Growing together: assessing equity and efficiency in a prenatal health program

Abstract

We study the acting mechanism of an early-life social safety net program and quantify its impact on child health outcomes at birth. We consider both the equity and efficiency implications of program impacts and provide a metric to compare such programs around the world. In particular, we estimate the impact of participation in Chile Crece Contigo (ChCC), Chile’s flagship early-life health and social welfare program, using a difference-in-differences style model based on variation in program intensity and administrative birth data matched to social benefits usage. We find that this targeted social program had significant effects on birth weight (approximately 10 grams) and other early-life human capital measures. These benefits are largest among the most socially vulnerable groups but shift outcomes toward the middle of the distribution of health at birth. We show that the program is efficient when compared to other successful neonatal health programs around the world and find some evidence to suggest that maternal nutrition components and increased links to the social safety net are important action mechanisms.

Appendices

Appendix 1. Tables and figures

Table 10 Test of FONASA coverage and ChCC rollout

Table 11 Difference-in-difference estimates using municipal variation in coverage and a rollout indicator

Table 12 Summary statistics by trimester: birth and Chile Crece Contigo data

Table 13 Difference-in-difference estimates with data collapsed by trimester

Table 14 Difference-in-difference estimates based on the year surrounding rollout

Table 15 Instrumental variables estimates based on the year surrounding rollout

Table 16 Examining robustness of impacts on birth weight to removal of extreme values

Table 17 Adjusting for multiplehypothesis testing

Table 18 Difference-in-difference estimates using municipal program availability

Table 19 IV Estimates using lagged ChCC enrollment

Table 20 Correction for Multiple Hypothesis Testing in Distributional Estimates

Table 21 Costs of ChCC per participant in gestational program

Table 22 Impact of Chile Crece Contigo on pregnancy inputs

Table 23 Gelbach (2016) decomposition of ChCC mechanism

Fig. 5

Program rollout by date. Chile consists of 346 municipalities (“comunas”) which are the lowest geographic administrative level with their own political administration. ChCC rollout started in June 2007, and reached 159 of the 346 municipalities in 2007 (chosen due to the availability of infrastructure) and then was rolled out to the remaining municipalities during 2008. Precise rollout dates are provided by the Ministry of Social Development of Chile. The full country is displayed in the left-hand panel, and only the Metropolitan Region of Santiago (from the centre of the country) is displayed in the right-hand panel

Fig. 6

ChCC usage in post-implementation period. The density of ChCC usage by municipality over the entire post-treatment period is displayed. Usage refers to the average proportion of all births in each municipality for which ChCC components were accessed by the mother during the gestational period. Usage data comes from The Ministry of Social Development’s administrative data on public program use, and is averaged at the level of each municipality. Refer to Fig. 8 for additional details regarding municipal-level usage of ChCC components and municipal characteristics

Fig. 7

Proportion of births attended in the public health system. Figures on the proportion of births in the public health system and all births nation-wide are provided monthly by the Department of Statistics and Health Information (DEIS) of the Ministry of Health of Chile. Monthly proportions are displayed for each month from January 2002 until December 2010. The first vertical dotted line is the beginning of ChCC rollout, while the second vertical dotted line is when ChCC reached the full country

Fig. 8

Municipal Characteristics and ChCC enrollment. Each panel presents the proportion of Chile Crece Contigo enrollees in each municipality after the introduction of the program along with municipal-level averages in a range of other social or political variables. In each case, ChCC enrollment is displayed on the horizontal axis, and alternative outcomes on the vertical axis. a Treated piped drinking water. b FONASA enrolments. c Proportion of FPS per year. d Poverty. e Education subvention. f Proportion of teen births. g Vote share (mayor). h Political association. i Maternal education

Fig. 9

Socioeconomic quintiles and health distributions at birth. Figures provide kernel density plots of birth weight (in grams) and weeks of gestation by quintiles of the Social Vulnerability Score. Quintile 1 is the most vulnerable, and quintiles 4 and above are grouped into a single plot. Means for birth weight are 3350 g, 3333 g, 3317 g, and 3298 g for quintiles 1, 2, 3, and 4+ respectively. Similar means for gestational period are 38.66 weeks, 38.61 weeks, 38.55 weeks, and 38.43 weeks. a Birth weight. b Gestational period

Fig. 10

Running variable (FPS) in RDD. Left-hand panel provides a histogram of all Social Protection Scores (“Ficha de Protección Social”) for mothers matched to their children’s birth records. The vertical dashed line indicates 13,484 points, the cutoff point for Chile Crece Contigo’s preferential benefits. This is defined as the top-end of the third quintile of vulnerability scores, though these quintiles are defined on all recipients of a score in the country, not just mothers. The right-hand panel documents (McCrary 2008)’s density test around 13,484, documenting the dispersion of observations within 1000 points on either side of the cutoff

Fig. 11

Event studies. Event studies present estimated models interacting ChCC treatment intensity with pre- and post-treatment indicators for each 3-month period pre- and post-reform. Here, the ChCC measure refers to average levels of ChCC use in the entire post-treatment period (to allow a constant treatment intensity for interaction), and this is interacted with indicators for the rollout of the ChCC program to each municipality. The precise specification is: ${InfantHealth}_{ct} = {\alpha }_{0} + \sum \limits _{j=-9}^{9} {\beta }_{j} 1\{\text {Time to Adoption} = j\}_{t} \times {ChCC}_{c} + \mu _{c} + \lambda _{t} + {\varepsilon }_{ct}$. As is standard, 1 period pre-treatment is the omitted reference group. Periods greater than 9 trimesters pre- or post-program are indicated in a single ≥ 9 term. a Birth weight. b LBW. c Prematurity. d Gestation. e Size at birth. f Fetal deaths

Fig. 12

Descriptive RD plot with smaller bins for Social Vulnerability Score (birth weight). Descriptive plot displays average birth weight outcomes in 5 point bins of the Social Protection Score, with a separate polynomial fitted on each side of the cutoff. This figure replicates Fig. 3a, however now using bins of 5 points, rather than 55 points, for the running variable

Fig. 13

Impact of FPS cutoff point on the probability of ChCC usage. Descriptive plot documents the probability that mothers are enrolled in the ChCC program around the official cutoff for the receipt of preferential benefits targeted at the bottom three quintiles of recipients of the Social Protection Score. When estimating a regression discontinuity specification in a local linear model with Calonico et al. (2014)’s optimal bandwidth, the additional likelihood of participating in ChCC when located just below the cutoff is 0.0065 (0.019) (coefficient and standard error)

Fig. 14

Variation in Home Visit Intensity by Municipality. Histogram documents the average quantity of “Integral Home Visits” received by each targeted family per municipality in Chile in 2013. A value of 1 refers to a situation where (on average) each family flagged to require a visit based on ChCC’s administrative criteria receives one visit during the gestational period. These data are averaged for each municipality, and are based on the year 2013 only. These data are released by the Ministry of Health (available at http://chcc.minsal.cl/indicadores/resultados/293) and are not available for earlier years. One small municipality with an average number of visits of 14.5 per flagged family was removed to simplify graphical presentation

Fig. 15

Health Services and Municipalities. Municipalities are indicated by municipal boundaries, and health services are indicated by colours. Each of Chile’s 346 municipalities belongs to one of 29 Health Services. The entire country is displayed at right, and the densely populated Metropolitan Region of Santiago is displayed at left

Fig. 16

ChCC rollout and pregnancy inputs disbursed. Solid blue line displays the rollout of ChCC and proportion of coverage of births as in Fig. 1. Dotted red lines display the total units of various components of the program disposed over time in whole of Chile. Each panel with the exception of Chile Solidario coverage in panel f presents the number of units divided by 1000. Additional discussion of variables and their measurement is provided in Section 5.3.a Prenatal check-ups. b Home visits. c Fortified milk (original formula). d Fortified milk (updated formula).e Social assistance appointments. f Chile Solidario

Appendix 2. Broader context: health system and birth outcomes chile

1.1 2.1 Birth outcomes and maternal characteristics

Following the return to democratic rule in 1990, full micro-data on all births in Chile has been available from the Ministry of Health’s Department of Statistics and Health Information (DEIS). These vital statistics include each child’s birth weight, weeks of gestation, and a number of characteristics of the mother and father (when the father is present). These data are recognised to be of high quality and very close to universal (see for example Mikkelsen et al. (2015)).

The average age of mothers in Chile has risen from slightly over 26 in 1990, to slightly under 28 in 2015 (Fig. 17). The average age of mothers increased constantly from 1990 until approximately 2004, before falling slightly, and ascending once again from 2009 onwards. This reduction in maternal age occurred during a considerable slow down in growth, and an uptick in the number of births each year (Fig. 18), in line with results suggesting countercyclicality in fertility. Panel b of Fig. 17 displays the proportion of teenage births (among all births), which rose until the early 2000s, began to fall until the growth slowdown in the mid-2000s, and has fallen sharply from 2007.

Fig. 17

Trends in Maternal Characteristics in Chile. Yearly averages of age and the proportion of all mothers aged under 20 years of age based on Ministry of Health (DEIS) micro-data covering all births in Chile between 1990 and 2015

Fig. 18

Number of Births per Year

We display descriptive plots of average birth outcomes across time in Fig. 19. These indicators, particularly birth weight, improved sharply following the transition to democracy in the early 1990s, and the implementation of a considerable public health reform. Average birth weight increased by more than 60 g, and the proportion of low birth weight babies fell by a full percentage point (refer to panels Fig. 19a and b). From the year 2000 onwards, average outcomes have gradually worsened, in line with increases in maternal age.

Fig. 19

Longer Term Trends in Birth Outcomes in Chile. Yearly averages of birth weight, the proportion of low birth weight births (< 2500 g), weeks of gestation, and the proportion of premature births (< 37 weeks) from Ministry of Health (DEIS) micro-data covering all births in Chile between 1990 and 2015

1.2 2.2 Prenatal health programs in Chile before ChCC

Prior to the implementation of ChCC, programs aimed at early childhood focused on health and education were already carried out in the country, separately.

With respect to the different health programs, the National Immunization Program (PNI) began in 1978, which is still in force at present. Its main objective is the reduction of morbidity and mortality, contributing to the reduction of infant mortality.

In 1987, the National Complementary Food Program (PNAC) was created, consisting of the delivery of milk to children under 6 years old and of food for pregnant women, delivered at primary care clinics. For the delivery of food, it must comply with health controls, controls for pregnant women and with the National Immunization Program.

In 1990, Chile ratified the Convention on the Rights of the Child, approved by the General Assembly of the United Nations, which promotes: non-discrimination, safeguarding the best interests, survival, development and protection of minors.

Since 1994 the government carries out the Program for the control of children Lower Respiratory Tract Infections (IRA, in Spanish), a campaign deployed every winter aimed at controlling these diseases.

In particular with regard to pre-formal education, there are two institutions with the longest history in the country. On the one hand, the National Board of Kindergartens (JUNJI) is a state institution created in 1979. On the one hand, the INTEGRA foundation, created in 1991, is a private non-profit educational institution whose objective is the integral development of children from 3 months to 4 years old (although they also have kindergartens that offer kindergarten and pre-kinder), belonging to families of the first and second income quintile.

1.3 2.3 The Chilean health system

Primary care in the public health system in Chile is provided by municipal health centres which, among other things, provide prenatal appointments for pregnant mothers and families. These municipal health centres exist in each municipality in Chile (refer to Fig. 20a for geographic distribution). These health centres are distributed much more sparsely in less populated northern and southern regions of the country. Secondary and tertiary care are provided in hospitals which are located in each region of the country. Births attended in the public health centre are delivered in these hospitals. The geographical distribution of hospitals is displayed in Fig. 20b, where once again these are concentrated in the central region of the country where the largest population resides.

Fig. 20

Geographic distribution of health centres and hospitals. Geo-referenced hospital and Health Clinic information from the Ministry of Health of Chile. All points represent public hospitals and health clinics

The health system in Chile is a mixed system,³² which consists of a public and private systems. In administrative terms, the public system operates thanks to the Sistema Nacional de Servicios de Salud (SNSS) that has autonomous services throughout the country, such as the Servicios Regionales Ministeriales (SEREMI), 29 Regional Health Services and the Servicio de Atención Primaria de Urgencia (SAPU). In this way, the Fondo Nacional de Salud (FONASA) is responsible for granting health care coverage as a financial institution with its own assets.

On the other hand, the private health system is composed of the Institutions of Provisional Health (ISAPRES). Currently there are 6 large private insurers and other smaller ones, that are empowered to capture and manage the mandatory health contribution of all formal workers that are not affiliated with FONASA, supplying the State in the granting and financing of health benefits.

Thanks to the contributions given to ISAPRES, they finance health services and the payment of medical licenses to their taxpayers. At present, the ISAPRES have achieved an increase in the supply and investment of private infrastructure in Chile. In addition, the main source of funding in ISAPRES is the contribution of its members, paying premiums based on the risks (sex and age) and their family responsibilities, thanks to an individual contract.

If an individual is enrolled in FONASA, they will be automatically assigned to one of the 4 groups depending on their disposable income, and their copayment will depend on this:

• Tranche/Section A: beneficiaries lacking resources to contribute, or in conditions of indigence (non-contributors).

• Tranche/Section B: Monthly taxable income less than or equal to $276,000 with co-payment equal to 0%.

• Tranche/Section C: Monthly taxable income greater than $276,000 and less than or equal to $402,960 with a copayment equal to 10% (with 3 or more family responsibilities is assigned to tranche B).

• Tranche/Section D: Monthly taxable income greater than $402,960 with a copayment equal to 20% (if 3 or more dependents, members in this group are assigned to tranche C).

The main difference between FONASA and ISAPRES is that FONASA is free or with low co-payments because the premiums do not depend on the risks or size of the family group, causing the state to make the largest contribution out of tax contributions.

The most recent data indicate the amount of the affiliated population in FONASA is 76% and in ISAPRE it is 18%.

Appendix 3. Additional program details and component data

Additional program details

The full Chile Crece Contigo program covers children from before birth (officially from the first planned gestational check-up at week 14 of pregnancy) until early childhood. Initially, with the design and rollout of the program in 2007, the program ended at age 4, once children enter the first transition level to primary school.³³ More recent extensions mean the program now follows children up until the age of 8, with mental health treatment for children with mental health disorders aged between 5 and 8.

The original program designed for children aged up to 4 years consisted of 5 components and various sub-components. We lay these out below in Table 24. Component 1, which is targeted to pregnant mothers, is the only component which can potentially impact birth outcomes, as the remainder of the components are entirely delivered in the birth to 4 year period of life. The components below are universal, with the exception of component 1B and component 5, which are preferential components received by families flagged as being among the 60% most vulnerable based on a social protection score.

Table 24 List of ChCC Policy Components and Phases

Each particular program item described in Table 24 consists of one or a series of check-ups, goods or other services. Each item also comes with a clear definition of how to deliver the item to the objective population, and key targets for public service workers. For example, Item 1A, Part i (pre-natal check-ups) specifies that 7 prenatal check-ups should be targeted in low-risk cases, and that the duration of these check-ups is 40 min. Particular check-ups also have their own requirements, such as specific diagnostic tests including the abbreviated psycho-social evaluation during the first and third trimester.

In this appendix we provide only a short summary of each component in Table 24. Full details regarding each component are available in the ChCC guide to services (Ministerio de Desarrollo Social 2014). Specific components targeted to vulnerable families consist of the generation of a personalised plan identifying availability of differential services, home visits lasting 1 h (which are targeted to families with specific risk factors), information related to other subsidies and local programs, and contact with local healthcare and social professionals. Additionally, all children in vulnerable families are guaranteed access to extended nursery and pre-school programs at no cost.

Data on program component coverage

The examination of program mechanisms of action in Section 5.3 relies on data recording program components, and their coverage over time. As laid out in the paper, we collect these data from public monthly administrative health statistics data. In each case we calculate the average level of component use for each birth in the 9 months prior to birth. Averages are always calculated at the health service and monthly level. In a number of cases, we linearly extrapolate coverage by month prior to 2005 only, given that data is not always available in 2003 and 2004. This period is entirely in the pre-program period, and time fixed effects also capture periods in which linear extrapolation is performed.

Fortified milk disbursed to pregnant women as part of the program was originally called “Leche Purita Fortificada” (Purita Fortified Milk). In 2008 this underwent a modification to better meet the dietary requirements of pregnant women, and was renamed to “Purita Mamá”. Purita Mamá thus replaced Leche Purita Fortificada, although a very small number of batches of the original formula was still disbursed post 2008. In Table 25 we show the change in composition between the two types of dietary supplements. The guidelines issued by the Ministry of Health provide a clear description of how this milk should be disbursed to pregnant women. For those who begin pregnancy with normal weight, are overweight, or are obese, 1 kilogram of milk powder is given per month. For those women who begin pregnancy with an underweight diagnosis, 3 kg of milk powder is delivered per month (Gobierno de Chile, 2008). cutoff Measures of home visits refer to “Integral Home Visits” to pregnant women. Government reports highlight that Chile Crece Contigo has increased the frequency of home visits to pregnant mothers by around 500%. These home visits are targeted particularly to families identified as being in “psycho-social risk”, which implies meeting the vulnerability cutoff, and also presenting a number of additional risk factors. Given that the demand for home visits varies considerably by income level of municipalities, the precise decision of which families to visit is made by municipal health centres, where visits should be targeted to families with the largest number of risk factors. A complete discussion of the goals and recommendations for social workers completing home visits is provided in Gobierno de Chile (2009).

Table 25 Changes in composition of complementary nutrition component

Remaining components such as prenatal check-ups and appointments with social assistants in local health centres are also reported in monthly health usage data. In this case the number of appointments completed are reported, and in Section 5.3 we calculate the average number of appointments per health service for a pregnancy in the 9 months prior to the birth.

Appendix 4. Maternal fixed effects

As a consistency check of the difference-in-difference results reported in the paper, we also undertake an analysis using the full matched micro-data observing each mother’s participation status in ChCC. Identification is driven by variation within mother’s exposure to the program over time. We estimate the following mother FE specification:

$$ Infant Health_{ijt} = \beta_0 + \beta_1 ChCC_{jt} + \boldsymbol{X}_{ijt}\beta_{x} + \phi_t + \mu_j + \varepsilon_{ijt} $$

(4)

where InfantHealth refers to the same measures of health at birth as discussed in the body of the paper of child i born to mother j at time t.

The matched administrative data allows us to construct a panel of mothers and their children, and the independent variable of interest in 4 is ChCC_jt. This measures for each mother at time t whether she participated in Chile Crece Contigo, and under typical (fixed effect) panel assumptions, β₁ identifies the effect of participation on infant health. We include maternal fixed effects μ_j and year fixed effects ϕ_t, as well as a series of time-varying controls for mothers including birth order dummies, mother’s age at birth dummies, and child year of birth dummies.³⁴ Identification takes advantage of the fact that there are mothers who (a) participated in ChCC and had births both before and after the introduction of the policy, and (b) never participated in the policy and also had births both before and after the policy’s introduction.

The matched mother and child data does not include the entire universe of births (we do use the entire universe of births in municipal-level regressions presented in the paper). As such, any estimated program impacts in the micro-level mother FE specification are at best suggestive of the average effects in the population. When matching vital statistics data with parental social program use data, approximately 50% of births were matched with fathers, rather than mothers, and in these cases we do not observe the mother’s ChCC participation status. We thus restrict the analysis with mother FE only to the population of children matched with mothers, noting that it is not a representative sample, and as such not directly comparable to the municipal-level difference-in-difference regressions presented in the paper based on the entire universe of births. Nevertheless, it acts as a useful robustness check of the impact of ChCC based on different identifying assumptions.³⁵

In Table 26 we present summary statistics of births to all mothers, births to mothers who were matched with their social program usage, and births to mothers who were not matched the mother’s social program usage data. While their observable measures are largely similar, matched mothers appear to be slightly younger (26.91 versus 27.19 years), and have births with slightly better health indicators (3333 g of birth weight versus 3324 on average).

Table 26 Summary statistics: matched mother, child and social security data

We present regression results using maternal fixed effects in Table 27. In this case identification is driven by mothers who have had more than one birth, and hence variation in program coverage. Despite the alternative methodology (and estimation sample) we observe results that are qualitatively similar to those reported using the municipal rollout to estimate program impacts. In this case we observe a larger impact on birth weight (19 g, versus 10 g), and significant impacts also when considering size at birth of each child. One result does not agree across specifications, and this is the estimate on the impact of ChCC on low birth weight children. In this specification we observe a weakly positive impact, while in the specification reported in Table 3 we observed a weakly negative impact. However, in Table 28 when we additionally include full time and municipal fixed effects, we observe that the result is no longer statistically distinguishable from zero, while remaining effects are largely unchanged. In panel B of Appendix Table 20 we present p values on the impact of ChCC when correcting for multiple hypothesis testing. For birth weight, birth size, and gestational length we observe that results remain statistically distinguishable from zero when controlling for the family-wise error rate using Romano and Wolf’s step-down correction.

Table 27 Estimated program effects with mother fixed effects

Table 28 Maternal FE estimates with additional controls

Finally, we briefly examine distributional impacts of the program on health at birth, as examined in Fig. 4. In this case we simply examine descriptive evidence, considering the distribution of birth weight between program recipients and non-program recipients prior and posterior to the program’s implementation. These are presented in Fig. 21, and we observe that in the pre-program period, the distribution of birth weight for recipient mothers is slightly below the corresponding distribution for non-recipient mothers, while post-program the reverse pattern is observed (both differences are observed in the rejection Kolmogorov-Smirnov of tests of the equality of distributions). Interestingly, the distribution appears to be most shifted from around 2500–4500 g, providing some descriptive support of the distributional results documented in Fig. 4.

Fig. 21

Birth weight distributions pre- and post-program implementation. Densities are plotted using an Epanechnikov kernel with a bandwidth of 5 g. Each panel separates distributions by whether the mother ever participates in Chile Crece Contigo. Panel a displays only pre-ChCC s, while panel b displays only post-ChCC time periods. In both cases, Kolmogorov-Smirnov tests reject equality of distributions (in different directions)