Participants
Sixty-nine pregnant women were recruited in and around Ames, Iowa, via a convenience sample as part of a larger observational study analyzing physical activity and omega-3 fatty acid intake in pregnant women living in a non-coastal community. Primary recruitment methods included mass emails to faculty, staff and students at Iowa State University, advertisements on campus and in the community, newspapers, and at local obstetric clinics. Ten women withdrew from the study or were excluded after enrollment due to time constraints, miscarriage, diagnosis of twins, or medical complications. In addition, seven women did not have complete data sets. Therefore, fifty-two women were included in this analysis.
Participants were recruited between May 2009 and May 2010. Non-smoking women between the ages of 18 and 45 with a singleton pregnancy were enrolled prior to their 18th week of gestation. Women were excluded if they had a history of chronic disease or planned to deliver outside of Ames, Iowa, or other partnering hospitals. Each participant's medical provider confirmed all aforementioned qualification criteria prior to their patient's participation in the study. Since the study was strictly observational, no medical pre-screens for exercise were required to participate. All participants provided written informed consent prior to participation. The study was approved by the Iowa State University Institutional Review Board.
Data Collection
All participants met with a staff member at week 18 of gestation to complete a medical history questionnaire and complete an interview assessing regular physical activity patterns since becoming pregnant. Each woman also received instructions on how to properly record her daily activity for two separate methods (described below) over the next 7-days. Following the 7-day data collection period, participants met with a staff member to return the PAR and activity monitor.
Physical activity record
A subjective PAR was completed by each participant to document all daily activity (24-hours per day for all 7 days), including sleep time and any activity that occurred between going to bed at night and waking the next day (i.e. trips to the restroom, to the kitchen, to children's bedrooms, etc.). Activities performed each day were listed chronologically with reference to start and end time of each event. A space for descriptions was included to discuss intensity and further details of each activity if necessary.
Activity monitor
The SenseWear® Mini Armband (Model Name: MF) (SWA), (BodyMedia, Inc., Pittsburgh, Pennsylvania) was worn on the upper left arm for all 7 days, 24-hours per day, to estimate energy expenditure. This device is an objective monitor that uses a combination of a triaxial accelerometer, skin temperature sensor, galvanic skin response sensor and thermometers (measure heat flux) to detect movement and predict energy expenditure using a proprietary algorithm (version 2.2). For each participant, a SWA monitor was configured with height, body weight, age, gender, smoking status, and handedness per the manufacturer's instructions. Participants were instructed to remove the armband only during times of water submersion (showering, swimming, etc.).
Participant characteristics
Anthropometric and demographic data collected included age, pre-pregnancy weight, height, parity, due date, ethnicity, education level and total household income. Body weight was measured using a Sunbeam (2008 Sunbeam® Products, Inc., Boca Raton, Florida) analog scale with participants not wearing shoes. Height and pre-pregnancy weight were self-reported in the medical history questionnaire and used as a descriptive characteristic only. Participants were asked to classify their ethnicity as American Indian or Alaska Native, African American, Caucasian, Asian, Hispanic, or other.
Data Analysis
Activity monitor
Raw data from the SenseWear analysis system were imported into MATLAB (Version R2008a, Mathworks, Natick, MA). Total time spent in moderate (3-5.9 METs), and vigorous (6-9 METs) physical activity were extracted from the data files, formatted, and totaled. Time discrepancies in the data were adjusted so all records were reflective of 7, 24-hour periods to equate to 1 week's time (n = 10,080 minutes). Time gaps were filled using an algorithm that first detected off-body time in the data and then filled these time gaps with 1 MET (equivalent to rest) [22]. Off-body time was present in the data mostly for the purposes of light-intensity activity such as bathing or showering and did not substantially alter the data.
Physical activity record
Data from the 7-day subjective PAR was entered into a spreadsheet (Microsoft, Redmond, WA) by a single individual. Each activity documented in the PAR, including sleep time, was assigned a metabolic equivalent (MET) using the Compendium of Physical Activities (CPA) [22]. A new version of the CPA was recently published [23]. We reviewed the PAR entries and MET levels that had been assigned for the specific activities from the old CPA and assessed if any of these would have changed categories (i.e. light to moderate or moderate to light) if we had used the new CPA. Although some of the specific MET values differed, none of the MET values for the activities entered in the PAR changed categories of intensity. Therefore our results would not have differed if the new CPA had been used in our original analysis of the PAR (data not shown). Each PAR was then verified to ensure it totaled 10,080 minutes, the time equivalent to a 7-day period. Due to intra-individual inconsistencies in start/end time of recording PAR data on the first and last day of data monitoring, eight records fell short of 10,080 minutes. Time was filled for these records on day 8 with CPA MET values equivalent to the previous identical week day until 10,080 minutes were reached.
Data from the PAR was then further analyzed on a day-by-day basis using algorithms developed in Microsoft Office Excel 2007. These algorithms allowed us to partition each 24-hour period into various levels of activity equivalent to the activity categorizations previously stated for the SWA. Total daily time spent in MVPA (> 3.0 METs) in the PAR was computed and compared to total daily time spent in MVPA (> 3.0 METs) as recorded by the SWA.
Data entry training
Since a MET level for each activity in the PAR had to be assigned by a data entry individual, it was imperative that each data entry individual was consistent with MET assignments. Those responsible for this task had to complete a training test. The test records consisted of PARs from this cohort of women. Total daily minutes spent in MVPA entered by the data entry individual had to be within five percent of the total daily MVPA minutes of the original analysis conducted by one of the authors. Upon successfully meeting these criteria on five test records, each data entry individual was allowed to enter and analyze data.
Reference standard
To identify the exercisers in our study population, the reference standard used the criterion of at least 3, 30 minute volitional sessions of MVPA. Data indicating frequency, duration, and type of exercise since becoming pregnant were self-reported in the medical history questionnaire. Additionally, each participant was verbally interviewed in person at the beginning of week 18 of gestation by a member of the research staff regarding physical activity since becoming pregnant. Each woman was asked "Do you regularly engage in at least 3, 30 minute sessions of moderate exercise per week?" Depending on the participant's response, the interviewer may have probed to determine if the activity was volitional or incidental, e.g. "I went for a walk" or "I walked from my parked car into the store", respectively. If the woman met the criterion of at least 3, 30 minute volitional sessions per week, she was provided a heart rate (HR) monitor (Polar E600, Polar Electro Oy, Kempele, Finland) and instructed to wear it while exercising during the following 7-day monitoring period. Not all women were given a HR monitor because we only used this information as an additional method to objectively confirm exercise sessions listed in the PAR. Furthermore we did not solely rely on the HR monitor data due to interference/noise associated with use of the HR monitor. At the follow-up visit, data collected by the participant during the preceding week was reviewed. An ACSM Certified Health Fitness Specialist (HFS) utilized the PAR and HR monitor data to confirm that the participant did or did not meet the exercise criterion. Therefore, the number of exercisers used as the reference standard was identified by the ACSM HFS using data from the initial interviews, the PAR, and the HR monitors, if applicable. In summary, if the woman stated that she met the criterion in the initial verbal interview, and this was confirmed by the PAR and the HR monitor, she was considered an "exerciser". If the woman answered "no" to regularly participating in at least 3-30 minute moderate exercise sessions per week, but her PAR indicated she completed at least this amount of activity, she was still considered an "exerciser." This criterion was established based on the ACSM Guidelines for Exercise Testing and Prescription [10], has been used in other prenatal physical activity studies [24, 25], and allowed us to identify as many exercisers as possible within our population.
Statistical analysis
The reference standard was used to evaluate the ability of the PAR and the SWA to accurately differentiate between exercisers and non-exercisers. The number of exercisers and non-exercisers identified with both modalities (PAR and SWA) for two different guidelines, 90 minutes (mirroring the reference standard) and 150 minutes MVPA (per DHHS guidelines) were determined. Next, each participant's categorization for each modality was compared to the reference standard, and the results were compared against the binomial distribution for categorical data.
Minute-accumulations of MVPA from both the PAR and SWA were used independently to differentiate between exercisers and non-exercisers. The participant categorization was compared to the reference standard using receiver operator characteristic (ROC) curves plotting sensitivity against the specificity at each cut-off for each modality. Models were compared by computing the area under the curve (AUC) of the ROC using the trapezoidal method. The AUC was computed and treated as the Mann-Whitney U statistic. The phi coefficient (Φ) was determined for each cut-off and this Φ was evaluated for significance of fit using the chi-square test as well as standard rules for effect size [26–28].
Data from the PAR and the SWA were combined factorially to form a composite measure so that each cut-off from the SWA was compared to each cut-off from the PAR. To be classified as an exerciser, both modalities had to agree that the participant met the criterion. Each of the approximately 2,000,000 categorizations was compared to the reference standard and the results were used to generate sensitivity, specificity, and Φ values. The chi-square test was used to determine a significant fit between the composite measure and reference standard.
Algorithms used to categorize the data through the use of cut-offs as well as algorithms to evaluate sensitivity, specificity, AUC, and Φ were developed and implemented in MATLAB R2008A. For all tests, significance was set at P < 0.05.