Understanding Outcomes in Behavior Change Interventions to Prevent Pediatric Obesity: The Role of Dose and Behavior Change Techniques

Abstract

Background. Behavioral interventions to prevent pediatric obesity have shown inconsistent results across the field. Studying what happens within the “black box” of these interventions and how differences in implementation lead to different outcomes will help researchers develop more effective interventions. Aim. To compare the implementation of three features of a phone-based intervention for parents (time spent discussing weight-related behaviors, behavior change techniques used in sessions, and intervention activities implemented by parents between sessions) with study outcomes. Methods. A random selection of 100 parent–child dyads in the intervention arm of a phone-based obesity prevention trial was included in this analysis. Sessions were coded for overall session length, length of time spent discussing specific weight-related behaviors, number of behavior change techniques used during the sessions, and number of intervention-recommended activities implemented by the parents between sessions (e.g., parent-reported implementation of behavioral practice/rehearsal between sessions). The primary study outcome, prevention of unhealthy increase in child body mass index (BMI) percentile, was measured at baseline and 12 months. Results. Overall session length was associated with decreases in child BMI percentile (b = −0.02, p = .01). There was no association between the number of behavior change techniques used in the sessions and decreases in child BMI percentile (b = −0.29, p = .27). The number of activities the parents reported implementing between sessions was associated with decreases in child BMI percentile (b = −1.25, p = .02). Discussion. To improve future interventions, greater attention should be paid to the intended and delivered session length, and efforts should be made to facilitate parents’ implementation of intervention-recommended activities between sessions (ClinicalTrials.gov, No. NCT01084590).

Keywords

behavior change techniques behavioral intervention dose pediatric obesity process evaluation

Childhood obesity prevention programs (Spiegel & Nabel, 2006) are a priority but have had limited effectiveness (Summerbell et al., 2005). These interventions are often delivered as a package of behavior change techniques (i.e., the active ingredients in an intervention, such as goal setting or problem solving) and activities for parents to complete between sessions. Little is known about the associations between intervention features (e.g., session length, in-session use of behavior change techniques, or participant implementation of between-session activities) and outcomes. Understanding these associations will allow researchers to design more effective and efficient interventions and begin to uncover the mechanisms by which interventions work or do not work. Tools to describe intervention implementation have been developed (JaKa, Seburg, Roeder, & Sherwood, 2015; Michie et al., 2013), but they are not consistently used to understand intervention effectiveness.

Intervention dose (e.g., number of sessions, session length) is a key factor often associated with better study outcomes in behavioral research (Gearing et al., 2011), but it is not often evaluated in pediatric obesity research (JaKa et al., 2016). The majority of obesity intervention studies that have examined the association between dose and study outcome have been treatment rather than prevention trials. A systematic review and meta-analysis of family-based lifestyle interventions for children who were already overweight or obese found that the treatment duration and number of treatment sessions were significantly associated with better weight outcomes (Janicke et al., 2014). In contrast, a second systematic review of obesity intervention trials did not find a clear association between dose and weight outcomes (Heerman et al., 2017). This second review and meta regression analysis included children across the weight spectrum, rather than only children with overweight or obesity, which may have contributed to the discrepant results. The specific methods for assessing dose may also have contributed to the mixed findings in the literature. Measuring the actual time spent in sessions or in discussing specific behaviors may lead to more consistent findings (Baranowski, Cerin, & Baranowski, 2009). It is possible that there is diminishing return as session length increases beyond a certain threshold. In fact, complex, nonlinear associations between overall session length and outcomes have been observed in other counseling interventions (Baldwin, Berkeljon, Atkins, Olsen, & Nielsen, 2009; Leblanc & Ritchie, 2001) but have not been consistently measured in obesity prevention or treatment research (JaKa et al., 2016).

The number of behavior change techniques (e.g., goal setting) used may also be associated with outcomes (Kazdin, 1974; Romanczyk, Tracey, Wilson, & Thorpe, 1973; Spencer, 1978; Stunkard, 1972). The Behavior Change Technique Taxonomy (BCTTv1; Michie et al., 2013; Michie et al., 2015) allows researchers to use a common vocabulary to identify behavior change techniques. Though session length and number of techniques used are associated (i.e., longer sessions allow for more techniques), it is proposed that the use of more techniques by interventionists within a given session length may be a stronger predictor of outcomes. Findings have been inconsistent. One review of pediatric obesity treatment interventions found that in effective interventions the interventionists used a greater number of unique behavior change techniques (Hendrie et al., 2012), whereas another showed that effective and noneffective interventions did not differ with respect to the number of behavior change techniques used by the interventionists (Martin, Chater, & Lorencatto, 2013). These reviews only evaluated planned techniques and did not assess the number of unique behavior change techniques delivered or used by the participants, nor did they control for session length (Lorencatto, West, Christopherson, & Michie, 2013). This work has been successfully conducted in other domains (Lorencatto, West, Bruguera, Brose, & Michie, 2016) and is warranted in the field of pediatric obesity prevention.

Another factor likely associated with intervention effectiveness is the degree to which parents are able to implement intervention-assigned activities between sessions. Interventionists help parents set goals during sessions, but whether these goals are successfully implemented by parents between sessions is rarely measured (JaKa et al., 2016). Methods to characterize participants’ role in the intervention process have been piloted in other areas like smoking cessation (Gainforth, Lorencatto, Erickson, West, & Michie, 2016) and adult physical activity (Michie et al., 2008). As with intervention dose and number of in-session techniques used, it is hypothesized that the number of between-session activities completed by parents will be associated with better outcomes in the context of an obesity prevention trial.

This study aimed to identify which intervention features are associated with outcomes by coding sessions from a completed behavioral obesity prevention intervention. The trial was designed to test the impact of a 14-session, phone-based parent-counseling intervention to improve weight-related behaviors and the home environment. It was hypothesized that (1) the amount of time parents spent in sessions with their phone coach would be inversely associated with child body mass index (BMI) percentile at 12 months, (2) the amount of time spent discussing specific behaviors (e.g., physical activity) would be associated with improvements in those behaviors, and (3) the number of behavior change techniques used within sessions would be inversely associated with child BMI percentile at 12 months. Exploratory analyses were also conducted to investigate whether the use of specific behavior change techniques and the number of between-session activities completed by parents were inversely associated with child BMI percentile at 12 months.

Method

Sample

One hundred participants from the intervention arm of Healthy Homes/Healthy Kids (HHHK 5-10) were randomly selected. The HHHK 5-10 trial (ClinicalTrials.gov, No. NCT01084590) is described elsewhere (Sherwood et al., 2013). A random selection of participants was used because of the cost of transcribing and coding all the intervention sessions. The trial recruited parents of children between 5 and 10 years of age at risk of becoming overweight or obese (70th to 95th BMI percentile). The children were identified via electronic medical records at 20 primary care clinics in Minnesota. Exclusion criteria included children and parents who were not able to read and write in English, children taking medications affecting growth, and any children participating in other health-related research studies. The trial tested a parent-delivered phone intervention to reduce child BMI at 12 months (immediately postintervention) and 24 months. Only data from baseline and 12 months were used in this analysis. The protocols in this study were approved by institutional review boards, and informed consent was obtained from all the participants.

Intervention

The intervention-arm parents agreed to participate in 14 phone sessions over 1 year. An intervention manual was used to standardize session length, format, and content. Planned session length was 45 minutes for Session 1 and 15 to 30 minutes for Sessions 2 to14. The interventionists and parents were allowed to determine the amount of time to be spent on each weight-related target behavior (Table 1). The sessions focused on behavioral and home environmental changes the parents could make to prevent unhealthy weight gain in their children, for example, choosing to remove the television from a child’s bedroom (environmental change) or to walk to school instead of drive (behavioral change). The phone sessions were supplemented with workbooks that gave a description of each of the weight-related target behaviors, tips, and example goals, and self-assessment worksheets. The sessions included a goal-setting activity, in which the parents and interventionists identified specific activities the parents would implement prior to the next session, (e.g., do something active as a family each weeknight after dinner.) At the beginning of subsequent sessions, the interventionists would check in to see if the activity was implemented. The intervention design was based on social-cognitive theory (Bandura, 2004), which attributes behavior to knowledge, the environment, attitude, and skills, and motivational Interviewing (Miller & Rollnick, 2002), which uses a participant-centered approach focused on self-determination.

Table 1.

Target Weight–Related Behaviors Covered in the Healthy Homes, Healthy Kids Intervention.

Increase fruit and vegetable intake	Decrease unhealthy snacks
Increase physical activity	Decrease sugary beverages
Increase breakfasts	Decrease TV and other screen time
Decrease eating at restaurants	Increase family meals

Outcome Measures

Independent, trained staff blinded to condition collected outcome data at baseline (prior to randomization) and 12 months (immediately postintervention). Separate staff were trained in coding protocols and coded data from the audio-recorded and transcribed intervention sessions as part of this analysis.

Anthropometry

Twelve-month change in child BMI percentile was calculated from staff-measured height and weight (Seca Corp., Hanover, MD) (Kuczmarski et al., 2002). Both height and weight were measured twice. If the first two measurements differed by more than 0.2 kg for weight or more than 1.0 cm for height, the process was repeated a third time, and the average measurement was used.

Accelerometry

Change in child physical activity from baseline to 12 months was assessed via accelerometers, a small device worn to measure vertical accelerations and estimate physical activity (GT1M, ActiGraph LLC, Pensacola, FL). Accelerometers were worn for 7 days, except while sleeping or doing water activities. The devices were set to collect data in 15-second epochs and aggregated to 1 minute for analysis. Accelerometry data were included if wear time criteria were met (three 10-hour days of wear, with non–wear time defined as 60-minute strings of 0 counts with 2-minute interruption intervals of 100 counts). Average daily accelerometer counts per minute of valid wear time were calculated as a marker of total activity.

Dietary Intake

Dietary intake was measured via a 24-hour recall (Nutrition Data System for Research, Minneapolis, MN) at baseline and 12 months. Portion size estimates were supplemented by an adapted food amounts booklet (van Horn et al., 1993) and three-dimensional cups, bowls, and measuring utensils. Change from baseline to 12 months was calculated for total energy intake (kcals) and servings of fruits/vegetables, unhealthy snacks, and sugary beverages by subtracting baseline values from 12-month values.

Additional Child Weight-Related Behaviors

Additional variables were measured via parent survey at baseline and 12 months. Child screen time was measured by averaging parent-reported amounts of weekday and weekend time spent watching TV or using other media (Schmitz et al., 2004). Survey items also asked parents to estimate how many days in the past week the child had family meals (McGarvey et al., 2004), restaurant meals (Boutelle, Fulkerson, Neumark-Sztainer, Story, & French, 2007), and breakfast meals. The response options for these items were the following: never, 1 to 2 times, 3 to 4 times, 5 to 6 times, and 7 or more times.

Intervention Measures

Session length, time spent discussing specific weight-related behaviors, in-session use of behavior change techniques, and parent implementation of intervention-recommended activities were coded from the audio-recorded and transcribed intervention sessions. All the coders (N = 5) were trained and certified in coding through practice intervention sessions. Weekly meetings were held to discuss coding decisions and prevent coder drift. A randomly selected portion of sessions from N = 20 participants were double coded by the lead coder to evaluate interrater reliability throughout the study.

Session Length

Overall session length was calculated as the total time the parents and interventionists spent talking in phone sessions. This was calculated by summing the lengths of the completed intervention sessions for a given participant.

Time Spent Discussing Weight-Related Behaviors

The time (in minutes) the parents and interventionists spent discussing specific weight-related behaviors lasting more than 1 minute was coded. Discussions covering more than one weight-related behavior were divided equally between behaviors. Time spent talking about each behavior was then summed across all sessions. The average intercoder reliability of time spent discussing each behavior was measured by Pearson correlation coefficient (mean r = .79). Time spent discussing “restaurant frequency” was excluded because of low reliability.

Behavior Change Techniques Used

The number of unique behavior change techniques used during the sessions was coded in five randomly selected session transcripts per participant. All the coders were trained and certified in coding using the 93-item BCTTv1 (Michie et al., 2013) through online training (www.bct-taxonomy.com) and two additional days of study-specific training. Twenty-six techniques identified during training by any coder in the intervention manual, workbooks, or practice sessions constituted the set of techniques that were coded. The coders read the transcript twice, the second time coding line by line any statement that qualified as a behavior change technique used during the session. As an example, the behavior change technique “Review behavior goals (review progress)” was coded in the following statement: “Last session, you set the goal of going to the farmers market to have your daughter pick out three new vegetables. Were you able to do that?” The number of unique behavior change techniques used was then calculated across all coded sessions for a given participant. Average reliability as measured by Cohen’s kappa (κ) was .91.

Activities Implemented by Parents Between Sessions

The intervention activities the parents reported implementing between sessions were coded from the transcripts. When the interventionists asked the parents about their goal progress, the parents reported whether or not they implemented the activity identified in the previous session. When a statement was identified, it was classified under one or more of 11 potential activity categories (Table 2). The definitions of these activity types correspond to behavior change techniques likely used when recommending the activity in the previous session. For example, during an intervention session, the interventionist could recommend the behavior change technique “self-monitoring.” During the next session, if a parent reports having done the self-monitoring strategy over the past week, that statement would be coded as parent implementation of self-monitoring. The number of unique activities reported as implemented by the parents was calculated across the transcribed sessions for each participant. For example, if a participant reported implementing “behavioral practice/rehearsal” at least once across sessions, this activity was identified as “present.” The number of present activities was then calculated for each participant. The average interrater reliability of these items, as measured by Cohen’s kappa (κ), was .92.

Table 2.

Definitions of Activities Parents Could Implement Between Sessions.

Activity category	Definition^a
Goal implementation	Participant reports implementing a goal defined in terms of the behavior achieved
Self-monitoring of behavior	Participant reports implementing the monitoring and recording of their or their child’s behavior(s) as part of a behavior change strategy
Social support, unspecified	Participant reports soliciting social support (e.g., from friends, relatives, colleagues, “buddies,” or staff) to perform the behavior
Social support, practical	Participant reports soliciting practical support (e.g., from friends, relatives, colleagues, “buddies,” or staff) to perform the behavior
Prompts/cues	Participant reports introducing or defining an environmental or social stimulus with the purpose of prompting or cueing their own or their child’s behavior
Behavioral practice/rehearsal	Participant reports practicing or rehearsing performance of the behavior one or more times in a context or at a time when the performance may not be necessary, to increase habit and skill
Behavior substitution	Participant reports substitution of an unwanted behavior with a wanted or neutral behavior
Habit formation	Participant reports rehearsing and repeating the behavior in the same context repeatedly so that the context elicits the behavior
Self-reward	Participant reports implementing self-praise or self-reward if and only if there has been effort and/or progress in performing the behavior
Restructuring the physical environment	Participant reports changing the physical environment to facilitate performance of the wanted behavior or create barriers to the unwanted behavior (other than prompts/cues, rewards, and punishments)
Adding objects to the environment	Participant reports adding objects to the environment to facilitate performance of the behavior

Definitions are based on the Behavior Change Technique Taxonomy (v1).

Statistical Analysis

Descriptive statistics are provided as means, standard deviations (SD), and frequencies. Time spent discussing weight-related behaviors was log transformed because of the largely right-skewed distributions. General linear regression was used to test the overall time in sessions associated with change in child BMI (Hypothesis 1), time spent discussing specific behaviors associated with change in those behaviors (Hypothesis 2), number of behavior change techniques associated with change in child BMI (Hypothesis 3), and number of parent activities implemented between sessions associated with change in child BMI (Hypothesis 5). Models were adjusted for baseline levels of the outcome. Time spent in sessions was considered as a potential confounder. Because of possible clustering by the interventionist, mixed regression models allowing for a random intercept by the interventionist were also assessed. Finally, nonlinear associations were tested using a quadratic term in Hypothesis 1.

Regression tree analysis (Lemon, Roy, Clark, Friedmann, & Rakowski, 2003) was used to test which combinations of behavior change techniques were most associated with change in child BMI percentile (Hypothesis 5). This analysis partitions the study sample into mutually exclusive subgroups defined by the presence or absence of unique behavior change techniques, based on variability in the outcome variable (change in child BMI percentile). Each subgroup continues to be partitioned until the between-subgroup variability in child BMI percentile change is maximized or until a prespecified subgroup sample size is reached. The tree for this analysis was generated such that the minimum subgroup size was n = 12 participants, which would yield up to 8 possible subgroups (96 participants/8 subgroups = 12 participants per subgroup) defined by up to 3 specific techniques (2³ = 8 subgroups). Additional pruning and growing parameters were also evaluated. The final model was adjusted for baseline child BMI percentile. Regression tree analysis is inherently data driven but helps identify variables to be tested in future research.

For all the analyses, significance was assessed using two-tailed tests with alpha set at .05. Regression coefficients, standard errors, and p values are presented and interpreted below. All the analyses were conducted in SAS Version 9.4 (SAS Institute Inc., Cary, NC, 2017).

Power Analysis

Using a general linear model approach (Lenth, 2006), a sample size calculation was conducted assuming 80% power, one predictor, a two-tailed alpha of .05, and a clinically meaningful difference of 2.5 units’ change in child BMI percentile from baseline to 12 months. A standardized minimal detectable effect size was multiplied by the standard deviation of child BMI percentile change from baseline to 12 months for the 181 participants in the intervention arm (SD = 7.7). By coding 100 participants, a child BMI percentile change of 2.2 units (standardized minimal detectable effect size or beta of 0.28) for every 1 standard deviation difference in intervention delivery measure could be detected. This 2.2-unit change is smaller than the clinically meaningful difference of 2.5 units selected above, therefore allowing analyses to detect a meaningful difference with 100 participants.

Results

Descriptive Characteristics

The children included in this analysis were an average of 6.7 years old (SD = 1.7 years), 49% female, and 78% non-Hispanic and White. The parents included in the analysis were an average of 27.4 years old (SD = 6.2 years), 91% female, and 58% employed. Table 3 provides descriptive statistics for time spent in intervention sessions, in-session use of behavior change techniques, and parent implementation of intervention-recommended activities, as well as change in study outcomes (child BMI percentile and weight-related behaviors) from baseline to 12 months. The participants completed 12.0 (SD = 3.9) sessions, lasting an average of 24.7 (SD = 5.1) minutes, for a total intervention time of 297.6 (SD = 89.8) minutes over the 12-month intervention. The interventionists and parents spent the most amount of time discussing “physical activity” (48.8 ± 30.6 minutes), followed by “fruit and vegetable intake” (25.8 ± 24.1 minutes) and “screen time” (18.2 ± 21.5 minutes). A total of 13.9 (SD = 2.8) unique behavior change techniques were used by the interventionists during sessions. Figure 1 shows the percentage of parents whose interventionist used a given behavior change technique in at least one coded session. Goal setting and information gathering were the two most common behavior change techniques, followed by the identifying barrier portion of problem solving and providing social reward in sessions. Other behavior change techniques such as those related to incentives or habit formation were used much less frequently. The parents reported implementing an average of 2.6 (SD = 1.3) unique activities throughout the intervention. Overall, child BMI percentile decreased by 4.0 (SD = 7.5) from baseline to 12 months.

Table 3.

Descriptive Characteristics for Selected Participants, N = 96.

Descriptive characteristics	M	SD
Intervention characteristics
Overall time spent in intervention sessions, minutes	297.6	89.8
Time spent discussing weight-related behaviors, minutes
Physical activity	48.8	30.6
Screen time	18.2	21.5
Energy intake	35.8	28.4
Sugary beverage intake	5.2	8.6
Fruit/vegetable intake	25.8	24.1
Breakfast frequency	4.6	7.8
Family meal frequency	4.4	5.6
Unique behavior change techniques used in sessions, 26 possible	13.9	2.8
Unique activities implemented by parents between sessions, 11 possible	2.6	1.3
Change in child variables from baseline to 12 months
Child BMI percentile	−4.0	7.5
Child physical activity (counts/day)	−34.5	194.7
Child screen time (hours/day)	−0.1	1.0
Child energy intake (kcal/day)	−57.4	538.1
Child sugary beverage intake (servings/day)	+0.1	1.4
Child fruit/vegetable intake (servings/day)	+0.2	2.1
Child breakfast frequency (0–4 scale)	+0.0	0.4
Family meal frequency (0–4 scale)	+0.0	0.9

Note. SD = standard deviation; BMI = body mass index.

Figure 1.

Behavior change techniques used in session transcripts, N = 96 participants.

Time Spent in the Intervention Sessions

Overall time spent in intervention sessions (Hypothesis 1) was significantly associated with change in child BMI percentile. Every 1 hour of time that the parents spent in intervention sessions corresponded to a 1.2 percentile reduction in child BMI from baseline to 12 months (b = −0.02, SE [standard error] = 0.01, p = .01). Results are presented in Figure 2. This association remained after adjusting for baseline level of child BMI percentile and after allowing for a random effect for interventionist. There was no evidence of a quadratic association between time spent in intervention sessions and change in child BMI percentile. There were no significant associations between time spent discussing specific weight-related behaviors and subsequent changes in those behaviors (Hypothesis 2), except in the model with time spent discussing breakfast predicting change in breakfast frequency. This statistical association was driven by only one participant who spent a large amount of time discussing breakfast during the intervention and substantially increased the frequency of breakfast consumption by the 12-month follow-up (Table 4).

Figure 2.

Time spent in intervention sessions compared with change in child body mass index (BMI) percentile, N = 96 participants.

Table 4.

Separate Univariate Models of Dose Delivered by Target Behavior Predicting Change in Related Child Weight-Related Behaviors, N = 96 Participants.^a

Independent variable	Dependent variable	b ^b	p value
Physical activity dose delivered	Change in child physical activity	−0.76	.3032
Screen time dose delivered	Change in child screen time	0.01	.2205
Energy intake dose delivered	Change in child energy intake	−1.76	.2971
Sugary beverage dose delivered	Change in child sugary beverage intake	0.00	.9654
Fruit/vegetable dose delivered	Change in child fruit/vegetable intake	0.01	.1565
Breakfast dose delivered	Change in child breakfast frequency	−0.07	<.0001
Family meal dose delivered	Change in family meal frequency	0.01	.3309

Adjusted for baseline values of specific child weight-related behaviors. ^bUnstandardized betas for separate regression models.

Behavior Change Techniques Used in Sessions

The number of unique behavior change techniques used was not associated with change in child BMI percentile after adjusting for total time in intervention sessions (Hypothesis 3, b = −0.29, SE = 0.26, p = .2748). No statistically significant results were found when testing the exploratory hypothesis that certain behavior change techniques would be associated with greater decreases in child BMI percentile (Hypothesis 4). The regression tree model using a minimum subgroup size of 12 participants that best explained the variance in child BMI did not include any splits (N = 1 leaf, average square error = 50.7). Reducing the minimum subgroup size, turning off pruning, and increasing the chi-square statistic parameters for splitting also did not result in any splits. Thus, none of the specific behavior change techniques significantly explained the variance in child BMI percentile change.

Intervention Activities Implemented by Parents Between Sessions

The number of unique activities the parents reported implementing between sessions was associated with change in child BMI percentile (Hypothesis 5). This remained after adjusting for time spent in intervention sessions. Each additional unique activity the parents reported implementing between sessions, regardless of the amount of time spent in intervention sessions, was associated with a 1.25-unit decrease in child BMI percentile between baseline and 12 months (b = −1.25, SE = 0.52, p = .02). This remained statistically significant after controlling for baseline child BMI percentile and after allowing for a random effect for interventionist.

Discussion

Efficacious behavior change interventions to prevent pediatric obesity are a major public health priority, yet these interventions have shown limited success thus far (Kamath et al., 2008). Measuring and reporting detailed intervention information may lead to a better understanding of the active, effective components of pediatric obesity prevention interventions. Promising intervention factors included in this analysis were time spent in intervention sessions, in-session use of behavior change techniques, and parent implementation of intervention-recommended activities. The results of the current study suggest that overall time spent in intervention sessions is an important consideration, as it was associated with decreases in child BMI percentile. These results are consistent with previous studies using crude measures of dose delivered, such as number of sessions delivered (Foster et al., 2012; Golan, Kaufman, & Shahar, 2006; Jelalian, Mehlenbeck, Lloyd-Richardson, Birmaher, & Wing, 2005; Kalarchian et al., 2009). The hypothesis that a more complex, nonlinear association between intervention dose and outcomes would exist was not supported by the current analysis. One explanation could be that this relatively low-intensity intervention (an average of 5 hours of intervention time over 12 months) was not sufficiently long and/or intense to demonstrate a quadratic effect. This requires further investigation.

The time spent discussing specific weight-related behaviors was not associated with changes in any of these behaviors from baseline to 12 months (e.g., spending more time discussing fruit and vegetable intake was not associated with greater increases in child fruit and vegetable intake). It is possible that the lack of association between these variables is due to measurement bias in these intermediate outcomes (i.e., dietary intake). It has been hypothesized that those participating in behavior change interventions become more accurate at reporting health behaviors postintervention (Senso, Anderson, Crain, Sherwood, & Martinson, 2014). However, this would not influence the objective measures, such as accelerometry, used in this study.

The hypothesis that the use of a greater number of unique behavior change techniques in sessions would be associated with better study outcomes was not observed after adjusting for overall intervention dose. Similarly, there were no specific behavior change techniques identified as important in explaining the variance in child BMI percentile change, as shown by the exploratory regression tree analysis. Though there was adequate sample size to detect a 9-point difference in child BMI percentile between the possible subgroups, a larger sample size would have been able to identify smaller between-subgroup differences. It is also possible that current measures of intervention content relating to the behavior change techniques are not capturing the active components of this type of intervention. Some have suggested that features such as the interpersonal style of the interventionist (Hagger & Hardcastle, 2014) or the interventionist–participant relationship (i.e., the therapeutic alliance; Martin, Garske, & Davis, 2000) may be necessary to explain the variability in the effectiveness of behavior change technique implementation. It is possible that the interactions between the behavior change techniques used and other features such as interventionist interpersonal and communication behaviors may be important in explaining the outcomes.

The last hypothesis in this analysis examined whether the number of unique activities that the parents reported implementing between intervention sessions would be positively associated with child BMI percentile change. Though there was relatively little reporting of the implementation of activities between sessions (only 2.6 of the 11 possible activities reported on average), the number of unique activities the parents reported implementing was positively associated with a reduction in child BMI percentile. This finding suggests that parental implementation of intervention-recommended activities is an important factor influencing the effectiveness of behavior change interventions. Similar findings have been reported in the adult weight loss literature, with implementation of intervention activities such as self-monitoring of diet or physical activity and self-weighing between sessions being consistently associated with improved weight loss (Burke, Wang, & Sevick, 2011). The specific reasons why the parents did not implement the recommended activities were not explored in this study, but they could include lack of support, lack of time, or competing priorities. An essential next step in this work is to identify key barriers to implementation and understand how various behavior change techniques could be delivered in sessions to address these barriers. Future research could also focus on the participant characteristics that may predict successful implementation of study goals and activities.

This study has a number of strengths, including the use of objective measures to identify features of intervention delivery; measurement of these features on an individual participant level, allowing for comparison with outcomes; and the reported high reliability of measuring these features. Still, there are some important limitations. One limitation is that since the data were from only one arm of a previously conducted trial, the participants were not randomly allocated to differing levels of intervention implementation. To address this limitation, a thorough evaluation of potential confounders, including baseline child BMI percentile and the possible random effect of interventionist, was used. Another limitation of this work is the acknowledgement that many of the additional features discussed above likely contribute to a participant’s success (e.g., the interventionist–participant relationship, participant engagement or intention, interventionist attributes, and additional demographic characteristics). The specific features were chosen for this study because of their measurability and their initial promise in the existing literature; other features may also be important.

The current analysis suggests that the amount of time spent in intervention sessions is an important factor in intervention success, as is the number of intervention activities implemented by parents between sessions. There appears to be more complexity in the association between study outcomes and the number or type of behavior change techniques delivered by interventionists. Though these two factors were not associated with intervention outcomes, this study has developed coding protocols that could be extended to cover other factors in future studies. Such studies would help identify the active components in behavior change interventions to prevent unhealthy weight gain in children.

Footnotes

Acknowledgements

The authors would like to acknowledge Dani M. Bredeson, Molly J. Colombo, Shannon N. Gerberding, and Ashley L. Barthel for their help with the data collection and coding.

Declaration of Conflicting Interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Author Susan Michie is Director of the Centre for Behaviour Change, University College London, which has received funds from industry and government agencies. All the other authors declare that they have no potential conflicts of interest.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (#R01DK084475, Co-PIs Sherwood and Levy; T32DK083250, PI Jeffery; P30DK050456, PI Levine; and P30DK092924 PI O’Connor).

ORCID iD

Meghan M. JaKa

References

Baldwin

S. A.

Berkeljon

Atkins

D. C.

Olsen

J. A.

Nielsen

S. L.

(2009). Rates of change in naturalistic psychotherapy: Contrasting dose-effect and good-enough level models of change. Journal of Consulting and Clinical Psychology, 77, 203-211.

Bandura

(2004). Health promotion by social cognitive means. Health Education & Behavior, 31, 143-164. doi:10.1177/1090198104263660

Baranowski

Cerin

Baranowski

(2009). Steps in the design, development and formative evaluation of obesity prevention-related behavior change trials. International Journal of Behavioral Nutrition and Physical Activity, 6, 6.

Boutelle

K. N.

Fulkerson

J. A.

Neumark-Sztainer

Story

French

S. A.

(2007). Fast food for family meals: Relationships with parent and adolescent food intake, home food availability and weight status. Public Health Nutrition, 10, 16-23.

Burke

L. E.

Wang

Sevick

M. A.

(2011). Self-monitoring in weight loss: A systematic review of the literature. Journal of the American Dietetic Association, 111, 92-102.

Foster

G. D.

Sundal

McDermott

Jelalian

Lent

M. R.

Vojta

(2012). Feasibility and preliminary outcomes of a scalable, community-based treatment of childhood obesity. Pediatrics, 130, 652-659.

Gainforth

H. L.

Lorencatto

Erickson

West

Michie

(2016). Characterizing clients’ verbal statements in behavioural support interventions: The case of smoking cessation. British Journal of Health Psychology, 21, 600-612.

Gearing

R. E.

El-Bassel

Ghesquiere

Baldwin

Gillies

Ngeow

(2011). Major ingredients of fidelity: A review and scientific guide to improving quality of intervention research implementation. Clinical Psychology Review, 31, 79-88.

Golan

Kaufman

Shahar

D. R.

(2006). Childhood obesity treatment: Targeting parents exclusively v. parents and children. British Journal of Nutrition, 95, 1008-1015.

10.

Hagger

M. S.

Hardcastle

S. J.

(2014). Interpersonal style should be included in taxonomies of behavior change techniques. Frontiers in Psychology, 5, 254.

11.

Heerman

W. J.

JaKa

M. M.

Berge

J. M.

Trapl

E. S.

Sommer

E. C.

Samuels

L. R.

. . . Barkin

S. L.

(2017). The dose of behavioral interventions to prevent and treat childhood obesity: A systematic review and meta-regression. International Journal of Behavioral Nutrition and Physical Activity, 14, 157. doi:10.1186/s12966-017-0615-7

12.

Hendrie

G. A.

Brindal

Corsini

Gardner

Baird

Golley

R. K.

(2012). Combined home and school obesity prevention interventions for children: What behavior change strategies and intervention characteristics are associated with effectiveness? Health Education & Behavior, 39, 159-171.

13.

JaKa

M. M.

Haapala

J. L.

Trapl

E. S.

Kunin-Batson

Olson-Bullis

B. A.

Heerman

. . . Sherwood

N. E.

(2016). Reporting of treatment fidelity in behavioural paediatric obesity intervention trials: A systematic review. Obesity Reviews, 17, 1287-1300.

14.

JaKa

M. M.

Seburg

E. M.

Roeder

A. M.

Sherwood

N. E.

(2015). Objectively coding intervention fidelity during a phone-based obesity prevention study. Journal of Obesity and Overweight, 1, 102.

15.

Janicke

D. M.

Steele

R. G.

Gayes

L. A.

Lim

C. S.

Clifford

L. M.

Schneider

E. M.

. . . Westen

(2014). Systematic review and meta-analysis of comprehensive behavioral family lifestyle interventions addressing pediatric obesity. Journal of Pediatric Psychology, 39, 809-825. doi:10.1093/jpepsy/jsu023

16.

Jelalian

Mehlenbeck

Lloyd-Richardson

Birmaher

Wing

(2005). “Adventure therapy” combined with cognitive-behavioral treatment for overweight adolescents. International Journal of Obesity, 30, 31-39.

17.

Kalarchian

M. A.

Levine

M. D.

Arslanian

S. A.

Ewing

L. J.

Houck

P. R.

Cheng

. . . Marcus

M. D.

(2009). Family-based treatment of severe pediatric obesity: Randomized, controlled trial. Pediatrics, 124, 1060-1068.

18.

Kamath

C. C.

Vickers

K. S.

Ehrlich

McGovern

Johnson

Singhal

. . . Montori

V. M.

(2008). Behavioral interventions to prevent childhood obesity: A systematic review and metaanalyses of randomized trials. Journal of Clinical Endocrinology & Metabolism, 93, 4606-4615.

19.

Kazdin

A. E.

(1974). Reactive self-monitoring: The effects of response desirability, goal setting, and feedback. Journal of Consulting and Clinical Psychology, 42, 704-716.

20.

Kuczmarski

R. J.

Ogden

C. L.

Guo

S. S.

Grummer-Strawn

L. M.

Flegal

K. M.

Mei

. . . Johnson

C. L.

(2002). 2000 CDC growth charts for the United States: Methods and development. Vital and Health Statistics: Series 11, No. 246, 1-190.

21.

Leblanc

Ritchie

(2001). A meta-analysis of play therapy outcomes. Counselling Psychology Quarterly, 14, 149-163.

22.

Lemon

S. C.

Roy

Clark

M. A.

Friedmann

P. D.

Rakowski

(2003). Classification and regression tree analysis in public health: Methodological review and comparison with logistic regression. Annals of Behavioral Medicine, 26, 172-181.

23.

Lenth

R. V.

(2006). Java applets for power and sample size [Computer software]. Retrieved from http://www.stat.uiowa.edu/~rlenth/Power

24.

Lorencatto

West

Bruguera

Brose

L. S.

Michie

(2016). Assessing the quality of goal setting in behavioural support for smoking cessation and its association with outcomes. Annals of Behavioral Medicine, 50, 310-318.

25.

Lorencatto

West

Christopherson

Michie

(2013). Assessing fidelity of delivery of smoking cessation behavioural support in practice. Implementation Science, 8, 40.

26.

Martin

D. J.

Garske

J. P.

Davis

M. K.

(2000). Relation of the therapeutic alliance with outcome and other variables: A meta-analytic review. Journal of Consulting and Clinical Psychology, 68, 438-450.

27.

Martin

Chater

Lorencatto

(2013). Effective behaviour change techniques in the prevention and management of childhood obesity. International Journal of Obesity, 37, 1287-1294.

28.

McGarvey

Keller

Forrester

Williams

Seward

Suttle

D. E.

(2004). Feasibility and benefits of a parent-focused preschool child obesity intervention. American Journal of Public Health, 94, 1490-1495.

29.

Michie

Hardeman

Fanshawe

Prevost

A. T.

Taylor

Kinmonth

A. L.

(2008). Investigating theoretical explanations for behaviour change: The case study of ProActive. Psychology and Health, 23, 25-39.

30.

Michie

Richardson

Johnston

Abraham

Francis

Hardeman

. . . Wood

C. E.

(2013). The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: Building an international consensus for the reporting of behavior change interventions. Annals of Behavioral Medicine, 46, 81-95.

31.

Michie

Wood

C. E.

Johnston

Abraham

Francis

Hardeman

(2015). Behaviour change techniques: The development and evaluation of a taxonomic method for reporting and describing behaviour change interventions (a suite of five studies involving consensus methods, randomised controlled trials and analysis of qualitative data). Health Technology Assessment, 19(99), 1-188. doi:10.3310/hta19990

32.

Miller

W. R.

Rollnick

(2002). Motivational interviewing: Preparing people for change (2nd ed.). New York, NY: Guilford Press.

33.

Romanczyk

R. G.

Tracey

D. A.

Wilson

G. T.

Thorpe

G. L.

(1973). Behavioral techniques in the treatment of obesity: A comparative analysis. Behaviour Research and Therapy, 11, 629-640.

34.

Schmitz

K. H.

Harnack

Fulton

J. E.

Jacobs

D. R.

Jr. Gao

Lytle

L. A.

Van Coevering

(2004). Reliability and validity of a brief questionnaire to assess television viewing and computer use by middle school children. Journal of School Health, 74, 370-377.

35.

Senso

M. M.

Anderson

C. P.

Crain

A. L.

Sherwood

N. E.

Martinson

B. C.

(2014). Self-reported activity and accelerometry in 2 behavior-maintenance trials. American Journal of Health Behavior, 38, 254-264.

36.

Sherwood

N. E.

Levy

R. L.

Langer

S. L.

Senso

M. M.

Crain

A. L.

Hayes

M. G.

. . . Jeffery

R. W.

(2013). Healthy Homes/Healthy Kids: A randomized trial of a pediatric primary care-based obesity prevention intervention for at-risk 5–10-year-olds. Contemporary Clinical Trials, 36, 228-243.

37.

Spencer

(1978). Two types of role playing: Threats to internal and external validity. American Psychologist, 33, 265-268.

38.

Spiegel

A. M.

Nabel

E. G.

(2006). NIH research on obesity and type 2 diabetes: Providing the scientific evidence base for actions to improve health. Nature Medicine, 12, 67-69.

39.

Stunkard

(1972). New therapies for the eating disorders: Behavior modification of obesity and anorexia nervosa. Archives of General Psychiatry, 26, 391-398.

40.

Summerbell

C. D.

Waters

Edmunds

Kelly

Brown

Campbell

K. J.

(2005). Interventions for preventing obesity in children. Cochrane Database of Systematic Reviews, 3(3), CD001871.

41.

van Horn

L. V.

Stumbo

Moag-Stahlberg

Obarzanek

Hartmuller

V. W.

Farris

R. P.

. . . Liu

. (1993). The Dietary Intervention Study in Children (DISC): Dietary assessment methods for 8- to 10-year-olds. Journal of the American Dietetic Association, 93, 1396-1403.