Caregiver participants were recruited between September 2018 and June 2019 from 6 Family Support Centers (FSCs) in under-resourced, urban, and suburban neighborhoods in Western Pennsylvania. The FSCs provide free, holistic home visiting and center-based interdisciplinary services for infants, children, and families living in communities that are heavily populated and experience high rates of poverty [21]. Using a partner-led strategy, FSC leaders headed recruitment efforts. Inclusion criteria were (1) caregiver of a child aged 0–5 years, (2) fluent in English, (3) willingness to participate in sessions. Participants were excluded if they had completed either program in the past. All participants provided written informed consent. This study was reviewed and approved by the University’s Institutional Review Board and is registered on ClinicalTrials.gov (NCT03559907).
This randomized longitudinal pilot study consisted of 3 treatment conditions: (1) CM; (2) Mealtime PREP; or (3) Combined: CM + Mealtime PREP. Prior to recruitment, six sites were identified to host a group based on prior success engaging members in similar programs. The randomization scheme was designed by a biostatistician using SAS PROC PLAN, and sites were randomized in blocks of 3 to ensure two sites (i.e., groups) were assigned to each treatment condition. Based on prior recruitment for community programming and attrition rates, we aimed to recruit 8 participants per site. An a priori power analysis determined that with a sample size of 48, we would be 80% powered to detect within-between interaction effects and between-group effects with an alpha of 0.05. Leaders recruited fifty-one caregivers within each FSC, but two were ineligible due to child age. Therefore, 49 eligible participants were included in our analyses. Assessments were caregiver-reported and occurred at baseline, directly following intervention (6 to 12 weeks after baseline), and approximately three months after intervention completion; compensation for each assessment session increased over time ($25, $50, $75). Incentives to boost attendance at group sessions included weekly groceries and a small appliance (e.g., blender, toaster oven) for attending ≥ 4 of 6 sessions for each program offered. All study-related activities (assessments and group sessions) occurred at the local FSCs and were scheduled on days and times identified by FSC leaders as convenient for members of their center. Trained graduate student research assistants collected assessment data using paper forms or the Research Electronic Data Capture (REDCap) mobile application on an iPad. Research assistants were available to help participants interpret questions as needed and were not blinded to intervention as measures were solely based on caregiver report and, therefore, unlikely to be influenced by bias related to the assessor’s knowledge of group assignment.
Interventions
CM for Parents was led by trained instructors and included six weekly, 2-h sessions to increase self-sufficiency in the kitchen. Each session was held in the participants’ local FSC and included meal preparation, didactic teaching, and sharing a meal as a group. Participants learned skills related to nutrition, healthy meal preparation, and cost-effective purchasing of healthy foods. Participants received groceries each week and practiced cooking skills within their homes. This course did not address child mealtime behavior, but healthy, child-friendly foods were discussed. Instructors had experience in nutrition education and culinary skills and completed seven hours of training to follow the manual of procedures using a gold standard checklist for each session [16].
Mealtime PREP was also delivered in the participants’ local FSC and led by a pediatric occupational therapist. It included 6-weekly, 2-h sessions to enhance daily child meals to promote healthy dietary variety. This intervention features a behavioral activation approach to facilitate the adoption of new, healthy family mealtime routines [19]. Caregivers selected healthy foods (e.g., vegetables, lean proteins, fruits) frequently offered in their homes to practice skills between sessions. Strategies to improve acceptance of healthy foods included: (1) frequent family meals; (2) positive reinforcement (i.e., verbal praise and positive attention); (3) repeated exposure; and (4) play to increase interaction [22,23,24]. Each week, participants received groceries and child-friendly dishware to promote proper portions (e.g., divided plate and measuring cups) or serving ware for family-style meals. The instructor completed 8 h of training and followed a manual of procedures. During each session, approximately 15 min were set aside to troubleshoot solutions to unique family circumstances (e.g., new baby in the home, scheduling conflicts, feeding problems) with individual participants.
Participants recruited to either of the two sites randomized to the Combined treatment received both programs in sequence, CM followed by Mealtime PREP, at their local FSC. This group received 12-weekly, 2-h treatment sessions with the same trained staff and established protocols.
Feasibility outcomes
Feasibility benchmarks were our primary outcome measures based on programmatic recommendations and preliminary data. They included: (1) Ability to recruit, on average, > 8 participants per site; (2) achieving an 80% intervention completion rate; (3) being rated as an acceptable intervention by 95% of intervention completers, and (4) treatment fidelity of ≥ 90% for CM and Mealtime PREP regardless of treatment condition. The intervention completion rate is the percentage of participants attending ≥ 4 of 6 sessions for each program. Therefore, a participant in the Combined treatment would need to complete at least four sessions of each program to meet this benchmark. Intervention acceptance was assessed using the Treatment Acceptability Questionnaire (TAQ), a validated eight-item Likert-type scale [25]. Participants completed the TAQ directly after the intervention, and a score ≥ 28 was used as a cut-off for intervention acceptability [19]. Treatment fidelity was assessed via video review of sessions and completion of checklists created using each program’s manual and established gold standards for CM [16].
Preliminary effects: intervention outcome
Data were collected on secondary outcomes of child dietary variety, child risk of nutritional problems, caregiver stress, and participant characteristics at baseline, post-intervention, and a 3-month follow-up. All data were collected in person at the FSCs apart from the 3-Day Food Diary, collected via mail.
Data on dietary variety were collected using two methods. Participants were provided with a 3-Day Food Diary and instructed to record all foods their child consumed over three days, including approximate amounts. They were asked to include at least one weekend day and weekday and given a self-addressed stamped envelope to return the diary. A 24-h Dietary Recall was collected at each time point, and the number of unique foods consumed each day was tallied.
The Nutrition Screening for Every Preschooler (NutriSTEP) assessed the risk of nutrition-related problems. This 17-item, caregiver-reported, multiple-choice assessment is a valid and reliable screen for risk of nutritional problems (e.g., obesity, limited dietary variety) in young children [26]. Questions cover food frequency, mealtime practices, screen time, and child growth.
The Parenting Stress Index, Short-Form (PSI-SF) assessed overall caregiver stress level. This reliable tool has been validated in diverse populations, and percentile scores ≥ 90% represent clinically significant caregiver stress [27]. Higher scores represent higher stress, with the total score representing the overall stress level experienced within the caregiver role [28].
Participant characteristics (e.g., gender, age, ethnicity) and caregiver education, employment, household income, and total number of siblings were collected using a demographic form. Race was self-reported by the caregivers of the children from a list including White, Black/African American, Asian, American Indian/Alaskan Native, Native Hawaiian/Pacific Islander, or Other (specify).
Data analysis
Descriptive statistics (e.g., means, frequencies, percentages) were used to determine whether feasibility benchmarks were met. Prior to modeling, detailed exploratory analyses were performed, including missing data analysis and screening for anomalies. Potential imbalances between the two treatment groups (combined vs. independent) were examined using chi-square analyses for categorical variables and t-tests or Mann–Whitney U tests for continuous variables as appropriate. Associations between the dependent variables at baseline and potential demographic confounders (e.g., race, education, employment status) were calculated.
A series of individual growth models were fitted to examine treatment effects over time. Mixed-effects models were used as they allow both fixed and random effects and modeling with missing outcome data, so long as they are missing at random [29]. For each dependent variable (dietary variety, risk of nutritional problems, total caregiver stress), an unconditional model was run with a random participant effect to examine change in the entire sample over time. Parameter estimates and measures of model fit (i.e., Akaike Information Criterion and Bayesian Criterion) were examined. Next, a conditional model was fitted, examining interaction and main effects over time. Potential covariates were identified based on significant correlations with outcome measures at baseline (i.e., employment status and education level) or likelihood to impact outcomes (i.e., intervention completion) and added to the best fitting model. An independent covariance structure with group and time considered as fixed effects, and a random effect for participants was used in all models. Differences in treatment group (combined vs independent) as well as condition (CM vs. Mealtime PREP vs Combined) were examined. Post hoc tests were performed for significant interaction or main effects. Statistical analyses were conducted with Stata SE (v.16) using an alpha of 0.05.