RFTS is an ongoing investigation of early pregnancy health. The study includes three waves of enrollment (RFTS1, 2 and 3). The physical activity questions differed slightly between waves 1 and 2/3, and thus only RFTS2/3 are included in this analysis. RFTS2/3 enrolled women from central North Carolina from 2004 to 2007.
RFTS invited women to participate through advertisements and community outreach. Study materials encouraged women planning a pregnancy or in early pregnancy to contact study staff through a toll free phone number. The recruitment methodology of RFTS2/3 is similar to that of RFTS1 and has been described previously.(28
) The study was approved by the Institutional Review Boards of the University of North Carolina and Vanderbilt University.
When women called to volunteer, study staff screened them to determine eligibility and, if eligible, collected the woman's age and pre-pregnancy weight. Women were eligible if they were currently trying to conceive or had been pregnant less than ten weeks based on self-report of their last menstrual period. Women also had to be at least 18 years of age, conceiving without assisted reproductive technology, willing to have a first trimester ultrasound at one of the study's ultrasound locations, intending to remain in the area for the next 18 months, intending to carry the pregnancy to term, able to access a telephone for the first trimester interview, fluent in either English or Spanish, and able to identify a prenatal or primary care provider at the time of screening. This portion of the study included 1,861 live births beyond 20 completed weeks of gestation. We further restricted eligibility to the first pregnancy among women who participated in RFTS2/3 more than once (N=1,735), singleton gestations (N=1,708), and women who completed the first trimester interview (N=1,647). Compared to women who completed the first trimester interview (N=1,647), women who did not complete the interview (N=61) were more likely to be under 30 years of age (67% vs 48%), Black Non-Hispanic (30% vs. 12%), unmarried (39% vs 6%), obese (21% vs 16%), and have less than a college education (56% vs 24%).
Multiple data sources were used to obtain and confirm infant date of birth. The hierarchy of the sources was hospital discharge summaries and prenatal care records (51%), birth and fetal death records (32%), and participant self-report (17%). Birthweight was obtained from vital record linkage for all participants. Gestational age at birth was estimated based on last menstrual period and corrected to the first trimester ultrasound estimate only if the difference between the two was more than seven days. The ultrasound was performed between six and twelve weeks of gestation.
In a telephone interview targeted for 14 weeks gestation (mean and median: 14 weeks, range: 7-20 weeks), women were asked to describe their physical activities by mode (recreational, occupational, indoor/outdoor household and child/adult care). This questionnaire was based on a modified version of a structured seven day recall, with evidence for validity and reliability among pregnant women.(29
) Women were asked to report if they do any “hard” or “very hard” recreational physical activities in a typical week. The description “hard or very hard” is a measure of vigorous intensity based on the Borg perceived exertion scale which has been found to correlate strongly with heart rate and oxygen uptake.(30
) Participants were then asked to describe the type of activity and how often and for how long they performed the activity. Women who reported engaging in recreational activity but did not describe it as “hard” or “very hard” were considered to engage in non-vigorous recreational activity.
Occupational, indoor/outdoor household, and child/adult care physical activities were assessed with analogous questions and coded similarly. Occupational activities included lifting or carrying boxes and lifting and transporting patients. Examples of household activities include washing, folding, and carrying laundry, vacuuming, washing floors, and gardening. Examples of child/adult care activities include lifting and carrying children, bathing children, and lifting or transporting adults.
We summed the minutes per week of each recreational activity (i.e., jogging + swimming + walking) to obtain the total minutes per week of vigorous recreational activity. Similarly, we summed the minutes per week of each activity, within each of the other modes, to obtain the total minutes of vigorous occupational, household, and child/adult care activity, respectively. Finally, we summed over all modes to obtain the total minutes of vigorous physical activity.
Metabolic equivalent (MET) values were assigned to recreational activities only, based on the Compendium of Physical Activities.(31
) The Compendium (originally published in 1993,(32
) updated in 2000(31
)) was developed to compare the intensities of different physical activities across participants. We multiplied the MET value for a given activity by the minutes per week of that activity and summed across activities to obtain total MET-minutes per week. METs were assigned by the first author (AMZJ) and reviewed by the second author (KRE). MET values are a measure of absolute intensity while the participants’ categorization of an activity as “hard or very hard” is a measure of perceived or relative intensity. Thus our analysis contained two assessments of intensity. MET values have not been measured in pregnant populations and thus may not be accurate for pregnant women. Given the numerous physiologic and metabolic changes that occur during pregnancy it is possible that the woman's characterization of the intensity of the activity is more accurate than MET values assigned to a given activity. Therefore, our presentation of results focused on perceived intensity and we present the results based on absolute intensity (MET values) only where they differed from the perceived intensity results.
The cumulative frequency of vigorous recreational activity sessions per week was calculated as the sum of the individual frequencies reported for each activity. For example, if she reported walking three times per week and swimming two times per week her cumulative frequency would be five sessions per week. The average duration of a recreational activity session was calculated by dividing the total reported minutes per week of vigorous recreational activity by the cumulative frequency of vigorous activity sessions. From the previous example, if she reported 60 minutes of walking per week and 40 minutes of swimming per week she would be assigned an average duration of 100/5 = 20 minutes/session.
Women were also asked to report if their overall current physical activity had increased, decreased or stayed the same compared with pre-pregnancy activity. They were also asked if they changed their behaviors in preparation for becoming pregnant. If she answered affirmatively she was asked what she changed. The interviewer did not read a list of responses, but some women responded that they started exercising and these responses were coded. A woman could give multiple responses.
The screening interview and the telephone interview collected information on important covariates including sociodemographics, reproductive history, presence of nausea and vomiting in early pregnancy, and lifestyle factors. Weight and height were measured at the first trimester ultrasound. If this measure was missing, then her self-reported weight and height from the first trimester questionnaire were used.
Covariates for these analyses were chosen if they were considered to be potential confounders based on directed acyclic graphs (33
) constructed for each outcome. In all models, we considered adjustment for maternal age, race/ethnicity, education, income, marital status, alcohol consumption, body mass index, cigarette smoking, illicit drug use, history of miscarriage, history of preterm birth, parity, vaginal bleeding, nausea/vomiting, and history of any type of diabetes. These variables were included in the models if their removal changed the estimates by more than 10% for preterm birth and 20% for birthweight.
Behavioral characteristics were reported in the first trimester questionnaire. Current smokers include women who were smoking at the time of interview and who reported quitting in the previous four months. Former smokers were women who reported quitting at least four months prior to the questionnaire. Alcohol use was categorized into women who have never used alcohol, current users, those who stopped drinking within four months of interview and those who stopped drinking more than four months from the interview.
Analyses were performed with SAS software, version 9.1. We used a standard multivariable logistic regression to examine the association between physical activity and preterm birth as a dichotomous variable (<37 completed weeks of gestation). For comparison with previous studies we also evaluated the association between physical activity and length of gestation using a discrete time survival model.(34
Among term births, we used a linear regression model to examine physical activity and birthweight, adjusted for gestational week at birth. Birthweight in preterm infants can reflect either their prematurity or growth restriction or both. Because the outcome is heterogeneous in preterm infants, we limited our analysis of birthweight to term infants. To improve the precision of our birthweight model, we included two strong predictors, maternal height and infant gender. The other outcomes were not modeled with linear regression, and thus adjustment for non-confounders is not warranted.(35
Continuous variables, including our exposures of interest, were finely categorized and examined with each outcome variable in an unadjusted analysis. The shape of the crude association of each variable with each outcome was visually inspected to determine the appropriate structure (linear, quadratic, categorical) and, if categorical, the number and location of cutpoints. More parsimonious models with fewer parameters were compared to the full model containing the highly categorized variable. Fewer parameters were used if information was not lost when compared to the highly parameterized model (likelihood ratio test p-value >0.05). For each mode of vigorous physical activity (recreational, occupational, household and child/adult care) the minutes of activity were categorized into tertiles resulting in five categories: no activity, non-vigorous activity, and tertiles of the minutes of vigorous activity. The total minutes of vigorous physical activity was divided into five categories: no physical activity reported or only non-vigorous activity reported, and four categories of the total minutes of vigorous activity. We combined the “none” and “non-vigorous” categories because the number of women who reported no physical activity in any mode was small (2%).
For vigorous recreational physical activity only, we conducted separate multivariable analyses for perceived intensity (minutes per week) and absolute intensity (MET-minutes per week), duration of vigorous recreational activity session, and frequency of vigorous recreational activity sessions. Duration and frequency were modeled separately and both were adjusted for the total minutes of recreational activity (as recommended by Lee and Skerret(36
)), the previously described covariates, and the other modes of physical activity (household, occupational, child/adult care). Women who perform zero minutes of recreational activity also had a frequency of zero recreational activity sessions per week. To avoid collinearity, women with a frequency of zero or one were combined to form the lowest frequency category. A similar strategy was employed with duration of recreational activity.