Recent Trends in Obstetric Practices, Gestational Age, and U.S. Birth Weight

Abstract

Birth weight in the United States declined substantially during the 1990s and 2000s. We suggest the declines were likely due to shifts in gestational age resulting from changes in obstetric practices. Using restricted National Vital Statistics System linked birth/infant death data for years 1990–2013, we analyze trends in obstetric practices, gestational age distributions, and birth weights among first-birth singletons born to U.S. non-Hispanic white, non-Hispanic black, and Latina women. We use life table techniques to analyze the joint probabilities of gestational age-specific birth and gestational age-specific obstetric intervention (i.e., induced cesarean delivery, induced vaginal delivery, not induced cesarean delivery, and spontaneous vaginal delivery) to fully document trends in obstetric practices by gestational age. We use simulation techniques to estimate counterfactual changes in birth weight distributions if obstetric practices did not change between 1990 and 2013. Results show that between 1990 and 2013 the likelihood of induced labors and cesarean deliveries increased at all gestational ages, and the gestational age distribution of U.S. births significantly shifted. Births became much less likely to occur above gestational week 40 and much more likely to occur during weeks 37–39. Overall, nearly 18% of spontaneous vaginal births at later gestational ages were replaced with births occurring at earlier gestational ages from obstetric interventions. Results suggest that if rates of obstetric practices did not change between 1990 and 2013, then average U.S. birth weight would have increased over this time. Findings strongly indicate that recent declines in U.S. birth weight were due to increases in induced labor and cesarean delivery at select gestational ages.

Since the 1990s, secular trends in U.S. births have been characterized by a large decline in mean birth weight and an uptick in the incidence of low birth weight (Martin et al. 2017). Average birth weights stabilized in the 2010s but have not increased. These recent trends have alarmed health researchers because low birth weight is often associated with adverse birth outcomes and poor health in early life (Aizer and Currie 2014; Lantos and Lauderdale 2011; Paneth 1995). Moreover, the trends might also have long-term consequences for U.S. population health. Life course scholars have long noted strong ties between early-life conditions and later-life health (Barker 2012; Ben-Shlomo & Kuh 2002; Berkman 2009; Hanson and Gluckman 2008). Declining birth weight in the United States might therefore lead to adverse health outcomes in both early life and later adulthood.

The recent declines in U.S. birth weight have puzzled researchers, who have sought to explain the trends as outcomes of changes in maternal characteristics and behaviors as well as possibly reflecting changes in gestational ages of U.S. pregnancies, which have been significantly altered by rising rates of cesarean deliveries and induced labor (Lantos and Lauderdale 2011; MacDorman et al. 2006, 2010; Martin et al. 2017). Yet extant evidence suggests that declines in U.S. birth weight are not accounted for by the changes in gestational age distributions, by changes in maternal characteristics, or by changes in obstetric practices (Catov et al. 2015; Donahue et al. 2010; Morisaki et al. 2013; Zhang et al. 2010). Evidence, in fact, suggests birth weight for gestational age is higher among cesarean deliveries and among births from induced labor than among births from spontaneous labor and vaginal deliveries (Catov et al. 2015). Thus, the declines in U.S. birth weight are anomalies and are suspected to reflect reduced fetal growth independent of gestational age.

Existing analytic approaches, however, have not fully considered how compositions of U.S. births at each gestation have been altered by changes in obstetric practices. The full effect of these changes on birth weight cannot be discerned by conventional statistical approaches that compare mean birth weight by gestational age in one time period to mean birth weight at the same gestational ages in another time period. This is because gestation-specific birth weight distributions depend on pregnancies lasting to a given gestational age. As such, one must compare the gestation-specific birth weights after accounting for the likelihood of a birth occurring at that gestational age. These likelihoods have changed substantially over time as a result of increases in labor inductions and cesarean deliveries, which have dramatically shifted when and how U.S. births occur. These changes have fundamentally transformed the composition of births across gestational ages in ways that have likely shifted U.S. birth weights downward. Only by attending to all of these changes, and how the changes are associated with one another, can we accurately estimate the effects of changing obstetric practices on trends in U.S. birth weight.

We use year-specific life tables to analyze changes in the joint probabilities of gestation-specific birth and gestation-specific obstetric intervention status (i.e., induced labor with cesarean delivery, induced labor with vaginal delivery, labor not induced with cesarean delivery, and labor not induced with vaginal delivery) to fully account for changes in the conditional probabilities of birth by a given obstetric intervention status. We then use simulation techniques to explore how these changes have affected birth weight distributions in the United States. We also use decomposition techniques to illustrate how changes in U.S. birth weight have been affected by (a) changes in gestational age exposures and (b) changes in gestational age compositions. Results suggest that recent U.S. birth weights have been affected by dramatic shifts in gestational ages, which have been driven by increases in labor induction and cesarean deliveries at increasingly select gestational ages. In 1990, cesarean deliveries accounted for about 25% of singleton first-births and they were common across a wide range of gestational ages. By 2013, cesarean deliveries accounted for over 31% of singleton first-births and the greatest absolute changes had occurred during gestational weeks 37–39. Similar changes in the timing and frequency of labor inductions occurred. In 1990, about 12% of singleton first-births were induced and the likelihood of induction spanned a wide range of gestational ages. By 2013, over 29% of singleton first-births were induced and the greatest increases had occurred during gestational weeks 37–40. These dramatic shifts have significantly altered the composition of U.S. births at particular gestational ages, and have likely affected distributions of birth weight in substantial ways. Indeed, results from simulations suggest that U.S. birth weight would have continued to increase between 1990 and 2013 if rates of inductions and cesarean deliveries had remained at 1990 levels.

Background

Average birth weight in the United States declined steadily across the 1990s and 2000s and has remained relatively stable since 2007 (Martin et al. 2017). Declines in average U.S. birth weight have been similar to declines reported in many other OECD member countries (OECD 2017). Investigations into the U.S. birth weight declines have thus far been unable to provide a satisfactory explanation for the trends, despite changes in several factors associated with birth weight (Catov et al. 2015; Donahue et al. 2010; Martin et al. 2017; Moriaski et al. 2014). For example, gestational age distributions among U.S. births shifted dramatically across the 1990s and 2000s. Between 1990 and the mid-2000s, preterm birth prevalence (less than 37 weeks) increased from about 10.6% to nearly 13% and then steadily declined to 11.4% in 2013. More recently, both early-term births (37 and 38 weeks) and post term births (42 or more weeks) declined. These changes at the low and high ends of the gestation distribution were coupled with dramatic increases in full term births (weeks 39–40) and indicate that U.S. birth weights have become increasingly influenced by term births (37–41 weeks) (Martin et al. 2017). Strong links between gestation and fetal growth suggest that these changes in the gestation distribution should likely have corresponded with increases in average birth weights. Indeed, intrauterine growth curves show substantive differences in birth weight between early full term (37–38 weeks) and late-term (41+ weeks) infants. For example, between 37 and 41 weeks, the 10^th percentile of birth weight among male infants has been documented to increase 500 grams (Olsen et al. 2010). Yet, average U.S. birth weights declined at all gestational ages, suggesting that the overall birth weight declines might have occurred independently of shifts in gestation.

Several maternal characteristics and behaviors in the United States have also changed in recent decades, indicating that the health of the childbearing population may have changed in important ways that can affect gestational age and birth weight (VanderWeele, Lantos, and Lauderdale 2012). In general, U.S. mothers are older and more racially/ethnically diverse. The average age of first-time mothers increased from 24.2 in 1990 to 26.3 in 2014 and the proportion of births to teenage mothers substantially declined (Mathews and Hamilton 2002, 2016). Age of the mother is strongly associated with birth weight, with younger and older mothers more likely to give birth to a low birth weight child (Restrepo-Méndez et al. 2015). In 1990, 64% of mothers were non-Hispanic white and 16% were non-Hispanic black, but by 2013 the percent decreased to 54% and 15%, respectively (Martin et al. 2017). Simultaneously, the percent of births to Latina mothers rose from 15% in 1990 to 23% in 2013. In 2013, low birth weight incidence rates were higher for non-Hispanic black women (13% of all births) than for non-Hispanic white and Latina women (7% of all births, each) (Martin et al. 2015).

In some respects, prenatal health behaviors among expectant mothers have improved considerably over time, while other behaviors may have worsened. For example, rates of smoking during pregnancy declined (Curtin and Mathews 2016), which it is a strong predictor of low birth weight (Kleinman and Madans 1985). Chomitz et al. (1995) find that mothers who smoke are twice as likely to give birth to a low birth weight infant than their nonsmoking counterparts. Conversely, excessive maternal weight gain, a major risk factor for obstetric intervention and predictor of fetal growth, has become more common in the United States (Goldstein et al. 2017). Between 1990 and 2013, the percentage of all mothers who gained over 40 pounds during pregnancy rose from 16% to 21% (Martin et al. 2007, 2013). Excessive gestational weight gain is associated with an elevated risk for cesarean delivery (Durie et al. 2011), and women who are overweight or obese are more likely to deliver via cesarean than women with normal weights (Poobalan et al. 2009).

Prenatal care expanded in the late-1990s and more expectant mothers in the United States began receiving care much earlier in their pregnancies (Kogan et al. 1998; Martin et al. 2005). Prenatal care is related to decreased risk of both preterm birth (Ickovicks et al. 2007) and low birth weight (Arima et al. 2009). Improvements in medical technologies and expansions to prenatal care at earlier gestations can better identify high-risk pregnancies. Obstetric interventions among these pregnancies have been found to increase infant survival and improve other early-life outcomes (Lantos and Lauderdale 2011). Thus, recent expansions of prenatal care may increase the link between obstetric interventions and preterm birth by increasing the ability to identify high-risk pregnancies.

Much attention has been given to recent trends in labor inductions and cesarean deliveries, especially as they might relate to birth timing and birth weight trends (Catov et al. 2015; Donahue et al. 2010; Martin 2014; Moriaski et al. 2014; VanderWeele et al. 2012). Rates of cesarean deliveries among all births increased from 22.7% in 1990 to 32.7% in 2013 (Osterman and Martin 2014a), and rates of induced labor more than doubled between 1990 and 2013 (Osterman and Martin 2014b). The increasing use of obstetric interventions may be responsive to the increasing ability to identify and intervene on high-risk pregnancies, although rising rates of obstetric interventions might also reflect trends in medical indications of risk during labor, maternal requests, and subjective indications of risk (i.e., “medically elective” interventions on pregnancies thought to be “high risk” due to maternal characteristics, nonreassuring fetal status, and/or labor arrest [Miesnik and Reale 2007]). Medical indications of risk are indicators of a medically necessary obstetric intervention, and arise most often from multiple births, malpresentation of the fetus, and/or other labor complications (Barber et al. 2011; Spong et al. 2012). Maternal requests for interventions have received much attention from researchers and media, but no evidence supports the suggestion that maternal requests occur frequently enough to affect trends in obstetric practices. Instead, subjective indications are thought to be most driving the increased use of obstetric interventions among U.S. pregnancies (Barber et al. 2011; Declercq et al. 2006; Gamble et al. 2007; MacDorman et al. 2008; Miesnik and Reale 2007; Weaver et al. 2007). Menacker (2006) for instance, reported increasing rates of cesarean deliveries among low-risk women in the United States regardless of age, race, or ethnicity; and Spong et al. (2012:9) noted that subjective indications for cesarean deliveries have been rising, “despite the fact that the number of women classified as high risk has not increased concomitantly.”

There are clear differences in obstetric intervention rates by gestational age and birth weight. In general, cesarean deliveries are most common at younger gestations while rates of induction are high among pregnancies at older gestations (Davidoff et al. 2006). High rates of cesarean deliveries among preterm births are largely driven by evidence suggesting improved infant survival rates among cesarean deliveries as well as lower risk of stillbirth (Ananth and Vintzileos 2011). Yet among preterm births with no indicated risk, MacDorman et al. (2010) finds higher mortality risk among births via cesarean deliveries. Further, the American College of Obstetricians and Gynecologists (ACOG) guidelines permit physicians and/or birth attendants a great deal of discretion in their decision to obstetrically intervene in a pregnancy, whereas the international guidelines from the World Health Organization (WHO) use strong language to discourage the use of obstetric interventions (Betran et al. 2016; Spong et al. 2012). The impression that obstetric interventions may reduce adverse birth outcomes and improve infant survival, coupled with the permissive and discretionary approach to U.S. obstetric interventions, may partly be responsible for the rising use of labor induction and cesarean deliveries among U.S. births, irrespective of risk or presentations of medical indications (Lantos and Lauderdale 2011).

Previous investigations into U.S. birth weight trends have been unable to fully explain the declines as outcomes of changes in obstetric practices, gestational age distributions, or maternal behaviors and demographic characteristics of the childbearing population (Donahue et al. 2010; Hong and Lee 2014; Morisaki et al. 2014). However, existing approaches have not fully considered how increases in cesarean deliveries and induced labors at increasingly select gestational ages have changed the composition of U.S. births across gestational age, which, in turn, have dramatically changed birth weight distributions in the United States.

Current Aim

We investigate recent trends in U.S. birth weight and estimate how increases in labor inductions and cesarean deliveries have likely changed (1) the gestational age distribution of U.S. births and (2) the birth weight distributions at each gestational age. We contend that compositional changes in birth are largely, if not entirely, responsible for the downward trend in average U.S. birth weights across the 1990s and 2000s. Specifically, we posit that the increasing use of obstetric interventions have disproportionately occurred at select gestational ages, dramatically shifting when U.S. births take place. Although obstetric interventions may be medically necessary in some instances (e.g., medically indicated high-risk pregnancies), they have been increasingly used in the United States at gestations where risk is relatively low (e.g., weeks 37–39) and among pregnancies with no indicated risks (MacDorman et al. 2006, 2010; Spong et al. 2012). Increases in these obstetric interventions have been common among pregnancies that likely would have resulted in births at older gestational ages and with higher birth weights if the obstetric interventions did not occur. These changes have altered both the distribution of births across gestational ages as well as the birth weight distributions within each gestation.

We explore these processes and estimate their effects on birth weight trends in the United States across several analytic steps. First, we document recent trends in U.S. birth weight among first-birth singletons. We limit our analytic sample to first-birth singletons to account for the confounding effects of multiple births and parity on risk for obstetric interventions and on birth weight (Blondel et al. 2002; Guise et al. 2010). We also limit our analytic sample to births among women who are likely healthier than the general U.S. childbearing population. Specifically, we limit our analyses to births with non-missing data on a number of key risk factors for high-risk pregnancies, poor maternal health, and labor complications. Second, we use Ordinary Least Squares (OLS) regression to estimate how U.S. birth weight trends across all years 1990–2013 were associated with changes in maternal characteristics and behaviors, gestational ages, and obstetric intervention status (i.e., onset of labor not induced with vaginal delivery, onset of labor induced with vaginal delivery, onset of labor not induced with cesarean delivery, and onset of labor induced with cesarean delivery). We fit a series of models to estimate trends in mean birth weight and demonstrate that accounting for changes in maternal characteristics and behaviors, gestational age, and obstetric practices in these models fail to account for the recent declines in U.S. birth weight. We do so to illustrate how conventional statistical approaches are ill-equipped to identify how changes in obstetric practices have affected trends in U.S. birth weight. We then contrast the gestational age distributions of U.S. first-birth singletons in 1990 and 2013 as well as the gestational age-specific rates of induced labors and cesarean deliveries in 1990 and 2013. We use survival analyses to estimate the joint probabilities of obstetric practices across all gestational ages. We generate gestational age-specific life tables for U.S. births in 1990 and 2013 to illustrate how changes in obstetric practices have shifted the timing and composition of U.S. births across these years¹. Next, we use simulations to generate 2013 birth weight distributions at each gestational age under an alternative scenario in which obstetric practices in the United States had not changed from 1990. Finally, we decompose the difference between the observed 2013 birth weight distributions and the simulated 2013 birth weight distributions to show that both the shift in birth timing as well as the composition of births at each gestational age contributed to birth weight declines in the United States.

Analytic Strategy

Data

We examine data from the National Vital Statistics Systems restricted linked birth/infant death data for the years 1990 through 2013 (NCHS n.d.). These data include all births occurring within the United States in each year, with the exception of the years 1992–1994 because data were not collected in those years and 2006 because the restricted version of the data is not available for that year. The data were restricted to include only first-birth singletons among U.S.-born non-Hispanic white, non-Hispanic black, and Latina mothers not missing data on gestational age, birth weight, or any additional maternal or birth characteristic listed below. This analytic sample contains 1,001,976 births in 1990 and 1,130,812 births in 2013, and is composed of 23,027,689 births in all years between 1990 and 2013.

We code obstetric intervention status into four groups: induced labor with cesarean delivery, induced labor with vaginal delivery, labor not induced with cesarean delivery, and labor not induced with vaginal delivery. Where noted, “intervention” encompasses any of the three possible obstetric interventions and “spontaneous vaginal” indicates births that had no obstetric interventions. Due to small cell sizes, births occurring earlier than 34 weeks of gestation are grouped together into “less than 34” (<34) and births occurring later than 43 weeks are grouped together into “44 weeks or more” (44+). We code maternal age into eight groups: less than 15 years, 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, and 45 years and older. We code marital status as married and unmarried because no additional marital statuses are collected on birth certificates. Month of prenatal care initiation is coded into monthly categories ranging from month 0 (indicating that care began prior to the pregnancy) through month 9.

Beginning in 2003, birth certificates include detail on trimester of tobacco use. To standardize the measure across all years, we code tobacco use dichotomously where a one indicates any use of tobacco products during the pregnancy. We code gestational weight gain into three groups: less than 21 pounds, 21 to 39 pounds, and 40 pounds or more. We code diabetes into a dichotomous variable where a one indicates that the mother had diabetes before and/or during the pregnancy. Finally, we code hypertension into a dichotomous variable where a one indicates that the mother had either chronic or pregnancy associated high blood pressure.

Methods

To predict yearly average birth weight among U.S. first-birth singletons, we fit a series of Ordinary Least Squares (OLS) regression models. We employ step-wise model building techniques to assess how year-specific birth weight averages change with the inclusion of model covariates. The baseline model includes fixed effects indicators of single calendar year: 1990, 1991, …, 2012, 2013. In the second model, we include maternal characteristics of race/ethnicity, five-year maternal age groups, marital status, maternal behaviors of month of prenatal care initiation and tobacco use during pregnancy, and other important measures associated with fetal growth and gestational age such as excessively high or excessively low gestational weight gain, diabetes, and hypertension. In the third model, we add indicators of obstetric interventions: no interventions, induced labor with cesarean delivery, induced labor with vaginal delivery, and not induced labor with cesarean delivery. In the final model, we include weekly indicators of gestational age.²

We use life table methods to estimate the conditional probabilities of birth via each obstetric intervention status at each gestational age. We define the radix (l₀) as number of pregnancies beginning at gestational week <34 (the first exposure time) in 1990 and in 2013. In all estimates we use a radix of 1,000,000 pregnancies. The probability of a birth occurring in each gestational age x is defined as q_x, estimated from the year-specific gestational week birth distributions in the official NVSS data. We define l_x as the number of pregnancies from the 1,000,000 radix that are carried to gestational week x, and we define d_x as the number of births that occur in gestational week x. Finally, among births in gestational week x, we define c_x as the probability of labor not induced with cesarean delivery, i_x as the probability of induced labor with vaginal delivery, ic_x as the probability of induced labor with cesarean delivery, and s_x as the probability of labor not induced with vaginal delivery.

We used the proportion of births per gestational week in each year (q_x) to calculate the number of births per gestational week (d_x), and we used the gestational week-specific prevalence for the probability of obstetric intervention status – cesarean delivery (c_x), induced labor (i_x), induced labor with cesarean delivery (ic_x), and spontaneous labor with vaginal birth (s_x) – to calculate the joint probabilities of giving birth in each gestational week via each obstetric practice.

We then simulated 2013 gestational age-specific birth weights that combined the observed 2013 gestational age-specific maternal characteristics and behaviors with the 1990 gestational age-specific rates of obstetric interventions. To do this, we fitted OLS regressions for all 1990 and 2013 births, separately, at each gestational week, predicting birth weight distributions from all covariates noted in the final OLS model above. We used the observed 1990 and 2013 gestational age-specific prevalence of each covariate in the model as the proportionate exposures. Using the estimated 2013 effect sizes, the 1990 exposures to obstetric interventions, and the 2013 exposures to all other covariates, we then simulated the 2013 birth weight distributions for each gestational age:

$$\begin{gathered} SimWg{t}_{G{A}_i}={\alpha }_{G{A}_{2013}}+IndVa{g}_{G{A}_{1990}}{\widehat{\beta }}_{G{A}_{2013}}+IndCe{s}_{G{A}_{1990}}{\widehat{\beta }}_{G{A}_{2013}}+Ce{s}_{G{A}_{1990}}{\widehat{\beta }}_{G{A}_{2013}} \\ +{\mathbf{\text{X}}}_{\mathit{G}{\mathit{A}}_{\mathbf{2013}}}{\mathbf{\text{β}}}_{\mathit{G}{\mathit{A}}_{\mathbf{2013}}}+e_{G{A}_i} \end{gathered}$$

where i indicates birth i in the radix of 1,000,000 births 1, …, 1,000,000 within each GA and GA indicates gestational age < 34 weeks, 34 weeks, …, 43 weeks, 44+ weeks; 𝛼 indicates the model constant; IndVag indicates induced labor with vaginal delivery, IndCes indicates induced labor with cesarean delivery, and Ces indicates not induced labor with cesarean delivery; ${\mathbf{\text{X}}}_{\mathit{G}{\mathit{A}}_{\mathbf{2013}}}$ indicates a set of covariates associated with 2013 birth weight and ${\mathbf{\text{β}}}_{\mathit{G}{\mathit{A}}_{\mathbf{2013}}}$ indicates the set of estimated coefficients associated with each covariate X; and $e_{G{A}_i}$ indicates the error term for birth i in gestational week GA. The mean squared error for each gestational week’s OLS model in 2013 was used to inform the amount of random error simulated for each birth weight distribution. The resulting counterfactual is the estimated 2013 U.S. birth weight distribution for all first-birth singletons if rates of induced labors and cesarean deliveries in 2013 were equal to rates in 1990.

Finally, we use decomposition techniques with two-component solutions (Kitagawa 1955) to decompose the difference between the observed 2013 U.S. birth weight distributions and the counterfactual 2013 U.S. birth weight distributions into (a) changes in exposure to gestational ages, or GA(Exp) (“gestational age exposure”), and (b) changes in gestational age composition, or GA(Wgt) (“gestational age composition”):

$$\text{GA}(\text{Exp})=\sum \left[\text{GA}{(\text{Exp})}_{Ob{s}_i}-\text{GA}{(\text{Exp})}_{Si{m}_i}\right]$$

$$\text{GA}(\text{Wgt})=\sum \left[\text{GA}{(\text{Wgt})}_{Ob{s}_i}-\text{GA}{(\text{Wgt})}_{Si{m}_i}\right]$$

Where

$$\text{GA}{(\text{Exp})}_{Ob{s}_i}=\left(\frac{{\text{Wgt}}_{Ob{s}_i}+{\text{Wgt}}_{Si{m}_i}}{2}\right){\text{Exp}}_{Ob{s}_i}$$

$$\text{GA}{(\text{Exp})}_{Si{m}_i}=\left(\frac{{\text{Wgt}}_{Ob{s}_i}+{\text{Wgt}}_{Si{m}_i}}{2}\right){\text{Exp}}_{Si{m}_i}$$

$$\text{GA}{(\text{Wgt})}_{Ob{s}_i}=\left(\frac{{\text{Exp}}_{Ob{s}_i}+{\text{Exp}}_{Si{m}_i}}{2}\right){\text{Wgt}}_{Ob{s}_i}$$

$$\text{GA}{(\text{Wgt})}_{Si{m}_i}=\left(\frac{{\text{Exp}}_{Ob{s}_i}+{\text{Exp}}_{Si{m}_i}}{2}\right){\text{Wgt}}_{Si{m}_i}$$

Where i indicates gestational age < 34 weeks, …, 44+ weeks; Obs indicates observed and Sim indicates simulated; and Wgt indicates birth weight and Exp indicates the proportion of births.

The two-component solution estimates the amount of the birth weight difference between the observed 2013 distribution, 2013_O, and the simulated 2013 distribution, 2013_S, that is attributable to changes in birth timing (i.e., differences in gestational age exposure between 2013_O and 2013_S) and attributable to changes in gestational age-specific birth weight (i.e., differences in gestational age composition between 2013_O and 2013_S)³.

Results

Table 1 shows characteristics of first-birth singletons born to U.S. non-Hispanic white, non-Hispanic black, and Latina women in 1990 and 2013 as well as the absolute differences between the characteristics in 1990 and 2013. The average birth weight of U.S. first-birth singletons in this analytic sample was 3314.5 grams in 1990, but by 2013 the average had declined over 67 grams to 3247.2 grams (among all births, the birth weight declined 69 grams from 3345.3 in 1990 to 3276.5 in 2013).

Table 1.

Characteristics of U.S. Singleton First-births, 1990 and 2013

	1990	2013	Change
Birth Weight (grams)	3314.5	3247.2	−67.3	***
Mother’s Age (years)
<15	0.6%	0.2%	−0.4%	***
15–19	23.6%	15.3%	−8.3%	***
20–24	30.9%	30.5%	−0.4%	***
25–29	27.8%	26.9%	−0.9%	***
30–34	13.1%	18.9%	5.9%	***
35–39	3.6%	6.6%	3.0%	***
40–44	0.4%	1.4%	1.0%	***
45–49	0.0%	0.1%	0.1%	***
Gestational Weight Gain
<21 lb.	16.1%	19.5%	3.4%	***
21–40 lb.	64.4%	53.5%	−10.9%	***
41+ lb.	19.5%	27.0%	7.5%	***
Month Prenatal Care Began
0	0.8%	1.1%	0.3%	***
1	17.6%	5.3%	−12.4%	***
2	41.3%	35.0%	−6.2%	***
3	21.0%	35.6%	14.6%	***
4	8.6%	10.9%	2.3%	***
5	4.9%	4.9%	0.0%
6	2.9%	2.9%	0.0%	*
7	1.8%	2.0%	0.2%	***
8	0.9%	1.5%	0.6%	***
9	0.3%	0.8%	0.5%	***
Tobacco Use	16.1%	7.6%	−8.4%	***
Mother’s Education
Less than Hight School	20.3%	13.0%	−7.2%	***
High School or Equal	38.3%	24.5%	−13.7%	***
Some College	21.5%	30.1%	8.6%	***
Bachelor’s Degree +	20.0%	32.3%	12.4%	***
Unmarried	31.7%	47.9%	16.2%	***
Mother’s Race/Ethnicity
Latina	7.9%	22.7%	14.8%	***
Non-Hispanic Black	16.0%	14.7%	−1.3%	***
Non-Hispanic White	76.1%	62.6%	−13.4%	***
Diabetes	2.0%	4.9%	2.9%	***
Hypertension	5.1%	8.0%	2.9%	***
Labor Onset and Delivery
Not Induced, Vaginal	66.4%	48.8%	−17.6%	***
Not Induced, Cesarean	21.1%	21.7%	0.6%	***
Induced, Vaginal	8.6%	20.0%	11.4%	***
Induced, Cesarean	4.0%	9.5%	5.5%	***
Gestational Age (weeks)
<34	3.0%	3.2%	0.2%	***
34	1.3%	1.3%	0.0%	*
35	2.1%	2.1%	−0.1%	***
36	3.5%	3.7%	0.2%	***
37	6.2%	7.6%	1.4%	***
38	11.7%	14.2%	2.5%	***
39	20.0%	26.7%	6.7%	***
40	23.2%	23.6%	0.4%	***
41	16.2%	11.3%	−4.8%	***
42	6.9%	3.2%	−3.7%	***
43	2.7%	1.5%	−1.2%	***
44+	3.1%	1.6%	−1.5%	***

Data come from National Vital Statistics Systems restricted linked birth/infant death data. Analytic Sample N(1990): 1,001,976; N(2013): 1,130,812.

^*p < .05;

^**p < .01;

^***p < .001

We also see differences in maternal, behavioral, and birth characteristics between 1990 and 2013, many of which are reflective of more general trends among all U.S. births (Martin et al. 2017). Maternal age has increased considerably in the United States, and an increasing proportion of first-birth singletons have been born to more highly educated women, unmarried women, and Latina women. Mothers of first-birth singletons in 2013 were less likely to have smoked during pregnancy than in 1990, were more likely to have gained excessively low or excessively high gestational weight, and more likely to have had diabetes or hypertension. Compared to 1990, labor onset in 2013 for first-birth singletons was much more likely to have been induced and delivery was more likely to have been cesarean. Gestational ages of first-birth singletons in 2013 were much less likely to have been over 40 weeks and much more likely to have been 37–39 weeks than in 1990.

Figure 1 shows yearly estimates of change in average birth weight among U.S. first-birth singletons from the baseline OLS model and the final OLS model that accounts for the maternal characteristics, obstetric practices, and gestational ages in Table 1. The trends estimated from the final model are similar to those estimated from the baseline model, both showing large declines in U.S. birth weight across the 1990s and 2000s. These OLS results suggest that U.S. birth weight trends occurred independently of the changes in maternal characteristics and behaviors, changes in obstetric interventions, and changes in gestational age distributions across these years. The finding is consistent with evidence from existing analyses suggesting that U.S. trends in birth weight may reflect changes in fetal growth independent of gestational age (Catov et al. 2015; Donahue et al. 2010; Moriaski et al. 2014).

Figure 1.

Change in U.S. Birth Weight, 1990 to 2013, Estimated Using OLS Regression

Baseline: single-year fixed effects

Full Model: single-year fixed effects + race, mother’s age, married, months of prenatal care, smoking during pregnancy, weight gain during pregnancy, diabetes, hypertension, gestational age, induced labor with vaginal delivery, induced labor with cesarean delivery, labor not induced with cesarean delivery.

Figure 2 plots the gestational age distributions of first-birth singletons in 1990 and 2013 (top) and the absolute difference in the probability of birth at each gestation in 1990 and 2013 (bottom). Three patterns in these distributions are worth noting. First, there are no substantive differences between the 1990 and 2013 rates of preterm births (<37 weeks) among U.S. first-birth singletons. This pattern is quite different from changes in gestational age distributions among all U.S. births, among which preterm births have increased considerably (Davidoff et al. 2006; Martin et al. 2017). The difference is likely due to the analytic sample being restricted to first-birth singletons among U.S.-born women who have no missing data. The sample contains no multiple births, likely contains few high-risk pregnancies, and thus likely is less affected by changes in obstetric interventions among preterm births (Lantos and Lauderdale 2011; MacDorman et al. 2010; VanderWeele et al. 2012).

Figure 2.

Gestational Age Distribution of U.S. First-birth Singletons (top) and Change in Probability of Birth by Gestational Age, 1990 and 2013 (bottom)

Second, compared to 1990, a far greater proportion of first-birth singletons in 2013 were born in gestational weeks 37–39. In 1990, about 37.9% of first-birth singletons were born between gestational weeks 37 and 39 and in 2013 the number had increased to about 48.5%. Also, the modal gestational age of first-birth singletons dropped from week 40 in 1990 to week 39 in 2013. And third, the proportionate increase in first-birth singletons between gestational weeks 37 and 39 was offset by sizeable decreases in births above gestational week 40. In 1990, nearly 29% of first-birth singletons were born above gestational week 40, but in 2013 only 18% were born at these older gestations.

Figure 3 shows the relative breakdown of all first-birth singletons in 1990 and 2013 by gestation and by obstetric intervention status. It plots the joint probabilities across gestations via each obstetric intervention status such that the three 1990 lines (gray) sum to 1.0 and the three 2013 lines (black) sum to 1.0. The lines are plotted separately by obstetric intervention status in order to highlight the 1990-to-2013 difference in the probabilities that first-birth singletons were born at a given gestational age via a specific intervention status. For instance, spontaneous vaginal births accounted for about 18% fewer births in 2013 than in 1990 (i.e., about 66% of all first-birth singletons in 1990 were spontaneous vaginal vs. about 48% in 2013). We see that the likelihood of a spontaneous vaginal delivery declined slightly among preterm births and declined substantially among gestational ages 39+ weeks. The modal gestational age of spontaneous vaginal births shifted down from week 40 in 1990 to week 39 in 2013. Overall, we see a striking reduction in the likelihood of spontaneous vaginal births from 1990 to 2013 and also see that nearly all of the change occurred among gestational weeks 39+. For cesarean deliveries among labors that were not induced we see very different patterns. Compared to 1990, we observe little substantive difference in the overall likelihood that 2013 first-birth singletons were delivered cesarean among labors that were not induced (i.e., about 21% in 1990 vs. about 22% in 2013). However, we do observe a sizeable shift in the gestational age distribution of these deliveries. Compared to 1990, the 2013 likelihood for a cesarean delivery that was not induced was higher for all gestational ages < 40 weeks and substantially lower for all gestational ages 40+ weeks. Thus, while the relative frequency of births from these deliveries did not change much between 1990 and 2013, the timing of this obstetric intervention concentrated these births to occur at younger gestational ages. Finally, we see in Figure 3 that births from labor induction accounted for a far greater share of first-birth singletons in 2013 than in 1990, and also that the gestational timing of these practices changed substantially over this time. In 1990, labor inductions accounted for only about 12% of U.S. first-birth singletons and it was most common among full term and post term gestations. Specifically, about 77% of labor inductions among first-birth singletons in 1990 were among gestational weeks 39+ with the modal gestational age 41 weeks. In 2013, labor was induced for 29% of first-birth singletons and was concentrated among gestational weeks 38–41, as nearly 80% of labor inductions occurred in those weeks, with a modal gestational age of 39 weeks.

Figure 3.

Joint probability of birth by gestational age and obstetric intervention status

Taken together, the patterns in Figure 3 strongly suggest that changes in obstetric interventions were the underlying cause for the changes in the gestational age distributions observed in Figure 2. The reductions in first-birth singletons above gestational weeks 40 occurred almost entirely among spontaneous vaginal births, and the declines were offset by increases in induced labor and cesarean deliveries concentrated around weeks 37–40. The increased use of these obstetric interventions changed how U.S. births were delivered, shifted when labor occurred, and likely altered the composition of births delivered full term and post term. This last change is vital for understanding how changes in obstetric practices – and their effects on changing gestational age distributions – likely drove the downward trends in U.S. birth weight. Although the OLS results suggest that changes in the gestational age distributions were unrelated to recent trends in U.S. birth weight, there are reasons to suspect the two trends are related. In fact, the evidence presented in Figure 2 and Figure 3 suggests that recent trends in U.S. birth weight may be directly shaped by the shifts in gestational age resulting from changes in obstetric interventions. First, the downward shift in the gestational age distribution was not uniform and, in fact, was overwhelmingly concentrated among term births (gestational ages 37–41 weeks). The shifts away from gestational weeks 41+ and toward weeks 37–39 likely were not due to worsening maternal health, increases in high-risk pregnancies, or trends in other medical indications of risk. Changes in these risk factors would most likely manifest as shifts among preterm births (Lantos and Lauderdale 2011; VanderWeele et al. 2012). The evidence in Figure 2 indicates the opposite of this, with no substantive changes occurring among preterm births and nearly all changes occurring among full term and post term births. Second, as shown in Figure 3, the gestational age shifts were intrinsically tied to specific changes in obstetric practices. From 1990 to 2013, induced labor increased threefold during gestational weeks 37–40, which increasingly concentrated U.S. first-birth singletons to be born during at these gestational ages – a shift that also occurred among cesarean deliveries. The joint probabilities show that combined effect of these obstetric interventions was to dramatically reduce the likelihood that U.S. pregnancies were carried to late-term gestational ages. And thus, third, the changes in obstetric practices have not only affected the timing of labor onset and the method of birth delivery in the United States, but they likely altered the composition of births at each gestational age. The effect of these compositional changes on birth weight trends arises from time-to-event phenomena and cannot be identified with conventional statistical techniques such as OLS regression. We next present results from simulation and decomposition exercises that explore how changing compositions of births across gestational ages may have been responsible for recent trends in U.S. birth weight.

Table 2 presents the differences in mean birth weight among U.S. first-birth singletons born in 1990 versus born in 2013, born in 1990 versus simulated in 2013, and simulated in 2013 versus born in 2013. Also shown in Table 2 is the proportion of the difference estimated to be attributable to differences in gestational age exposure and attributable to differences in gestational age weight (i.e., compositional changes within gestational ages). The observed mean birth weight among U.S. first-birth singletons in 2013 was 67.3 grams less than the observed mean birth weight in 1990. Decomposition results suggest about 41% of the difference was due to shifts in the gestational age distribution of births (27.6 grams) and 59% was due to compositional changes of birth weight within gestational ages (39.7 grams). Results from the simulated birth weights in the counterfactual 2013 indicate that average birth weight among U.S. first-birth singletons would have increased nearly 12 grams if rates of obstetric interventions had remained at the 1990 levels. Thus, the observed mean birth weight among U.S. first-birth singletons in 2013 was 79.1 grams less than the simulated mean birth weight in the counterfactual 2013. Decomposition results suggest that about 34% of the difference is attributable to shifts in gestational age exposure (27 grams) and 66% is attributable to compositional changes of birth weight within gestational ages (52.1 grams). The birth weight differences attributable to gestational age exposure and attributable to composition are plotted across all gestational ages in Figure 4.

Table 2.

Differences in birth weight 1990 vs 2013 (observed), 1990 vs. 2013 (simulated), 2013 (observed) vs. 2013 (simulated), and proportion of change due to change in gestational age exposure and due to change in gestational age birth weight.

		Decomposition
	Birth Weight Δ	GA Exposure	GA Weight
1990 (O) to 2013 (O)	−67.3	0.41	0.59
1990 (O) to 2013 (S)	11.7	0.00	1.00
2013 (S) to 2013 (O)	−79.1	0.34	0.66

Note: (O) indicates observed, (S) indicates simulated.

Figure 4.

Birth Weight Difference, 2013 Observed versus 2013 Simulated, Attributable to Changes in Gestational Age Exposure (top) and Changes in Gestational Age Composition (bottom)

The top panel of Figure 4 shows the difference between the observed 2013 birth weight and the simulated 2013 birth weight attributable to differences in gestational age exposures. We see that the observed 2013 birth weight increased due to more births occurring in gestational weeks 37–39, but that the observed 2013 birth weight also decreased due to fewer births occurring in gestational weeks 41+. The cumulative weight lost due to fewer births occurring in gestational weeks 41+ (393 grams) is greater than the cumulative weight gained by more births occurring in gestational weeks 37–39 (366 grams). And thus, the higher rates of obstetric interventions in 2013 dramatically changed gestational age exposure of U.S. births and decreased birth weight by 27 grams. The bottom panel of Figure 4 shows the difference between the observed 2013 birth weight and the simulated 2013 birth weight attributable to differences in birth weight composition within gestational ages. There are two key points worth emphasizing in the figure. One, birth weight in the observed 2013 distribution is lower at every gestational age than in the simulated 2013 distribution. This finding is consistent with existing works suggesting birth weight declines at all gestational ages (Catov et al. 2015; Donahue et al. 2010; Moriaski et al. 2014). However, two, the differences vary in size across gestational age in two distinct and important ways. First, lower mean birth weight among preterm births in the observed 2013 distribution is consistent with the birth weight effects of rising rates of obstetric interventions among high-risk preterm pregnancies. Specifically, very early preterm births (< 34 gestational weeks) in the observed 2013 distribution weigh on average about seven grams less than the counterfactual 2013 distribution. Such a large reduction in birth weight likely reflects, on the one hand, the increased use of prenatal technologies to identify and monitor at-risk pregnancies and, on the other hand, the rising use of cesarean deliveries to intervene on these pregnancies in hopes of improving infant and mother outcomes (Lantos and Lauderdale 2011). Second, gestation-specific differences in birth weight between the observed 2013 distribution and the counterfactual 2013 distribution exhibit a distinct pattern across gestational weeks 34 and 42. The differences in birth weight are greatest among full term and post term gestations, which are the gestations at which pregnancies have been most affected by changes in obstetric practices. The finding is consistent with our suspicion that the composition of births at these gestational ages have been substantially altered by the increased use of elective obstetric interventions, and that these compositional changes have had profound effects on birth weight trends in the United States.

Discussion

Average birth weight in the United States substantially declined across the 1990s and 2000s. The decline may have consequences for the immediate and future health of the U.S. population due to the strong ties between gestational growth, birth weight, and health in early life and adulthood. The decline has been thought to be an anomaly and perhaps reflective of reduced fetal growth. Changes in some maternal characteristics, improvements in prenatal care across the 1990s and 2000s, and shifts in gestational age distributions likely have changed U.S. birth weight distributions. Yet, existing work suggests that the birth weight declines have been independent of changes in obstetric interventions (i.e., cesarean delivery or induction of labor), changes in gestational age distributions, and changes in maternal behaviors (Donahue et al. 2010; Moriaski et al. 2013; Zhang et al. 2010).

Existing research has overlooked the important ways that rising use of obstetric interventions has changed the composition of births at any given gestational week. Previous approaches have not been able to measure these changes because gestation-specific birth weight distributions are strongly shaped by the conditional likelihood of pregnancies being carried to each gestational age. This point is similar to time-to-event processes in mortality studies, where the risk of a time-specific event, in this case birth, is conditional upon surviving to that age (Vaupel and Yashin 1985). We account for this conditional likelihood by using classic demographic techniques that allow us to estimate (1) conditional likelihoods of gestation-specific event (i.e., birth), (2) the gestation-specific likelihood of an important exposure (i.e., obstetric intervention), and thus (3) the joint probability of an event happening by way of the exposure. Only by using these joint probabilities can we fully account for changes in the event, changes in the exposure, and changes in how the exposure conditional on the event affects changes in birth weight distributions.

We find that rates of labor induction and cesarean deliveries both increased between 1990 and 2013 for all gestational ages and, importantly, that the increases were especially high among full term births. We also find that the gestational age distribution has changed in two significant ways. First, a decreasing percentage of births are occurring after gestational week 40. Second, the gestational age distribution has compressed around week 39, and around term pregnancies more generally. Overall, 18% of all births in 2013 were reallocated from spontaneous vaginal births at older gestational ages (i.e., 40+) to earlier gestational ages from obstetric interventions. When we simulated the birth weight distribution in 2013 in the absence of increasing intervention rates, we found that average U.S. birth weight would have increased between 1990 and 2013 by nearly 12 grams. Finally, decomposition results indicate that the reduction in birth weight reflected both a shift in birth timing as well as compositional changes in births.

Our findings have two key takeaways for public health and obstetrics and for demography, respectively. The first is that changes in obstetric practices likely account for the recent decline in U.S. birth weight. Our findings suggest that the declines in U.S. birth weight are tied to dramatic shifts in gestational ages of U.S. births as well as compositional changes with gestational ages. Our findings further show that these shifts were due to the increased use of labor inductions and cesarean deliveries at increasingly select gestational ages, with the greatest increases having occurred among full term pregnancies. Though some of these interventions are necessary, evidence suggests that many may be elective and increasingly so (Ashton 2010; Miesnik and Reale 2007; Simpson and Thorman 2005). Our results point to how changing cultural norms, preferences, and institutional practices can have serious and wide-ranging consequences at the population level, as well as potentially for women’s and infant’s health. Although there have been efforts to raise awareness about the potential dangers of overusing obstetric interventions, the rates in the United States remain persistently high. Future research ought to consider how norms, preferences, and practices at various levels of U.S. social life work to influence and even encourage obstetric interventions.

Our second major takeaway relates to methods. Specifically, the advantages to studying trends in birth outcomes from a demographic perspective. This story is largely about compositional changes and multiple scales of risk. Although various methodological approaches can control for varying compositions of who receives an exposure, most approaches cannot adequately account for how changing exposures across multiple time metrics affect who remains at risk of receiving the exposure in the future. Fundamentally, the question about trends in U.S. birth weight is best approached as a time-to-event (i.e., survival) question. Further, it is a survival question about changes involving two different time metrics – gestational age and time period – and how differential selection processes affect trends in each. All of these changes affect the likelihood of birth at a given gestation, the likelihood of obstetric intervention, and gestation-specific birth weight conditional upon an intervention. In short, the associations between likelihood of obstetric interventions and birth weight at specific gestational ages have been changing over time as a result of interdependent processes. These conditional probabilities are beyond the capabilities of standard statistical approaches to fully measure, so we drew from classic demographic approaches to estimate the joint probabilities of interventions and gestational age. We then used simulations to model the counterfactuals and decomposition methods to attribute differences to both changes in exposure and changes in composition. Such demographic methods may be useful for future studies related to trends in obstetrics and fertility because these techniques are well-equipped to model conditional likelihoods relevant to many fertility questions.

Our analyses are limited in a number of ways. One limitation is that there is no way to measure changes in obstetric decision-making processes in official administrative data. We assume that our coded intervention statuses adequately captures important variation in these processes as it relates to shifts in gestational ages and birth weights. Yet, heterogeneity within categories of obstetric intervention status remains (e.g., labor not induced with cesarean delivery contains a) women who entered labor spontaneously and who delivered cesarean and b) women who had a scheduled cesarean). Relatedly, the administrative data allows us to only measure changes in an exposure (i.e., intervention) and outcomes from this exposure (i.e., shifts in gestational age distributions). The data do not allow us to measure the possible reasons for the increased likelihood of the exposures. It is possible that the increases in labor inductions and cesarean deliveries among U.S. first-birth singletons were responding to a number of unobserved risk factors that we do not measure in our analyses. That said, we believe this is improbable as the analytic sample is limited to first-birth singletons and we control for changes in key demographic and maternal behavioral risk factors. Further, the greatest changes in obstetric interventions were observed among births at full term gestational ages. It is unlikely that increased use of obstetric interventions at these gestational ages (where “high-risk” pregnancies are uncommon) in this analytic sample (which is limited to first-birth singletons) reflected trends in unobserved risk factors.

Individual measures of maternal behaviors, characteristics, and other risk factors for obstetric interventions were also quite limited in the NVSS data. Potentially key details about maternal health risk factors related to obstetric decisions may be missing. A final limitation is that we analyze a limited analytic sample restricted to singleton first-births. If we include multiple births or multiparous births then we risk confounding the birth weight effect of the exposure (i.e., intervention) because the intervention and birth weight may be driven by a common cause (e.g., multiple birth). Yet, while the analytic sample reduces bias in estimates of the birth weight effect of obstetric interventions, it also reduces our ability to generalize our findings to all U.S. births.

Although results from this study must be interpreted with these limitations in mind, they suggest that the puzzling declines in U.S. birth weight may, in fact, have a simple answer. In the United Sates, increases in obstetric interventions at select gestations have dramatically shifted how and when labor occurs. These shifts have significantly altered the composition of births at each gestational age. It is likely that increasingly elective use of labor induction and planned cesarean deliveries means that some births may not realize their full birth weight potential. Indeed, we find that if U.S. pregnancies and births in 2013 had experienced the rates of intervention in 1990, then average birth weight of first-birth singletons would have increased across this time period. Identifying the underlying contributors to the decline is U.S. birth weight is just the first step toward improving maternal and child health. Research ought to explore the institutional and social mechanisms that underlie the persistently high rates of labor induction and cesarean deliveries among U.S. births.