All procedures were approved by the University of California San Francisco, California Pacific Medical Center, University of California Berkeley, and Contra Costa Regional Medical Center and Health Centers Institutional Review Boards.
Study Population
Our inclusion criteria included English-speaking women with singleton pregnancies, aged 18–45, with a self-reported pre-pregnancy BMI between 25 and 41 kg/m2, and with a household income less than 500% of the federal poverty level.Footnote 1 Women had to be 12–19 weeks’ gestation at the start of the intervention, and those in the intervention had to be able to attend 8 weekly two hour intervention classes. We limited gestational age to ensure that women were past their first trimester, when risk of miscarriage is high, and early enough in pregnancy for the intervention to still impact gestational weight gain. All women enrolled had to be from 19 to 45 years, able to complete forms in English, not have a substance abuse, mental health, or medical condition that, in the opinion of investigators, would make it difficult for the potential participant to participate in a group intervention. Exclusion criteria included the inability to complete forms in English, needle phobia or fainting response, substance abuse, medical conditions that might affect gestational weight gain (including known diabetes, HIV, hypertension, and eating disorders), polycystic ovarian syndrome treated with metformin, a regular meditation practice (20 or more minutes two times or more a week), recent weight loss (more than 5% within 6 months), chronic use of corticosteroids, or a history of gastric bypass surgery. We did not exclude women who have had previous children.
Recruitment
We first conducted a pilot trial to determine feasibility and to refine recruitment strategies [44]. For the current study, we recruited women from hospital-based clinics, community health centers, Supplemental Nutrition Assistance Program (SNAP) and Women, Infants, and Children (WIC) offices, organizations providing services to pregnant women, and through online advertisements (e.g., Craigslist) from August 2011 to June 2013 with the goal of enrolling 220 women. Women were recruited and enrolled in sequential “waves” to form groups of 8–12 women with expected dates of delivery within 2 months of each other. Details of our recruitment strategy have been published previously [44]. We included an orientation session where we described the pros and cons of being in a study, and the importance of adherence to the study, to reduce dropouts [45].
Non-Randomized Control Group (Treatment as Usual or TAU)
While it is always preferable to have a randomized control group, this is less critical in the early stages of intervention development. The current study was an initial efficacy study where we were piloting a new intervention. As described in the “Introduction” section above, based on the ORBIT model, studies testing the development of a new intervention are encouraged to use a socio-demographically similar comparison group. Because our group class format required enrolling women at the same stage of pregnancy, it was difficult to recruit sufficient numbers for a wave, and it would have been problematic to use only half for the treatment group, given the desired group class size of 10 women. Therefore, women unable to attend intervention classes due to their schedules, or women with gestational age of 20–23 weeks who otherwise met the eligibility criteria, were eligible for being part of the “Treatment As Usual” (TAU) group. The treatment as usual included whatever prenatal medical care the individuals received on their own. Therefore, both the treatment group and the TAU group received medical care as usual, with the only difference being that the treatment group received the MMT classes as well. Due to this recruitment design, the TAU group had slightly later gestational age than the intervention group, a limitation addressed in the discussion.
Study Design
Participants were asked to complete study questionnaires regarding psychological distress, eating behavior, and exercise at baseline and 8 weeks later (post-intervention). Participants were paid $25 for completing the measurement battery. Participants in the intervention were additionally compensated $25 for attending each session to help cover their time, transportation, and any childcare. After delivery, medical records were reviewed by trained research assistants to confirm pre-pregnancy BMI, when possible, and gestational age, assess total gestational weight gain based on pre-pregnancy BMI, code medical complications, and assess which participants may have developed gestational diabetes (based on glucose levels from the OGTT). Women in the MMT group came in person for their baseline assessment, but TAU did not.
Intervention
The intervention development process was based on the ORBIT model [40]. This model guides early stages of intervention development, applying basic behavioral mechanisms of eating and optimizing delivery and dose before moving on to further stages of more formal testing. After completing a pilot study to confirm that the intervention we developed, the MAMAS (Maternal Adiposity, Metabolism, and Stress) Mindful Training, was affecting the purported mechanisms (stress and stress eating), we moved to this efficacy trial. The final intervention, the Mindful Moms Training (MMT), included 8 weekly 2-h sessions, two “booster” telephone sessions, and one postpartum group session with mothers and babies. The intervention was led by two practitioners who worked in pairs. They had graduate degrees (MA, certified nurse midwife, and PhDs), with additional training in both mindfulness and Mindfulness-Based Eating Awareness Training (MB-EAT) [36].
For sessions that included both experiential and didactic components integrated material from three empirically supported interventions: Mindful Motherhood [41], Mindfulness-Based Stress Reduction [46], and MB-EAT [36]. The sessions focused on three commitments, represented by the slogan “Mindful Eating, Move My Body, Breathe!” Participants were given a list of physical activities safe for pregnancy, which focused on ways to increase amount of daily walking and stretching.
Classes began with mindful movement and a check-in where each person shared their experiences with mindfulness practices in the past week. Didactic discussions then covered (1) stress reduction, focused on acceptance-based coping, awareness of breath, body, thoughts, and emotions; (2) mindful eating, focused on heightening awareness of hunger, fullness, taste experience, and thoughts and emotions leading to reactive and/or automatic eating; and (3) nutrition, focused on optimal foods to eat more of (e.g., whole foods), what to eat less of (e.g., processed foods), reading labels, identifying healthy portion sizes, and introducing both the plate method and food pyramid as resources. Each class ended with a minute mindfulness practice and a review of homework for the upcoming week. The curriculum model, development and content, and the high level of fidelity and adherence is described in detail in Vieten et al. [43].
Measures
Weight, Metabolic Health, and Physical Activity
Gestational Weight Gain
The primary outcome was gestational weight gain category, based on the 2009 IOM recommendations. Total gestational weight gain was calculated as the difference between weight at the last prenatal visit before delivery, and self-reported pre-pregnancy weight that had been recorded in the medical record. If the last recorded prenatal weight occurred more than 30 days before delivery, total gestational weight gain was coded as missing (n = 23). When pre-pregnancy weight was not recorded in the prenatal record (n = 69), we used self-reported pre-pregnancy weight from the study eligibility screener.
Secondary Outcomes
Six-Month Postpartum Weight Retention
This secondary outcome was calculated as the difference between the participant’s measured 6-month postpartum weight and their pre-pregnancy weight. For analysis, this was assessed both as a continuous variable and categorical: the difference in weight at 6 months was dichotomized as higher than their pre-pregnancy weight vs. equal to or less than their pre-pregnancy weight.
Psychosocial Distress, Mindfulness Outcomes, and Eating Behavior
The primary psychosocial outcomes were three standardized distress measures, with Cronbach’s α reliability for this sample as stated: (1) global perceived stress, using Cohen’s Perceived Stress Scale (α = 0.87), a 10-item measure of stress perceptions, including ratings of feeling overwhelmed, out of control, and stressed [47]; (2) depressive symptoms, using the Patient Health Questionnaire (PHQ-9) (α = 0.84), a 9-item scale of depressive symptoms used in primary care settings [48, 49]; and (3) pregnancy-related anxiety, using the Pregnancy-Related Anxiety Scale (α = 0.87), a 10-item scale, that assesses the extent to which pregnant women are concerned about their health, their baby’s health, labor and delivery, and caring for their baby [50].
Secondary measures included the acceptance of negative experiences using the Acceptance and Action Questionnaire-II (α = 0.85). This measures the extent to which people are bothered by having negative thoughts and feelings, versus being able to accept them [51]. The eating behaviors measured included emotional eating (α = 0.96) and external eating behavior (α = .84) using the Dutch Eating Behavior Questionnaire [52], and level of food addiction, using the Yale Food Addiction Scale (α = 0.80) [53]. We also measured trait mindfulness in the intervention group, as reported elsewhere [54].
Oral Glucose Tolerance Test (OGTT)
As part of their usual prenatal care, 141 study participants (141/180 = 78% of final sample) completed an oral glucose tolerance test between 24 and 28 weeks’ gestation. Glucose levels from after the 1-h test were abstracted from the prenatal medical record. Level of impaired glucose tolerance was assessed using continuous values of glucose and impaired tolerance was categorically defined as a glucose level above 130 mg/dL. Comparison of OGTT cases to those without an OGTT showed no significant differences on BMI nor any of the demographic variables listed in Table 1.
Table 1 Baseline characteristics
Physical Activity
This was assessed at baseline and at post-intervention using the Stanford Brief Activity Survey, which provides a brief assessment of the amount and intensity of activity completed during leisure time [45]. Level of activity was reported as six levels, ranging from 1 (inactive) to 6 (very active). We used both the continuous score as well as a categorical analysis. The responses were further collapsed into three mutually exclusive categories: lowest level = inactive or light activity, medium level = moderate or vigorous activity 3 days per week, and highest level = moderate or vigorous activity at least 5 days per week.
Sociodemographics and Covariates
A parsimonious set of covariates were selected a priori for the two physiological outcomes (weight and glucose control) that are well-established correlates of gestational weight gain. Covariates included age at enrollment (continuous), parity (continuous), gestational age because weight gain is a function of the length of gestation, and pre-pregnancy BMI category (<30 kg/m2, 30.0–34.9 kg/m2, and ≥ 35.0 kg/m2). We used pre-pregnancy BMI category as a covariate rather than actual BMI because the IOM weight gain guidelines are based on pre-pregnancy BMI category, and the relationship between pre-pregnancy BMI and weight gain is non-linear.
We measured additional sociodemographic variables to further describe the sample. This included race and ethnicity (categorized as Caucasian, African American, Latino, and other/multiracial), educational attainment (<12th grade, high school graduate, any college/vocational training, or college graduate or higher), marital status (single versus married or in a committed relationship), household income (reported value), and smoking status (current, former, or never smoked). Household food security was measured with the 10-item U.S. Adult Food Security Module [55], which includes questions about limiting food intake or not eating balanced meals due to lack of money. Food insecure households were defined as answering “yes” to three or more questions. These additional sociodemographic variables are shown in Table 1; since they did not vary by group, they were not used as covariates.
Statistical Analyses
All statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC). Statistical significance was determined at p < 0.05, two-tailed. The analysis plan followed an intent-to-treat principle by comparing participants’ outcomes in the intervention (regardless of how many classes they attended) to outcomes of all participants in the TAU group. However, an as-treated analysis, comparing women who attended a minimum of five of the eight classes to women in the TAU group, yielded highly similar results. To deal with missing data, for gestational weight gain, analyses were performed on available data, where we removed cases in the presence of missing/incomplete data. For the self-report scales, for missing items, mean substitution was used, unless > 30% of scale items were missing.
Given that in our theoretical model (Fig. 1), changes in distress precede and will lead to changes in eating and weight, we first report the psychological variables, then weight change, our primary outcome, and lastly the other physical health outcomes. To examine the intervention effect on outcomes of psychosocial distress, eating behaviors, and mindfulness within each group, we first used paired t tests to compare baseline to post-intervention scores on our measurement scales. For our primary analysis, to test whether the change scores were significantly different when compared to the TAU group, multivariate linear regression models were used to compare the between-group changes, further adjusting for age and pre-pregnancy BMI.
To examine the intervention effect on gestational weight gain, a multinomial logistic regression model was fit for IOM gestational weight gain category, using “normal” weight gain (e.g., within the IOM guidelines based on a woman’s pre-pregnancy BMI) as the referent group. A sensitivity analysis that excluded women who developed gestational diabetes mellitus (GDM) after enrolling in the study was also conducted (n = 23). Covariates included age, pre-pregnancy BMI, and parity.
For secondary analyses, a multinomial logistic regression model was fit for post-intervention physical activity categories, adjusted for age, pre-pregnancy BMI, and parity. Multivariate logistic regression models were fit for outcomes of impaired glucose tolerance and 6-month postpartum weight retention, adjusted for the study covariates.
We examined the statistical power for this size of a sample. For a group means comparison, with our actual sample size of roughly 105 per group, we had 80% power to detect a small medium effect size (around d = 0.39). Another way of considering power for our primary analysis is that, given empirical data on similar samples [28], we expected around 60% to gain excessive weight, in which case we would have 80% power to detect an additional 20% of the MMT sample not gaining excessive weight compared to TAU group (so only 40% total with excessive weight gain), associated with MMT with regard to excessive weight gain, equivalent to an odds ratio of up to − 0.44 for excessive weight gain and a Cohen’s d of small to medium effect.