Data collected from Family Meals LIVE! [15], a direct observational, mixed methods study, were used to measure the healthfulness of foods served at 8 meal occasions. The University of Minnesota’s Institutional Review Board Human Subjects Committee approved the study protocol. Families (N=120) were recruited between 2012 and 2013 from 4 primary care clinics in the Minneapolis–St Paul metropolitan area that serve a racially/ethnically diverse, urban population of primarily low-income families. Participating families were provided an iPad (Apple Inc) and asked to video-record 8 consecutive days of family dinner meals with a minimum of two weekend meals. Only dinner meals in the home were recorded because of privacy issues. At the start of each meal, families spoke into the camera to indicate what foods were being served. After the meal, families completed a meal screener, which is a self-reported, open-ended measure of the foods served at the meal. Comprehensive study procedures have been described elsewhere [11,13].

In total, 800 meals were available for analysis [13]. Families were asked to record meal occasions over consecutive days, and recordings were taken every 1.8 days on average (SD 0.89), indicating good participant compliance with data collection procedures and minimal lack of family meals or meals outside of the home. A 1-day washout period was employed to allow families to acclimate to the study procedures and recording equipment.

Previous studies have shown that direct observational research conducted in the home using unstructured observations (eg, play, routines) has more predictive validity and reliability compared to laboratory settings using structured observations (eg, tasks given to participants) and allows participants to acclimate and exhibit less reactivity [16-18]. The lengthened, 8-day observation window has been shown to offer advantages over cross-sectional designs, which include the measurement of weekday and weekend meals, the capture of variability in the healthfulness of weekly meals, and more reliable and objective measurement of family meal occasions [16-18].

The HOM, created for the Family Meals Live! study and adapted from the Healthy Eating Index 2010 [19], was used to assess family meal healthfulness [13,15,20]. The HOM assesses 7 categories of foods served at meals: fruit, vegetables, dark green vegetables, dairy, protein, high sodium foods (reverse scored), and added sugars (reverse scored). A present-or-absent format is used to score the HOM, the components are summed, and a total of 9 points are available (the fruit and vegetable categories can each receive a total of 2 points). A higher total score is reflective of a more healthful family meal with regards to foods served. To calculate the HOM score, 3 research members (including 2 registered dietitians) watched each video-recorded meal to code the foods present [13]. The self-report meal screener was also used to corroborate the foods seen in the videos. Because the HOM evaluates meal dietary healthfulness, all foods present were coded even if they were not consumed by all family members.

Permutations were constructed to evaluate study conditions (timing of measurement and duration of measurement period) that researchers implemented at the design stage of direct observational studies. First, a permutation was calculated to examine how adding observation days affects the HOM relative to a measure that incorporates all observation days. In all, 13 HOM permutations were calculated: a full-week index of average meal healthfulness (this was the primary reference permutation), 6 indices adding 1 additional day on the front end of the observation window (permutation 1: day 1 only; permutation 2: average of days 1 and 2; permutation 3: average of days 1 through 3; etc.), and 6 permutations adding 1 additional day beginning with the last observation day (measure 1: day 7 only; measure 2: average of days 7 and 6; measure 3: average of days 7, 6, and 5 etc). The primary reference permutation was computed assuming that capturing more dinner meals would reduce the random variation in the composition of foods that are served across days to obtain a measure of typical meal healthfulness. Relative to this comprehensive direct assessment of meal healthfulness, a measure containing fewer observation days that is highly correlated with the full measure may sufficiently characterize typical family meal healthfulness without excess resource investment.

Survey estimation procedures were performed for each permutation of the HOM to determine whether the means differed by day of week, with sampling weights being applied to obtain population average meal healthfulness measures generalizable to the 4 clinics from which families were recruited. Effect consistency in the relationship between the HOM and dietary intake and family meal frequency were examined in sensitivity analyses to evaluate the presence of measurement error in permuted variables with a fewer number of observation days. A third correlational analysis was performed to evaluate the strength of the linear relationship between each HOM permutation. Comparisons between each reduced measure and the full reference measure were examined to determine how many days of additional meal recordings were needed to approximate the full reference measure. The intraclass correlation coefficient (ICC; 0.663) was calculated to evaluate consistency across the permutations within families. Pearson correlation coefficients above the ICC were used to visually evaluate at what points the permutations with fewer measurement days approximated the measure incorporating all days. All analysis and data management were performed in Stata 13.1 SE (StataCorp).

Study results were consistent with our hypothesis that a fewer number of direct observation days would be sufficient to characterize typical weekly meal healthfulness. We also found evidence that including both weekday and weekend day family dinner meals differed in healthfulness across a week-long observation period. Single-day and 2-day observations of meal healthfulness may be inappropriate for generalizing about the healthfulness of foods served at dinner meal occasions over the course of a week. In addition, correlational analyses indicated that when using just 2 days of data, the fewer-day permutations were strongly correlated (r>0.70) with the full 7-day measure. This is in part because meal healthfulness was moderately to highly correlated within the family. Thus, it is not surprising that adding a fourth, fifth, and sixth day of observational data provided little additional information about the healthfulness of foods served. Using 3- or 4-day observations of family meal healthfulness appeared to maximize measurement reliability and to minimize the cost of data collection and respondent burden.

The study had several strengths, including the use of direct observational methods, consecutive observation of family meals, and a substantial number of meals (N=800) observed. Practical advantages are also noted, such as assessing measurement variability, providing new information about how to allocate staff time, and minimizing respondent burden. Replication studies are needed to provide support for the finding that relatively few observation days (ie, 1 weekend day and 1 weekday) are required, with the ideal number of days possibly being as few as 4; to test findings in a population with heterogeneous characteristics; and to assess meal healthfulness in multiple ways to avoid social desirability bias, recall error, and participant reactivity.

Findings from the current study suggest that relatively few direct observational days of family meals data are needed to approximate the healthfulness of meals across 1 week. Specifically, 1 weekday and 1 weekend observation at a minimum, along with 3-4 days of direct observational data, are needed. Findings from the current study may inform future direct observational study designs to reduce both research costs and participant burden.