Introduction

Hypertensive disorders of pregnancy (HDP), which include gestational hypertension, preeclampsia, and chronic hypertension, affect up to 10% of all pregnancies. Manifestations range from mild elevations in blood pressure with minimal or no systemic effects to severe increases and organ damage. Several studies investigating modifiable risk factors found that HDP is more likely to develop in women with greater gestational weight gain (GWG). However, edema could not be excluded as a confounder because researchers looked at weight gain throughout the pregnancy. MacDonald-Wallis and colleagues avoided this statistical dilemma by looking at weight gain up to 18 weeks’ gestation as a risk factor for development of HDP.

Laura Anne Parks, MD and George A. Macones, MD, MSCE, Associate Editor

Introduction

Hypertensive disorders of pregnancy (HDP), which include gestational hypertension, preeclampsia, and chronic hypertension, affect up to 10% of all pregnancies. Manifestations range from mild elevations in blood pressure with minimal or no systemic effects to severe increases and organ damage. Several studies investigating modifiable risk factors found that HDP is more likely to develop in women with greater gestational weight gain (GWG). However, edema could not be excluded as a confounder because researchers looked at weight gain throughout the pregnancy. MacDonald-Wallis and colleagues avoided this statistical dilemma by looking at weight gain up to 18 weeks’ gestation as a risk factor for development of HDP.

Laura Anne Parks, MD and George A. Macones, MD, MSCE, Associate Editor

Study Design

Parks: What were the primary and secondary objectives of this study?

Liu: MacDonald-Wallis et al sought to better characterize the relationship between GWG and HDP. Specifically, the primary objective was to determine whether GWG up to 18 weeks’ gestation is a risk factor for subsequent development of preeclampsia or gestational hypertension. The secondary objective was to demonstrate whether GWG occurs prior to increases in blood pressure in normotensive individuals.

Parks: What type of study design was used? Who were the participants?

Meister: The study was an analysis of data obtained as part of the Avon Longitudinal Study of Parents and Children, a prospective birth cohort study established in 1991. Based at the University of Bristol in the United Kingdom, it aims to investigate the health and development of children. Originally, participants were pregnant women who had an estimated delivery date between April 1991 and December 1992. Over 14,000 women from the Avon region, which includes the city of Bristol and adjacent areas, were recruited. Their offspring continue to be followed today. For this study, MacDonald-Wallis and colleagues used the following inclusion criteria: women with singleton live births and at least 1 blood pressure and weight measurement during pregnancy. Women were excluded if they reported a history of hypertensive disorders outside of pregnancy. That left a study pool of 12,522 women for the current analysis.

Statistical Analysis

Parks: Please outline the analytic plan for this study and comment on its appropriateness.

O’Neil: The authors attempted to examine the impact of GWG prior to 18 weeks on the incidence of HDP late in gestation. As noted, data were extracted from a prospective birth cohort study that included women with a singleton birth, those without a history of chronic hypertension, and those with at least 1 available blood pressure and weight measurement during pregnancy. The researchers determined a predicted weight at 0 weeks’ gestation using statistical models derived from prenatal weight measurements, and these were comparable with the women’s self-reported prepregnancy weight. They computed absolute weight gain in pregnancy by subtracting the patient’s predicted weight at 8 weeks’ gestation (few actual measurements were available before this point) from the predicted weight at 40 weeks’ gestation; they used a multilevel model for weight change across pregnancy based on data from women with term pregnancies.

To examine early pregnancy GWG and risk for developing HDP, they then incorporated all weight data obtained during weeks 0-18 into a linear model for change in weight with gestational age. The outcomes were classified into 3 different categories: normotensive, gestational hypertension, or preeclampsia. Adjustments were made for maternal height, age, parity, smoking, education, sex of offspring, and prepregnancy weight. Nonlinear relationships were estimated with linear splines, values identified through a form of interpolation.

GWG and changes in blood pressure were evaluated in women categorized as normotensive. To accomplish this, the authors used previously-derived multilevel linear spline models for changes in systolic blood pressure, diastolic blood pressure, and weight with gestational age. These models, along with prespecified points known as knots, were incorporated into spline models for changes in systolic blood pressure with changes in weight and changes in diastolic blood pressure with changes in weight. The new models were used to derive associations between prepregnancy weight and GWG with changes in blood pressure.

The analytic plan for examining early pregnancy GWG and the risk for developing HDP allowed the authors to group data into 3 prespecified categories: normotensive, gestational hypertension, and preeclampsia. To this aim, they established anticipated trajectories in GWG for each expected outcome. The disadvantage to this method is that the data were generalized, excluding individual factors that might contribute but are not apparent as the general linear trend is established.

In the second portion of their analysis, the researchers sought to identify associations between GWG and prepregnancy weight and changes in blood pressure in normotensive women. Again, linear models were constructed from a large amount of abstracted data to establish a normalized curve or trend. These were used to form generalized conclusions regarding associations between changes in blood pressure as they correlate to changes in weight. For the particular goal of their analysis, this method was appropriate for setting up a generalized curve or trend for comparison over a broad dataset. While these curves may estimate an average effect, they may not be generalizable.

Stout: True. The statistical analysis for this study included methods that can account for temporal associations; multiple measurements of a single type (eg, weight and blood pressure) and relationships between these measurements over time. Multinomial regression is similar to logistic regression, which is used to assess an association between an exposure and an outcome while controlling for confounders. Simple logistic regression is used to find a single endpoint from a single categorical variable from multiple ordered categories.

The authors used multinomial regression to assess the relationship between GWG and hypertensive categories (normotensive, gestational hypertension, and preeclampsia) while controlling for confounders. Next, the authors examined GWG and hypertensive disorders using multilevel models with random effects. This type of modeling evaluates the correlation between 2 variables (weight and blood pressure) at baseline and over time. However, this methodology does not account for whether a change in 1 variable precedes a change in the other variable. The authors wanted to investigate whether weight gain heralds the development of HDP, so the relationship of time is important.

To address this, the authors used linear splines. Linear splines employ multivariate modeling to gauge the rates of change over time in processes; in this case, weight gain and blood pressure. In this way, they could investigate whether the weight gain predates the blood pressure changes. Lastly, the authors performed a sensitivity analysis including only patients who had at least 8 measurements of blood pressure and weight. This ensures that the relationship in the patients with the best-documented blood pressure and weight parameters was consistent with the findings of the whole cohort.