SYSTEMATIC REVIEW
Self-Help for Weight Loss in Overweight and Obese Adults: Systematic Review and Meta-Analysis We conducted a systematic reviewand meta-analysis investigating the components and effectiveness of self-help weight-lossinterventionsand their applicability to lessadvantaged populations. We searched (November 2013) for randomized controlled trials comparing selfhelp interventions with each other or with minimal controls in overweight and obese adults, with 6 months or longer follow-up. We calculated mean difference between intervention and control for 6- and 12-month weight change. Twenty-three studies met the inclusion criteria (9632 participants; 39 intervention arms). Intervention participants lost significantly more weight than controls at 6 months (mean difference –1.85 kg; 95% confidence interval [CI] = –2.86, –0.83; 7 studies). No significant effect was detected at 12 months but results were sensitive to the inclusion of 1 study at high risk of bias. Interactive programs appeared more effective than standard ones at 6 months (mean difference –0.94 kg; 95% CI = –1.50, –0.38). Evidence is insufficient to reach conclusions on effectiveness in socioeconomically disadvantaged people, but suggests self-help interventions may be less effective in this group. (Am J Public Health. 2015;105:e43–e57. doi:10. 2105/AJPH.2014.302389)
Jamie Hartmann-Boyce, MA, Susan A. Jebb, PhD, Ben R. Fletcher, MPH, and Paul Aveyard, MD, PhD
OVERWEIGHT AND OBESITY are a major cause of preventable morbidity and mortality worldwide, with the World Health Organization estimating that they cause at least 35.8 million disability-adjusted life years and 2.8 million deaths annually.1 At any one time, more than a quarter of US women are trying to lose weight (27%), with men not far behind (22%), the large majority of whom are doing so without professional support and outside the context of formal weight-loss programs.2 By contrast with more intensive interventions,3---5 very little is known about these selfdirected efforts to lose weight. The high prevalence of overweight and obesity presents a resource challenge to health care systems. Trials of self-help interventions for weight loss can inform our knowledge about what types of self-directed weight loss strategies are most effective and which might be usefully highlighted to the public as a scalable, low-cost public health intervention. Moreover, understanding the effective components of self-help interventions for weight loss may enable future tailoring of other weight management programs to enhance effectiveness. Interventions that are selfdirected and do not require professional input to deliver (“self-help”) come in a variety of formats, including but not limited to print, Internet, and mobile phone---delivered programs. Data from the past 5 years suggest that such interventions are being widely used: 33% of US
March 2015, Vol 105, No. 3 | American Journal of Public Health
Internet users look online for information about weight management and 19% of smartphone owners are estimated to have a health app on their phone, with the most popular type being diet, exercise, and weight apps.6---8 Most reviews of trials of selfhelp interventions for weight loss are defined by the format of the programs rather than the content.9---11 Results from formatspecific reviews suggest that such interventions may be promising, but results across these reviews are mixed. In addition, there is limited research on the longerterm effectiveness of these efforts, with longer-term data frequently only emerging from uncontrolled program evaluations.11 Moreover, this format-specific focus, though informative for questions about delivery medium, limits the opportunity to examine the behavioral components across format types, and also leads to the inclusion of some more resource-intensive interventions alongside true self-help programs. For example, some Internet interventions may include input from trained personnel. We therefore conducted a systematic review of self-help interventions across all relevant formats, but limited to those that do not require professional input to deliver. It is also possible that the effectiveness of self-help interventions could vary by population group.12 In middle- and high-income countries, obesity is more common among lower socioeconomic groups,13,14 but there is concern that some interventions tackling
obesity are taken up more effectively by more advantaged groups, thus further widening existing inequalities.15 Of particular relevance to this review, it may be that self-help programs are more effective for socioeconomically advantaged groups. There are a number of possible mechanisms through which socioeconomic status (SES) may influence the effectiveness of self-help programs. Executive functioning, a theorized control network regulating behaviors, has been linked with poverty in childhood, with more disadvantaged groups associated with lower strength of executive functioning in adulthood. Weaker executive functioning could conceivably diminish an individual’s ability to enact the recommended actions in self-help programs, as lower levels of executive functioning have been found to be associated with uncontrolled eating and a lower level of inhibitory control.16,17 It has also been theorized that cultural capital, expressed through certain attitudes, knowledge, and competencies, may increase the ability of more socioeconomically advantaged groups to enact behavior change to improve health outcomes, whereas the corresponding lack of cultural capital may diminish health-related behavior change in less-advantaged populations.18 In addition, it may be that the environmental factors that contribute to higher levels of obesity in socioeconomically disadvantaged groups provide further barriers to responding to the recommended actions in self-help
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e43
SYSTEMATIC REVIEW
programs, including issues such as access to healthy food and safe and affordable recreation.19,20 Accordingly, the provision of selfhelp materials, which draw on executive functioning and cultural capital to effect change in health outcomes, could exacerbate existing health disparities by leading to greater weight loss in more socioeconomically advantaged groups while having less of an impact on disadvantaged groups. The aims of this review, therefore, were 3-fold. We set out to evaluate the effectiveness of selfhelp interventions for weight loss in overweight and obese adults and to identify the most effective components in terms of behavioral processes. We then looked at data on socioeconomic variables to assess if the observed effects were likely to be transferable to both socioeconomically disadvantaged and advantaged populations.
METHODS We agreed on a review protocol before commencing work.21 We ran systematic searches of 11 electronic databases in November 2013 (MEDLINE, EMBASE, PsychINFO, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Conference Proceedings Citation Index, Database of Abstracts of Reviews and Effects, Health Technology Assessment Database, Science Citation Index, Cumulative Index to Nursing and Allied Health Literature, and SPORTDiscus) for randomized controlled trials with terms relating to obesity, weight loss, diet, exercise, behavior change, selfcare, and formats relevant to selfcare, including Web-based and printed interventions. (The MEDLINE search strategy can be found in supplemental file A,
available as a supplement to the online version of this article at http://www.ajph.org.) We also identified studies through screening reference lists of relevant trial reports and systematic reviews and through contacting authors in the field. We searched clinical trials registries (ClinicalTrials.gov and the ISRCTN Register) to identify unpublished and ongoing studies. To be included, studies had to be randomized controlled trials comparing a self-help intervention for weight loss with another selfhelp intervention or with a minimal control, and to be published in English. There were no restrictions on publication date. Studies had to include adults (aged ‡ 18 years) with a body mass index (BMI; defined as weight in kilograms divided by the square of height in meters [kg/m2]) of greater than or equal to 25 kg/m2 (or a BMI of ‡ 23 kg/m2 in Asian populations) and measure weight change at 6 months or longer, either via self-report or in person. We excluded studies exclusively in pregnant women or in people with eating disorders. To be included, interventions had to aim to achieve weight loss through changes in diet or physical activity behavior (as opposed to through medications or dietary supplements). We excluded interventions that involved other lifestyle changes (e.g., smoking cessation). We defined self-help as interventions that could feasibly be delivered in a self-management context (i.e., used by individuals for a weight loss attempt not assisted by health care professionals, counsellors, or any other kind of person-to-person support). We only included interventions involving human input in program delivery if input was limited to the research component or to advising a participant on how to use the
e44 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
resource being tested or if input was brief (1 session), was easily able to be automated or was judged not to be integral to the intervention, and was provided to both intervention and control participants. Following methods used in the Cochrane review of self-help interventions for smoking cessation, we did not include leaflets limited to general health information as interventions, but considered these as a control if compared with more substantial self-help materials.22
Data Collection and Outcome Measurement A single reviewer screened titles and abstracts, with a sample checked by a second. Two reviewers independently extracted data, with discrepancies resolved by discussion. Two reviewers assessed each study for risk of bias based on criteria developed by the Cochrane Collaboration.23 We judged studies to be at low, high, or unclear risk of bias on the basis of random sequence generation, allocation concealment, detection bias, attrition, and other risks of bias. As blinding of participants is often unfeasible in trials of behavioral interventions, we assessed risk of detection bias based on the method of outcome measurement used (e.g., selfreport or objective measurement). In the absence of validated criteria with which to judge risk of attrition bias in weight loss trials, we considered studies to be at high risk of attrition bias if less than 50% of the sample was followed up at 6 months or if follow-up rates differed by greater than 20% between groups at 6 months. We judged studies that employed wait-list controls to be at high risk of other bias if participants in the control arm knew they had been placed on a wait list, as this could
have an impact on the initiation of a weight loss attempt. If further detail were required on study design or outcome, we contacted authors for further information and sought related publications and trial protocols. We had planned to assess risk of publication bias by using funnel plots, but were unable to do so because of insufficient data. Alongside data extraction, 2 reviewers independently coded each intervention against a predefined list of self-management strategies for weight loss (the Oxford Food and Activity Behaviours taxonomy, referred to from hereon as the taxonomy), grouped by domain into independent behavioral processes (Table 1). This taxonomy was used for the first time in this review, and was developed before data extraction through qualitative analysis of commonly used self-management weight loss resources, drawing on a grounded framework approach.24 Before its use in this review, we also evaluated the taxonomy against existing behavior change theories and taxonomies. We coded each intervention as yes, no, or unclear for use of strategies in each domain. Two reviewers also extracted detail on other program characteristics, including whether the intervention was tailored, interactive, or both (see protocol21). We defined tailored interventions as those in which participant characteristics were used to provide individualized content (e.g., tailored based on information provided by participants at baseline), and defined interactive interventions as those programs in which participants could actively engage with intervention content (e.g., through online quizzes or entering their own content). We also extracted data on socioeconomic variables,
American Journal of Public Health | March 2015, Vol 105, No. 3
SYSTEMATIC REVIEW
TABLE 1—Self-Management Weight Loss Strategies by Domain Domain
Definition
Energy compensation
Conscious adjustment of behaviors to alter energy intake, expenditure, or both to control weight in light of
Goal setting
Setting of specific behavioral or outcome targets
Impulse management:
Respond to unwanted impulses through awareness and acceptance of the feeling that generates the impulse
acceptance Impulse management:
and reacting without distress or over-analysis Respond to unwanted impulses by evaluating personal motives behind that impulse before acting
previous energy intake or expenditure
awareness of motives Impulse management:
Respond to unwanted impulses through distraction in an attempt not to act on the impulse
distraction Information seeking
Seek specific information to enhance knowledge to help manage weight
Motivation
Strategies to increase the desire to control weight
Planning content Regulation: allowances
Plan types of food or physical activity in advance of performing behavior Unrestricted consumption of or access to prespecified foods or behaviors
Regulation: restrictions
Avoid or restrict prespecified foods, behaviors, or settings
Regulation: rule setting
Mandate responses to specific situations
Restraint
Conscious restriction over the amount that is eaten
Reward
Reinforcement of achievement of specific behavior or outcome through reward contingent on the meeting
Scheduling of diet
Plan timing and context or location of food or physical activity in advance of performing behavior
and activity Self-monitoring
Record specific behaviors or outcomes on regular basis
Stimulus control
Alter personal environment such that it is more supportive of target behaviors (adapted from CALO-RE25)
Support: buddying
Perform target behaviors with another person
Support: motivational
Discussing, pledging, or revealing weight loss goals, plans, or achievements or challenges to others to
Support: professional
Seek help to manage weight from someone with specific expertise
Weight management aids
Use of or purchase of aids to achieve weight loss in any other manner (including, but not limited to,
RESULTS
of that target
bolster motivation
reducing energy intake and increasing energy output)
including baseline education, income, and deprivation indexes, where reported, and any reported analyses comparing effects by SES. Our primary outcome was mean difference in weight change between study arms at 6 months, calculated by using baseline observation carried forward (BOCF; an intention-to-treat analysis that imputes baseline weight for participants missing at follow-up) to facilitate comparisons between studies. In studies in which BOCF data were not presented in the paper, we used complete case data (and in 1 case, last observation
carried forward data) to calculate BOCF.26 In studies in which weight change data were not available in any of these forms, we report results narratively in the text. Where available, we also extracted intermediate outcome data related to the behavioral processes listed in the previously mentioned domains.
Statistical Analysis We pooled studies in which self-help interventions were compared with minimal control interventions and studies that compared self-help interventions with respect
March 2015, Vol 105, No. 3 | American Journal of Public Health
50% and 90% are viewed as an indication of substantial heterogeneity.23 We conducted sensitivity analyses to test if effect sizes differed when we excluded studies at high risk of bias or if they differed when we used complete case data as opposed to BOCF. We had originally planned to supplement our meta-analyses with metaregression to explore associations between intervention characteristics and weight change, and associations between the degree to which participants enacted specific behavioral processes and weight change, but did not have sufficient data with which to do so.
to a variable of interest. We conducted meta-analyses by using a random effects model in Review Manager version 5.2 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark). We conducted separate analyses for weight change at 6 and 12 months. We presented pooled results as mean differences (kg) with 95% confidence intervals (CIs). We used the I2 statistic to present statistical heterogeneity between studies.27 In general, values between 30% and 60% are viewed as indicating moderate heterogeneity and values between
After we removed duplicates, our search retrieved 3883 references, 3697 of which were excluded at title---abstract stage. We assessed 186 full-text articles for eligibility, and excluded 143 of them, with the most common reason for exclusion being that the self-help intervention was confounded by an additional, more intensive intervention. (Other reasons for exclusion are listed in Supplemental File B PRISMA diagram; available as a supplement to the online version of this article at http://www.ajph.org). We included 43 references, representing 23 studies, 18 of which had sufficient data to be included in statistical analyses. Twelve studies compared self-help interventions with minimal controls (no additional contact with or without printed information on consequences of obesity), 7 compared tailored or interactive programs with fixed (nontailored, noninteractive) programs, and 2 compared tailored and interactive programs with nontailored interactive programs.
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e45
SYSTEMATIC REVIEW
Characteristics of Included Studies Characteristics of the included studies are summarized in this section, covering participants, intervention format and content, and risk of bias. Table 2 provides further detail on the participants and interventions in individual studies. Participants. In total, relevant study arms represented 9672 participants, with study samples ranging from 33 to 2862 participants and a mean of 420 participants per study. Twelve studies were conducted in the United States, 4 were conducted in Australia, 3 each were conducted in Japan and the United Kingdom, and 1 was conducted in Finland. The mean age of study participants ranged from 36 to 60 years and the majority were female (61% average). Four studies recruited men only39,40,42,46 and 1 recruited women only.28 Mean BMI at baseline ranged from 26.2 (Japanese studies) to 36.0 kg/m2, with an average of 31.8 kg/m2. Intervention. Overall, the included studies represented 39 intervention arms. Of the interventions being tested, 23 included print materials; 22 included access to a Web site; 10 involved some form of self-monitoring equipment (most commonly pedometers); 5 included automated text messages; 2 each employed podcasts, DVDs, and smartphone apps; and 1 each included e-mail, computer software, and phone recordings. Eighteen interventions were both tailored and interactive, 6 were interactive but not tailored, 3 were tailored but not interactive (e.g., print materials tailored based on baseline data), and 12 were neither tailored nor interactive (i.e., the same content would be viewed or heard by all participants, referred
to from hereafter as “fixed”). In cases in which access to the intervention was limited (e.g., nonprint interventions), length of access ranged from 2 to 12 months. Independent from length of access, some programs also specified the frequency with which participants accessed the intervention content (for example, recommending participants access a program once a week). In instances in which such recommendations were made, this ranged from daily to 4 times over 3 months. In 2 studies, intervention material could only be accessed at specified sites.49,50 In all other studies, interventions could be accessed at the participants’ discretion. Individuals were given set energy intake targets in 8 studies, and a further 5 studies generally specified “healthy eating.” One study explicitly set no specific dietary recommendations and 9 did not report any details of the dietary advice provided. Eleven studies did not report what physical activity advice, if any, was given to participants. Of the remaining studies, 5 specifically encouraged walking and 7 recommended an increase in physical activity but did not specify an activity type. Eight studies identified social cognitive theory as the theoretical basis for their intervention, 3 were designed with the transtheoretical model, 2 employed cognitive behavioral theory, and 1 each used habit formation theory and the behavioral determinants model. Eleven studies did not state the theoretical basis for their interventions. The most commonly recommended self-management strategies were goal setting (32 of 39 interventions), self-monitoring (36 of 39), and seeking social support from nonprofessionals (19 of 39).
e46 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
Fourteen interventions recommended restrictions, and 11 each recommended strategies to boost motivation, strategies to reward oneself for hitting targets, and strategies involving conscious “energy compensation” (most commonly offsetting energy intake by increasing energy output). Figure 1 shows the number of interventions employing each group of strategies. Risk of bias. We judged 3 studies to be at low risk of bias across all domains assessed.38,39,41 We judged 8 studies to be at high risk of bias in at least 1 domain, and the remainder of included studies were judged to be at unclear risk of bias in at least 1 domain, most commonly because of lack of detail in the reporting of methods of random sequence generation and allocation concealment (16 studies). With the exception of 1 study that used self-report (high risk)42 and 1 study in which the method of outcome assessment was unclear (unclear risk),28 all included studies measured weight objectively and hence were judged to be at low risk of detection bias. We judged 3 studies to be at high risk of attrition bias (2 in which £ 20% of participants were followed up at 6 months36,43 and 1 in which attrition was differential between groups [93% vs 53%]).31 We judged 2 that did not report the number followed up at 6 months to be at unclear risk.29,33 We judged 3 studies to be at high risk of other bias because of the use of a wait-list control40,44,49 and we judged a further 3 to be at unclear risk as participants were offered the intervention at study end, but it was not clear if they knew they would be offered the intervention in advance.28,30,42 We judged an additional study to be at high risk of other bias as a participant in the control group
used a weight loss app and lost 32 kg, which the authors stated had a strong influence on the mean weight loss in the control group.31 Another was judged to be at high risk of other bias as the intervention arm was followed up at 3-month intervals whereas the control arm was only followed up at 12 months.34 (Supplemental File C, available as a supplement to the online version of this article at http://www.ajph.org, contains risk of bias judgments for each included study across all domains, along with reasons for each judgment.)
Weight Change Pooled results of studies comparing self-help interventions with controls detected a significant effect in favor of the intervention at 6 months (mean difference –1.85 kg; 95% CI = –2.86, –0.83), though statistical heterogeneity was substantial (I 2 = 76%). Heterogeneity was not explained by subgroup differences between tailored or interactive versus fixed programs, and all subgroups detected significant effects in favor of the intervention with no evidence of a difference between subgroups (Figure 2). Sensitivity analysis using complete case data were not significantly different (Supplemental File D, available as a supplement to the online version of this article at http://www.ajph. org) and the effect estimate remained similar when we excluded studies at high risk of bias (mean difference –1.42 kg; 95% CI = –2.57, –0.28). However, at 12 months, a significant effect was not detected in pooled results from the 5 studies with data at this point (all tailored and interactive programs; mean difference –0.76 kg; 95% CI = –1.73, 0.20; Figure 3), with substantial statistical heterogeneity (I 2 = 62%). Results were
American Journal of Public Health | March 2015, Vol 105, No. 3
March 2015, Vol 105, No. 3 | American Journal of Public Health
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e47
pedometer)
2. Interactive (tailored unclear; Web)
3. Fixed nontailored (print—book)
1. Tailored and interactive (Web)
2. Tailored and interactive (as per arm 1 plus automated personalized
(n = 128)
Collins et al. 2013,32
Australia (301)
2. Minimal control
1. Tailored interactive (Web, DVD, pedometer)
2. Fixed (DVD, pedometer) 3. Minimal control
Australia (65)
Australia (159)
1. Tailored interactive (Web)
Morgan et al. 2011,39
Morgan et al. 2013,40
1. Tailored interactive (Web)
2. Minimal control
consumption at 6 mo
Weight, portion size and alcohol
Weight at 6 mo and 12 mo
Weight at 6 mo and 12 mo
None reported
3 mo; 4 d/wk
3 mo; frequency NS
minimum weekly
12 mo; frequency
3-mo intervals
individualized feedback sheets using standardized set of feedback
None in arm 2; in arm 1,
control
One information session as per
site
Brief instruction on how to use Web
No
Continued
intervention primarily automated
Some therapist contact but
authors
Further outcome data provided by
United States (1277)
United Kingdom (221)
Weight measured but not reported
2. Minimal control
Johnson et al. 2008,37
McConnon et al 2007,38
reported—no usable data in group 3 9 mo; 4 reports at
Weight at 8 mo measured but not
3. Fixed (print plus contract plus phone recording) 1. Tailored noninteractive (print reports and manual)
1.5 mo; weekly in
No groups 1 and 2, daily
None reported
2. Fixed (print plus contract)
No
1. Fixed (print)
weekly feedback
United States (33)
Length NS; automated
No
Jeffery et al. 1982,36
Weight at 6 mo and 12 mo
NS
Authors provided additional outcome data
of intervention as cited in text
Additional data from YouTube videos
1. Interactive tailored (Web, print)
at 12 mo
Weight and information seeking
No
prompts
and did not respond to other
Phone prompts if did not enter weight
No
available at time of reporting
2. Fixed (Web, print)
2. Minimal control
Finland (125)
NS
6 mo; 4 d/wk
then users’ choice
6 mo; daily for first wk,
Session with standard advice as per
United States (1177)
1. Interactive tailored (SMS, Web)
Haapala et al. 2009,34
Weight at 6 mo
scale at 6 mo
Weight and cognitive restraint
at 6 mo
Weight and self-monitoring
NS; frequency: daily
presentation; results at 6 mo not
Data from unpublished conference
Hersey et al. 2012,35
1. Interactive nontailored (Web, print, accelerometer) 2. Minimal control
Greene et al. 2013,33 United States (349)
reports, feedback, and reminders via Web and SMS)
1. Tailored and interactive (smartphone app, SMS, Web)
United Kingdom
mail research staff with technical problems
heart rate transmitter belt, print materials) 2. Minimal control
Carter et al. 2013,31
control group, training and told to e-
watch with in-built program,
Australia (74)
Weight at 6 mo
1. Tailored and interactive (computerized
manual using defined script
Byrne et al. 2006,30
to use daily
indefinitely; instructed
1. Fixed (print leaflet)
2. Minimal control
30 min talking patients through study
place of 6-mo data
3 mo, have leaflet
comparisons; 7-mo data used in
Arm 3 not included in any
Notes
NS
No
Personal Contact
indefinitely; frequency
1 mo, but have booklet
Length and Frequency of Accessc
United Kingdom (527)
None reported
and restrictions at 7 mo
Weight, eating behavior,
Outcomes Relevant to This Review
Beeken et al. 2013,29
4. Fixed (booklet)
2. Fixed (print—booklet and letters) 3. Interactive not tailored (print—booklet, pedometer)
1. Tailored and interactive (print—booklet and letters,
Japan (205)
Study Arms for This Reviewb
Adachi et al. 2007,28
Study ID, Year, Country, (No.a)
TABLE 2—Characteristics of Included Studies in Systematic Review and Meta-analysis of Self-Help Interventions for Weight Loss
e48 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
American Journal of Public Health | March 2015, Vol 105, No. 3
1. Interactive tailored (Web)
2. Minimal control
1. Interactive tailored (computer software, print)
2. Interactive tailored (print)
2011,48 United States (96)
Winett et al. 2007,49
United States (707)
Wylie-Rosett et al.
2001,50 United States
Weight at 12 mo
3 mo, monthly after
12 mo; weekly for first
No
accessed in foyer of HMO (group 1)
Computer material can only be
accessed in church
modeling; material can only be
Cluster randomized, analysis via ITT
author; low level of app and Twitter use in intervention group
Additional outcome data provided by
using generalized mixed model
Unclear
messages per day to Twitter groups (not individualized)
In group 1, study coordinator sent 2
social support at 12 mo
3 mo; weekly
group 1 asked to log on to Twitter daily
6 mo; podcasts 2·/wk,
following standard script
Instructions on how to use Web site
intake, planning and tracking,
Fiber, fruit and vegetable, fat
monitoring at 6 mo
Weight, social support and self-
arm 1
weekly reporting in
6 mo; frequency NS but
Notes. BOCF = baseline observation carried forward; HMO = health maintenance organization; ITT = intention to treat; LOCF = last observation carried forward; NS = not specified; SMS = short message service. a No. is total of only those participants relevant to this review (excludes additional arms testing more intensive interventions). b Fixed are programs that are neither tailored nor interactive. c As recommended.
(352)
1. Interactive nontailored (podcasts, print, app, Twitter)
2. Fixed (podcasts, print)
Turner-McGrievy and Tate
2. Interactive nontailored (Web)
States (128)
Weight at 6 mo
received from authors
1. Tailored interactive (Web, e-mail)
data; additional outcome data
received from authors 7-mo data used in place of 6-mo
data; additional outcome data
9-mo data used in place of 6-mo
provided
calculated BOCF using LOCF data
Complete case data not available,
reported in useable form
Weight at 6 mo measured but not
data
Authors provided additional outcome
frequency NS
troubleshooting No
equipment training and
Brief health education session,
No
be publicly posted on study Web site No
study staff with questions that would
Orientation face-to-face; could e-mail
providing materials
assignment; 2-h group session
Motivational lecture before
indefinitely);
Tate et al. 2006,47 United
meals at 3 mo
2. Fixed nontailored (booklet)
Japan (51)
4 mo; armband 16 h/d
1 mo (have
Weight at 7 mo, eating between
Weight at 9 mo
newsletters monthly
Tanaka et al. 2010,46
Shuger et al. 2011,45
arm 2; frequency NS 12 mo; SMS 4·/d,
and upload info daily
1. Interactive nontailored (print, monitoring armband with
United States (170)
Weight at 6 mo and 12 mo
wrist watch display, Web)
2. Minimal control (info only)
Shapiro et al. 2012,44
2 mo in arm 1, NS in
intake at 6 mo and 12 mo Weight at 6 mo
2. Fixed (print) 1. Tailored interactive (print—booklet and letters, pedometer)
1. Tailored interactive (Web, SMS)
United States (2862)
mo; weekly homework
NS but presumably 12
daily
NS, but record weight
vegetables, saturated fat
Weight at 12 mo, fiber, fruit and
Weight at 6 mo
United States (99)
1. Tailored interactive (Web with tailored action plans)
2. Tailored noninteractive (Web)
Rothert et al. 2006,43
1. Tailored interactive (Web, pedometer)
2. Minimal control
United States (441)
Japan (125)
Patrick et al. 2011,42
1. Tailored noninteractive (print and pedometer)
2. Minimal control
Nakata et al. 2011,41
TABLE 2—Continued
SYSTEMATIC REVIEW
Self-monitoring Goal setting Support: Motivational Regulation: Restrictions Motivation Energy compensation Reward Regulation: Rule setting Planning content Stimulus control Imitation (modelling) Regulation: Allowances Scheduling of diet and activity Impulse management: Awareness of motives Information seeking Weight management aids Support: Buddying Flexible restraint Support: Professional Impulse management: Distraction Impulse management: Acceptance 0
10
20
30
40
Number of Interventions Recommending Use of Strategy
FIGURE 1—Frequency of use of recommended strategies (by domain) in included studies: systematic review and meta-analysis of self-help interventions for weight loss.
heavily influenced by 1 study, which was at high risk of attrition bias for 12-month data (49% of intervention participants vs 70% control participants followed up at 12 months); removing this study from the analysis reduced heterogeneity to moderate and yielded a significant effect in favor of the intervention (mean difference –1.10 kg; 95% CI –1.97, –0.23; I 2 = 38%). Three further studies, not included in any meta-analyses, also compared self-help interventions with minimal controls. One study of a tailored and interactive intervention did not publish data on weight change, but reported that the intervention significantly increased healthy eating and exercise behaviors.37 A second study
was not included because it was a cluster randomized trial of a church-based intervention. Participants in the intervention group lost an adjusted average of 0.2 kg (SD = 0.4) at 7 months, whereas control participants had gained 0.1 kg (SD = 0.5).49 In the third study, in which a print leaflet was compared with a minimal control, 6-month data, though collected, were not available at the time of publication. Unpublished data from 3-month follow-up detected a significant effect of the intervention.52
Comparisons on the Basis of Intervention Type Interactive versus fixed. Results from 7 studies directly comparing interactive (with and without
March 2015, Vol 105, No. 3 | American Journal of Public Health
tailoring) with fixed interventions detected a significant effect in favor of the interactive programs (mean difference –0.94 kg; 95% CI = –1.50, –0.38; I 2 = 2%). In subgroup analysis, the significant effect remained in the group of tailored and interactive studies but the pooled effect was not significant in 2 studies of interactive nontailored programs (Figure 4). A sensitivity analysis using complete case data revealed similar results (Supplemental File D, available as a supplement to the online version of this article at http://www.ajph.org) and results continued to significantly favor the interactive programs when we removed studies at high risk of bias (mean difference –0.83 kg; 95% CI = –1.53, –0.14). In
the only study that reported 12-month data, a significant effect remained (mean difference –0.70; 95% CI = –1.37, –0.03).36 Tailored and interactive versus interactive only. Two studies directly compared tailored and interactive programs with nontailored but interactive programs; pooled results detected a small but significant effect in favor of tailored programs (mean difference –0.41 kg; 95% CI = –0.61, –0.20; I 2 = 0; n = 2990). However, the large majority of participants in this comparison came from 1 study that suffered very high loss to follow-up, limiting confidence in this finding.43 Addition of specific features. Four studies tested the addition of adjuncts to self-help interventions, but only one detected a significant impact. One added a pedometer to both a fixed and a tailored and interactive program; pooled results comparing arms with pedometers with those not providing pedometers revealed a nonsignificant mean difference of –0.27 kg (95% CI = –0.93, 0.40; I 2 = 0%; n = 205).28 One study compared standard with enhanced versions of a Web intervention. The enhanced version included the addition of personalized reports and an escalating reminder schedule to use the program Web site; a significant difference was not detected between arms at 6 months (mean difference –0.90 kg; 95% CI = –2.12, 0.32; n = 301).32 The third tested the addition of an expert software program written to guide participants through a tailored workbook and detected a significant difference between participants using the software and those using the workbook only (mean difference –0.80 kg; 95% CI = –0.98, –0.62; n = 352; data at 12 months).50 The fourth tested the addition of a weight loss
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e49
SYSTEMATIC REVIEW
Intervention Control Mean SD Total Mean SD Total Weight Study or Subgroup 1.1.1 Tailored and interactive Byrne 200630 McConnon 200738 Morgan 201139 Morgan 201340 Shapiro 201244 Subtotal (95% CI)
-4.8 -0.6 -5.3 -5.1 -1.3
3.9 3 5.8 5.4 3.8
41 111 34 53 81 320
-1.9 -0.9 -3.5 -0.5 -0.6
3.4 4.5 5.6 3.4 3.3
Mean Difference IV, Random (95% CI)
33 110 30 26 89 288
12.0% 15.0% 7.5% 10.7% 14.7% 59.9%
-2.90 (-4.56, -1.24) 0.30 (-0.71, 1.31) -1.80 (-4.60, 1.00) -4.60 (-6.55, -2.65) -0.70 (-1.77, 0.37) -1.81 (-3.50, -0.13)
169 63 232
15.6% 13.2% 28.8%
-1.70 (-2.58, -0.82) -1.60 (-3.00, -0.20) -1.67 (-2.42, -0.93)
26 26
11.3% 11.3%
-3.00 (-4.81, -1.19) -3.00 (-4.81, -1.19)
546 100.0%
-1.85 (-2.86, -0.83)
Mean Difference IV, Random, 95% CI
Heterogeneity: τ²=2.94; χ²=24.96; df=4 (P
Self-Help for Weight Loss in Overweight and Obese Adults: Systematic Review and Meta-Analysis We conducted a systematic reviewand meta-analysis investigating the components and effectiveness of self-help weight-lossinterventionsand their applicability to lessadvantaged populations. We searched (November 2013) for randomized controlled trials comparing selfhelp interventions with each other or with minimal controls in overweight and obese adults, with 6 months or longer follow-up. We calculated mean difference between intervention and control for 6- and 12-month weight change. Twenty-three studies met the inclusion criteria (9632 participants; 39 intervention arms). Intervention participants lost significantly more weight than controls at 6 months (mean difference –1.85 kg; 95% confidence interval [CI] = –2.86, –0.83; 7 studies). No significant effect was detected at 12 months but results were sensitive to the inclusion of 1 study at high risk of bias. Interactive programs appeared more effective than standard ones at 6 months (mean difference –0.94 kg; 95% CI = –1.50, –0.38). Evidence is insufficient to reach conclusions on effectiveness in socioeconomically disadvantaged people, but suggests self-help interventions may be less effective in this group. (Am J Public Health. 2015;105:e43–e57. doi:10. 2105/AJPH.2014.302389)
Jamie Hartmann-Boyce, MA, Susan A. Jebb, PhD, Ben R. Fletcher, MPH, and Paul Aveyard, MD, PhD
OVERWEIGHT AND OBESITY are a major cause of preventable morbidity and mortality worldwide, with the World Health Organization estimating that they cause at least 35.8 million disability-adjusted life years and 2.8 million deaths annually.1 At any one time, more than a quarter of US women are trying to lose weight (27%), with men not far behind (22%), the large majority of whom are doing so without professional support and outside the context of formal weight-loss programs.2 By contrast with more intensive interventions,3---5 very little is known about these selfdirected efforts to lose weight. The high prevalence of overweight and obesity presents a resource challenge to health care systems. Trials of self-help interventions for weight loss can inform our knowledge about what types of self-directed weight loss strategies are most effective and which might be usefully highlighted to the public as a scalable, low-cost public health intervention. Moreover, understanding the effective components of self-help interventions for weight loss may enable future tailoring of other weight management programs to enhance effectiveness. Interventions that are selfdirected and do not require professional input to deliver (“self-help”) come in a variety of formats, including but not limited to print, Internet, and mobile phone---delivered programs. Data from the past 5 years suggest that such interventions are being widely used: 33% of US
March 2015, Vol 105, No. 3 | American Journal of Public Health
Internet users look online for information about weight management and 19% of smartphone owners are estimated to have a health app on their phone, with the most popular type being diet, exercise, and weight apps.6---8 Most reviews of trials of selfhelp interventions for weight loss are defined by the format of the programs rather than the content.9---11 Results from formatspecific reviews suggest that such interventions may be promising, but results across these reviews are mixed. In addition, there is limited research on the longerterm effectiveness of these efforts, with longer-term data frequently only emerging from uncontrolled program evaluations.11 Moreover, this format-specific focus, though informative for questions about delivery medium, limits the opportunity to examine the behavioral components across format types, and also leads to the inclusion of some more resource-intensive interventions alongside true self-help programs. For example, some Internet interventions may include input from trained personnel. We therefore conducted a systematic review of self-help interventions across all relevant formats, but limited to those that do not require professional input to deliver. It is also possible that the effectiveness of self-help interventions could vary by population group.12 In middle- and high-income countries, obesity is more common among lower socioeconomic groups,13,14 but there is concern that some interventions tackling
obesity are taken up more effectively by more advantaged groups, thus further widening existing inequalities.15 Of particular relevance to this review, it may be that self-help programs are more effective for socioeconomically advantaged groups. There are a number of possible mechanisms through which socioeconomic status (SES) may influence the effectiveness of self-help programs. Executive functioning, a theorized control network regulating behaviors, has been linked with poverty in childhood, with more disadvantaged groups associated with lower strength of executive functioning in adulthood. Weaker executive functioning could conceivably diminish an individual’s ability to enact the recommended actions in self-help programs, as lower levels of executive functioning have been found to be associated with uncontrolled eating and a lower level of inhibitory control.16,17 It has also been theorized that cultural capital, expressed through certain attitudes, knowledge, and competencies, may increase the ability of more socioeconomically advantaged groups to enact behavior change to improve health outcomes, whereas the corresponding lack of cultural capital may diminish health-related behavior change in less-advantaged populations.18 In addition, it may be that the environmental factors that contribute to higher levels of obesity in socioeconomically disadvantaged groups provide further barriers to responding to the recommended actions in self-help
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e43
SYSTEMATIC REVIEW
programs, including issues such as access to healthy food and safe and affordable recreation.19,20 Accordingly, the provision of selfhelp materials, which draw on executive functioning and cultural capital to effect change in health outcomes, could exacerbate existing health disparities by leading to greater weight loss in more socioeconomically advantaged groups while having less of an impact on disadvantaged groups. The aims of this review, therefore, were 3-fold. We set out to evaluate the effectiveness of selfhelp interventions for weight loss in overweight and obese adults and to identify the most effective components in terms of behavioral processes. We then looked at data on socioeconomic variables to assess if the observed effects were likely to be transferable to both socioeconomically disadvantaged and advantaged populations.
METHODS We agreed on a review protocol before commencing work.21 We ran systematic searches of 11 electronic databases in November 2013 (MEDLINE, EMBASE, PsychINFO, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Conference Proceedings Citation Index, Database of Abstracts of Reviews and Effects, Health Technology Assessment Database, Science Citation Index, Cumulative Index to Nursing and Allied Health Literature, and SPORTDiscus) for randomized controlled trials with terms relating to obesity, weight loss, diet, exercise, behavior change, selfcare, and formats relevant to selfcare, including Web-based and printed interventions. (The MEDLINE search strategy can be found in supplemental file A,
available as a supplement to the online version of this article at http://www.ajph.org.) We also identified studies through screening reference lists of relevant trial reports and systematic reviews and through contacting authors in the field. We searched clinical trials registries (ClinicalTrials.gov and the ISRCTN Register) to identify unpublished and ongoing studies. To be included, studies had to be randomized controlled trials comparing a self-help intervention for weight loss with another selfhelp intervention or with a minimal control, and to be published in English. There were no restrictions on publication date. Studies had to include adults (aged ‡ 18 years) with a body mass index (BMI; defined as weight in kilograms divided by the square of height in meters [kg/m2]) of greater than or equal to 25 kg/m2 (or a BMI of ‡ 23 kg/m2 in Asian populations) and measure weight change at 6 months or longer, either via self-report or in person. We excluded studies exclusively in pregnant women or in people with eating disorders. To be included, interventions had to aim to achieve weight loss through changes in diet or physical activity behavior (as opposed to through medications or dietary supplements). We excluded interventions that involved other lifestyle changes (e.g., smoking cessation). We defined self-help as interventions that could feasibly be delivered in a self-management context (i.e., used by individuals for a weight loss attempt not assisted by health care professionals, counsellors, or any other kind of person-to-person support). We only included interventions involving human input in program delivery if input was limited to the research component or to advising a participant on how to use the
e44 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
resource being tested or if input was brief (1 session), was easily able to be automated or was judged not to be integral to the intervention, and was provided to both intervention and control participants. Following methods used in the Cochrane review of self-help interventions for smoking cessation, we did not include leaflets limited to general health information as interventions, but considered these as a control if compared with more substantial self-help materials.22
Data Collection and Outcome Measurement A single reviewer screened titles and abstracts, with a sample checked by a second. Two reviewers independently extracted data, with discrepancies resolved by discussion. Two reviewers assessed each study for risk of bias based on criteria developed by the Cochrane Collaboration.23 We judged studies to be at low, high, or unclear risk of bias on the basis of random sequence generation, allocation concealment, detection bias, attrition, and other risks of bias. As blinding of participants is often unfeasible in trials of behavioral interventions, we assessed risk of detection bias based on the method of outcome measurement used (e.g., selfreport or objective measurement). In the absence of validated criteria with which to judge risk of attrition bias in weight loss trials, we considered studies to be at high risk of attrition bias if less than 50% of the sample was followed up at 6 months or if follow-up rates differed by greater than 20% between groups at 6 months. We judged studies that employed wait-list controls to be at high risk of other bias if participants in the control arm knew they had been placed on a wait list, as this could
have an impact on the initiation of a weight loss attempt. If further detail were required on study design or outcome, we contacted authors for further information and sought related publications and trial protocols. We had planned to assess risk of publication bias by using funnel plots, but were unable to do so because of insufficient data. Alongside data extraction, 2 reviewers independently coded each intervention against a predefined list of self-management strategies for weight loss (the Oxford Food and Activity Behaviours taxonomy, referred to from hereon as the taxonomy), grouped by domain into independent behavioral processes (Table 1). This taxonomy was used for the first time in this review, and was developed before data extraction through qualitative analysis of commonly used self-management weight loss resources, drawing on a grounded framework approach.24 Before its use in this review, we also evaluated the taxonomy against existing behavior change theories and taxonomies. We coded each intervention as yes, no, or unclear for use of strategies in each domain. Two reviewers also extracted detail on other program characteristics, including whether the intervention was tailored, interactive, or both (see protocol21). We defined tailored interventions as those in which participant characteristics were used to provide individualized content (e.g., tailored based on information provided by participants at baseline), and defined interactive interventions as those programs in which participants could actively engage with intervention content (e.g., through online quizzes or entering their own content). We also extracted data on socioeconomic variables,
American Journal of Public Health | March 2015, Vol 105, No. 3
SYSTEMATIC REVIEW
TABLE 1—Self-Management Weight Loss Strategies by Domain Domain
Definition
Energy compensation
Conscious adjustment of behaviors to alter energy intake, expenditure, or both to control weight in light of
Goal setting
Setting of specific behavioral or outcome targets
Impulse management:
Respond to unwanted impulses through awareness and acceptance of the feeling that generates the impulse
acceptance Impulse management:
and reacting without distress or over-analysis Respond to unwanted impulses by evaluating personal motives behind that impulse before acting
previous energy intake or expenditure
awareness of motives Impulse management:
Respond to unwanted impulses through distraction in an attempt not to act on the impulse
distraction Information seeking
Seek specific information to enhance knowledge to help manage weight
Motivation
Strategies to increase the desire to control weight
Planning content Regulation: allowances
Plan types of food or physical activity in advance of performing behavior Unrestricted consumption of or access to prespecified foods or behaviors
Regulation: restrictions
Avoid or restrict prespecified foods, behaviors, or settings
Regulation: rule setting
Mandate responses to specific situations
Restraint
Conscious restriction over the amount that is eaten
Reward
Reinforcement of achievement of specific behavior or outcome through reward contingent on the meeting
Scheduling of diet
Plan timing and context or location of food or physical activity in advance of performing behavior
and activity Self-monitoring
Record specific behaviors or outcomes on regular basis
Stimulus control
Alter personal environment such that it is more supportive of target behaviors (adapted from CALO-RE25)
Support: buddying
Perform target behaviors with another person
Support: motivational
Discussing, pledging, or revealing weight loss goals, plans, or achievements or challenges to others to
Support: professional
Seek help to manage weight from someone with specific expertise
Weight management aids
Use of or purchase of aids to achieve weight loss in any other manner (including, but not limited to,
RESULTS
of that target
bolster motivation
reducing energy intake and increasing energy output)
including baseline education, income, and deprivation indexes, where reported, and any reported analyses comparing effects by SES. Our primary outcome was mean difference in weight change between study arms at 6 months, calculated by using baseline observation carried forward (BOCF; an intention-to-treat analysis that imputes baseline weight for participants missing at follow-up) to facilitate comparisons between studies. In studies in which BOCF data were not presented in the paper, we used complete case data (and in 1 case, last observation
carried forward data) to calculate BOCF.26 In studies in which weight change data were not available in any of these forms, we report results narratively in the text. Where available, we also extracted intermediate outcome data related to the behavioral processes listed in the previously mentioned domains.
Statistical Analysis We pooled studies in which self-help interventions were compared with minimal control interventions and studies that compared self-help interventions with respect
March 2015, Vol 105, No. 3 | American Journal of Public Health
50% and 90% are viewed as an indication of substantial heterogeneity.23 We conducted sensitivity analyses to test if effect sizes differed when we excluded studies at high risk of bias or if they differed when we used complete case data as opposed to BOCF. We had originally planned to supplement our meta-analyses with metaregression to explore associations between intervention characteristics and weight change, and associations between the degree to which participants enacted specific behavioral processes and weight change, but did not have sufficient data with which to do so.
to a variable of interest. We conducted meta-analyses by using a random effects model in Review Manager version 5.2 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark). We conducted separate analyses for weight change at 6 and 12 months. We presented pooled results as mean differences (kg) with 95% confidence intervals (CIs). We used the I2 statistic to present statistical heterogeneity between studies.27 In general, values between 30% and 60% are viewed as indicating moderate heterogeneity and values between
After we removed duplicates, our search retrieved 3883 references, 3697 of which were excluded at title---abstract stage. We assessed 186 full-text articles for eligibility, and excluded 143 of them, with the most common reason for exclusion being that the self-help intervention was confounded by an additional, more intensive intervention. (Other reasons for exclusion are listed in Supplemental File B PRISMA diagram; available as a supplement to the online version of this article at http://www.ajph.org). We included 43 references, representing 23 studies, 18 of which had sufficient data to be included in statistical analyses. Twelve studies compared self-help interventions with minimal controls (no additional contact with or without printed information on consequences of obesity), 7 compared tailored or interactive programs with fixed (nontailored, noninteractive) programs, and 2 compared tailored and interactive programs with nontailored interactive programs.
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e45
SYSTEMATIC REVIEW
Characteristics of Included Studies Characteristics of the included studies are summarized in this section, covering participants, intervention format and content, and risk of bias. Table 2 provides further detail on the participants and interventions in individual studies. Participants. In total, relevant study arms represented 9672 participants, with study samples ranging from 33 to 2862 participants and a mean of 420 participants per study. Twelve studies were conducted in the United States, 4 were conducted in Australia, 3 each were conducted in Japan and the United Kingdom, and 1 was conducted in Finland. The mean age of study participants ranged from 36 to 60 years and the majority were female (61% average). Four studies recruited men only39,40,42,46 and 1 recruited women only.28 Mean BMI at baseline ranged from 26.2 (Japanese studies) to 36.0 kg/m2, with an average of 31.8 kg/m2. Intervention. Overall, the included studies represented 39 intervention arms. Of the interventions being tested, 23 included print materials; 22 included access to a Web site; 10 involved some form of self-monitoring equipment (most commonly pedometers); 5 included automated text messages; 2 each employed podcasts, DVDs, and smartphone apps; and 1 each included e-mail, computer software, and phone recordings. Eighteen interventions were both tailored and interactive, 6 were interactive but not tailored, 3 were tailored but not interactive (e.g., print materials tailored based on baseline data), and 12 were neither tailored nor interactive (i.e., the same content would be viewed or heard by all participants, referred
to from hereafter as “fixed”). In cases in which access to the intervention was limited (e.g., nonprint interventions), length of access ranged from 2 to 12 months. Independent from length of access, some programs also specified the frequency with which participants accessed the intervention content (for example, recommending participants access a program once a week). In instances in which such recommendations were made, this ranged from daily to 4 times over 3 months. In 2 studies, intervention material could only be accessed at specified sites.49,50 In all other studies, interventions could be accessed at the participants’ discretion. Individuals were given set energy intake targets in 8 studies, and a further 5 studies generally specified “healthy eating.” One study explicitly set no specific dietary recommendations and 9 did not report any details of the dietary advice provided. Eleven studies did not report what physical activity advice, if any, was given to participants. Of the remaining studies, 5 specifically encouraged walking and 7 recommended an increase in physical activity but did not specify an activity type. Eight studies identified social cognitive theory as the theoretical basis for their intervention, 3 were designed with the transtheoretical model, 2 employed cognitive behavioral theory, and 1 each used habit formation theory and the behavioral determinants model. Eleven studies did not state the theoretical basis for their interventions. The most commonly recommended self-management strategies were goal setting (32 of 39 interventions), self-monitoring (36 of 39), and seeking social support from nonprofessionals (19 of 39).
e46 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
Fourteen interventions recommended restrictions, and 11 each recommended strategies to boost motivation, strategies to reward oneself for hitting targets, and strategies involving conscious “energy compensation” (most commonly offsetting energy intake by increasing energy output). Figure 1 shows the number of interventions employing each group of strategies. Risk of bias. We judged 3 studies to be at low risk of bias across all domains assessed.38,39,41 We judged 8 studies to be at high risk of bias in at least 1 domain, and the remainder of included studies were judged to be at unclear risk of bias in at least 1 domain, most commonly because of lack of detail in the reporting of methods of random sequence generation and allocation concealment (16 studies). With the exception of 1 study that used self-report (high risk)42 and 1 study in which the method of outcome assessment was unclear (unclear risk),28 all included studies measured weight objectively and hence were judged to be at low risk of detection bias. We judged 3 studies to be at high risk of attrition bias (2 in which £ 20% of participants were followed up at 6 months36,43 and 1 in which attrition was differential between groups [93% vs 53%]).31 We judged 2 that did not report the number followed up at 6 months to be at unclear risk.29,33 We judged 3 studies to be at high risk of other bias because of the use of a wait-list control40,44,49 and we judged a further 3 to be at unclear risk as participants were offered the intervention at study end, but it was not clear if they knew they would be offered the intervention in advance.28,30,42 We judged an additional study to be at high risk of other bias as a participant in the control group
used a weight loss app and lost 32 kg, which the authors stated had a strong influence on the mean weight loss in the control group.31 Another was judged to be at high risk of other bias as the intervention arm was followed up at 3-month intervals whereas the control arm was only followed up at 12 months.34 (Supplemental File C, available as a supplement to the online version of this article at http://www.ajph.org, contains risk of bias judgments for each included study across all domains, along with reasons for each judgment.)
Weight Change Pooled results of studies comparing self-help interventions with controls detected a significant effect in favor of the intervention at 6 months (mean difference –1.85 kg; 95% CI = –2.86, –0.83), though statistical heterogeneity was substantial (I 2 = 76%). Heterogeneity was not explained by subgroup differences between tailored or interactive versus fixed programs, and all subgroups detected significant effects in favor of the intervention with no evidence of a difference between subgroups (Figure 2). Sensitivity analysis using complete case data were not significantly different (Supplemental File D, available as a supplement to the online version of this article at http://www.ajph. org) and the effect estimate remained similar when we excluded studies at high risk of bias (mean difference –1.42 kg; 95% CI = –2.57, –0.28). However, at 12 months, a significant effect was not detected in pooled results from the 5 studies with data at this point (all tailored and interactive programs; mean difference –0.76 kg; 95% CI = –1.73, 0.20; Figure 3), with substantial statistical heterogeneity (I 2 = 62%). Results were
American Journal of Public Health | March 2015, Vol 105, No. 3
March 2015, Vol 105, No. 3 | American Journal of Public Health
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e47
pedometer)
2. Interactive (tailored unclear; Web)
3. Fixed nontailored (print—book)
1. Tailored and interactive (Web)
2. Tailored and interactive (as per arm 1 plus automated personalized
(n = 128)
Collins et al. 2013,32
Australia (301)
2. Minimal control
1. Tailored interactive (Web, DVD, pedometer)
2. Fixed (DVD, pedometer) 3. Minimal control
Australia (65)
Australia (159)
1. Tailored interactive (Web)
Morgan et al. 2011,39
Morgan et al. 2013,40
1. Tailored interactive (Web)
2. Minimal control
consumption at 6 mo
Weight, portion size and alcohol
Weight at 6 mo and 12 mo
Weight at 6 mo and 12 mo
None reported
3 mo; 4 d/wk
3 mo; frequency NS
minimum weekly
12 mo; frequency
3-mo intervals
individualized feedback sheets using standardized set of feedback
None in arm 2; in arm 1,
control
One information session as per
site
Brief instruction on how to use Web
No
Continued
intervention primarily automated
Some therapist contact but
authors
Further outcome data provided by
United States (1277)
United Kingdom (221)
Weight measured but not reported
2. Minimal control
Johnson et al. 2008,37
McConnon et al 2007,38
reported—no usable data in group 3 9 mo; 4 reports at
Weight at 8 mo measured but not
3. Fixed (print plus contract plus phone recording) 1. Tailored noninteractive (print reports and manual)
1.5 mo; weekly in
No groups 1 and 2, daily
None reported
2. Fixed (print plus contract)
No
1. Fixed (print)
weekly feedback
United States (33)
Length NS; automated
No
Jeffery et al. 1982,36
Weight at 6 mo and 12 mo
NS
Authors provided additional outcome data
of intervention as cited in text
Additional data from YouTube videos
1. Interactive tailored (Web, print)
at 12 mo
Weight and information seeking
No
prompts
and did not respond to other
Phone prompts if did not enter weight
No
available at time of reporting
2. Fixed (Web, print)
2. Minimal control
Finland (125)
NS
6 mo; 4 d/wk
then users’ choice
6 mo; daily for first wk,
Session with standard advice as per
United States (1177)
1. Interactive tailored (SMS, Web)
Haapala et al. 2009,34
Weight at 6 mo
scale at 6 mo
Weight and cognitive restraint
at 6 mo
Weight and self-monitoring
NS; frequency: daily
presentation; results at 6 mo not
Data from unpublished conference
Hersey et al. 2012,35
1. Interactive nontailored (Web, print, accelerometer) 2. Minimal control
Greene et al. 2013,33 United States (349)
reports, feedback, and reminders via Web and SMS)
1. Tailored and interactive (smartphone app, SMS, Web)
United Kingdom
mail research staff with technical problems
heart rate transmitter belt, print materials) 2. Minimal control
Carter et al. 2013,31
control group, training and told to e-
watch with in-built program,
Australia (74)
Weight at 6 mo
1. Tailored and interactive (computerized
manual using defined script
Byrne et al. 2006,30
to use daily
indefinitely; instructed
1. Fixed (print leaflet)
2. Minimal control
30 min talking patients through study
place of 6-mo data
3 mo, have leaflet
comparisons; 7-mo data used in
Arm 3 not included in any
Notes
NS
No
Personal Contact
indefinitely; frequency
1 mo, but have booklet
Length and Frequency of Accessc
United Kingdom (527)
None reported
and restrictions at 7 mo
Weight, eating behavior,
Outcomes Relevant to This Review
Beeken et al. 2013,29
4. Fixed (booklet)
2. Fixed (print—booklet and letters) 3. Interactive not tailored (print—booklet, pedometer)
1. Tailored and interactive (print—booklet and letters,
Japan (205)
Study Arms for This Reviewb
Adachi et al. 2007,28
Study ID, Year, Country, (No.a)
TABLE 2—Characteristics of Included Studies in Systematic Review and Meta-analysis of Self-Help Interventions for Weight Loss
e48 | Systematic Review | Peer Reviewed | Hartmann-Boyce et al.
American Journal of Public Health | March 2015, Vol 105, No. 3
1. Interactive tailored (Web)
2. Minimal control
1. Interactive tailored (computer software, print)
2. Interactive tailored (print)
2011,48 United States (96)
Winett et al. 2007,49
United States (707)
Wylie-Rosett et al.
2001,50 United States
Weight at 12 mo
3 mo, monthly after
12 mo; weekly for first
No
accessed in foyer of HMO (group 1)
Computer material can only be
accessed in church
modeling; material can only be
Cluster randomized, analysis via ITT
author; low level of app and Twitter use in intervention group
Additional outcome data provided by
using generalized mixed model
Unclear
messages per day to Twitter groups (not individualized)
In group 1, study coordinator sent 2
social support at 12 mo
3 mo; weekly
group 1 asked to log on to Twitter daily
6 mo; podcasts 2·/wk,
following standard script
Instructions on how to use Web site
intake, planning and tracking,
Fiber, fruit and vegetable, fat
monitoring at 6 mo
Weight, social support and self-
arm 1
weekly reporting in
6 mo; frequency NS but
Notes. BOCF = baseline observation carried forward; HMO = health maintenance organization; ITT = intention to treat; LOCF = last observation carried forward; NS = not specified; SMS = short message service. a No. is total of only those participants relevant to this review (excludes additional arms testing more intensive interventions). b Fixed are programs that are neither tailored nor interactive. c As recommended.
(352)
1. Interactive nontailored (podcasts, print, app, Twitter)
2. Fixed (podcasts, print)
Turner-McGrievy and Tate
2. Interactive nontailored (Web)
States (128)
Weight at 6 mo
received from authors
1. Tailored interactive (Web, e-mail)
data; additional outcome data
received from authors 7-mo data used in place of 6-mo
data; additional outcome data
9-mo data used in place of 6-mo
provided
calculated BOCF using LOCF data
Complete case data not available,
reported in useable form
Weight at 6 mo measured but not
data
Authors provided additional outcome
frequency NS
troubleshooting No
equipment training and
Brief health education session,
No
be publicly posted on study Web site No
study staff with questions that would
Orientation face-to-face; could e-mail
providing materials
assignment; 2-h group session
Motivational lecture before
indefinitely);
Tate et al. 2006,47 United
meals at 3 mo
2. Fixed nontailored (booklet)
Japan (51)
4 mo; armband 16 h/d
1 mo (have
Weight at 7 mo, eating between
Weight at 9 mo
newsletters monthly
Tanaka et al. 2010,46
Shuger et al. 2011,45
arm 2; frequency NS 12 mo; SMS 4·/d,
and upload info daily
1. Interactive nontailored (print, monitoring armband with
United States (170)
Weight at 6 mo and 12 mo
wrist watch display, Web)
2. Minimal control (info only)
Shapiro et al. 2012,44
2 mo in arm 1, NS in
intake at 6 mo and 12 mo Weight at 6 mo
2. Fixed (print) 1. Tailored interactive (print—booklet and letters, pedometer)
1. Tailored interactive (Web, SMS)
United States (2862)
mo; weekly homework
NS but presumably 12
daily
NS, but record weight
vegetables, saturated fat
Weight at 12 mo, fiber, fruit and
Weight at 6 mo
United States (99)
1. Tailored interactive (Web with tailored action plans)
2. Tailored noninteractive (Web)
Rothert et al. 2006,43
1. Tailored interactive (Web, pedometer)
2. Minimal control
United States (441)
Japan (125)
Patrick et al. 2011,42
1. Tailored noninteractive (print and pedometer)
2. Minimal control
Nakata et al. 2011,41
TABLE 2—Continued
SYSTEMATIC REVIEW
Self-monitoring Goal setting Support: Motivational Regulation: Restrictions Motivation Energy compensation Reward Regulation: Rule setting Planning content Stimulus control Imitation (modelling) Regulation: Allowances Scheduling of diet and activity Impulse management: Awareness of motives Information seeking Weight management aids Support: Buddying Flexible restraint Support: Professional Impulse management: Distraction Impulse management: Acceptance 0
10
20
30
40
Number of Interventions Recommending Use of Strategy
FIGURE 1—Frequency of use of recommended strategies (by domain) in included studies: systematic review and meta-analysis of self-help interventions for weight loss.
heavily influenced by 1 study, which was at high risk of attrition bias for 12-month data (49% of intervention participants vs 70% control participants followed up at 12 months); removing this study from the analysis reduced heterogeneity to moderate and yielded a significant effect in favor of the intervention (mean difference –1.10 kg; 95% CI –1.97, –0.23; I 2 = 38%). Three further studies, not included in any meta-analyses, also compared self-help interventions with minimal controls. One study of a tailored and interactive intervention did not publish data on weight change, but reported that the intervention significantly increased healthy eating and exercise behaviors.37 A second study
was not included because it was a cluster randomized trial of a church-based intervention. Participants in the intervention group lost an adjusted average of 0.2 kg (SD = 0.4) at 7 months, whereas control participants had gained 0.1 kg (SD = 0.5).49 In the third study, in which a print leaflet was compared with a minimal control, 6-month data, though collected, were not available at the time of publication. Unpublished data from 3-month follow-up detected a significant effect of the intervention.52
Comparisons on the Basis of Intervention Type Interactive versus fixed. Results from 7 studies directly comparing interactive (with and without
March 2015, Vol 105, No. 3 | American Journal of Public Health
tailoring) with fixed interventions detected a significant effect in favor of the interactive programs (mean difference –0.94 kg; 95% CI = –1.50, –0.38; I 2 = 2%). In subgroup analysis, the significant effect remained in the group of tailored and interactive studies but the pooled effect was not significant in 2 studies of interactive nontailored programs (Figure 4). A sensitivity analysis using complete case data revealed similar results (Supplemental File D, available as a supplement to the online version of this article at http://www.ajph.org) and results continued to significantly favor the interactive programs when we removed studies at high risk of bias (mean difference –0.83 kg; 95% CI = –1.53, –0.14). In
the only study that reported 12-month data, a significant effect remained (mean difference –0.70; 95% CI = –1.37, –0.03).36 Tailored and interactive versus interactive only. Two studies directly compared tailored and interactive programs with nontailored but interactive programs; pooled results detected a small but significant effect in favor of tailored programs (mean difference –0.41 kg; 95% CI = –0.61, –0.20; I 2 = 0; n = 2990). However, the large majority of participants in this comparison came from 1 study that suffered very high loss to follow-up, limiting confidence in this finding.43 Addition of specific features. Four studies tested the addition of adjuncts to self-help interventions, but only one detected a significant impact. One added a pedometer to both a fixed and a tailored and interactive program; pooled results comparing arms with pedometers with those not providing pedometers revealed a nonsignificant mean difference of –0.27 kg (95% CI = –0.93, 0.40; I 2 = 0%; n = 205).28 One study compared standard with enhanced versions of a Web intervention. The enhanced version included the addition of personalized reports and an escalating reminder schedule to use the program Web site; a significant difference was not detected between arms at 6 months (mean difference –0.90 kg; 95% CI = –2.12, 0.32; n = 301).32 The third tested the addition of an expert software program written to guide participants through a tailored workbook and detected a significant difference between participants using the software and those using the workbook only (mean difference –0.80 kg; 95% CI = –0.98, –0.62; n = 352; data at 12 months).50 The fourth tested the addition of a weight loss
Hartmann-Boyce et al. | Peer Reviewed | Systematic Review | e49
SYSTEMATIC REVIEW
Intervention Control Mean SD Total Mean SD Total Weight Study or Subgroup 1.1.1 Tailored and interactive Byrne 200630 McConnon 200738 Morgan 201139 Morgan 201340 Shapiro 201244 Subtotal (95% CI)
-4.8 -0.6 -5.3 -5.1 -1.3
3.9 3 5.8 5.4 3.8
41 111 34 53 81 320
-1.9 -0.9 -3.5 -0.5 -0.6
3.4 4.5 5.6 3.4 3.3
Mean Difference IV, Random (95% CI)
33 110 30 26 89 288
12.0% 15.0% 7.5% 10.7% 14.7% 59.9%
-2.90 (-4.56, -1.24) 0.30 (-0.71, 1.31) -1.80 (-4.60, 1.00) -4.60 (-6.55, -2.65) -0.70 (-1.77, 0.37) -1.81 (-3.50, -0.13)
169 63 232
15.6% 13.2% 28.8%
-1.70 (-2.58, -0.82) -1.60 (-3.00, -0.20) -1.67 (-2.42, -0.93)
26 26
11.3% 11.3%
-3.00 (-4.81, -1.19) -3.00 (-4.81, -1.19)
546 100.0%
-1.85 (-2.86, -0.83)
Mean Difference IV, Random, 95% CI
Heterogeneity: τ²=2.94; χ²=24.96; df=4 (P
