THE EFFECTIVENESS OF PRENATAL CARE IN URUGUAY’S LOW-INCOME POPULATION: A PANEL DATA APPROACH^*

 

Ana Inés Balsa^**

Patricia Triunfo^***

* The authors would like to thank Graciela Sanromán and Partha Deb for their comments and methodological contributions and the anonymous LAJE referees who have improved this version of the paper. This work would not have been possible without data from the Perinatal Information System Division of the Epidemiology Area (National Information Unit of the Ministry of Public Health).
** University of Montevideo.
*** Faculty of Social Sciences, University of the Republic (Uruguay). Email: patricia.triunfo@cienciassociales.edu.uy.

---

This paper studies the effectiveness of prenatal care on low-income women’s birth outcomes. We analyze all births between 1995 and 2011 in Uruguay’s largest public maternity ward. We use mother-specific first differences to circumvent biases due to time-invariant, unobserved heterogeneity and implement robustness checks that reduce concerns about time variant shocks and feedback effects. We find that adequate use of prenatal care, as defined by early initiation and at least 9 visits, decreases the probability of low birth weight by 6 percentage points and the probability of pre-term birth by 11 percentage points, and increases birth weight by 149 grams.

JEL classification: I12, J13, C14

Keywords: Prenatal care, low birth weight, low-income populations, first differences, panel data

---

 

1. Introduction

Preterm birth (< 37 weeks gestation) and low birth weight (< 2500 grams) are commonly used as proxies for infant health (McCormick, 1985; Institute of Medicine, 1985). Low birth weight (LBW) has been associated with increased morbidity and mortality during the life course, higher health care costs, lower educational attainment, and decreased lifetime income (Petrou et al., 2000; Boardman et al., 2002; Black et al., 2007). Several authors have stressed that LBW serves as an important mechanism for the intergenerational transmission of economic status (Currie and Madrian, 1999; Grossman, 2000; Case et al., 2004; Behrman and Rosenzweig, 2005; Currie and Moretti, 2005). To reduce the burden that LBW imposes on society, the medical community has underscored prenatal care as a key input. Through prenatal care, it is argued, mothers at risk of premature delivery or babies with intrauterine growth retardation (IUGR) can be identified, thus enabling a variety of medical, nutritional, and educational interventions aimed at reducing poor birth outcomes.

Estimation of the relationship between prenatal care and birth outcomes is complicated by difficulties in controlling for maternal characteristics associated with both the demand for prenatal care and the infant’s health at birth. Without adequately controlling for health endowments, the mother’s health habits, her propensity to engage in risky behaviors, or the extent to which the pregnancy is desired, an association between prenatal care and infant health cannot be regarded as causal. In the past 15 years the economic literature has proactively pursued the identification of such a causal relationship. Most investigations have exploited the association between exogenous variation in health care coverage and prenatal care use by employing either reduced-form models or two-stage least squares (2SLS) techniques (Kaestner, 1999; Brien and Swann, 2001; Currie and Grogger, 2002; Figlio et al., 2009; Habibov and Fan, 2011). A few authors have used 2SLS with arguably exogenous instruments, such as input prices, the availability of prenatal clinics in the area (Grossman and Joyce, 1990; Gajate-Garrido, 2013), public transportation strikes in the county (Evans and Lien, 2005), and average distance to the nearest health facility (Awiti, 2014). Even after addressing confounders, this literature has been mixed regarding the effects of prenatal care on birth weight. The evidence is divided between those who find slight or no effects (Grossman and Joyce, 1990; Kaestner, 1999; Currie and Grogger, 2002; Kaestner and Lee, 2003; Evans and Lien, 2005; Figlio et al., 2009) and those who find positive effects of significant magnitude (Rosenzweig and Schultz, 1983; Conway and Deb, 2005; Wehby et al., 2009; Habibov and Fan, 2011; Awiti, 2014). This lack of consistency, together with other findings from clinical investigations (McDuffie et al., 1996; Clement et al., 1999; Villar et al., 2001), has led some researchers to question whether the benefits of prenatal care have been "oversold" (Misra and Guyer, 1998).

A recent critique posits that past research has only been able to identify average effects, losing sight of the differential impact of prenatal care in different types of pregnancies (Conway and Deb, 2005). The results of randomized clinical trials, for example, are only valid externally for populations with low-risk pregnancies.¹ Moreover, 2SLS estimates local effects that are valid only for the population marginally affected by changes in the instrument. In addition, Conway and Deb argue that significant differences between healthy and risky pregnancies, when not explicitly accounted for, create a bimodal distribution of errors that downwardly biases the estimated effects of prenatal care.

Another problem with the literature is that, until recently, it almost exclusively focused on developed countries. The effectiveness of prenatal care may be quite different in developing countries, where women are, in general, less informed about the health consequences of certain conditions and behaviors, and have, arguably, fewer resources to address nutritional and hygiene needs. The role of the health care provider, particularly in low-income contexts, may be critical to promote healthy pregnancies and decrease the incidence of LBW. In line with this argument, preliminary evidence for Uruguay shows that the transition from no prenatal care to nine prenatal visits significantly increases the birth weight of the child (Jewell and Triunfo, 2006; Jewell et al., 2007). More recent research based in Argentina, Azerbaijan, Cebu, and Kenya provides evidence of the strong effects of prenatal care (Wehby et al., 2009; Habibov and Fan, 2011; Gajate-Garrido 2011; Awiti, 2014).

This paper analyzes the impact of prenatal care on the likelihood of preterm birth and LBW in a low-income population in Uruguay, a middle-income country in South America with a population of 3.3 million. Our estimation strategy addresses time-invariant, unobserved confounders by exploiting intra-mother variation in prenatal care for women who had at least two births between 1995 and 2011 in the main maternity hospital in the capital city Montevideo (Pereira Rossell Hospital). We also perform several robustness checks that reduce concerns about biases due to time-varying, unobserved heterogeneity and to feedback effects from previous pregnancies.

Our analysis sheds new light on the effectiveness of prenatal care for poor populations in developing countries. The evidence is very timely for Uruguay, given the recent efforts by the Ministry of Public Health (MPH) to improve the coverage of prenatal care in the country. Since 2010, the Uruguayan government has been financially rewarding providers affiliated with the National Social Health Insurance system for increasing the fraction of pregnant mothers who initiate care during the first trimester and attend at least six prenatal visits. While all health-care providers are, in principle, eligible to receive supplemental payments, in general public providers are not receiving the incentive payments because most of the population they serve is not affiliated with the National Social Health Insurance system.² Our results have clear public policy implications. If the State Health Services Administration in Uruguay, which offers health services to low SES women, were to achieve the prenatal care goals proposed by the MPH, the rate of low birth weight in this population would drop by 4 percentage points, from 10% to 6.2%, increasing the number of newborns with adequate birth weight by 800 per year.³ The effects would be even stronger (two additional percentage points) if the goals were aimed at achieving nine rather than six prenatal visits in a full-term pregnancy. The number of children born with adequate weight would increase, in that case, by 1,200 per year.

2. Conceptual Framework and Methodology

The production of birth weight is usually modeled in economics as the result of a process of parental utility maximization. Parents’ utility is a function of their children’s well-being, which depends directly on the children’s health. Conditional on genetic endowments and household resources, parents are indirect producers of children’s health and decide which inputs to invest in to maximize their children’s health status. These inputs include the use of prenatal care, use of substances during pregnancy, exercise, and nutrition, among others (Grossman, 2000). As mentioned earlier, the analyst’s inability to observe the full set of maternal preferences, resources, and information involved in the household production of health may lead to estimation biases. Unobserved characteristics such as the health endowment of the fetus and/or the mother, the mother’s health habits, her propensity to engage in risky behavior, or the extent to which the pregnancy is desired, could bias the estimates if their influence is not taken into consideration.

In this paper, we address this endogeneity using differencing techniques. Our methodology exploits the availability of longitudinal information for the same mother over several pregnancies. The underlying model is of the form: 

where Y_ij reflects the outcome of mother i’s pregnancy j (preterm birth, birth weight, very low birth weight, and low birth weight), j denotes the birth order, CP_ij is an indicator of the adequacy of prenatal care, and X_ij includes other determinants of the newborn’s health (such as mother’s age, education, marital status, tobacco use, body mass index, history of past births, and quarter and year of pregnancy). The term α_i captures the time-invariant unobserved heterogeneity in i, namely, personality characteristics of the mother that affect her habits, her involvement in risky behavior, her health endowment, her knowledge about the benefits of prenatal care, her preferences, and so forth. Finally, ε_i3 is an idiosyncratic error term independent of α_i and other explanatory variables.

A naive estimate of resulting from the regression of the outcome variable Y_ij on the indicator of adequacy of prenatal care CP_ij is potentially biased (even after adjusting for other controls X_ij) if it does not account for the unobserved heterogeneity component, α_i, which is associated both with the explanatory variable of interest and with the dependent variable. This has been a common estimation error in the biomedical literature, which has resulted in unreliable estimates of the effectiveness of prenatal care.

We begin our analysis by projecting deviations in neonatal outcomes on deviations in prenatal care that happen across the same mother’s different pregnancies (adjusting for within variations in other characteristics). In order to eliminate the potential correlation between the mother-specific fixed effect α_i and inputs in the production function of child health, the methodology requires a transformation of the data into within-mother deviations. We work with first differences in our core model⁴, and for sensitivity, we also re-estimate the models using the "within transformation" (fixed effects) and orthogonal deviations (Arellano and Bover, 1995). Once the data are transformed, the model identifies the effect of interest, α₁, by getting rid of the idiosyncratic time invariant term α_i:

The nice feature of this method relative to 2SLS is that it estimates average treatment effects. One of the problems with the extant literature on the effectiveness of prenatal care is that it tends to rely exclusively on the population of compliers, i.e., those who increase their use of prenatal care when confronted with an exogenous policy shock (an increase in health care coverage, for example). These local treatment effects may provide a distorted picture of the effectiveness of prenatal care if the impacts are heterogeneous across the different subpopulations. If policy changes do not modify, in the margin, the behavior of those most likely to benefit from prenatal care, the 2SLS estimates will underestimate its impact. Moreover, even if the group of compliers includes complicated and normal pregnancies, combining them in a single 2SLS estimation may yield bimodal residuals that will result in insignificant estimates. Using a finite mixture model, Conway and Deb (2005) find estimates of prenatal care that have a consistent, substantial effect on normal pregnancies. Using a Monte Carlo experiment, they show that ignoring even a small proportion of complicated pregnancies can cause prenatal care to appear as insignificant.

While fixing biases due to time-invariant, unobserved heterogeneity, the differencing technique proposed in (2) may fail to produce consistent estimates in two scenarios: a) if there are unobserved time-variant shocks associated with both the use of prenatal inputs and birth outcomes (a problem we refer to as time-variant, unobserved heterogeneity); and b) if past birth outcomes affect the current demand for prenatal inputs (a problem we refer to as feedback effects). The former would include any changes in preferences, resources, or information between deliveries that are not captured by the time-variant adjustors used in the analysis. For example, this would be the case if an unobserved negative shock on the fetus’ health endowment decreased the expected prenatal outcomes and led the mother to increase the use of prenatal care. Or, if the government implemented an information campaign that encouraged the use of prenatal care as well as other changes in maternal behavior. The second problem would occur if past shocks affected the contemporaneous demand for inputs. For example, a mother may react to an adverse shock to a previous pregnancy (a pregnancy that ended in a preterm birth or that had some risk of miscarriage) by increasing the demand for inputs or for the quality of inputs in the current pregnancy. In either case, working with deviations will not lead to consistent estimates of the coefficients of interest.

To address these problems Arellano and Bond (1991) devised a difference-GMM technique that uses the level of the explanatory variable lagged two periods as the instrument for the first difference.⁵ This approach is, in principle, able to address feedback effects and endogeneity due to time-variant, unobserved heterogeneity. We initially attempted to take this avenue, but as in Abrevaya (2006), the two-period-lag instruments were too imprecise. Thus, in this paper we proceed to formally address the feedback effects problem, which requires using one-period lags of the explanatory variables as instruments, and discuss later why we think the problem of time-variant, unobserved heterogeneity may be, at most, moderate.

To see the feedback effects problem formally, suppose that the model is as in (1), but past shocks pre-determine the level of inputs in t:

Under this assumption, the first difference transformation in (2) generates an endogenous relationship between the deviations in prenatal care and the differenced error term. We propose to address this feedback problem by running GMM on first differences (difference-GMM) and by using a one-period (and eventually deeper) lag(s) of the predetermined variable as "GMM-style" instrument(s) of the contemporary deviations in that variable (Holtz-Eakin et al., 1988; Arellano and Bond, 1991; Roodman, 2009). Specifically, we use the level of prenatal care in pregnancy j — 1, and deeper lags when available, as instruments for the difference in prenatal care use between pregnancies j and j - 1. The orthogonality conditions in our GMM model are: E[X_i1 (ε_i2 - ε_i1)] = 0 for mothers with two deliveries in the period, E[X_i1 (ε_i2 - ε_i1)] = E[X_i1 (ε_i3 - ε_i2)] = E[X_i1 (ε_i3 - ε_i2)] = 0 for mothers with three deliveries, and so forth with mothers with more than three deliveries.

Our first specification assumes no feedback effects: it relies on the assumption that past shocks are orthogonal to the current demand for prenatal inputs. Next, we allow for feedback effects to play a role and estimate the model using the lagged levels of prenatal care as instruments of the first difference in prenatal care. All the regressions control for the year of birth dummies and compute robust standard errors that are clustered at the mother’s level. We run two specifications of the model: one without adjusting for the duration of the pregnancy, and the other controlling for the number of weeks of gestation at delivery. This latter specification accounts for the fact that the effects of prenatal care on birth weight can occur through the probability of reaching full term. In the GMM specification, we instrument deviations in gestational weeks with two lags of the number of gestational weeks in levels.

3. Data

We analyze births registered in the Perinatal Information System (CLAP-OP-OMS, 1999) of the Pereira Rossell Hospital between 1995 and 2011. Pereira Rossell is the public teaching hospital for the only public university in the country, and is part of the State Health Services Administration (ASSE). The hospital is a referral center for acute care of mothers and children for the whole country, concentrating 70% of the births that take place in public wards in Montevideo, 33% of all births in Montevideo, and 15% of births nationwide. The hospital serves women who have no access to private insurance through their employment or who cannot afford individual insurance. These women are entitled to access to prenatal and obstetric care free-of-charge at public facilities.

The Perinatal Information System at the Pereira Rossell Hospital covers approximately 98.5% of all births that take place in the hospital, according to national birth registries. These data are quite unique in that they allow for identification of mothers over a period of 17 years. The dataset is larger than similar ones used in other medical and epidemiological studies, and provides information on a population of women who have not been studied intensively, with SES, cultural, and geographic differences relative to women in developed countries.

The outcomes in our analysis are preterm birth, birth weight (BW), low birth weight (LBW), and very low birth weight (VLBW). We consider a delivery to be preterm if it occurs before the 37th week of gestation. BW is a continuous variable in grams. LBW is a binary variable that takes the value of 1 if the birth weight is less than 2,500 grams and 0 otherwise. VLBW is a binary variable that takes the value of 1 if the birth weight is less than 1,500 grams and 0 otherwise.

We analyze three alternative measures of prenatal care. Our core measure is based on the Kessner Index, a widely used indicator of adequacy of care (Kotelchuck, 1994). According to Kessner’s criterion, a woman has "adequate" prenatal care if she has her first visit during the first trimester (week 13 or earlier) and has at least nine visits at term, or between 4 and 8 visits in the case of a preterm birth. Prenatal care is deemed "inadequate" if the mother initiates the control visits in the third trimester or if care was initiated before but she has fewer than 4 control visits by the time of delivery, or between 1 and 3 visits when the birth is premature. All other combinations of initiation and visits belong to an "intermediate" category.

The second categorization of prenatal care is based on MPH guidelines established in 2010, which serve as the reference for incentive payments received by providers. The variable takes the value of 1 if the woman initiated prenatal care in the first trimester of her pregnancy and attended at least 6 control visits by the time of delivery. Finally, we analyze the timing of initiation of prenatal care, another variable analyzed in the literature that takes the value of 1 if visits were initiated in the first trimester and 0 otherwise.

We use the following covariates to adjust for mother or pregnancy-specific observed heterogeneity. Tobacco use is an indicator of any smoking, as reported by the mother at her first prenatal visit. Smoking during pregnancy has been associated with low birth weight (Permutt and Hebel, 1989; Veloso da Veiga and Wilder, 2008; Reichman et al., 2009). Previous research has also shown an inverse, U-shaped relationship between maternal age and birth weight: Women who are either under the age of 26 or above 35 have higher rates of low birth weight children compared to other childbearing women (Abel et al., 2002). To capture this pattern, we include five categories of maternal age: under 17, between 17 and 19, between 20 and 34 (optimal age group), between 35 and 39, and over 40. As proxies for socioeconomic status, we consider mothers’ marital status (married, single, cohabitating, other), and education (whether the mother completed primary education, middle school, or high school). Among the mother’s epidemiologic risk factors, we consider body mass index (BMI) before the pregnancy (underweight, overweight, or obese) and the presence of hypertension, preeclampsia, and eclampsia in each pregnancy. Both high BMI and chronic hypertension have been associated with low birth weight (Ehrenberg et al., 2003, Haelterman et al., 1997). Finally, the epidemiology literature shows that the experience of previous births is associated with anatomical changes that may impact the health of a newborn (Khong et al., 2003). Among these variables, we consider the number of prior births, episodes of mortality in prior deliveries, and prior abortions. For biological reasons, girls generally weigh less than boys (Thomas et al., 2000), so we include a dummy variable that equals one if the newborn is a boy.

Because of the differencing methodology used in this paper, we work with a sample of low-SES Uruguayan women who gave birth to at least two children between 1995 and 2011. The data include information about the mother, the pregnancy, and the newborn’s health. Of the 143,228 total births registered in the hospital in the period (about 8,400 births a year), 995 were discarded because of unviable pregnancies (less than 25 weeks of gestation or birth weights below 500 grams), 3,343 were not considered because of multiple pregnancies, and 7,227 were not included because proper identification of the mother was lacking. Of the remaining observations (131,663), we discarded those with inconsistent information or missing values for relevant variables and considered only births to mothers who delivered at least twice during the period (33% of births). Finally, we excluded pregnancies that did not show variation in the use of prenatal care across the same mother.

The final sample comprises 40,729 births. Altogether, there are 17,278 mothers in the sample, of which 12,719 had two births, 3,335 had three, 911 had four, 253 had five, and 60 had six or more. Table 1 presents descriptive statistics for mothers with at least two deliveries in the period of analysis. The first and second columns show means and standard deviations for first pregnancy characteristics, while the third and fourth present data on all pregnancies. The rate of prematurity in all deliveries is 14%, the average birth weight is 3,145 grams, 1.5% of the newborns have very low birth weight, and 10% have low birth weight. Only 13% of all pregnancies have a proper follow-up according to the Kessner criterion, 43.6% are inadequately followed-up, and the rest are in between. Barely 22% of pregnancies meet the prenatal care goals set by the Uruguayan Ministry of Public Health. While half of the women attend the target six control visits by the time of delivery, only 26% initiate care in the first trimester. These figures are quite surprising when considering that pregnancy care is free in Uruguay and that there are few geographic barriers to obstetric care facilities. Similar behavior, however, has been found among persons eligible for public assistance programs in the United States (Currie and Grogger, 2002; Kaestner and Lee, 2003). Most women in the sample are between 20 and 34 years old, unmarried, and have not finished middle school. One out of three mothers smokes during the pregnancy.

 

Table 1. Summary statistics, analysis sample 1995-2011
Mothers with at least two deliveries

Table 1. (continued)

Source: Authors’ estimations. Data from the Perinatal Information System
of Pereira Rossell Hospital (Ministry of Public Health).

 

In Table A1 of the appendix we compare the first pregnancy characteristics of mothers in the analysis sample with those of the excluded mothers (i.e., mothers observed only once during the period). Mothers in the analysis sample are younger, less educated, more likely to be smokers, and less likely to have adequate prenatal care during their first observed pregnancy. On the other hand, they are less likely to have low weight or very low weight children, are less likely to be overweight or obese, and have a lower likelihood of having hypertension and preeclampsia during the pregnancy. While mothers with two or more deliveries are clearly different than mothers with only one delivery, the above comparison does not suggest obvious discrepancies in the quality of pregnancies or in mothers’ social vulnerability between the two samples. For externai validity purposes, we also compare women delivering at Pereira Rossell with low-income women delivering at other public hospitals in Uruguay. We make this comparison for the 2000-2011 period, as we do not have good information on other hospitals before the year 2000.⁶ While the age distribution is similar in both samples, women delivering at Pereira Rossell Hospital are slightly less educated and less likely to be married. They are also more likely to have premature or low birth weight children. This comparison suggests that women in our study sample may be somehow more vulnerable than other low-income women in Uruguay and that our results are more likely to apply to the poorest fractions of the population.

4. Results

4.1 Kessner criterion for impact of adequate prenatal care

Table 2 shows the results of the estimation when the adequacy of prenatal care is defined in terms of the Kessner index. Each column in Table 2 depicts the estimates of a linear regression model on a panel of observations that have been transformed to within-mother first differences. The defined categories of prenatal care already acknowledge that reduced duration of gestation truncates the time available to make visits, thus avoiding a problem of mechanical reverse causation.⁷ Results show large effects of prenatal care on all birth outcomes considered. An adequate use of prenatal care decreases the likelihood of preterm birth by 11.3 percentage points, the likelihood of VLBW by 1.1 percentage points, and LBW by 6 percentage points, and increases birth weight by 149 grams. Even if prenatal care is not fully adequate in the sense of Kessner, women initiating care before the third trimester and showing at least four prenatal visits by the end of the pregnancy (intermediate prenatal care) are 7.6 percentage points less likely to experience a preterm birth and 3.4 percentage points less likely to deliver a baby below 2,500 grams than women with inadequate use. Once we adjust for gestational age (columns 3, 5, and 7), the estimated impact of adequate and intermediate care on LBW and birth weight decreases substantially, suggesting that prenatal care improves birth weight mainly through a reduction in the likelihood of a preterm birth. Our estimates show that 4 percentage points out of the 6-percentage-point decrease in uncontrolled LBW are due to the fall in prematurity.

 

Table 2. Ef fects of adequate prenatal care (Kessner index) on birth outcomes
Estimation using mother-specific first dif ferences (N = 23,451)

Table 2. (continued

Table 2. (continued)

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health). Note: Clustered standard errors in parentheses.
* p < 0.1, ** p < .05, *** p < .01.

 

Both the probability of LBW and the likelihood of preterm birth are higher for teenage mothers (19 years or younger), for mothers with prior obesity, and for mothers with preeclampsia or eclampsia during pregnancy; they are lower for mothers who had a still-birth in a prior pregnancy. LBW is also positively associated with being a smoker, with a first pregnancy, and with a female baby.

We repeat the analysis using within-mother fixed effects and orthogonal differences instead of first differences, and the results are strongly robust to these variations.

4.2 Other measures of prenatal care adequacy: early initiation and MPHc guidelines

Tables 3 and 4 present the findings for alternative measures of adequate prenatal care. Table 3 shows the effects of initiation of care during the first trimester on the likelihood of preterm delivery and LBW. When compared to the results in Table 2, these estimates give a sense of the relative importance of early initiation versus the number of control visits in the overall effect of adequacy of care. Early initiation has some impact on the likelihood that the pregnancy reaches full term, but this impact is much smaller than the aggregate effect of early initiation plus an adequate number of visits: initiation during the first trimester decreases a preterm birth by 1.3 percentage points. This finding suggests that the positive effect of prenatal care on the duration of the pregnancy and on birth weight would be due mostly to an adequate number of visits. It also highlights the importance of quantifying the number of visits in addition to the time of initiation in studies analyzing the effectiveness of prenatal care.

 

Table 3. Ef fects of early initiation of prenatal care on birth outcomes
Estimation using mother-specific first differences

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health).
Note: Ali estimations include the full set of Controls depicted in Table 2. Clustered standard
errors in parentheses. * p < 0.1, ** p< .05, *** p < .01.

 

Table 4. Effects of adequacy of prenatal care according to Uruguayan MPH guidelines
Estimation using mother-specific first differences

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health).
Note: Ali estimations include the full set of Controls depicted in Table 2. Clustered standard
errors in parentheses. * p < 0.1, ** p< .05, *** p < .01.

 

Table 4 shows the effects of compliance with the standards of prenatal care set by the Uruguayan Ministry of Public Health. Complying with the MPH standards decreases the likelihood of prematurity by 4.8 percentage points. The effects on LBW are also statistically significant and large. Initiating care during the first trimester and making at least six visits by the end of the pregnancy reduces the likelihood of LBW by 4 percentage points (a 40% decrease).

The comparison between the Kessner effects and those estimated with the MPH guidelines suggests that increasing the target number of control visits beyond those required by the MPH may lead to more pronounced decreases in the probability of LBW. Note that there is about a 2 percentage-point-difference between the impact of the Uruguayan MPH guidelines on LBW (-0.04) and the coefficient on the Kessner measure of adequate care (-0.06). The additional three visits required in the Kessner measure of adequacy appear to have a large impact on birth outcomes. Moreover, much of the beneficial impact of this increased number of control visits seems to operate through a smaller probability of preterm birth.

5. Robustness Analysis

The main contribution of first differencing techniques is to reduce biases due to time-invariant, unobservable heterogeneity. However, as mentioned before, this technique may fail to produce consistent estimates in two scenarios: a) if there are feedback effects in the decision to use prenatal care; and b) if there are time-variant shocks associated with both the use of prenatal inputs and birth outcomes. As we show below, several checks support the robustness of our core findings even when considering these other confounders.

5.1 Feedback effects in the demand for prenatal care

Results for the feedback effects model are presented in Table 5. The likelihood of a preterm birth in this specification falls by 12.6 percentage points when prenatal care is adequate and by 7.9 percentage points when care is intermediate, relative to inadequate care. These coefficients are very similar to those in the model in first differences, suggesting a small role for feedback effects. Regarding the effects on LBW, the coefficient on adequate care is now -0.058 and that on intermediate care is -0.029. While slightly smaller than the coefficients in the core model, the estimates are still significant, large, and not statistically distinguishable from the magnitudes previously described.

 

Table 5. Effects of adequacy of prenatal care, as defined by the Kessner índex,
on birth outcomes
Difference GMM with feedback ef fects

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health).
Note: All estimations inelude the full set of Controls depicted in Table 2. Clustered standard
errors in parentheses. * p < 0.1, ** p < .05, *** p < .01.

 

5.2 Time-variant heterogeneity

Time-variant shocks would include any changes in preferences, endowments, or information between deliveries that are not captured by the time-variant adjustors included in the analysis. Our models already address time-variant heterogeneity stemming from aggregate shocks and from mother’s health endowments. We capture time-variant aggregate shocks, such as information campaigns, improvements in the quality of prenatal care, or a better economy by introducing year fixed effects.⁸

In our core specification we account for the mother’s pregnancy-specific health endowments by including controls for preeclampsia, eclampsia, and hypertension, the three most important pregnancy-related conditions. To assess whether the estimates are sensitive to maternal health shocks, we rerun the above models excluding these conditions. We find that prenatal care coefficients are not affected by these exclusions (results are available upon request). Overall, we are not particularly concerned about biases due to mother-specific or fetus-specific adverse shocks, because we expect them to shift our estimates towards zero. A less healthy mother or a less healthy fetus would be associated with worse pregnancy outcomes and with more use of prenatal care. In such a case, our findings would be conservative.

We are more concerned about time-variant unobserved heterogeneity not uniform across all women that may potentially inflate our estimates. For example, both the demand for health care and fetus health may be compromised by a negative income shock at the household level. While prenatal care is provided free of charge in Uruguay, there are other costs involved (i.e., transportation, child care) that could decrease its use. Alternatively, women who engage in alcohol or drug consumption between two pregnancies are expected to have poorer birth outcomes and, at the same time, to decrease their demand for prenatal care (due to indolence or fear of such consumption being detected).

While we are not able to address each of these conjectures directly⁹, we check for robustness by rerunning the analysis for different periods of time and different samples of women. Our assumption is that the shocks that are prevalent in an economic downturn are not necessarily the same as shocks occurring in a stable or growing economy.¹⁰ For example, we should expect income shocks to be larger between 2000 and 2005, when Uruguay experienced a deep crisis (income fell in Uruguay from 1999 until 2002, and only recovered to pre-crisis levels in 2005). Furthermore, national trends in substance use show sizeable increases in the use of marijuana and cocaine after 2001, and discrete increases in the use of alcohol after 2006. We thus expect biases due to changes in substance use to be more critical in the 2000s than in the last years of the 1990s. To encompass these trends, we compare first-difference estimates for the periods 1995-1999, 2000-2005, and 2006-2011 and display them in Table 6. The first three columns compare prenatal care coefficients associated with the likelihood of preterm birth. The coefficients are fairly consistent for the 1995-1999 and 2006-2011 periods, and slightly higher for the 2000-2005 period (the recession and recovery timeframe). The last three columns report the effects of prenatal care on low birth weight without controlling for gestational age. As expected, we find larger coefficients for the period after 2000, suggesting potential biases due to income or substance use shocks. Still, the effects for 1995-1999 are large and statistically significant. Intermediate use of prenatal care decreases low birth weight by 3.2 percentage points and adequate use decreases it by 6 percentage points, relative to inadequate use.

 

Table 6. Effects of prenatal care (Kessner) on birth outcomes by delivery period
First-difference estimates

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health).
Note: Ali estimations include the full set of Controls depicted in Table 2 only for selected
outcomes. Clustered standard errors in parentheses. * p < 0.1,, ** p < .05, *** p < .01.

 

Finally, we distinguish women who increased their use of prenatal care over two subsequent pregnancies from women who decreased their use of prenatal care. Our conjecture is that women who increase prenatal care use over time are likely to experience different types of shocks than women who reduce its use (that is, the shocks experienced by those reducing use are not necessarily the reversal of shocks that result in positive changes). Any difference could be attributed either to differential biases due to distinct types of shocks or to asymmetric effects. A finding of symmetric effects, on the other hand, would support the consistency of our estimates. The results of this comparison are shown in Table 7. While the coefficients are slightly higher for positive changers, the differences do not exceed 1 percentage point. This similarity in the estimates across women with opposing trends in the use of care is an additional indication of the robustness of our findings.

 

Table 7. Effects of prenatal care (Kessner) on birth outcomes, by sign of change in
care use over time. First-difference estimates
Mothers with at least two deliveries and changes in prenatal care between 1995 and 2011

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira Rossell
Hospital (Ministry of Public Health).
Note: Ali estimations include the full set of Controls depicted in Table 2. Clustered standard
errors in parentheses. * p < 0.1, ** p < .05, *** p < .01.

 

5.3 Effectiveness of prenatal care over successive pregnancies

Our core estimates reflect average impacts of adequate vs. inadequate prenatal care on birth outcomes. An interesting question is whether this impact changes as the number of visits gets larger. Women could learn from past experiences and improve outcomes in future pregnancies, leaving lower margins for a direct effect of prenatal care in later pregnancies. We assess this question by rerunning the first-difference estimates only on first and second births, on second and third births, and on third and fourth births. Our findings (Table A2 of the appendix) confirm that the effects of prenatal care become smaller as the parity gets larger. Adequate prenatal care, according to the Kessner index, reduces the likelihood of preterm birth by 6.4 percentage points in the fourth pregnancy, versus a decrease of 11.8 percentage points in a second pregnancy. On the other hand, adequate prenatal care drops the likelihood of a low birth weight by 3.9 percentage points in a fourth pregnancy vs. 6.1 percentage points in a second pregnancy.

5.4 Marginal benefits of successive prenatal care visits

The analysis using the Kessner measures as indicators of adequacy of care has the advantage of adjusting prenatal care for the pregnancy’s gestational length and for the timing of initiation of care. One drawback, however, is that it does not allow for a marginal evaluation of the benefits of successive visits. We explore this issue by running a semi-parametric regression of the number of prenatal care visits on birth outcomes, only for full-term pregnancies and conditioning on the number of gestational weeks to avoid endogenous mechanical relationships. The results are reported in Table A3 in the appendix. We find that prenatal care visits increase birth weight linearly, with changes of around 20 grams for every couple of visits. In contrast to the first four visits, the fifth and sixth visit contribute significantly to reducing the likelihood of low birth weight. Further visits continue to reduce the probability of low birth weight, but in smaller magnitudes.

6. Discussion and concluding Remarks

In this paper we estimate the impact of prenatal care on infant health in a developing country by exploiting intra-mother variations in pregnancy inputs and outcomes. We analyze a longitudinal panel of births that took place between 1995 and 2011 in the largest public maternity ward in Uruguay, representative of the population of lower socioeconomic status in the country. The data set is quite unique in its ability to identify mothers throughout the period, its large size, and its reliance on clinical history (rather than self-reports).

Using within-mother first differences, we find that adequate use of prenatal care, as defined by early initiation and a minimum number of visits throughout the pregnancy, has a significant positive impact on neonatal outcomes. Our estimates indicate that the probability of low birth weight falls by between 3 and 6 percentage points (from a baseline prevalence of 10%), depending on the minimum number of control visits that are considered "adequate" (six or nine respectively), and the likelihood of preterm birth decreases by between 5 and 11 percentage points respectively (from a baseline of 14%).

By using first-differences estimation and difference-GMM estimation we avoid biases due to time-invariant, unobserved heterogeneity and feedback effects from prior births. We cannot claim causality, though, as these techniques do not account for time-variant shocks in preferences or health over different pregnancies. Still, we find our results to be strongly consistent when we run the analysis over different time periods (arguably subject to varying time shocks¹¹) and when we restrict the estimation to mothers experiencing only positive variation or only negative variation in prenatal care use across pregnancies. Furthermore, one of the main sources of time-varying concerns-health shocks to the fetus or the mother-would bias our results towards zero.

Our estimates are larger than those obtained in other international studies using two-stage least squares (2SLS) and exploiting health policy changes as instruments in developed countries (Grossman and Joyce, 1990; Kaestner, 1999; Kaestner and Lee, 2003; Evans and Lien, 2005). We see two potential explanations for the discrepancy with prior studies. First, prenatal care may have a significantly larger effect among lower-SES women in a developing country than among women in a high-income country. In fact, our findings are in line with those of Wehby et al. (2009) for Argentina and Habibov and Fan (2011), Gajate-Garrido (2011), and Awiti (2014) for Azerbaijan, Cebu, and Kenya. Second, our statistical approach avoids the problems inherent in prior approaches that rely on instrumental variables to measure only local average treatment effects.

One potential concern with our analysis has to do with biased attrition of mothers from the Pereira Rossell Hospital. Mothers who leave the sample (deciding to deliver their second, third, or fourth child at a hospital other than Pereira Rossell) may differ from mothers who remain in the sample. To address this concern, we explore changes in the fraction of nationwide births occurring at the Pereira Rossell Hospital between 1995 and 2011. We find two significant events that affected the number of deliveries at the hospital during the analysis period: the 2002 economic crisis and implementation of the health care reform in 2008. In 2002, when Uruguay experienced a severe economic crisis, the Pereira Rossell Hospital reached its peak coverage of 35% of all deliveries in Montevideo. Because only formal workers were entitled to receive care from private providers, the crisis shifted relatively better-off women from the private to the public sector. Once the economy began to recover in 2004, this fraction gradually returned to pre-crisis levels. The second important event was the implementation of the health care reform in 2008, which extended private health care coverage to spouses and dependents of formal workers. Two years after this event, births at Pereira Rossell Hospital fell to a minimum of 30% of births in Montevideo, the reform having shifted relatively better-off women out of the sample. While the impact of these sample variations is difficult to assess directly, the robustness of our estimates across different periods (Table 6) mitigates these concerns somewhat. In effect, prenatal care estimates are very similar across the 2000-2005 and the 2006-2011 periods, when attrition was working in opposite directions.

One of the reasons that our analysis focused on the Pereira Rossell Hospital is that this university hospital has good quality of information and a high level of coverage when compared to national birth registries, and it covers 70% of all births in the public sector in Montevideo. The external validity of our findings is closely related to the representativeness of this hospital’s patients of the population of low-income mothers in Uruguay. We showed that women delivering at Pereira Rossell are slightly less educated and less likely to be married than other low-socioeconomic status women in the country, suggesting that our analysis applies to the poorest women in the country.

Several factors mediate the effectiveness of prenatal care on neonatal outcomes: the physician’s influence on the woman’s behavior during pregnancy (protein supplementation, recommendations to abstain from the use of alcohol, tobacco, and other drugs, among other behaviors), detection and treatment of conditions associated with low birth weight (syphilis, anemia, hypertension, urinary infections), the ability to detect early labor and prevent premature birth through bed rest and pharmacological agents, and preparation for delivery. In Uruguay, all prenatal care is provided by obstetricians. The usual paraclinical studies required in low-risk pregnancies are: 1) complete hemogram during the first visit and in the third trimester; 2) urine tests in all visits; 3) glycemia analysis in the first visit; 4) VDRL in the first visit and in the third trimester; 5) blood type and RH; 6) serology for toxoplasmosis; 7) testing for Hepatitis B antigen; 8) serology for Chagas disease; 9) HIV serology; 10) three obstetrical ultrasounds (one per trimester); 11) screening for gestational diabetes; 12) urine culture in the second and/or third trimester. Unfortunately, in this paper we are unable to disentangle the channels through which a prenatal care visit translates into better health outcomes at birth. Future research should explore these mechanisms.

Our findings have direct policy implications for developing countries, and particularly for the Uruguayan population. If the low-income population in Uruguay were to achieve the prenatal care goals set by the MPH, the impact on neonatal health would be substantial. However, only 22% of the population in our sample complied with the standards suggested by the MPH and only 13% displayed adequate use of prenatal care as defined by the Kessner criterion. These low figures, together with the magnitude of our findings, present a strong case for the design of policies aimed at encouraging the use of prenatal care in low-SES populations. The financial incentives that the Uruguayan Ministry of Health currently offers to health care providers for achieving a set of prenatal care goals may not be sufficient to promote significant changes among lower-income populations. Any policy aimed at improving prenatal care among low-SES women, either through financial incentives to providers or directly through conditional cash transfers or other demand-centered initiatives, must focus differentially on the segments of the population least likely to use prenatal care adequately, i.e. teenagers or older women, women who are single, uneducated, or those who have many children. Conditional cash transfer programs, which provide financial aid to low-income individuals upon compliance with health and education goals, should also align their incentives to elicit better compliance with prenatal care optimal standards.¹²

NOTES

1 Most randomized trials have been conducted in populations with low-risk pregnancies.

2 Uruguay has a mixed health insurance system. The population formally engaged in the labor market and their families are covered by a National Social Health Insurance that provides services either through private or public providers. Most beneficiaries are enrolled with private providers. The public provider, the State Health Services Administration (ASSE), covers low-income populations that are not formally inserted in the labor market. This population shows the lowest rates of initiation of care during the first trimester and the highest non-compliance with recommended standards of care.

3 This back-of-the-envelope calculation is based on the approximately 20,000 births in the public health care sector in Uruguay per year.

4 This transformation involves subtracting the observation in j — 1 from that in j for the same mother.

5 Abrevaya (2006) recognized these problems and attempted a similar correction while analyzing the effect of tobacco use on birth outcomes.

6 Results are available upon request.

7 Women who deliver before the 37^th week of gestation are categorized as compliers with adequate standards of care if they have completed the number of visits that corresponds to that gestation week.

8 This approach is valid only if the time-variant effects are constant across all women in the population.

9 Abrevaya (2006) recognized these problems while analyzing the effect of tobacco on birth outcomes, and suggested using a difference-GMM model with the level of the explanatory variable lagged two periods as the instrument for the first difference. We initially attempted to take this avenue, but as in Abrevaya the instruments were too imprecise.

10 If these shocks occur and disappear at the same rate and if their effects are symmetric, they should not be a cause of concern. However, there may be periods in which the proportion of women experiencing these shocks may outnumber the fraction overcoming them (and vice versa).

11 One of these shocks may have to do with the conditional cash transfer programs implemented in Uruguay after the 2002 crisis (PANES). Amarante et al. (2011) show that PANES had some positive impact on infant birth weight. However, the effects did not seem to be related to improved prenatal care.

12 There is some evidence in Uruguay that the PANES, an emergency financial aid plan that took place after the 2002 crisis, had some positive impacts on infant birth weight (Amarante et al., 2011). However, the effects did not seem to work through improved prenatal care.

 

REFERENCES

Abel, E.L., M. Kruger and L. Burd (2002), "Effects of maternal and paternal age on Caucasian and Native American preterm births and birth weights," American Journal of Perinatology 19: 49-54.

Abrevaya, J. (2006), "Estimating the effect of smoking on birth outcomes using a matched panel data approach," Journal of Applied Econometrics 21: 489-519.

Amarante, V., M. Manacorda, A. Vigorito and M. Zerpa (2011), "Social assistance and labor market outcomes: Evidence from the Uruguayan PANES," Inter-American Development Bank.

Arellano M. and S. Bond (1991), "Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations," Review of Economic Studies 58(2): 277-97.

Arellano M. and O. Bover (1995), "Another look at the instrumental variable estimation of error-components models," Journal of Econometrics 68(1): 29-51.

Awiti, J. O. (2014) "A multilevel analysis of prenatal care and birth weight in Kenya," Health Economics Review 4(1): 1-16.

Barros, F.C., Z.A. Bhutta, M. Batra, T.N. Hansen, C.G. Victora and C.E. Rubens (2010), "Global report on preterm birth and stillbirth (3 of 7): Evidence for effectiveness of interventions," BMC Pregnancy Childbirth 10, Suppl 1:S3.

Behrman, J.R. and M.R. Rosenzweig (2005), "Returns to birth weight," Review of Economic and Statistics 86(2): 586-601.

Black, S., P. Devereux and K. Salvanes (2007), "From the cradle to the labor market? The effect of birth weight on adult outcomes," Quarterly Journal of Economics 122(1): 409-39.

Boardman, J.D., D.A. Powers, Y.C. Padilla and R.A. Hummer (2002), "Low birth weight, social factors, and developmental outcomes among children in the United States," Demography 39(2): 353-63.

Brien, M.J. and C.A. Swann (2001), "Prenatal WIC participation and infant health: Selection and maternal fixed effects." Manuscript, SUNY-Stony Brook Department of Economics.

Case, A., A. Fertig and C. Paxson (2004), "The lasting impact of childhood health and circumstance." Center for Health and Wellbeing, Woodrow Wilson School, Princeton University.

Centro Latinoamericano de Perinatología y Desarrollo Humano (CLAP-OPS/OMS), available at http://www.paho.org/clap/index.php?option=com_content&view=article&id=84&Itemid=242&lang=es.

Clement, S., et al. (1999), "Does reducing the frequency of routine antenatal visits have long term effects? Follow up of participants in a randomized controlled trial," British Journal of Obstetrics and Gynecology 106(4), 367-70.

Conway, K.S. and P. Deb (2005), "Is prenatal care really ineffective? Or, is the ‘devil’ in the distribution?" Journal of Health Economics 24(3): 489-513.

Currie, J. and B. Madrian (1999), "Health, health insurance, and the labor market," in O. Ashenfelter and D. Card, eds., Handbook of Labor Economics, Volume 3, 3309-415.

Currie, J. and E. Moretti (2005), "Biology as destiny? Short and long-run determinants of intergenerational transmission of birth weight" National Bureau of Economic Research Working Paper 11567.

Currie, J. and J. Grogger (2002), "Medicaid expansions and welfare contractions: Offsetting effects on prenatal care and infant health?" Journal of Health Economics 21:313-36.

Ehrenberg, H.M., L. Dierker, C. Milluzzi and B.M. Mercer (2003. "Low maternal weight, failure to thrive in pregnancy, and adverse pregnancy outcomes," American Journal of Obstetrics and Gynecology 189:1726-30.

Evans, W. N. and D.S. Lien (2005), "The benefits of prenatal care: Evidence from the PAT bus strike," Journal of Econometrics 125(1): 207-39.

Figlio, D., S. Hamersma and J. Roth (2009), "Does prenatal WIC participation improve birth outcomes? New evidence from Florida" Journal of Public Economics 93(1): 235-45.

Gajate-Garrido, G. (2013), "The impact of adequate prenatal care on urban birth outcomes: An analysis in a developing country context," Economic Development and Cultural Change 62(1): 95-130.

Grossman, M. and T.J. Joyce (1990), "Unobservables, pregnancy resolutions, and birth weight production functions in New York City," Journal of Political Economy 98: 983-1007.

Grossman, M. (2000) "The human capital model," in A. J. Culyer and J. P. Newhouse, eds., Handbook of Health Economics Vol. 1A. Amsterdam: North-Holland, Elsevier Science, 347-408.

Habibov, N.N. and L. Fan (2011), "Does prenatal healthcare improve child birthweight outcomes in Azerbaijan? Results of the National Demographic and Health Survey," Economics & Human Biology 9(1): 56-65.

Haelterman, E., G. Breart, J. Paris-Llado, M. Dramaix and C. Tchobroutsky (1997), "Effect of uncomplicated ypertension on the risk of small-for-gestational age birth," American Journal of Epidemiology 145: 689-95.

Holtz-Eakin, D., W. Newey and H.S. Rosen (1998), "Estimating vector autoregressions with panel data," Econometrica 56: 1371-95.

Institute of Medicine (1985), Preventing low birth weight. Washington, D.C.: National Academy.

Jewell, R. T. (2007), "Prenatal care and birthweight production: Evidence from South America," Applied Economics 39(4): 415-26.

Jewell, R.T. and P. Triunfo (2006), "The impact of prenatal care on birthweight: The case of Uruguay," Health Economics 15: 1245-50.

Jewell, R.T., P. Triunfo and R. Aguirre (2007), "El peso al nacer de los niños en la principal maternidad del Uruguay: 1995 a 2004," Revista Desarrollo y Sociedad 59:1-20, Universidad de los Andes, Colombia.

Joyce, T.J. and M. Grossman (1990), "Pregnancy wantedness and the early initiation of prenatal care," Demography 27:1-17.

Kaestner, R. and W.C. Lee (2003), "The effect of welfare reform on prenatal care and birth weight," National Bureau of Economic Research Working Paper 9769.

Kaestner, R. (1999) "Health insurance, the quantity and quality of prenatal care, and infant health," Inquiry 36(2): 162-75.

Khong, T.Y., E.D. Adema and J.J.H.M. Eerwich (2003), "On an anatomical basis for the increase in birth weight in second and subsequent born children," Placenta 24: 348-53.

Kleinman, J.C. and S.S. Kessel (1987), "Racial differences in low birth weight: Trends and risk factors," New England Journal of Medicine 317:749-53.

Kotelchuck, M. (1994), "An evaluation of the Kessner adequacy of prenatal care index and a proposed adequacy of prenatal care utilization index," American Journal of Public Health 84: 1414-20.

McCormick, M.C. (1985), "The contribution of low birth weights to infant mortality and childhood morbidity," New England Journal of Medicine 312: 82-90.

McDuffie, R., et al. (1996), "Effect of frequency of prenatal care visits on perinatal outcome among low-risk women," The Journal of the American Medical Association 275: 847-51.

Ministry of Public Health (MPH) (2010), Instructivo de metas asistenciales JUNASA_MSP (Health care goals manual), at www.msp.gub.uy.

Misra, D. and B. Guyer (1998), "Benefits and limitations of prenatal care: from counting visits to measuring content," The Journal of the American Medical Association 279: 1661-2.

Permutt T. and J.R. Hebel (1989), "Simultaneous-equation estimation in a clinical trial of the effect of smoking on birth weight," Biometrics 45: 619-22.

Petrou, S., T. Sach and L. Davidson (2001), "The long-term costs of preterm birth and low birth weight: Results of a systematic review," Child: Care, Health and Development, 27: 97-115.

Reichman, N.E., H. Corman, K. Noonan and D. Dave (2009) "Infant health production functions: What a difference the data make?" Health Economics 18(7): 761-82.

Roodman, D. (2009), "How to do xtabond2: An introduction to difference and system GMM in Stata," Stata Journal 9: 86-136.

Rosenzweig, M.R. and T.P. Schultz (1982), "The behavior of mothers as inputs to child health: The determinants of birth weight, gestation, and rate of fetal growth," in V.R. Fuchs, ed., Economic Aspects of Health. Chicago: University of Chicago Press.

Rosenzweig, M.R. and T.P. Schultz (1983), "Estimating a household production function: Heterogeneity, the demand for health inputs, and their effects on birth weight," Journal of Political Economy 91: 723-46.

Rosenzweig, M.R. and T.P. Schultz (1988), "The stability of household production technology: A replication," Journal of Human Resources 23: 535-49.

Thomas, P., J. Peabody, V. Turnier and R.H. Clark (2000), "A new look at intrauterine growth and the impact of race, altitude, and gender," Pediatrics 106: E21.

Veloso da Veiga, P. and R. Wilder (2008), "Maternal smoking during pregnancy and birth weight: A propensity score matching approach," Maternal and Child Health Journal 12: 194-203.

Villar, J., et al. (2001), "WHO antenatal care randomized trial for the evaluation of a new model of routine antenatal care," Lancet 357 (9268), 1551-1564.

Wehby G.L., J.C. Murray, E.E. Castilla, J.S. Lopez-Camelo and R.L. Ohsfeldt (2009), "Quantile effects of prenatal care utilization on birth weight in Argentina," Health Economics 18(11): 1307-21.

 

APPENDIX

 

Table A1. Comparison of mothers in analysis sample (with at least two births
in 1995-2011) with mothers in excluded sample (only one birth in the period)

Mother’s characteristics at first observed pregnancy only

Table A1. (continued)

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira
Rossell Hospital (Ministry of Public Health).
a. * p < 0.1, ** p < .05, *** p < .01.

 

Table A2. Effects of prenatal care (Kessner) on birth outcomes for different
number of pregnancies

Selected outcomes

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira
Rossell Hospital (Ministry of Public Health).
Note: All estimations include the full set of controls depicted in Table 2. Clustered
standard errors in parentheses. * p < 0.1, ** p < .05, *** p < .01.

 

Table A3. Effects of prenatal care on birth outcomes for categorical ranges of
prenatal care visits

Only full-term pregnancies and selected outcomes

Source: Authors’ estimations. Data from the Perinatal Information System of Pereira
Rossell Hospital (Ministry of Public Health).
Note: All estimations include the full set of controls depicted in Table 2. Clustered
standard errors in parentheses. * p < 0.1, ** p < .05, *** p < .01.