Overall study design
The HT project uses a randomized experimental, short-term longitudinal design with a convenience sample of participants from each of the three states. Participants are randomly assigned to the HT intervention treatment group or control group with a goal of N = 300 participants per group at Time 3 data collection. There are three assessment periods: baseline (Time 1), post intervention (Time 2), and delayed post (6 months after intervention) (Time 3). The USDA has funded this project for 5 years and supports the three states, making this multi-state collaboration between Colorado, Michigan, and Wisconsin possible.
Development of the intervention
The Healthy Toddlers intervention is developed and refined based on a pilot program, Nutrition Education Aimed at Toddlers (NEAT) , that provided initial evidence of increased nutrition knowledge within the intervention group as well as decreased television watching during meals and is based on the Social Cognitive Theory (SCT) . The HT addresses core nutrition concepts but moves well beyond basic nutrition to address maternal self-efficacy during feeding, appropriate feeding styles, and practices, including skill development to increase success in making these behavioral changes. The curriculum pinpoints specific concerns of mothers of toddlers evident in our previous work, that economically and educationally disadvantaged mothers need help introducing fruits and vegetables to their toddlers . Staff members in Cooperative Extension in all three states reviewed the lessons.
Participants and recruitment
Participants with the following characteristics are being recruited: economically disadvantaged (≤ 185% federal poverty level in the U.S. and eligible for federal food assistance programs), 18 years of age or over, mothers with a toddler between 12 and 32 months of age, and neither child nor mother having diagnosed eating, physical, or chronic health problems. Mothers and toddlers are targeted for recruitment from local community programs such as: the Special Supplemental Nutrition Program for Women, Infants and Children (WIC), a Federal assistance program of the Food and Nutrition Service of the United States Department of Agriculture for healthcare and nutrition of low-income pregnant women, breastfeeding women, and infants and children under the age of five; immunization clinics; and food pantries. Staff from these programs provide information about the study and encourage mothers with toddlers to enroll in the project. Once eligibility is established, families are contacted for the initial data collection home visit where written informed consent is obtained. Newly recruited participants are not previously enrolled in either Building Strong Families (BSF) or Expanded Food and Nutrition Education Program (ENFEP). The study is approved by each state's University Committee on Research Involving Human Subjects and informed consent for voluntary participation is obtained prior to commencement in the study.
Following consent and baseline data collection, mothers are randomized into either the HT or control group. This is done via computer minimization procedure balancing groups with respect to the county recruitment locations. According to CONSORT guidelines , allocation concealment is used to prevent participants, data collectors and educators from knowing which group the participants would be assigned. Only after enrollment in the program do researchers assign participants to intervention or control groups. The assignment of groups is withheld from data collectors throughout the entire program.
The HT intervention consists of eight in-home visits by a specially trained paraprofessional instructor plus four weekly telephone follow-up reinforcement contacts. Particularly for high-risk families with young children, providing services within the context of the family's home environment appears to be a useful and effective strategy to provide parents with information, emotional support, access to other services and direct education . The home-visitation model also engages families who lack transportation or child care, a challenge frequently reported by families with low incomes.
Paraprofessional instructors are peer educators who can relate to the target audience. Research shows that people learn best from their peers (people like themselves). Eight home visit sessions have been found to produce behavioral change . At each visit, the paraprofessional spends approximately 1 hour with the mother and toddler dyad. The HT lessons use a variety of techniques and materials to enhance each mother's learning experience and help reinforce knowledge. Each lesson includes opportunities for discussion, hands-on activities, and an opportunity for mothers to practice skills covered in the lesson. The eight lessons include a lesson plan, handouts, and recipes. Mothers receive a notebook binder at the beginning of Lesson 1.
The control group families receive the usual services provided by BSF or EFNEP in respective states. These families are newly enrolled into BSF or EFNEP as part of the HT study and have not received home visitation previously. The control lessons are similarly delivered as the HT lessons, such that, a paraprofessional instructor provides eight lessons during an in-home visit, which last approximately 60 minutes. However, the control lessons focus on parenting (BSF) or nutrition (EFNEP) and do not include extensive content on feeding toddlers. Paraprofessionals who provide the lessons for the control group families are different to prevent cross contamination between the two groups.
Data are obtained by trained data collectors using various methods, including paper-and-pencil questionnaires, children's 3-day dietary records, anthropometric assessment, and mealtime observations. Data collection in all three states occur at baseline (prior to the first lesson of the HT or the control intervention), at post (after completion of the eighth lesson, after 3 months for controls), and at 6 months after completion of the 2nd data collection for both the intervention and control groups. The intervals for data collection were chosen to reflect points in time critical to determining outcomes (immediately following the in-home learning, and then again 6 months later to determine sustainability). We expect that the 6-month follow-up for children who were enrolled when they were 12-32 months of age, a critical developmental stage, will be sufficient to detect change in children's eating skills and food intake.
Data collectors are trained to use all of the tools, including the mealtime observation, and are trained to teach mothers how to keep 3-day dietary records. All data collectors are trained to be as unobtrusive as possible when they are recording observational data and are trained to conduct data collections in an interactional style least likely to provoke social desirability characteristics on the part of the mother.
To ensure that observation and diet record data are accurately recorded, a 2-day intensive training session is held for data collectors prior to the collection of data, followed by booster sessions in years 2 and 3. Training consists of a standardized training guide and protocol including a review of the instruments, recording the first day of the diet record, role-playing of data collection, and recording observations using videotapes and volunteer mother-toddler dyads until inter-rater reliability is confirmed at ≥ 90%. Data collectors are monitored for quality and fidelity to the protocol on a quarterly basis.
The Child-Parent Mealtime Behavior Questionnaire (CPMBQ) was adapted from the Children's Eating Behavior Inventory  to assess toddler feeding self-regulation and was tested with 400 low-income mother-toddler dyads. Six subscales comprise the CPMBQ. All subscales have a moderate to high internal consistency between .67 -.89. The Child-Parent Mealtime Observation is the companion to the CPMBQ and assesses maternal child-feeding interaction and child eating behavior. It consists of 44 behaviors rated on a three-point scale (Yes/Sometimes/No) where the observer scores the extent to which a behavior was either observed or not observed. Inter-rater reliability is .90 with a high internal consistency reliability.
The Nutrition Attitudes Questionnaire assesses parent attitudes related to feeding children and consists of 15 items and has been validated with parents of toddlers, 1-3 years of age. For each item, the parent indicates on a 5-point Likert response scale to what extent they agree with the item from 1 (strongly agree) to 5 (strongly disagree). The coefficient alpha for this scale is α = .83.
The Self-Efficacy Questionnaire consists of 8 items and provides a measure of parents' self report of their self-efficacy in feeding their toddler and mealtime interaction and has been validated with parents of toddlers, 1-3 years of age. The instrument takes approximately 5 minutes to complete. The mother indicates on a 5-point Likert response scale how confident she feels in feeding her child (i.e., 1, "Not confident at all" to 5, "Very Confident"). The coefficient alpha for this scale is α = .83.
The Child Feeding Questionnaire (CFQ) assesses maternal child feeding style. It consists of 31 items and provides a measure of mother's self-report of parental feeding control practices. The instrument takes approximately 10 minutes to complete and has been used with limited-resource mothers of color. Confirmatory factor analysis was used in two non-Hispanic white samples and a Hispanic sample of parents (over 600 parents total) to establish the seven factors in the CFQ. Internal consistency coefficient alphas for the factors range from 0.70 to 0.88 .
Three-day Dietary Record. The mother receives detailed instructions by data collectors on how to complete a written record for 3 days (2 week days, 1 weekend day) of her child's food and beverage intake at 3 time points (baseline, post, delayed post). Mothers are asked to include, when possible, brand names of foods, preparation techniques, a description of the foods, and to submit labels of various foods they prepare for their children. Mothers are given measuring cups and spoons and a portion estimation tool to assist in estimating the amounts of foods offered to toddlers and the amounts of foods their toddlers consume.
Sample size, power calculations and Data analysis
Sample size and power calculations
Requirements are based on power calculations for two primary outcome variables: average toddler fruit servings (without juice) and average toddler vegetable servings. An Analysis of Covariance (ANCOVA) model is assumed, in which 9-month follow-up means are compared by study group, with baseline measures as covariates. Formally, the model can be stated as Yij=μ + Ti + β (cij-c..) + Єij; where Yij is the follow-up response of the ith study participant in the jth study group (j = I for intervention and j = C for control group), and cij is the corresponding baseline value, treated as a covariate. This model can easily be expanded to include other covariates, including site, sociodemographic variables, etc. The effect of the intervention is represented in the model by T I - T C, which is the difference between study group means at follow-up adjusted for baseline differences. Sample size is estimated with the objective that power for the test of the null hypothesis that Ti - Tv = 0, versus a two-sided alternative with Type I error rate 0.05, which would be at least 0.9, when the true difference between the groups is 0.33 (one-third of a serving per day).
NEAT  data were used to estimate the covariance parameters needed for the power calculation. Based on the experience in the NEAT study, 9-month retention can be expected to be about 75%, thus initial sample size must be inflated by a factor of 1.33 (= 1/0.75), yielding sample size requirements of 384 and 436 for fruit and vegetables, respectively. These numbers are approximately satisfied by the plan to recruit 390 subjects per group (130 subjects per state per group). The plan is to use mixed model analysis which uses all of the data, even data from individuals that do not have follow-ups. Some checking of that method can be done by comparing the means base on complete data only to the estimate from the mixed model.
The effect size does have public health significance, such that even a small increase in a serving of fruit and/or vegetables is significant for a toddler where portion sizes are small and the intent is to seek to increase fruit and vegetable intake even if in small increments.
Baseline demographic variables (education, ethnicity, household composition) will be compared by group using a two-factor (group-by-state) model, with continuous responses assumed to be approximately normally distributed and categorical responses assumed to be binomially distributed. When supported by the data, two-factor model will be simplified to one-factor by eliminating the state effect and group by state interaction.
Outcomes (such as toddler fruit, vegetable, and beverage intake; toddler eating skills; and maternal feeding style) will be analyzed using repeated measures ANCOVA. Repeated measures at two time points (post-treatment and follow-up) will be compared by group using a two-factor (group by state) model. Repeated measures of Analysis of Variance (ANOVA), in which the pre-treatment data as a third time point, rather than a covariate, will be considered as an alternative model. However, it is anticipated that the ANCOVA model will be preferred to the ANOVA model if the intrasubject correlations between pre-treatment and follow-up measures of fruit and vegetable intakes are low. Additionally, the ANCOVA model provides a more flexible pre-treatment adjustment than the ANOVA model because it allows covariate by period and covariate by group interactions to be estimated and tested. The latter interaction is of interest in this study, because the difference between groups maybe smaller among families who serve more fruits and vegetables at the outset than among families who serve fewer fruits and vegetables at the outset.
Additional covariates will be included where supported by the analysis, such as breast versus formula feeding, age of introduction of solids, and maternal weight. Interval level intermediate variables also will be examined using the same factorial ANOVA design. Transformation of response variables to more closely approximate normality will be considered and evaluated using residual plots. The final analysis will be done in a scale that more conforms to the assumption of normality, if warranted. The normality assumptions will be assessed using standard statistical tools, primarily plots of residuals versus predicted values and residuals versus normal scores.