treated households that are more similar in observable characteristics, something that may still be useful, but doesn't get you closer to causality. The current wording of the paper is in line with that logic, but being explicit about what the propensity score brings you in this application, and what it doesn't, would seem important to add.

3) With regard to the point on multiple hypotheses, the concerns stand whether the different hypotheses are tested in the same regression or not. See, for instance, Young (2019), List et al (2019) for related discussions. This is exactly one of the issues that pre-analysis plans aim to address (see Olken 2015).

In addition, the main concern here is what all other potential regressions you may have run in the exploratory analysis before setting on the ones reported. All of these tests are additional hypotheses and p-values would need to be adjusted for the fact that those additional hypotheses were tested. I suggest to at least acknowledge this in the manuscript.

4) Finally, the specific suggestion made in my earlier comments regarding the matching estimations, following the references cited earlier and common practice in economics, is to show robustness of results to various alternative matching estimators (1 or 5 nearest neighbour, kernel, etc.. ). This will help to demonstrate the robustness (or not) of your results.

Other comments

Line 218: you motivate not including self-reported income in the propensity score because it would violate the exclusion restriction of the instrumental variable approach but you are not using an IV approach! The assumptions needed for IV are different than for PSM. Matching estimators rely on the assumption that it is possible to account for selection on observables that drive program participation (independently on whether they also affect the outcome or not), so that one can compare 2 groups that according to those observables have equal likelihood of participation (but some for an exogenous reason participated and others didn't). See earlier provided references. Including self-reported income in the estimations gets you closer (in this case) to the assumptions needed for PSM. Therefore, I encourage the authors to show how results change when including self-reported income in the propensity score.

There is a similar confusion on line 426: IV and RDD are two separate quasi-experimental methods, each with separate sets of assumptions (even if discontinuity in eligibility criteria can sometimes be used as instrument, there are many other types of instruments). In the case of Bolsa, given the evidence of manipulation of income (which would be the running var in RDD estimate), assumptions needed to use a RDD estimation are probably not valid. I hence suggest taking out the reference to RDD.

References

List, J., A. Shaikh, and Y. Xu, (2019), "Multiple Hypothesis Testing in Experimental Economics," Experimental Economics (22): 773-793.

Young, A., (2019) "Channelling Fisher: Randomization Tests and the Statistical Insignificance of Seemingly Significant Experimental Results," Quarterly Journal of Economics, 134(2).

Olken, B., (2015), "Promises and Perils of Pre-Analysis Plans," Journal of Economic Perspectives, Vol 29(3): 61-80.

Any attachments provided with reviews can be seen via the following link:

[LINK]

13 May 2021

Submitted filename: response to editors and reviewers_May13.pdf

Click here for additional data file.

14 Jul 2021

Dear Dr. Ramos,

Thank you very much for submitting your manuscript "Conditional cash transfer program and child mortality, a cross-sectional analysis nested within the 100 Million Brazilian Cohort" (PMEDICINE-D-20-05781R3) for consideration in PLOS Medicine’s Special Issue: Global Child Health: From Birth to Adolescence and Beyond.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with an academic editor with relevant expertise. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

We would like to consider an additional revised version that addresses the editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response.

In revising the manuscript for further consideration, your revisions should address the specific points made by the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Jul 21 2021 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

1. From the academic editor: We request that you please more thoroughly acknowledge in the manuscript the limitations pointed out by reviewer 3, regarding the missing data (records not available for linkage) and the potential for this to impact the findings and conclusions of the study.

2. Title: Please revise your title according to PLOS Medicine's style. We suggest: “Conditional cash transfer program and child mortality: A cross-sectional analysis nested within the 100 Million Brazilian Cohort”

3. Abstract: Line 24: We suggest changing to (1-4 years of age) or similar.

4. : Line 38 and line 43: Please report this as p<0.001.

5. Abstract: Line 45: Please be more specific if possible regarding the “small over-estimate” attributable to the loss of nameless death records before linkage.

6. Author summary: Why was this study done? In the first point, we suggest “reaching” rather than “completing” if appropriate.

7. Introduction: Line 112: Please replace “developed countries” with high-income countries.

8. Methods: Line 248: Please list the perinatal conditions adjusted for here (although they are also noted in Table 4).

9. Results: Throughout the Results section, please report p<0.0001 as p<0.001.

10. Results: Please include more description and interpretation of the results presented in supporting information tables S1a, S1b, and S1c.

11. Discussion: Limitations Lines 477-484: In the sentence describing the loss of nameless records (“...loss of nameless records before linkage seems to be resulting in a possible over-estimation of the association…”) it would be helpful here, and in the appropriate section of the Results, to quantify or define the over-estimation. You mention in this paragraph that the majority of observations do come from municipalities where associations were consistent with the sample as a whole. However, as brought up previously by Reviewer 3, we request that you more thoroughly acknowledge the limitation of missing data and the impact on results, and discuss the potential for this to confound the conclusions.

12. Figure 1: Please include a descriptive legend for this figure.

13. Figure 2: Please define MHDI-R and DMI in the legend. We suggest changing the y axis to something more descriptive. If possible, please present all panels on the same range of y axis values, or note the different y axis scales in the legend.

14. Table 3: Please define the abbreviation (MHDI-R) and please describe in the legend the factors that distinguish each of the models.

15. Table 4: Please define the abbreviation DMI in the legend, and please describe the factors that distinguish each of the models.

16. Tables 5 and 6: Please describe the factors that distinguish each of the models.

17. Study Protocol: We suggest renaming the file to “S1_Protocol” and double checking to be sure all information is intended to be included (e.g. the contact information on the first page, and the investigator CVs at the end of the document).

18. RECORD Checklist: Please replace the location of the Background rationale (currently noted as pages 3 and 4) with the appropriate section/paragraph of the main text (for example, within the Introduction).

19. Figure S1a: Please include a legend, with all abbreviations used within the figure defined.

20. Figure S2: We suggest revising “treated” and “untreated” to “beneficiaries” and “non-beneficiaries” respectively.

Any attachments provided with reviews can be seen via the following link:

[LINK]

20 Jul 2021

Submitted filename: response letter July20_MB.docx

Click here for additional data file.

4 Aug 2021

Dear Dr. Ramos,

Thank you very much for re-submitting your manuscript "Conditional cash transfer program and child mortality: A cross-sectional analysis nested within the 100 Million Brazilian Cohort" (PMEDICINE-D-20-05781R4) for consideration in PLOS Medicine’s Special Issue: Global Child Health: From Birth to Adolescence and Beyond.

I have discussed the paper with my colleagues and the academic editor. I am pleased to say that provided the remaining editorial and production issues are dealt with we are planning to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

Please note, when your manuscript is accepted, an uncorrected proof of your manuscript will be published online ahead of the final version, unless you've already opted out via the online submission form. If, for any reason, you do not want an earlier version of your manuscript published online or are unsure if you have already indicated as such, please let the journal staff know immediately at plosmedicine@plos.org.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by Aug 11 2021 11:59PM.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine

plosmedicine.org

------------------------------------------------------------

Requests from Editors:

1. Data availability statement: Please check that the DOI for the data is accurate/working. At this time, following the link leads to an error message (“Error: Cannot connect to server”).

2. Throughout: Please carefully check the grammar and language throughout the text for errors.

3. Abstract: Introduction: At line 25, we suggest replacing “gestational age” with “preterm birth” if this helps to clarify the objective. At line 24-26, we suggest revising to “...also examining how this association differs by maternal race/skin color, preterm birth, municipality income level, and indexes of quality of BFP management.” or similar.

4. Abstract: Methods and Findings: Line 38: Please remove the extra space in the reported p value.

5. Abstract: Methods and Findings: Line 38-39: Please also provide weighted OR, 95% CI and p values for the BFP-mortality associations for preterm birth and children of Black mothers, as is reported for municipality income status and BFP management status.

6. Abstract: Methods and Findings: Line 44-45: Please remove the word “small” and revise the sentence to: “Furthermore, sensitivity analyses showed that loss of nameless death records before linkage may have resulted in an over-estimate of the measured associations between BFP and mortality…”

7. Author summary: Why was this study done? In the first point, we suggest adding “globally” if this is meant. In the third bullet point, we suggest revising to: “...the world’s largest conditional cash transfer program)...” if this is accurate.

8. Author summary: What did the researchers do and find? In the second point, we suggest revising to: “...gestational age at birth…” or similar.

9. Author summary: What do these findings mean? In the first point, please revise to “Our findings are consistent with previous studies…” or similar. We suggest revising the wording of the second point to: “The greater association among more vulnerable groups of children suggests conditional cash transfers may help to promote equity, with stronger results among those more in need.”

10. Introduction: In the first paragraph, please make sure that all statements/reported statistics are appropriately referenced with citations.

11. Introduction: Line 112: Please revise to: “...they are also present in countries including....”

12. Introduction: Line 115: We suggest changing “speedily reached” to “...rapidly implemented throughout the country…”

13. Introduction: Line 139: Please clarify to “gestational age at birth” if this is meant. It may be more straightforward throughout to refer to preterm birth status as opposed to gestational age, as gestational age was categorized into preterm or term births in your analyses.

14. Introduction: Line 139-140: Please move this sentence to the Methods, where you describe your analysis plan: “The subgroup analysis was established by our research protocol (see S4. Research Protocol, objective 3).” - this seems to be approximately lines 211-214.

15. Methods: Line 164: Please clarify which supporting information file you are referring to with “supplementary material 1” as this is not clear from the supporting information file names.

16. Methods: Line 198: We suggest changing “before the child’s 5th anniversary…” to “prior to the child reaching 5 years of age” or similar.

17. Methods: 203: Please provide a more complete reference (or a numbered reference from the References list) for “Bolsa Família. Perguntas Frequentes” - we note that the embedded link does not work.

18. Methods: Line 223: Please note how race/ethnicity was defined/reported and by whom?

19. Methods: Line 261: We suggest using “As outlined in our research protocol…” rather than “forseen” in this sentence.

20. Methods: Line 287: Please group all of these references within one set of brackets, if appropriate ([20,21,22-25,26,27]).

21. Methods: Line 293: Please change “will be based” to “was based” in this sentence.

22. Results: Line 370-372: Please avoid using italics for emphasis.

23. Discussion: Line 409: Please revise the first sentence to “We found that receiving Bolsa Família is significantly associated with a reduced probability of deaths…”

24. Discussion: Line 412: Please revise the sentence to “These findings are consistent with previous studies…”

25. Discussion: Line 417: Please revise the sentence to “We believe that interpreting individual-level findings…” or similar.

26. Discussion: Line 431: Please revise to “...an in access to health care in Brazil…” if this is meant.

27. Discussion: Line 435: Please change “premature” to “preterm” in this sentence.

28. Discussion: Line 461: Please add “...to the best of our knowledge” to the end of the sentence, or similar: “...making this the largest study on CCTs and child mortality to date.”

29. Discussion: Line 466: Please clarify the heterogeneities that you are referring to here: “We have also accounted for heterogeneities not yet assessed regarding the association of BFP with child health outcomes.”

30. Discussion: Line 471: Please change “once” to “because” or similarly clarify what is meant here.

31. Discussion: Line 485: Please remove the hyphen from “evidence-based” in this sentence, it seems it should read “...providing evidence based on a large population…”

32. Discussion: Line 486-490: We suggest revising to: “Limitations also arise from the linkage process, as loss of nameless records before linkage may have resulted in a possible over-estimation of the association between the program and child mortality, since we observed that in the municipalities without such losses, the direction of the measured association changed, and the lower bound of the confidence interval is very close to the null value of 1.00.”

33. Discussion: Line 500: Please change “superior to” to “greater than 40%” in this sentence.

34. References: Please check if reference 29 and 30 are duplicates.

35. Table 2: Please include the definition for the abbreviation BFP in the legend or elsewhere in the table.

36. Figure 2: We suggest including an X axis label for panel C such as “Maternal race” or similar, and revising “Whites” to “White” and “Blacks” to “Black” along the labels. For panel D, we suggest including a note in the legend that Term is defined as greater than or equal to 37 weeks gestation, and Preterm is defined as less than 37 weeks gestation.

37. Supporting information: The purpose of the file “Export_20210505_25908PM” is unclear, we suggest removing.

38. Supporting information figure S3a: Please define the abbreviation PBF / please clarify if this should be BFP.

39. Tables S2a and S2b: Please check if the commas should be decimal points, “88,00” for example.

40. Supporting information file: It might be helpful to separate this document into “S2 Supporting information” and “S3 Supporting Information” to make access easier.

Any attachments provided with reviews can be seen via the following link:

[LINK]

10 Aug 2021

Submitted filename: response letter_Aug4.docx

Click here for additional data file.

11 Aug 2021

Dear Dr Ramos,

On behalf of my colleagues and the Academic Editor, Zulfiqar A. Bhutta, I am pleased to inform you that we have agreed to publish your manuscript "Conditional cash transfer program and child mortality: A cross-sectional analysis nested within the 100 Million Brazilian Cohort" (PMEDICINE-D-20-05781R5) in PLOS Medicine’s Special Issue: Global Child Health: From Birth to Adolescence and Beyond.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. Please be aware that it may take several days for you to receive this email; during this time no action is required by you. Once you have received these formatting requests, please note that your manuscript will not be scheduled for publication until you have made the required changes.

In the meantime, please log into Editorial Manager at http://www.editorialmanager.com/pmedicine/, click the "Update My Information" link at the top of the page, and update your user information to ensure an efficient production process.

Please also complete the following editorial request:

Abstract: Line 25: Please change “(term x preterm)” to “(full term or preterm birth)” if that is intended.

PRESS

We frequently collaborate with press offices. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximise its impact. If the press office is planning to promote your findings, we would be grateful if they could coordinate with medicinepress@plos.org. If you have not yet opted out of the early version process, we ask that you notify us immediately of any press plans so that we may do so on your behalf.

We also ask that you take this opportunity to read our Embargo Policy regarding the discussion, promotion and media coverage of work that is yet to be published by PLOS. As your manuscript is not yet published, it is bound by the conditions of our Embargo Policy. Please be aware that this policy is in place both to ensure that any press coverage of your article is fully substantiated and to provide a direct link between such coverage and the published work. For full details of our Embargo Policy, please visit http://www.plos.org/about/media-inquiries/embargo-policy/.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Thank you again for submitting to PLOS Medicine. We look forward to publishing your paper.

Sincerely,

Caitlin Moyer, Ph.D.

Associate Editor

PLOS Medicine