The association between gestational weight gain z-score and stillbirth: a case-control study

The Stillbirth Collaborative Research Network (SCRN) Study was a multicenter case−control study conducted from 2006 to 2008 throughout Rhode Island and selected counties in Georgia, Massachusetts, Utah, and Texas. SCRN’s study methodology has been described in detail elsewhere [21]. Women with stillbirths (cases) and live births (controls) were enrolled at the time of delivery, with oversampling of women with preterm live births and non − Hispanic black women with live births at later gestational ages. Women had to be ≥13 years of age to participate and enrolled prior to hospital discharge. Data collection of consenting women included medical record abstraction, placental pathology, fetal autopsy, and a postpartum maternal interview [21]. Most interviews were completed face−to−face before hospital delivery discharge; a few interviews were completed by telephone or other method within 4 weeks of delivery. Sociodemographic information was derived from the maternal interview. Maternal height, pre − pregnancy weight, weight at last prenatal visit, and weight at delivery were abstracted from medical records. If maternal height or pre − pregnancy weight were unavailable in the medical record, women’s self−reported height and pre − pregnancy weight data were taken from the maternal postpartum interview (we used postpartum interview data on height and pre-pregnancy weight for only n = 1 woman in our analysis). Maternal pre-pregnancy body mass index (BMI) was categorized as underweight (BMI < 18.5 kg/m²), normal weight (BMI 18.5 − < 25.0 kg/m²), overweight (BMI 25.0 − < 30.0 kg/m²), class 1 obese (BMI 30.0 − < 35.0 kg/m²), and classes 2–3 obese (BMI ≥35.0 kg/m²).

Gestational age at delivery for both live births and stillbirths was determined via an algorithm that incorporated date and reliability of last menstrual period, ultrasound estimates of GA, and GA at study screening [22]. Gestational age at fetal death was determined via an algorithm based on fetal foot length and other measures [22]. Fourteen singleton stillbirths had an estimated GA at fetal death that was later than the estimated GA at delivery; GA at delivery was re-coded to GA at fetal death for these observations. Cause of fetal death was determined via a thorough review of autopsy reports, placental pathology, laboratory findings, and medical records [23]. Causes of fetal death were classified as probable causes, possible causes, or present conditions. Cause-of-death categories included placental abnormalities, obstetric conditions, fetal genetic/structural abnormalities, infection, umbilical cord abnormalities, hypertensive disorders, maternal medical conditions (excluding hypertension), and other causes [23]. Stillbirths could be classified as having multiple causes of death. Seventy-six percent of stillbirths in SCRN were assigned a probable or possible cause of death, and 31% of stillbirths had more than one possible or probable cause [23].

We excluded women with deliveries < 20 weeks, missing or implausible GWG (weight loss > 50 pounds or gain > 150 pounds), or missing pre − pregnancy BMI or covariates. Due to a lack of GWG z−score charts for certain groups, we excluded women with triplet gestations, monochorionic/monoamniotic twin pregnancies, and monochorionic/diamniotic twin pregnancies. We excluded women who had dichorionic/diamniotic twin pregnancies and underweight BMI but included women who had dichorionic/diamniotic twin pregnancies and normal weight, overweight, or obese BMI (Fig. 1).

Total GWG was defined as maternal weight at delivery minus pre − pregnancy weight. We calculated GWG z−scores using published charts that were developed from a follow−up study of healthy Pittsburgh women who delivered term live births [18‐20]. Gestational weight gain (in kg) was transformed to a z−score using the published formula \( \frac{\ln \left( GWG+c\right)- mean\ \left(\ln (GWG)\right)}{standard\ deviation\left(\ln (GWG)\right)} \) [18‐20]. The mean and standard deviation depended on pre − pregnancy BMI category and GA at delivery, while c was a constant that transformed ln(GWG) to a positive value (means, standard deviations, and the constant were taken from published references [18‐20]). GWG z−scores can be calculated for either ongoing or completed pregnancies and require only three measurements: pre − pregnancy weight, weight at delivery (or weight at the GA in question), and GA at delivery (or at the time in question) [18‐20]. Among women with singleton pregnancies, published GWG z−score charts end at 40 weeks for women with normal weight and at 41 weeks for women with overweight or obesity; among women with dichorionic twin pregnancies, published charts end at 39 weeks for women with normal weight and at 38 weeks for women with overweight/obesity. In our main analyses, gestational ages above these cutoffs were rounded down to the last gestational week available in the chart (e.g., 40 weeks in singleton pregnancies to women with normal weight).

We conducted data analysis in SAS (Cary, NC). To examine associations between GWG z−score and stillbirth, we used univariable and multivariable logistic regression models with generalized estimating equations that accounted for correlation between multiple gestations. We chose an independent correlation structure after initial models showed a negligible difference between exchangeable and independent correlation structures. All analyses were weighted to account for SCRN’s sampling design and individuals’ probabilities of participating and completing all parts of the data collection process [21]. Models were adjusted for maternal age at delivery, maternal race and ethnicity, study site (Rhode Island/Massachusetts, Georgia, Galveston (Texas), San Antonio (Texas), Utah), maternal education, marital status/cohabitating, health insurance type, trimester prenatal care began, family income in the last 12 months, WIC enrollment, smoking or alcohol consumption during the 3 months prior to pregnancy, lifetime drug use, pregnancy history, history of hypertension, history of preexisting diabetes, history of thyroid disorder, and pre-pregnancy BMI category. Potential confounders, including maternal race and ethnicity, were selected a priori using directed acyclic graphs based on evidence of their associations with GWG and stillbirth [2].

We modeled GWG z−score as a restricted cubic spline with three knots corresponding to the 5th, 50th, and 95th percentiles (percentiles were calculated among live births only using SCRN analysis weights in SAS) [24]. We then calculated odds ratio contrasts of interest by comparing selected GWG z−scores (ranging from − 2.5 to 2.5, in intervals of 0.5) to a referent GWG z−score of 0. In addition, we used GWG z−score models stratified by pre − pregnancy BMI category (normal weight, overweight, obese (BMI ≥30.0 kg/m²)) [25, 26]. We were unable to analyze women with underweight BMI separately due to inadequate sample size. To use a restricted cubic spline in models stratified by pre − pregnancy BMI category, we re − calculated the 5th, 50th, and 95th percentiles of GWG z−score separately in each BMI category among live birth controls using SCRN analysis weights. Models among women with BMI ≥30.0 kg/m² were adjusted for obesity class.

To examine how different assumptions affected results, we conducted various sensitivity analyses. These included: 1) using separate models for women with class 1 obesity versus classes 2–3 obesity; 2) restricting to women with stillbirths estimated to have died ≤1 day before delivery in order to limit the potential for reverse causality (i.e., to limit the risk of the fetal death itself influencing total GWG), and comparing these results to a model restricted to women with stillbirths estimated to have died > 1 day before delivery [22]; 3) restricting to stillbirths estimated to have been alive at their last prenatal visit [22], as well as re-calculating GWG z−scores for these stillbirths using weight and GA at last prenatal visit, in order to limit potential for reverse causality (a small number of stillbirths without prenatal care were excluded from this sensitivity analysis); 4) excluding mummified stillbirths (grade IV or higher maceration among fragmented fetuses and grade V or higher maceration among intact fetuses) because these stillbirths may have a significant discrepancy between fetal weight at death and delivery; 5) analyzing stillbirths by timing of delivery (< 28 weeks, ≥28 weeks, < 37 weeks, and ≥ 37 weeks) in order to explore potential etiologic differences; 6) restricting to women with non − anomalous antepartum stillbirths or non − anomalous live births; 7) analyzing intrapartum stillbirths separately due to their differing pathophysiology [23]; 8) excluding stillbirths with causes of death related to fetal genetic, structural, or karyotypic abnormalities or maternal/fetal hematologic conditions [27] because fetal weight may be driven more by congenital abnormalities than by maternal nutritional status in these pregnancies [28]; 9) excluding stillbirths that had an estimated GA at fetal death < 20 weeks [22], despite having a GA at delivery ≥20 weeks; 10) excluding women with a GA at delivery that exceeded the limit on the GWG z−score charts [18‐20]; 11) using weight at last prenatal visit as an estimate of weight at delivery for women who were missing delivery weight (last prenatal visit is typically a few days before delivery [29] for term pregnancies); 12) controlling for weight and height squared as separate variables because of concern about introducing bias with the use of ratio measures [30]; and 13) analyzing stillbirths by three cause-of-death categories [23]. For this final sensitivity analysis, we first restricted to stillbirths with a probable cause of death, possible cause of death, or present condition related to placental abnormalities (e.g., uteroplacental insufficiency, maternal vascular disorders) [23]. Second, we restricted to stillbirths with a probable cause of death, possible cause of death, or present condition related to maternal medical conditions, excluding hypertension (e.g., diabetes, antiphospholipid syndrome, thyroid disorder) [23]. Third, we restricted to stillbirths with a probable cause of death, possible cause of death, or present condition related to obstetric conditions (e.g., placental abruption; complications of multiple gestations; the combination of preterm labor, preterm premature rupture of membranes, and cervical insufficiency) [23].

This study was reviewed and approved by the Institutional Review Boards of each of the participating sites (Brown University, Emory University, University of Texas Health Science Center at San Antonio, University of Texas Medical Branch at Galveston, University of Utah) and by the data coordinating center (RTI International). Written informed consent was obtained from participants or from their legal guardians (if participants were minors).