We recruited women attending their initial prenatal visit at one of eight urban and suburban obstetric offices in a multispecialty group practice in eastern Massachusetts from 1999 to 2002, as summarized previously.16,17
All mothers provided written informed consent, and all procedures were in accordance with ethical standards for human experimentation.18
The human subjects committee of Harvard Pilgrim Health Care approved all study protocols.
We enrolled 63% of eligible participants, resulting in 2,128 women who delivered live infants. We excluded from analysis women with a history of type 1 or type 2 diabetes (n = 20), who had no measurement of blood glucose concentration during pregnancy (n = 24), who did not report any information on physical activity or television viewing (n = 259), or who had missing information on prepregnancy body mass index (BMI) or history of GDM (n = 20), leaving 1,805 available for inclusion in this analysis. Women not included were less likely to be white (37% versus 71% of those included), but rates of GDM and abnormal glucose tolerance did not differ among excluded compared with included white and nonwhite women. Among the 1,805 included participants, 1,638 women provided information about behaviors during the year before pregnancy, and 1,581 women provided information about behaviors during pregnancy.
At the initial study visit (mean 10.4 weeks of gestation), participants completed a questionnaire regarding their physical activity habits during the 12 months before pregnancy. They reported average weekly hours spent in three classes of recreational activity, namely walking (“for fun or exercise, including to or from work, but not at work”), light-to-moderate physical activities (“such as yoga, bowling, stretching classes, and skating, not including walking”), and vigorous physical activities (“such as jogging, swimming, cycling, aerobic class, skiing, or other similar activities”). We used a questionnaire modified from the leisure time activity section of the Physical Activity Scale for the Elderly (PASE).19
Modifications included the following: Instead of using the previous 7 days as the time referent, we asked women to average their weekly activity over the year before pregnancy and to report average hours per week rather than both days per week and hours per day. We also combined light and moderate activities. Our choice of examples of activities was influenced by the PASE, the Paffenbarger physical activity questionnaire,20
and knowledge of activities common to women in the northeastern United States. Women also reported the average number of hours per week they had spent watching television or videos. At 26–28 weeks of gestation, using the same questions, women reported average physical activity and television viewing during the preceding 3 months.
To allow comparison with previously published studies, we classified light-to-moderate activity, vigorous activity, and combined light-to-moderate and vigorous activity as none versus any reported. Subdividing these variables into additional categories did not yield substantively different results. We defined total physical activity as time spent in walking, light-to-moderate activities, and vigorous activities combined and analyzed this exposure in categories that might translate to specific recommendations: 2 hours or less, 3–6 hours, 7–13 hours, and 14 hours or more weekly (less than 30 minutes, 30 minutes to less than 1 hour, 1 to less than 2 hours, and 2 or more hours daily). We defined sedentary lifestyle as 2 or fewer weekly hours of total physical activity, based upon current recommendations that adults, including pregnant women, should accumulate at least 30 minutes of exercise a day.21,22
In the study population women were routinely screened for gestational diabetes at 26–28 weeks of gestation with a nonfasting oral glucose challenge test in which venous blood was sampled 1 hour after a 50-g oral glucose load. If the 1-hour glucose result was at least 140 mg/dL, the participant was referred for a 100-g fasting glucose 3-hour tolerance test. Normal results were a blood glucose below 95 mg/dL at baseline, below 180 mg/dL at 1 hour, below 155 mg/dL at 2 hours, and below 140 mg/dL at 3 hours.8
We categorized participants with a normal screening glucose challenge as having normal glucose tolerance and those who failed the challenge test as having abnormal glucose tolerance. We classified those with at least two abnormal results on the fasting glucose tolerance test as having GDM. For the 30 participants with an abnormal glucose challenge test but no 3-hour tolerance test performed, we reviewed all laboratory results, including random finger-stick glucose and the text of the clinical medical record, to identify 19 cases of GDM.
We collected information on maternal age, race/ ethnicity, parity, education, marital status, and smoking history, all reported by the women at their first study visit. We calculated prepregnancy BMI (kg/m2) from self-reported height and weight. Each woman reported her history of gestational diabetes in prior pregnancies and whether her mother had a history of diabetes mellitus. We obtained clinically measured prenatal weights from the medical record and calculated gestational weight gain before 26 weeks as the difference between the last pregnancy weight before 26 weeks of gestation and the self-reported prepregnancy weight.
We used multivariable logistic regression to fit separate models investigating associations of physical activity with risk of abnormal glucose tolerance and GDM. Exposures of interest were walking, light-to-moderate activity, vigorous activity, combined light-to-moderate and vigorous activity, time spent in total physical activity, sedentary lifestyle, and time spent in television viewing, all assessed both before and during pregnancy. We evaluated associations between television viewing and physical activity using Spearman correlation.
We also evaluated the combined effects of activity before and during pregnancy. Because we believed it likely that many women who engaged in vigorous activity before pregnancy would continue activity during pregnancy but might reduce the intensity, we compared women who reported vigorous activity before pregnancy and light-to-moderate or vigorous activity during pregnancy with women who reported no vigorous activity before pregnancy and no light-to-moderate or vigorous activity during pregnancy.
We included as covariates participant characteristics that have previously been reported to be important predictors of gestational diabetes, namely age (less than 25, 25–34, 35–39, and 40 years or older), race/ethnicity (white, nonwhite), history of gestational diabetes (yes, no, nulliparous), prepregnancy BMI (continuous), and mother’s history of diabetes (yes, no, unknown). Adjustment for other characteristics, including education, marital status, smoking, and weight gain before 26 weeks of gestation, did not result in material changes in the magnitudes of the observed associations between physical activity and outcomes, and thus we did not include these variables in the final models. Although we did assess diet during pregnancy, we did not include dietary factors such as intake of total energy, fat, carbohydrates, fiber, or glycemic load in the present analysis because diet was not associated with glucose tolerance in this cohort.
We explored whether prepregnancy BMI might modify associations of physical activity and television viewing with risk of abnormal glucose tolerance using multiplicative interaction terms and stratification.23
Because of small numbers within strata, we could not perform analyses stratified by BMI using GDM as an outcome. Because parous women might change their behaviors based on experiences in prior pregnancies, we also performed analyses restricted to nulliparous women.
We report adjusted odds ratios for each exposure compared with the reference category. To obtain the P value for trend across categories of total physical activity, we analyzed the median number of hours of activity within each category as a continuous predictor. We used estimates from the multivariate models and the observed population means and prevalence to calculate the adjusted risk difference and its reciprocal, the “number needed to treat.” We used SAS 8.2 (SAS Institute, Cary, NC) for all analyses.