Amy H. Herring

Missing-Data Methods for Generalized Linear Models: A Comparative Review

Missing data is a major issue in many applied problems, especially in the biomedical sciences. We review four common approaches for inference in generalized linear models (GLMs) with missing covariate data: maximum likelihood (ML), multiple imputation (MI), fully Bayesian (FB), and weighted estimating equations (WEEs). There is considerable interest in how these four methodologies are related, the properties of each approach, the advantages and disadvantages of each methodology, and computational implementation. We examine data that are missing at random and nonignorable missing. For ML, we focus on techniques using the EM algorithm, and in particular, discuss the EM by the method of weights and related procedures as discussed by Ibrahim. For MI, we examine the techniques developed by Rubin. For FB, we review approaches considered by Ibrahim et al. For WEE, we focus on the techniques developed by Robins et al. We use a real dataset and a detailed simulation study to compare the four methods.

Exploring associations between physical activity and perceived and objective measures of the built environment.

The built environment may be responsible for making nonmotorized transportation inconvenient, resulting in declines in physical activity. However, few studies have assessed both the perceived and objectively measured environment in association with physical activity outcomes. The purpose of this study was to describe the associations between perceptions and objective measures of the built environment and their associations with leisure, walking, and transportation activity. Perception of the environment was assessed from responses to 1,270 telephone surveys conducted in Forsyth County, NC and Jackson, MS from January to July 2003. Participants were asked if high-speed cars, heavy traffic, and lack of crosswalks or sidewalks were problems in their neighborhood or barriers to physical activity. They were also asked if there are places to walk to instead of driving in their neighborhood. Speed, volume, and street connectivity were assessed using Geographic Information Systems (GIS) for both study areas. Locations of crashes were measured using GIS for the NC study area as well. Objective and perceived measures of the built environment were in poor agreement as calculated by kappa coefficients. Few associations were found between any of the physical activity outcomes and perception of speed, volume, or presence of sidewalks as problems in the neighborhood or as barriers to physical activity in regression analyses. Associations between perceptions of having places to walk to and presence of crosswalks differed between study sites. Several associations were found between objective measures of traffic volume, traffic speed, and crashes with leisure, walking, and transportation activity in Forsyth County, NC; however, in Jackson, MS, only traffic volume was associated with any of the physical activity outcomes. When both objective and perceived measures of the built environment were combined into the same model, we observed independent associations with physical activity; thus, we feel that evaluating both objective and perceived measures of the built environment may be necessary when examining the relationship between the built environment and physical activity.

Prevalence of uterine leiomyomas in the first trimester of pregnancy: An ultrasound-screening study

OBJECTIVE: To estimate the proportion of pregnant women with one or more leiomyomas detected by research-quality ultrasound screening in the first trimester, to describe the size and location of leiomyomas identified, and to report variation in prevalence by race/ethnicity. METHODS: Within an ongoing prospective cohort, we conducted 4,271 first-trimester or postmiscarriage ultrasound examinations. Sonographers measured each leiomyoma three separate times, recording the maximum diameter in three perpendicular planes each time. Sonographers and investigators classified type and location. RESULTS: Among 458 women with one or more leiomyomas (prevalence 10.7%), we identified a total of 687 leiomyomas. The mean size of the largest leiomyoma was 2.3 cm (95% confidence interval [CI] 1.8–2.8). Mean gestational age at ultrasonography was 61±13 days from last menstrual period. Prevalence varied by race/ethnicity: 18% in African-American women (95% CI 13–25), 8% in white women (95% CI 7–11), and 10% in Hispanic women (95% CI 5–19). The proportion of women with leiomyomas increased with age much more steeply for African-American women than for white women. CONCLUSION: Leiomyomas are common in pregnancy and occur more often among African-American women. Given the limited research on effects of leiomyomas on reproductive outcomes, the degree to which race/ethnic disparities in prevalence of leiomyomas may contribute to disparities in events such as miscarriage and preterm birth warrants investigation. LEVEL OF EVIDENCE: II

A Time Series Analysis of Air Pollution and Preterm Birth in Pennsylvania, 1997--2001

Preterm delivery can lead to serious infant health outcomes, including death and lifelong disability. Small increases in preterm delivery risk in relation to spatial gradients of air pollution have been reported, but previous studies may have controlled inadequately for individual factors. Using a time-series analysis, which eliminates potential confounding by individual risk factors that do not change over short periods of time, we investigated the effect of ambient outdoor particulate matter with diameter ≤10 μm (PM10) and sulfur dioxide on risk for preterm delivery. Daily counts of preterm births were obtained from birth records in four Pennsylvania counties from 1997 through 2001. We observed increased risk for preterm delivery with exposure to average PM10 and SO2 in the 6 weeks before birth [respectively, relative risk (RR) = 1.07; 95% confidence interval (CI), 0.98–1.18 per 50 μg/m3 increase; RR = 1.15; 95% CI, 1.00–1. 32 per 15 ppb increase], adjusting for long-term preterm delivery trends, co-pollutants, and offsetting by the number of gestations at risk. We also examined lags up to 7 days before the birth and found an acute effect of exposure to PM10 2 days and 5 days before birth (respectively, RR = 1.10; 95% CI, 1.00–1.21; RR = 1.07; 95% CI, 0.98–1.18) and SO2 3 days before birth (RR = 1.07; 95% CI, 0.99–1.15), adjusting for covariates, including temperature, dew point temperature, and day of the week. The results from this time-series analysis, which provides evidence of an increase in preterm birth risk with exposure to PM10 and SO2, are consistent with prior investigations of spatial contrasts.

Ethnicity and gestational diabetes in New York City, 1995-2003

Objective  To characterise the patterns of occurrence of gestational diabetes among a wide range of ethnic groups that reside in New York City.

Physical activity patterns during pregnancy.

PURPOSE The aim of the study was to describe the mode, frequency, duration, and intensity of physical activity among pregnant women, to explore whether these women reached the recommended levels of activity, and to explore how these patterns changed during pregnancy. METHODS This study, as part of the third phase of the Pregnancy, Infection, and Nutrition Study, investigated physical activity among 1482 pregnant women. A recall of the different modes, frequency, duration, and intensity of physical activity during the past week was assessed in two telephone interviews at 17-22 and 27-30 wk of gestation. RESULTS Most women reported some type of physical activity during both periods. Child and adult care giving, indoor household, and recreational activities constituted the largest proportion of total reported activity. The overall physical activity level decreased during pregnancy, particularly in care giving, outdoor household, and recreational activity. Women who were active during the second and the third trimesters reported higher levels of activity in all modes of activity than those who became active or inactive during pregnancy. The majority did not reach the recommended level of physical activity. CONCLUSION These data suggest that self-reported physical activity decreased from the second to the third trimesters, and only a small proportion reached the recommended level of activity during pregnancy. Further research is needed to explore if physical activity rebounds during the postpartum period.

Antimüllerian Hormone as a Predictor of Natural Fecundability in Women Aged 30-42 Years

OBJECTIVE: To generate estimates of the association between markers of ovarian aging and natural fertility in a community sample at risk for ovarian aging. METHODS: Women aged 30–44 years with no history of infertility who had been trying to conceive for less than 3 months provided early-follicular phase serum and urine (N=100). Subsequently, these women kept a diary to record menstrual bleeding and intercourse and conducted standardized pregnancy testing for up to 6 months. Serum was analyzed for estradiol, follicle-stimulating hormone (FSH), antimüllerian hormone, and inhibin B. Urine was analyzed for FSH and estrone 3-glucuronide. Diary data on menstrual cycle day and patterns of intercourse were used to calculate day-specific fecundability ratios. RESULTS: Sixty-three percent of participants conceived within 6 months. After adjusting for age, 18 women (18%) with serum antimüllerian hormone levels of 0.7 ng/mL or less had significantly reduced fecundability given intercourse on a fertile day compared with women with higher antimüllerian hormone levels (fecundability ratio 0.38; 95% confidence interval [CI] 0.08–0.91). The day-specific fecundability for women with early-follicular phase serum FSH values greater than 10 milli-international units/mL compared with women with lower FSH levels was also reduced, although nonsignificantly (11% of women affected; fecundability ratio 0.44; 95% CI 0.08–1.10). The association with urinary FSH was weaker (27% women affected; fecundability ratio 0.61; 95% CI 0.26–1.26), and the associations for the other markers were weaker still. CONCLUSION: Early-follicular phase antimüllerian hormone appears to be associated with natural fertility in the general population. LEVEL OF EVIDENCE: II

Sociodemographic, Perinatal, Behavioral, and Psychosocial Predictors of Weight Retention at 3 and 12 Months Postpartum

Postpartum weight retention plays an important role in the pathway leading to obesity among women of childbearing age. The objective of this study was to examine predictors of moderate (1–10 pounds) and high (>10 pounds) postpartum weight retention using data from a prospective pregnancy cohort that followed women into the postpartum period; n = 688 and 550 women at 3 and 12 months, respectively. Analysis included descriptive statistics and predictive modeling using log‐binomial techniques. The average weight retained at 3 and 12 months postpartum in this population was 9.4 lb (s.d. = 11.4) and 5.7 lb (s.d. = 13.2), respectively. At 3 months postpartum, prepregnancy weight, gestational weight gain, and hours slept during the night were associated with moderate or high weight retention, whereas having an infant hospitalized after going home and scoring in the upper 75th percentile of the Eating Attitudes Test (EAT) were associated only with high weight retention. At 12 months postpartum, prepregnancy weight, gestational weight gain, and maternal education were associated with moderate weight retention; and gestational weight gain, maternal age, race, employment status, and having an infant hospitalized at birth were associated with high weight retention. The results of this study illustrate the importance of prepregnancy weight and gestational weight gain in predicting postpartum weight retention. Furthermore, given the lack of successful intervention studies that exist to date to help women lose weight in the postpartum period, the results of this study may help to inform future interventions that focus on such aspects as hours of sleep, dealing with stress associated with a hospitalized infant, and nonclinical eating disorder symptomatology.

Comparison of gestational age at birth based on last menstrual period and ultrasound during the first trimester

Reported last menstrual period (LMP) is commonly used to estimate gestational age (GA) but may be unreliable. Ultrasound in the first trimester is generally considered a highly accurate method of pregnancy dating. The authors compared first trimester report of LMP and first trimester ultrasound for estimating GA at birth and examined whether disagreement between estimates varied by maternal and infant characteristics. Analyses included 1867 singleton livebirths to women enrolled in a prospective pregnancy cohort. The authors computed the difference between LMP and ultrasound GA estimates (GA difference) and examined the proportion of births within categories of GA difference stratified by maternal and infant characteristics. The proportion of births classified as preterm, term and post-term by pregnancy dating methods was also examined. LMP-based estimates were 0.8 days (standard deviation = 8.0, median = 0) longer on average than ultrasound estimates. LMP classified more births as post-term than ultrasound (4.0% vs. 0.7%). GA difference was greater among young women, non-Hispanic Black and Hispanic women, women of non-optimal body weight and mothers of low-birthweight infants. Results indicate first trimester report of LMP reasonably approximates gestational age obtained from first trimester ultrasound, but the degree of discrepancy between estimates varies by important maternal characteristics.

Bayesian methods for highly correlated exposure data

Studies that include individuals with multiple highly correlated exposures are common in epidemiology. Because standard maximum likelihood techniques often fail to converge in such instances, hierarchical regression methods have seen increasing use. Bayesian hierarchical regression places prior distributions on exposure-specific regression coefficients to stabilize estimation and incorporate prior knowledge, if available. A common parametric approach in epidemiology is to treat the prior mean and variance as fixed constants. An alternative parametric approach is to place distributions on the prior mean and variance to allow the data to help inform their values. As a more flexible semiparametric option, one can place an unknown distribution on the coefficients that simultaneously clusters exposures into groups using a Dirichlet process prior. We also present a semiparametric model with a variable-selection prior to allow clustering of coefficients at 0. We compare these 4 hierarchical regression methods and demonstrate their application in an example estimating the association of herbicides with retinal degeneration among wives of pesticide applicators.