Study oversight

The National Institute of Health and Welfare (https://thl.fi/en/web/thlfi-en/about-us) in Finland, in accordance with national data protection legislation, authorized the release of the anonymized data for this register-based study and provided ethical approval. Individual informed consent was not required, as only deidentified data were used without any additional patient contact. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist).

Study population

We use prospective data registered in FMBR, covering all births within the period January 1, 2006, to December 31, 2010, in Finland. Antenatal care in Finland is tax-paid and offered without fees to all women residing in the country. Standardized information is recorded prospectively during antenatal, obstetric, and neonatal care for all pregnant women and neonates in all clinics and hospitals in Finland. These data are copied after delivery and sent to FMBR for archiving. The FMBR includes data on all mothers and their live-born infants and stillbirths weighing ≥500 g or with a gestational age of ≥22 weeks.

The total number of births recorded during the study period was 289,722. We excluded births <24 gestational weeks, multiple births (because of risk of decreased birth size), and stillbirths. Fig 1 shows the entire study population from which we drew the subsample of matched participants for the main analysis. Results for the entire sample are presented in the supporting information.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Pregnancies, ACT treatment, and timing of births.

ACT, antenatal corticosteroid therapy.

https://doi.org/10.1371/journal.pmed.1002746.g001

ACT

ACT administration during the current pregnancy was recorded in FMBR as either received treatment or not. The recommended treatment consisted of betamethasone 12 mg administered twice, 24 hours apart. According to national guidelines, delivery can be postponed through tocolytic therapy to enable corticosteroid prophylaxis and referral to tertiary hospital, if necessary.

Birth size

Using a uniform method across Finland, medical staff measure BWT accurate to ±1 g, BL, and HC in centimeters immediately after birth. We calculated PI using the standard formula (BWT [kg] / BL [m³]).

Confounding factors

The most important confounder is timing of birth, as the length of gestation directly sets the limits or opportunity for fetal growth. We therefore examined the impact of ACT on birth size stratified by the timing of birth: very preterm (gestational weeks 24–29), preterm (30–34), near term (35–37), at term (38–41), and postterm (≥42). Gestational age at birth is recorded in FMBR as the best estimate including last menstrual period and ultrasound examination. Women routinely undergo an early-pregnancy ultrasound scan by a specially trained midwife at maternity outpatient clinics across the country between 11 + 0 and 13 + 6 weeks of pregnancy. This scan is used to date pregnancies and is offered free of charge to all women registered at the antenatal clinics. Uptake is voluntary, and the majority use the service [20].

The FMBR includes information of clinical relevance on sociodemographics, pregnancy medical history, and gestational complications, which we used as potential confounders, depicted in Fig 2 and listed in Table 1. These variables were adjusted for in multiple regression analyses using the full sample. In the main analysis, the confounders were used to create a propensity score–matched subset of participants for matched comparison of exposed and unexposed infants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. DAG illustrating factors potentially influencing birth size.

DAG, directed acyclic graph.

https://doi.org/10.1371/journal.pmed.1002746.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of study population available in the FMBR and inclusion in PSM.

https://doi.org/10.1371/journal.pmed.1002746.t001

Factors used in additional analyses

We examined the factors available in FMBR that could be used as indicators of neonatal health to shed light on whether any growth restriction was due to ACT or to poor fetal health, which would both limit growth and precipitate treatment.

The Apgar score was implemented in 1952 as a quick way to evaluate the physical condition of the newborn by assessing heart rate, respiratory effort, muscle tone, reflex irritability, and color. The Apgar score conducted at 5 minutes after birth has been found to be an excellent predictor of infant survival in the first 28 days of life and thus considered a meaningful measure of neonatal health [21].

Neonatal care is recorded in FMBR for the first 7 days of life. We used these data to compare ACT-exposed and ACT-unexposed infants. Data were also available indicating whether infants were still hospitalized beyond 7 days.

Lastly, we present in the supporting information results by mode of delivery. Medically indicated preterm labor reflects a complication with the mother or fetus and occurs mostly because of preeclampsia, placental abruption or dysfunction, fetal distress, and fetal growth restriction [22]. In contrast, spontaneous preterm labor starts because of physiologic changes in the cervix, uterus, or amniotic sac. Causal risk factors associated with spontaneous preterm birth include intrauterine infections, psychosocial stress, sociodemographic characteristics, and maternal illness [23]. About 70% of preterm births are considered spontaneous [22–23]. Planned cesarean at term is typical for breech presentation, whereas unplanned or emergency is common for fetal distress.

Statistical analyses

All statistical analyses were planned a priori and carried out using SAS software, version 9.4 (SAS Institute). Our a priori thinking is depicted in Fig 2, which illustrates the potential causal pathways in a directed acyclic graph (DAG) that guided our work. All models were stratified by timing of birth. ACT as a single exposure was examined for each of the four birth size outcomes stratified by five categories of gestational age at birth. Randomized controlled trials (RCTs) provide firm evidence of causal associations. However, the risk of bias is not completely eliminated. Analytic strategies in RCTs do not always realistically reflect clinical complexity or report potential biases. We set out to examine a very large number of pregnancies in order to be able to carry out matched comparisons. The purpose of random assignment is to equate participants in the treatment and control groups in order to isolate the effectiveness of treatment. However, outside of an RCT, this equality among participants is not known, because specific characteristics can predispose individuals to either receive treatment or not. In such a case, the treatment and control groups are different; thus, the treatment effect cannot be isolated, and firm conclusions cannot be drawn about the causal effect of treatment. PSM technique attempts to equate the treatment and control groups after allocation in an observational study in an effort to produce equal groups. We adopted the propensity score approach to match participants and minimize the imbalance with regards to the confounding factors. We match subsamples on all data archived in the FMBR and thus available for analysis in this study. The two groups (treated versus control) were carefully balanced on baseline characteristics associated with treatment allocation, thereby reducing confounding associated with receipt of treatment. Our a priori strategy was to conduct stratified unadjusted and adjusted regression analyses using the entire sample (presented in the supporting material), followed by the formation of matched subsamples and analysis by stratified regression models using PSM score (presented as the main results). In this way, we can compare results using two analytic strategies and are able to ascertain whether any observed differences are driven by bias.

Because all of the background and medical variables recorded in FMBR bear clinical relevance, all were included as confounders, including, e.g., those as identified in previous work such as cohabitation with expectant father [24–25]. Thus, all available background and medical factors recorded in FMBR were used. Few missing values are present in the dataset, and in such cases, pairwise deletions were made. We analyzed the entire cohort using multiple regressions to calculate the association between ACT exposure and birth size (BWT, BL, PI, and HC) and in the fully adjusted model including all the confounders. We then used the confounders to generate propensity scores to match ACT patients with controls. ACT patients were matched to controls on the logit propensity score, using “nearest neighbor matching” [26]. The large sample size permitted a subsample to be generated for analysis consisting of a matching ratio of 5:1, controls to patients, which has been shown to be the optimum matching ratio [27]. However, because relatively few infants are born very preterm, we used 1:1 matching for this stratum. The main analysis comprised direct comparison in the matched samples consisting of 3,077 infants prenatally exposed to ACT (treated) and 11,821 unexposed infants (control).

Lastly, we performed additional analyses to assess whether potential differences in birth size could be attributed to poorer health, as indexed by available data, among the exposed versus unexposed infants. We examined Apgar scores at 5 minutes (additionally stratified by infant sex, as sex differences have been previously reported), neonatal care, and mode of delivery.

Study strengths and limitations

This study has several unique strengths afforded by the use of objective population-based register data. As far as we know, no other study is based on such a large sample or has focused on birth size as the primary outcome. Though observational in nature, the size of this study provided the opportunity to carefully match infants on baseline characteristics that could have influenced clinical decisions to treat, including risk factors/complications (e.g., preeclampsia) that can lead to both spontaneous and elective preterm birth as well as intrauterine growth retardation. The PSM technique enabled us to limit possible bias and include five matched controls per patient for most analyses, which has been found to be the optimum number for bias reduction [27]. In this way, difference between groups on severity of gestational complications is minimized, as five controls are compared to one patient.

We stratified our analysis by gestational age. Stratification is the method of choice for controlling potential confounding and for identifying effect modification related to timing of birth [41], which would not be captured had we merely controlled for gestational age. Our stratified analysis revealed a dose-response-like association with greatest deficits in growth seen in all measurements (BWT, BL, and HC) for preterm infants and reducing in magnitude over time until eventually disappearing for infants born post term, thus suggesting a potential opportunity for catch-up growth in utero or changing contribution of medically indicated preterm births. At term, the growth deficits were still statistically and clinically significant, indicating that even though some degree of catch-up growth may have occurred, the impact of ACT was long-lasting. Regardless of analytic strategy (PSM or adjusted regression analyses), the results were consistent and in the same direction, thus indicating that collider stratification bias is not likely to operate [42]. Further, it is likely that because the sample size was so large, the effect of ACT on birth size was not attributable to bias in treatment allocation. In fact, the entire sample produced estimates of greater deficits in birth size, which are probably inflated as compared to the matched analyses. Thus, we provide here robust estimates.

An advantage of carrying out our study in Finland is the high-quality and uniform antenatal care that is offered free of charge to women residing in Finland. Most women have on average 15 antenatal care visits [43]. There are no sociodemographic differences in the uptake of care [43], so it is unlikely that complications were missed because of insufficient care by a more vulnerable group; on the contrary, poor pregnancy outcomes are associated with a higher number of visits, indicating greater surveillance [43]. Thus, confounding due to inequalities in access to care is highly unlikely in this sample. Our results show that there were no differences in socioeconomic status associated with treatment. This is particularly important because socioeconomic inequalities are likely to confound birth size yet are rarely mentioned in previous research. Understandably, the rate of preterm birth is low in Finland, as the standard of antenatal care is among the best in the world. This fact helps to discern well the impact of ACT from potential confounders related to access to care. However, countries where preterm rates are higher may also have care conditions that are less favorable for the preterm neonate who would benefit from ACT to promote survival. Indeed, care decisions should also consider issues of morbidity, neonatal care availability, and quality of life beyond survival.

We concentrated on live-born infants because our emphasis is on growth restriction among survivors. Stillbirths in high-income countries are also associated with avoidable maternal lifestyle factors such as obesity and smoking [44], factors that we controlled for in our analyses. Stillbirths can be related to major complications or malformations, which would make it very difficult to discern the impact of ACT on such infants.

Lack of detail in FMBR concerning ACT is the main limitation of this study. For example, we lacked information on the time lag between ACT and birth (which would have enabled examination of critical periods of growth), drug name, dose, or number of doses (which would enable drug-specific conclusions). Betamethasone remains biologically active for a period ranging between 36 and 54 hours after administration. Thus, the infants who were born shortly after ACT administration would be unlikely to receive the full effect of the drug. Any growth deficit would be due to compromised fetal health. In contrast, those born at least 3 days after treatment are likely to have had a full effect of the drug. Growth deficits in these infants would be due to both compromised health that led to treatment in the first place and the ACT itself. However, we expect that national guidelines were implemented that recommend a single dose of betamethasone and use tocolytic therapy to delay birth when needed. Data on serial ultrasound scans are not recorded in FMBR. Studies with serial ultrasound scanning would elucidate this issue by monitoring growth trajectory before and after ACT to determine the independent effects attributable to restricted growth due to poor health versus due to treatment. As seen in our DAG representation, unmeasured factors including genetic liability and lifestyle characteristics, such as health-related behaviors and mental health, are important factors to include. Thus, the information gained herein is useful for designing future cohort studies, which would not be restricted by population-based register data but would likely be of a smaller sample size.

Our data underline the need to establish clinical practices to predict parturition with greater accuracy to avoid inessential treatment, as nearly 45% of exposed infants were born at term. Our report should serve as an impetus for identifying techniques to more accurately predict preterm labor. Such advances have been preliminarily reported for a blood test that could cheaply and accurately predict parturition [45] and go beyond fibronectin tests, which are costlier and less often used worldwide. Appropriately targeting high-risk women for ACT would result in less medical over-intervention. Our report of consistent reductions in growth (BWT, BL, HC) coupled with increased medical interventions at birth for infants exposed to ACT calls for further study and reevaluation of clinical practice.