Classifying Intervention Components Used in Sleep Duration Interventions for Children: A Systematic Review and Meta-Analysis

SUMMARY

The Multiphase Optimization Strategy (MOST) is a three-phase iterative framework that could accelerate the development of behavioral interventions. This systematic review and meta-analysis was conducted within the MOST preparation phase and aimed to classify components included in pediatric sleep duration interventions, using the Behavior Change Technique (BCT) Taxonomy. Across 37 interventions, 46 out of 93 BCTs have been used, with an average of 8 techniques used per study. The most common BCTs used were instruction on how to perform the behavior (N=29; code 4.1), practical social support (N=22; code: 3.2), and behavioral practice/rehearsal (N=22; code: 8.1). A latent class analysis identified two classes of interventions, distinguished by the presence of BCTs falling within the following behavior change groups: shaping knowledge, natural consequences, comparison of behavior, and repetition and substitution. Our meta-analysis revealed that interventions belonging to the latent class with these behavior change groups (N=15) had a pooled positive intervention effect of 14 minutes (95% CI: 8–21) versus 8 minutes (95% CI: 1–15) for interventions without these behavior change groups (N=19). This systematic review and meta-analysis will enhance the development of sleep promotion interventions and guide the selection of candidate intervention components for future optimization and randomized control trials.

Introduction

Age specific sleep duration guidelines have been established for infants, children, and adolescents [1]. Sufficient sleep duration has been linked to multiple developmental and health outcomes including improved alertness [2], neurodevelopment [3], mental health [4], and metabolic health [5–7]. With sleep duration having such broad impacts, getting adequate sleep has implications for enhancing cognition and attention [8]; preventing risky behavior [9]; reducing anxiety and depressive symptoms [4]; and lowering the risk of obesity [6, 7] and the development of type 2 diabetes [10] as well as other cardiovascular disease risk factors [11] [12]. It is therefore critical to increase the proportion of children who acquire sufficient sleep duration regularly.

Estimates suggest that 53% of middle school aged and 65% of high school aged children in the United States obtain insufficient sleep duration on school nights [13], and a Healthy People 2030 goal is to lower the prevalence to 29% [14]. Further, there is evidence that increasing sleep duration by 30 minutes per night can lead to improvements in clinical and health related quality of life outcomes [2, 15, 16]. However, a recent meta-analysis revealed that nonpharmacological sleep duration interventions for youth have an overall effectiveness of increasing sleep duration by 12 minutes per night [17], indicating that the current collection of intervention approaches are not yet sufficiently effective. Importantly, a detailed classification of the intervention components used in prior sleep duration interventions has not been performed [17, 18]. This knowledge gap limits our ability to fully understand the specific behavior change techniques employed in past studies, and therefore our ability to design novel interventions needed to surpass current levels of effectiveness.

The Multiphase Optimization Strategy (MOST) framework is an engineer-inspired framework for designing and testing multi-component interventions that could accelerate the development of more effective pediatric sleep duration interventions [19]. This iterative, three-phase framework is used to identify and pilot test candidate intervention components (preparation phase), determine which of the candidate components should be retained using an optimization trial (optimization phase), and then to confirm the effectiveness of the optimized intervention using a randomized controlled trial (evaluation phase) [19]. We are using the MOST framework to design a sleep promotion intervention in children for potential use in pediatric healthcare [20–22]. From our experience, the identification of candidate intervention components could be enhanced during the preparation phase of the MOST framework. More specifically, under the preparation phase of the MOST framework, the Behavior Change Technique Taxonomy (BCTT) can be used as a method to classify intervention components, in a standardized manner, to help with the selection of candidate intervention components [23].

The BCTT includes 93 behavior change techniques falling within 16 behavior change groups [23]. To the best of our knowledge only one pediatric sleep duration intervention used the BCTT to explicitly state the behavior change techniques used in their intervention [24]. Further, while past reviews have incorporated BCTT to summarize intervention components in sleep related interventions [25–28], only one focused on sleep duration interventions that included pediatric studies (adolescents and young adults) [29]. Our overall objective was therefore to assign behavior change groups and techniques using the BCTT to classify intervention components used in pediatric sleep duration interventions. Within the preparation phase of the MOST framework, this systematic review and meta-analysis aimed to 1) document and report the frequency of behavior change techniques used in non-pharmacological sleep duration interventions for children and 2) determine if certain behavior change techniques are more commonly included in the most effective pediatric sleep duration interventions. This systematic review and meta-analysis provides important knowledge on intervention components used in past research to assist with the selection of candidate intervention components in future optimization trials and randomized controlled trials. This knowledge will enhance the development of sleep promotion interventions and help surpass current levels of effectiveness.

Methods

This study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines.

Inclusion & Exclusion Criteria

Trials that included randomization to a study arm, or study condition, that aimed to increase sleep duration in typically developing youth aged 1–18 years were targeted for inclusion. For a randomized controlled trial (RCT), the control group could be any non-exposed control group, including a waitlist control group or a minimally exposed comparator group (e.g., standard of care). Optimization trials, with no control group, were eligible for inclusion, but were not included in the meta-analysis. Studies were excluded if: 1) they targeted infants <1 year of age or adults >18 years of age; 2) they targeted youth with a previously diagnosed sleep disorder, such as insomnia; 3) the intervention was pharmacological in nature; 4) the study did not include randomization to a study arm or study condition; 5) the study did not report sleep duration as an outcome; or 6) the study reported sleep duration data from a prior study.

Search Strategy

The following databases were used to identify articles: EMBASE, PubMed, Web of Science, MEDLINE, CENTRAL, and PsycINFO. In brief, the following phrases were used to build each database specific search strategy: child; pediatrics; adolescent; sleep duration, quality, pattern, or hygiene; bedtime; time in bed; sleep time; randomized controlled trial; humans. The complete search strategies used for each database are included in Supplementary Table 1; these strategies match those used in the systematic review and meta-analysis by Magee et al. [17] Our searches were month and year filtered to identify articles published between January 2019 and July 2023. We used this date range because the searches by Magee et al. identified sleep duration interventions from database initiation to November 2019 [17]. Therefore, articles published before November 2019 were sourced from the Magee et al article. We also queried reference lists of other relevant review papers to identify applicable articles [29, 30].

Selection Process

The articles identified from the search strategies were managed and selected using Covidence (Veritas Health Innovation, Melbourne, Australia). The 6 lists of articles identified from each database were uploaded into Covidence and duplicate references were automatically identified and removed. The titles and abstracts were then screened independently by two of three reviewers (JD, MF, or GV), followed by a full article screen (if applicable) independently by two of four reviewers (JD, MF, GV, or SS). The group convened an additional reviewer (JM) to resolve conflicts identified when screening titles and abstracts and full articles.

Data Collection

Study Characteristics

The following characteristics from each retained article were entered into Covidence: first author, year of publication, age of research participants, target population, setting, inclusion criteria, exclusion criteria, mode of intervention delivery, intervention overview, intervention dose, intervention duration, and study duration.

Risk of Bias Assessment

Version 2 of the Cochrane risk-of-bias for randomized trials (RoB2) was used to assess the risk of bias and overall quality of all studies included in the review. RoB2 assesses the quality of studies across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result.[31] Bias domains and overall study quality could be judged as 1) low risk of bias, 2) some concerns, and 3) high risk of bias [31]. Two researchers (MF, GV) independently extracted data to assess the risk of bias and quality of the studies included and jointly resolved discrepancies.

Behavior Change Groups and Techniques

The BCTT includes 16 groups of 93 hierarchically clustered techniques [23]. The 16 groups include: goals and planning, feedback and monitoring, social support, shaping knowledge, natural consequences, comparison of behavior, associations, repetition and substitution, comparison of outcomes, reward and threat, regulation, antecedents, identity, scheduled consequences, self-belief, and covert learning [23]. The coding scheme indicates the behavior change group and the behavior change technique ([group integer].[technique integer]).

A total of four reviewers (JD, MF, GV, JM) contributed to assigning BCT codes. Each reviewer carefully read the primary BCTT article[23] and the BCTT manual. In pairs to reach consensus, the specific BCT codes were assigned based on the description of the intervention components from each primary article (or a related protocol paper, if available). Each reviewer referred to the BCTT manual when assigning the BCT codes and the manual was also leveraged to resolve disputes between reviewing pairs. The primary reviewer and agreed upon codes were documented in Excel. Brief notes were kept that explained why each code was assigned. Using these notes, we derived themes to provide examples of sleep-related intervention approaches used for given code assignments.

Latent Class Analysis

Certain behavior change techniques may be more likely to be used in combination, as it may make scientific sense to use them together. Indeed, the BCTT manual indicates that some behavior change techniques should be coded in tandem (e.g., goal setting behavior is applied in tandem with action planning if the goal defines a specific context) [23]. To determine if combinations of behavior change strategies do exist, we utilized a latent class analysis (LCA). At the behavior change technique level, there were 46 codes for 37 studies. We considered reducing the data at the behavior change technique level, but there was insufficient data at this resolution. We therefore reduced the data at the behavior change group level (14 codes). The indicator for class membership was later used as a covariate in the meta-analysis to examine differences in the pooled effect by classes. The LCA was performed using Mplus version 8 [32].

Quantitative Data for Meta-Analysis

The following data points were extracted from each article for the meta-analysis: sample size, mean sleep duration (standardized to minutes), the reported metric of variance (standard deviation, standard error, or 95% confidence interval), and duration of the intervention (standardized to days). To be included in the meta-analysis, the study had to be an RCT and post-intervention sleep duration data in hours and/or minutes had to be provided. If a study reported sleep duration data from actigraphy and self-report, the actigraphy reported data were used in the meta-analysis as this is a device-based approach that does not rely on participant recall or caregiver recall. The summary statistics for sleep were captured for the control/comparison group and the intervention group at baseline and all follow-up time points.

We used standard methods to synthesize the abstracted data depending on the type of measurements provided. For studies reporting sample size, mean, and standard deviations (or standard error) for intervention and control groups, we estimated the mean difference (MD) in sleep duration in minutes and standard error of the MD for each follow-up time point. Some studies [21, 33, 34] reported the intervention effect and 95% confidence interval which was taken as the MD and standard errors calculated from the confidence interval.

Given the potential for multiple effect size estimates from a single study, we applied a multilevel meta-regression model using random effect and robust standard errors to account for the correlation among estimates from the same study [35]. We examined the sensitivity of the pooled effect with regard to the latent class group by adding this variable to the model. Publication bias was examined using a contour-enhanced funnel plot [36]. The meta-analysis was performed in Stata version 18 using the meta suite of commands.

Results

Search Results

The search strategy yielded 9,050 research articles (Figure 1). After removing duplicates and completing the screening process of titles and abstracts, the full text of 87 articles were reviewed (Figure 1). Of these, 37 articles were included in the systematic review and 34 were included in the meta-analysis (Figure 1).

Figure 1.

Flow chart showing the identification and screening of eligible research articles.

Study Characteristics

The characteristics of all studies included in the systematic review are outlined in Table 1. Overall, the age of the participants from the included articles ranged from 2 to 18 years, and the mean age across all studies was 12.0 (SD=4.3) years. The interventions were delivered in school only (N=23) [24, 33, 34, 38–57], home only (N=11) [16, 21, 22, 58–64], school and home (N=1) [65], and laboratory and home (N=2) [66, 67]. The duration of the interventions ranged from one day to eight weeks; the mean duration was 6.5 (SD: 8.8) weeks and the median duration was 4 weeks. The total study duration ranged from approximately 3-to 52-weeks; the mean study duration was 17 (SD: 21) weeks, and the median study duration was 10 weeks. The approach for measuring sleep duration included sleep diaries (N=11) [34, 44, 45, 50–52, 54, 58, 60, 61, 67], self-report questionnaires (N=9) [24, 39–41, 49, 56, 59, 65], actigraphy and sleep diaries (N=8) [16, 22, 33, 38, 42, 46, 62, 64], actigraphy only (N=6) [21, 53, 57, 63, 66], parent-report questionnaires (N=2) [43, 47], and actigraphy and self-report questionnaires (N=1) [48].

Table 1.

Characteristics of studies included in the systematic review.

First Author	Year	Country	N	Age (Mean ± SD [or see footnote])	Setting	Intervention Duration (weeks)	Study Duration (weeks)	Sleep Measurement	Total Groups	Total Techniques
Beijamini [38]	2012	Brazil	21	13–14	School	1	1	Actigraphy Sleep Diary	5	6
Blake [33]	2016	Australia	144	14 ± 1.13	School	7	9	Actigraphy Sleep Diary	10	14
Bonuck [34]	2022	USA	519	3 ± 0.1	School	2	52	Sleep Diary	6	7
Cain [39]	2011	Australia	104	16 ± 0.4	School	4	6	Questionnaires	7	13
Champion [40]	2023	Australia	6640	13 ± 0.5	School	6	104	Questionnaires	9	13
Crowley [66]	2023	USA	46	16 ± 0.8	Laboratory Home	2	4	Actigraphy	5	11
Das-Friebel [41]	2019	Switzerland	352	15 ± 1.7	School	0	4	Questionnaires	8	9
Dewald-Kaufmann [42]	2013	The Netherlands	55	15 [12.8 – 18.5]a	School	2	3	Actigraphy Sleep Diary	6	9
Dong [58]	2020	USA	176	15 ± 1.8	Home	6	26	Sleep Diary	9	12
Hart [16]	2017	USA	78	10 ± 1.0	Home	8	10	Actigraphy Sleep Diary	4	9
Hiscock [43]	2019	Australia	334	6 ± 0.4	School	4	52	Questionnairesb	6	10
Inhulsen [44]	2022	The Netherlands	972	13 ± 0.7	School		12	Sleep Diary	6	6
Kaplan [67]	2019	USA	87	16 ± 1.1	Laboratory Home	4	5	Sleep Diary	7	9
Kira [45]	2014	New Zealand	29	15 ± 1.1	School	4	10	Sleep Diary	6	6
Lin [24]	2018	Iran	2841	116 ± 1.1	School	8	24	Questionnaires	9	12
Lufi [46]	2011	Israel	47	14 ± 0.3	School	1	2	Actigraphy Sleep Diary	3	3
Mindell [60]	2009	USA	405	0.6 – 3c	Home	2	3	Sleep Diary	5	5
Mindell [59]	2016	USA	152	6 ± 3.0	Home	4	4	Questionnaires	4	6
Mitchell [21]	2021	USA	30	11 ± 0.8	Home	7	9	Actigraphy	5	6
Mitchell [21]	2021	USA	43	11 ± 0.8	Home	7	9	Actigraphy	6	7
Mitchell [22]	2023	USA	97	11.5	Home	7	44	Actigraphy Sleep Diary	4	7
Mousarrezaei [61]	2020	Iran	206	9.1 ± 1.7	Home	4	13	Sleep Diary	4	4
Quach [47]	2018	Australia	452	7 ± 0.4	School	6	26	Questionnairesa	2	3
Rigney [48]	2015	Australia	296	12 ± 0.6	School	4	18	Actigraphy Questionnaires	6	8
Santiago [49]	2022	Brazil	30	16 ± 1.5	School	12	12	Questionnaires	6	8
Sermet Kaya [65]	2022	Turkey	44	16c	School Home	16	40	Questionnaires	4	4
Sousa [50]	2013	Brazil	34	17 ± 0.6	School	1	5	Sleep Diary	5	6
Tamura [51]	2014	Japan	148	4th, 5th, 6th grade d	School	2	2	Sleep Diary	7	8
Tamura [52]	2016	Japan	243	12–13c	School	2	2	Sleep Diary	8	12
Uhlig [53]	2019	The Netherlands	75	8–13c	School	16	16	Actigraphy	5	6
Van Dyk [62]	2017	USA	76	16 ± 1.1	Home	5	5	Actigraphy Sleep Diary	6	15
van Rijn [57]	2020	Singapore	210	14 ± 0.3	School	4	11	Actigraphy	7	10
Willliamson [63]	2023	USA	27	3 ± 1.2	Home	1	4	Actigraphy	4	5
Wilson [54]	2014	USA	142	4 ± 0.5	School	2	4	Sleep Diary	8	11
Wing [55]	2015	China	3713	15 ± 1.5	School	8	13	Questionnaires	9	10
Wolfson [56]	2015	USA	143	7th graded	School	52	52	Questionnaires	9	12
Yoong [64]	2019	Australia	76	4 ± 0.5	Home	12	12	Actigraphy Sleep Diary	6	7

Abbreviations: BCT, behavior change techniques.

^aRange was reported.

^bParent-reported questionnaires.

^cAge was reported as a range only.

^dChronological age in years was not reported.

Risk of Bias in Studies

The risk of bias ratings overall and by domain for each study are shown in Figure 2. The risk of bias assessment revealed that the majority of articles had a low risk of bias due to the randomization process (N=21; 57%). Nearly all articles had some concern for risk of bias due to deviations from the intended interventions (N=35, 95%). For example, in most studies, both participants and personnel were aware of the intervention, did not provide enough information as to whether non-protocol interventions were balanced across groups, but did describe an appropriate analysis to estimate the adherence effect. All articles (N=37, 100%) had low risk of bias due to missing outcome data and selection of the reported result. The majority (N=26, 70%) of the articles had some concern for risk of bias due to measurement of the outcome. For example, one study utilized a non-validated questionnaire for assessment of the primary outcome. For the overall risk of bias, most (N=31, 84%) of the articles were rated as having “some concerns”; five (14%) were rated as high risk of overall bias; and one (3%) was rated as low risk of bias.

Figure 2.

Cochrane Risk of Bias 2 (RoB 2.0) domain and overall ratings by study. D1: Bias risk from the randomization process; D2: Bias due to deviations from intended interventions; D3: Bias due to missing outcome data; D4: Bias in measurement of the outcome; D5: Bias in selection of the reported result. Green indicates low risk; Yellow indicates some concerns; Red indicates high risk.

Behavior Change Technique Taxonomy (BCTT)

Across the pediatric sleep duration interventions included, 14 out of 16 behavior change groups were used (Figure 3) and 46 out of 93 behavior change techniques were used (Figure 4). The average number of groups used per study was 6 and the average number of techniques used per intervention was 8 (Table 1). The 2 behavior change groups and 47 behavior change techniques not used in current pediatric sleep interventions are in Supplementary Table 2.

Figure 3.

Frequency of behavior change groups used in pediatric sleep duration interventions (N=37).

Figure 4.

Frequency of behavior change techniques used in pediatric sleep duration interventions (N=37).

The most frequently used groups were shaping knowledge (N=29; group #4), goals and planning (N=26; group #1), and social support (N=26; group #3; Figure 3); and the most frequently used techniques were instruction on how to perform the behavior (N=29; code: 4.1), social support, practical (N=22; code: 3.2), and behavioral practice/rehearsal (N=22; code: 8.1; Figure 4). The least utilized groups were identity (N=1; group #13), scheduled consequences (N=3; group #14), and regulation (N=4; group #11; Figure 3); and the least utilized techniques included review outcome goal(s) (N=1; code: 1.7), remove aversive stimulus (N=1; code: 7.5), and body changes (N=1; code: 12.6; Figure 4).

Themes are listed in Table 2 to provide context linking behavior change techniques to specific intervention approaches. Using the most common behavior change techniques as examples, for several studies that were assigned the behavior change technique code 4.1 (instruction on how to perform the behavior), the intervention often included an education component that focused on sleep hygiene (Table 2). For several studies that were assigned behavior change techniques code 3.2 (social support, practical), the intervention often provided intervention materials to parents (Table 2). And for several studies that were assigned behavior change technique code 8.1 (behavioral practice/rehearsal), the intervention often included a component that encouraged practicing bedtime routines (Table 2).

Table 2.

Behavior Change Technique Taxonomy (BCTT) groups, techniques, and themes (N=37).

Behavior Change Group		Behavior Change Technique		Themes
1	Goals and planning	1.1	Goal setting (behavior)	● Goals targeting better planning [43, 66] ● Goals targeting sleep duration [21, 41] ● Goals targeting bedtime [16, 21, 39, 41, 42, 51, 52, 60] ● Goals set but unspecified to sleep behaviors [40, 56, 67]
		1.2	Problem solving	● Problem solving barriers and challenges to getting sufficient sleep [16, 21, 22, 24, 43, 50, 57, 62, 65, 66] ● Problem solving to understand if sleep behaviors align with recommendations [39, 52, 58, 64] ● Problem solving to prevent setbacks in sleep improvements [33, 58]
		1.3	Goal setting (outcome)	● Goals targeting specific sleep outcomes [16, 22, 33, 53, 58]
		1.4	Action planning	● Action planning to establish schedules and routines [16, 21, 22, 24, 34, 40, 49, 57–59, 64, 66, 67] ● Action planning to remove distractions and reduce media consumption before bed [53, 59, 64] ● Action planning to resolve barriers to achieve sleep goals [16, 33, 67]
		1.5	Review behavior goal(s)	● Reviewing sleep behavior goals [43, 66]
2	Feedback and monitoring	2.2	Feedback on behavior	● Feedback on changes in sleep behaviors [22, 51, 62, 66]
		2.3	Self-monitoring of behavior	● Self-monitoring using sleep diaries [24, 39, 42, 43, 46, 48, 51, 52, 55, 57, 58, 60, 62, 66, 68] ● Self-monitoring of sleep using actigraphy [16, 46] ● Self-monitoring of sleep using other tools [33, 40, 44, 51, 56]
		2.6	Biofeedback	● Self-monitoring of sleep using actigraphy with real-time feedback [16, 21, 22, 62]
		2.7	Feedback on outcome(s) of behavior	● Review sleep duration outcomes following the intervention [16, 55]
3	Social support	3.2	Social support (practical)	● Practical support from parents, no materials [21, 24, 33, 34, 43, 48, 53, 59–61, 65, 67] ● Practical support from parents, with materials [22, 41, 45, 54, 56] ● Practical support through group activities [40, 45, 50, 53]
		3.3	Social support (emotional)	● Emotional support from parent attendance at intervention sessions [24, 48] ● Emotional support through motivational interviewing [58, 64, 66, 67]
4	Shaping knowledge	4.1	Instruction on how to perform a behavior	● Sleep education delivered in person [16, 24, 33, 34, 38–45, 48, 50–52, 54–58, 60, 62, 65, 67] ● Sleep education delivered digitally [21, 40, 61]
		4.2	Information about antecedents	● Information about causes of poor sleep [33, 38, 39, 42, 50, 52–57, 69]
5	Natural consequences	5.1	Information about health consequences	● Information about consequences of suboptimal sleep and benefits of optimal sleep [24, 33, 34, 38–41, 43–45, 48, 50–52, 54, 55, 57, 64, 67]
		5.6	Information about emotional consequences	● Information on suboptimal sleep and mental health [39, 40, 43, 48]
6	Comparison of behavior	6.1	Demonstration of the behavior	● Sleep education class/program [24, 33, 38–41, 44, 45, 48, 51, 52, 55, 57] ● Role-playing to practice bedtime routines [34, 39, 40, 44, 49, 54, 56]
		6.2	Social comparison	● Peer comparisons to improve sleep [21, 40]
7	Associations	7.1	Prompts/cues	● Physical prompts to encourage behavior change for optimal sleep [38, 54, 55, 59, 63] ● Digital prompts to encourage behavior change for optimal sleep [40, 64]
8	Repetition and substitution	8.1	Behavioral practice/rehearsal	● Sleep education class/program [24, 33, 38–41, 44, 45, 48, 51, 52, 55, 57] ● Development of time management and bedtime routines to put into practice later for habit formation [34, 41, 53, 54, 57, 58, 66]
		8.3	Habit formation	● Development of time management and bedtime routines to put into practice later for habit formation [34, 41, 51–54, 56, 57, 59, 61, 66]
9	Comparison of outcomes	9.1	Credible source	● Professionals with knowledge in sleep health involved in intervention delivery [16, 24, 33, 39, 41–43, 51, 52, 54–56, 58, 61, 64, 65, 67]
		9.2	Pros and cons	● Pros and cons of optimizing sleep health [24, 39]
11	Regulation	11.2	Reduce negative emotions	● Managing stress/anxiety to improve sleep [33, 55, 58]
12	Antecedents	12.1	Restructuring the physical environment	● Physical changes made to the bedroom environment to promote sleep [24, 41, 42, 52, 54, 59, 64]
		12.2	Restructuring the social environment	● Sleep related time management advice given in a group/ social setting [55, 57]
		12.3	Avoidance/ reducing exposure to cues for the behavior	● Restricting the use of electronics before bedtime [42, 66]
		12.5	Adding objects to the environment	● Physical changes made to the bedroom environment to promote sleep [59, 66, 67]
14	Scheduled consequences	14.1	Behavior cost	● Loss-framed incentives designed to promote sleep [21, 22]

Latent Classes

For the latent class analysis, a 2-class model resulted in the best fit. The Vuong-Lo-Mendell-Rubin adjusted likelihood ratio test comparing the 3-class model to the 2-class model did not achieve statistical significance (p=0.084). The 2-class model yielded a better fit compared to the 1-class model (p=0.009).

The two classes were distinguished by four behavior change groups (Table 3): shaping knowledge (group: #4), natural consequences (group: #5), comparison of behavior (group: #6), and repetition and substitution (group: #8). Studies with a high probability of class I membership were more likely to be sleep duration interventions that included components that fit within these four behavior change groups. In contrast, studies with a high probability of class II membership we less likely to be sleep duration interventions that included components that fit within these four behavior change groups. Specifically, interventions belonging to class I are more likely to include behavior change techniques such as information about antecedents, information about health consequences, prompts/cues, and graded tasks. Examples of approaches related to these techniques include educating participants on causes and consequences of poor sleep, providing participants with physical cues to remind them of bedtime routines, and advising participants to advance their time in bed by five minutes each day for the duration of the study (Table 2).

Table 3.

Probability of including the behavior change group for each class.

Behavior Change Group	Class I 15 studies (44%)	Class II 19 studies (56%)
1. Goals and planning	0.624	0.762
2. Feedback and monitoring	0.689	0.523
3. Social support	0.561	0.81
4. Shaping knowledge	>0.999	0.621
5. Natural consequences	>0.999	0.242
6. Comparison of behavior	>0.999	0.100
7. Associations	0.313	0.286
8. Repetition and substitution	>0.999	0.432
9. Comparison of outcomes	0.626	0.381
10. Reward and threat	0.187	0.190
11. Regulation	0.126	0.095
12. Antecedents	0.375	0.428
13. Identity	<0.001	0.047
14. Scheduled consequences	<0.010	0.142

Meta-Analysis

The meta-analysis included 74 effect size estimates from 34 studies (Table 4 and Figure 5). The overall pooled intervention estimate was 11 (95% CI: 7, 16; p<0.001) minutes, in favor of the intervention arm compared to the control arm. The Q statistic for the test of homogeneity was Chi-square(df=73)=10110.68, p<0.001. The I² indicated that the study level explained 97.7% of the total variation. The funnel plot did not reveal concerns of publication bias (Figure 6).

Table 4.

Meta-analysis results for all studies pooled and by the two latent classes.

	Pooled Analysis			Latent Class Comparison
	N	Estimate (95% CI)	P-value	Estimate (95% CI)	P-value
All Studies Combined	34	11 (6, 16)	<0.001	--	--
Latent Class
Class I	15	8 (1, 15)	0.017	−6 (−15, 3)	0.174
Class II	19	14 (8, 21)	<0.001

Figure 5.

Forest plot of the mean difference in sleep duration (minutes per night) between intervention and control groups (N=34).

Figure 6.

Funnel plot for the sleep duration interventions included in the meta-analysis.

We assessed if the pooled estimate differed by the two latent classes (Table 4). Interventions belonging to class I (i.e., interventions with behavior change groups 4, 5, 6 and 8 highly represented) had a pooled positive intervention effect of 14 minutes (95% CI: 8, 21) versus 8 minutes (95% CI: 1, 15) for interventions belonging to class II. While this indicates a stronger effect for class I interventions, the 95% confidence interval when comparing the difference in these pooled estimates included zero (beta=−6; 95% CI: −15, 3, P=0.17).

Discussion

In this systematic review and meta-analysis, we classified intervention components used in prior pediatric sleep duration interventions using the BCTT. We found that techniques under the shaping knowledge, goals and planning, and social support groups were the most common, and the three most common techniques were instruction on how to perform the behavior (e.g., education component focused on sleep hygiene), practical social support (e.g., intervention material provided to parents), and behavioral practice/rehearsal (e.g., practicing bedtime routines). Our latent class analysis revealed two classes distinguished by the presence (class I) or absence (class II) of the following four behavior change groups: shaping knowledge, natural consequences, comparison of behavior, and repetition and substitution. Interestingly, our meta-analysis revealed that class I interventions had a slightly higher pooled intervention effect. These data provide detailed insight into the frequency and effectiveness of behavior change techniques used in prior pediatric sleep duration interventions and this knowledge will help in the identification of candidate intervention components to be studied in future optimization and randomized controlled trials, per the MOST framework.

To the best of our knowledge, two reviews of adolescents and young adults [25, 29], one review of adults [26], and two reviews of mobile health applications for sleep [27, 28] described components used in prior sleep duration interventions using the BCTT. Baron et al., Pegado et al., and Arroyo et al. found that techniques under the shaping knowledge group were the most common [25, 28, 29]. Scheduled consequences [29], social support [25], and comparison of behavior [28] were among the least frequently used behavior change groups and covert learning was not used at all [25, 28, 29]. This aligns with our systematic review in an updated and exclusive set of pediatric studies, in which we also found shaping knowledge to be among the most frequently used behavior change group, scheduled consequences was one of the behavior change groups infrequently used, and covert learning was not used at all. The Murawski et al. and Lancaster et al. systematic reviews did not report behavior change groups, so comparisons cannot be drawn at the group level. However, similar to Pegado et al., Murawski et al., and Lancaster et al., we found that instruction on how to perform the behavior was the most commonly used behavior change technique [25–27]. The Baron et al. and Arroyo et al. systematic reviews did not report behavior change techniques, so no comparisons can be drawn with our technique level data. These alignments across six independent systematic reviews are encouraging and indicate that the BCTT method can be used to classify intervention components used in sleep duration interventions.

A key activity performed under the preparation phase of the MOST framework is the selection of candidate intervention components. By classifying intervention components used in pediatric sleep duration interventions using the BCTT, we have developed a resource that can assist with the candidate intervention selection process. Investigators may look to the high frequency behavior change groups and techniques used in past research for candidate component selection. If there are gaps in knowledge regarding the utility of high frequency behavior change groups and techniques, then relevant components could be designed and included as experimental factors in optimization trials. High frequency behavior change groups and techniques may also signal that components targeting these areas are important (for effectiveness or context) and so could be included as constant components delivered to all participants in optimization trials. For example, shaping knowledge was the most frequently used group and instruction on how to perform the behavior was the most frequently used technique. A typical intervention with this common technique often included an education component that focused on sleep hygiene. Depending on the research question, such a component could be experimentally assessed for effectiveness in an optimization trial or included as a constant to provide contextual information to participants.

Alternatively, behavior change groups used less frequently (or not at all) may provide insight into novel intervention component approaches, which is needed if we are to develop interventions that surpass the effectiveness of current interventions. For example, identity was the least frequently used group and monitoring of behavior by others without biofeedback was among the least frequently used techniques. Future interventions could consider designing components relevant to infrequently used behavior change techniques. For example, the monitoring of behavior by others without biofeedback technique was used in a prior study for a component that used actigraphy to measure sleep, but participants did not have access to this data [48]. This is an example of a more novel behavior change technique that could be more widely investigated.

Candidate intervention components could also be selected based on associations between behavior change techniques and intervention effectiveness. We applied a latent class analysis which revealed two classes of interventions, where class I interventions had a high likelihood of having components related to four specific behavior change groups (shaping knowledge, natural consequences, comparison of behavior, and repetition and substitution). Given that class I interventions may be more effective (per our meta-analysis), this may be a source for which to identify candidate components for future optimization trials. From the notes and themes we developed, the specific behavior change techniques associated with class I interventions include: educating participants on causes of poor sleep (code: 4.2), educating parents on consequences of poor sleep (code: 5.1), providing participants with a refrigerator magnet to serve as a visual cue of a sample bedtime routine (code: 7.1), and advising participants to advance their time in bed by five minutes each day for the duration of the study (code: 8.7).

Improvements in clinical and health related quality of life outcomes are expected if children can increase their sleep duration by 30 minutes per night [2, 15, 16]. However, current sleep duration interventions generally fall short of this threshold. A meta-analysis by Magee et al. reported that nonpharmacological sleep duration interventions for youth have an overall effectiveness of increasing sleep duration by 12 minutes per night [17]. This aligns with our meta-analysis which revealed an average effect size of 11 minutes in favor of pediatric sleep duration interventions over controls groups in RCTs. This alignment is to be expected given that several studies included in the meta-analyses were overlapping, and further underscores the need to develop novel interventions to increase sleep duration in childhood.

Our systematic review and meta-analysis has strengths. We focused exclusively on pediatric studies and did not include young adults; we used the latest RoB2 tool; we assigned classification codes at the behavior change group and technique levels; notes and themes were maintained to provide sleep specific contextual information; and we assessed effectiveness by intervention classes distinguished by the presence/absence of four behavior change groups. Our study also has limitations. The retrospective assignment of behavior change techniques is a subjective process contingent on adequate descriptions of the intervention components in the published literature. Similarly, the notes and themes developed involved a degree of subjective decision making and may also vary across study teams. From a data management perspective, code assignment documentation was done in Excel, and it would have been more efficient to use a formal database (e.g., REDCap or Covidence). Ideally, behavior change technique codes would be assigned by the primary investigative team in their primary research articles (including protocol papers), with a clear description of how each intervention component was designed and delivered. We focused on interventions targeting sleep duration; future reviews assigning behavior change technique codes should be done for other sleep health outcomes. While we included 37 interventions, there was insufficient data to perform the latent class analysis at the behavior change technique level and we could only consider a two-class model; this approach can be revised as more interventions are completed. In the meta-analysis, we only included actigraphy estimated sleep duration for those studies that also provided self-report sleep duration data; actigraphy methods are not standardized and the validity may be lower for participants who have more wake/sleep transitions during a sleep period, such as is more typical with younger children. However, there are also limitations with self-reported sleep duration assessment approaches, particularly recall bias and social desirability bias. Finally, we did not assess for differences in outcomes according to sleep assessment approach, Magee et al previously reported that using objective or subjective measures did not impact sleep duration outcomes.[17]

Conclusion

Within the preparation phase of the MOST framework, we used the BCTT to characterize and analyze the effectiveness of behavioral sleep promotion interventions in children. This systematic review and meta-analysis provides critical information on candidate intervention components for pediatric sleep duration interventions that will help to facilitate future optimization trials and randomized control trials using the MOST framework and thus enhance the development of sleep promotion interventions.

PRACTICE POINTS.

1. Current behavioral interventions to increase sleep duration in youth have a pooled overall effectiveness of 12-minutes.

2. This systematic review classified behavior change techniques used in pediatric sleep duration interventions; this serves as a resource to help identify novel intervention components that, in theory, could surpass the current level of effectiveness.

RESEARCH AGENDA.

1. MOST is a three-phase iterative framework that can be used to develop effective pediatric sleep duration interventions.

2. Candidate intervention components are selected under the MOST preparation phase, and are then experimentally assessed for effectiveness in optimization and randomization trials.

3. By classifying behavior change techniques used in pediatric sleep duration interventions, we provide a resource to help select candidate intervention components that can be experimentally assessed for effectiveness.

Abbreviations:

BCTT

behavior change technique taxonomy

LCA

latent class analysis

MOST

multiphase optimization strategy

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-analyses

RoB2

Version 2 of the Cochrane risk-of-bias