Maternal and Neonatal Outcomes in Obese Women Who Lose Weight During Pregnancy

CM Cox Bauer; KA Bernhard; DM Greer; DC Merrill


J Perinatol. 2016;36(4):278-283. 

In This Article

Materials and Methods

Study approval was obtained from the Institutional Review Board of Aurora Health Care, which provided a determination of non-human subjects research. Maternal and neonatal outcomes associated with weight change in pregnancy were retrospectively investigated for women with obesity (BMI≥30 kg m−2). We included women in the study who gave birth at any of 12 hospitals in a single, regional health-care system during January 2008 through December 2013. Women with multiple gestations or incomplete records across predefined explanatory, matching (confounder) and outcome variables (except umbilical cord pH) were excluded. Women with multiple pregnancies, however, were included. We retrieved most data directly from an internal database that housed maternal and neonatal information for all deliveries in the health-care system; the gathering of missing data was attempted through review of individual patient medical records. Based on the maternal weight lost or gained during pregnancy, we identified women as belonging to one of the four weight change categories, including: (1) weight lost (WtL; change in weight ≤0 pounds), (2) weight maintained (WtM; change in weight >0 and <11 pounds), (3) IOM recommended weight gained (that is, weight gain appropriate, WtGA; change in weight 11 to 20 pounds), and (4) weight gain excessive (WtGE; change in weight >20 pounds). Prepregnancy weight was patient reported if undocumented during a medical appointment prior to pregnancy. Height was measured at the first obstetric appointment, and final weight was recorded upon admission prior to delivery.

To control for intrinsic and prepregnancy maternal characteristics, we matched women across weight change categories in a 1:1:1:1 design by maternal age (>35 years (yes/no); then <±3 years), race/ethnicity (white non-Hispanic, black non-Hispanic, other non-Hispanic, Hispanic/Latina), prepregnancy BMI (<±10%), chronic hypertension (yes/no), diabetes mellitus (yes/no) and smoking status (smoking ever/never). The resultant matched population was used to study the effects of weight change on gestational age at delivery (continuous) and incidence of preterm birth (<37 weeks; yes/no). For all other non-gestational age outcomes of interest, we created a second 1:1:1:1 matched population by additionally incorporating gestational age at delivery (<±1 week) in the matching algorithm. These other maternal and neonatal outcomes included cesarean delivery (yes/no), gestational hypertension or preeclampsia (yes/no), gestational diabetes (yes/no), infant birth weight (continuous), low birth weight (<2500 g; yes/no), macrosomia (birth weight ≥4500 g; yes/no), small-for-gestational age (yes/no), infant admission to NICU (yes/no), APGAR score at 5 min (continuous), low APGAR score at 5 min (<7; yes/no) and estimated blood loss (ml; continuous). Small-for-gestational age births were defined by infant birth weight below the 10th percentile of gestational age-specific birth weight.[15] With data limited to <45% of births, the outcomes of umbilical cord pH (continuous) and low cord pH (<7.15; yes/no) were examined across weight change categories in the full study population rather than either of the matched populations.

We characterized the full population using frequencies and means with 95% confidence intervals (CI). Differences in proportions and means across weight change categories were tested using Pearson's test of independence and analysis of variance, respectively. We report analysis of variance results only if assumptions of normality and homogeneity of variances were satisfied. Within matched populations, associations between maternal weight change and the maternal and neonatal outcomes of interest (except umbilical cord pH) were examined using linear and logistic regression models for continuous and binary responses, respectively. Methods of generalized estimating equations parameter estimation and use of an exchangeable working correlation structure allowed for consistent and unbiased estimation of given repeated measures of subjects within the same matched cluster and identified correlations among all pairs of matched cases within a single cluster as equal.[16] We report model-based estimates herein, but empirical and model-based estimates of s.e.s. were first compared to assess adequacy of correlation structure.

Effect sizes corresponding to maternal weight change categories were represented by parameter estimate (mean value) differences and odds ratios for continuous and binary responses, respectively. Both value differences and odds ratios, computed as the exponentiated difference between parameter estimates, are reported for WtL, WtM and WtGE, each relative to the reference category, WtGA (IOM recommendations). We also defined contrasts a priori in order to identify differences in effect sizes among all pairs of maternal weight change categories. Effect sizes were adjusted by the inclusion of all matching variables in the models of gestational age, cesarean delivery, infant birth weight, APGAR score at 5 min and estimated blood loss. Matching variables included maternal age in the models of low birth weight and NICU admission; maternal age and BMI in the model of preterm birth; and maternal age, BMI and gestational age in the models of gestational hypertension/preeclampsia, gestational diabetes and small-for-gestational age. Effect sizes in the models of macrosomia and low 5-min APGAR score were unadjusted. The number of variables included for adjustment was based on the conservative rules-of-thumb of 30 observations required per parameter estimate for continuous outcomes and 10 events required per parameter estimate for binary outcomes.[17] Patterns in umbilical cord pH across weight change categories were explored using analysis of variance, and patterns in low cord pH across categories were explored using the Cochran–Mantel–Haenszel test for linear trend (M 2 ). We performed all statistical analyses using the SAS statistical software (version 9.4; SAS Institute, Cary, NC, USA).