Moderators of Theory-Based Interventions to Promote Physical Activity in 77 Randomized Controlled Trials

Abstract

A meta-analysis of randomized controlled trials (RCTs) has recently showed that theory-based interventions designed to promote physical activity (PA) significantly increased PA behavior. The objective of the present study was to investigate the moderators of the efficacy of these theory-based interventions. Seventy-seven RCTs evaluating theory-based interventions were systematically identified. Sample, intervention, methodology, and theory implementation characteristics were extracted, coded by three duos of independent investigators, and tested as moderators of interventions effect in a multiple–meta-regression model. Three moderators were negatively associated with the efficacy of theory-based interventions on PA behavior: intervention length (≥14 weeks; β = −.22, p = .004), number of experimental patients (β = −.10, p = .002), and global methodological quality score (β = −.08, p = .04). Our findings suggest that the efficacy of theory-based interventions to promote PA could be overestimated consequently due to methodological weaknesses of RCTs and that interventions shorter than 14 weeks could maximize the increase of PA behavior.

Keywords

exercise health behavior change meta-regression methodological quality scoring method theoretical implementation

Physical inactivity is currently the fourth leading cause of death worldwide (Kohl et al., 2012). It is a recognized risk factor of many noncommunicable diseases, for example, cardiovascular disease, type 2 diabetes, breast cancer, cognitive impairment, and depression (Khan et al., 2012)—generating a tremendous burden in terms of care cost. To prevent these human, social, and economic consequences, interventions aiming to increase regular physical activity (PA) represent nowadays a political and research priority. Previous research indicated that achieving at least 2.5 hours per week of moderate-intensity PA leads to significant health benefits in adults (Haanstra & Kamper 2012; Samitz, Egger, & Zwahlen, 2011).

Numerous trials promoting PA in inactive or insufficiently active adults have been published in the past decade (Avery, Flynn, van Wersch, Sniehotta, & Trenell, 2012; Ng, Mackney, Jenkins, & Hill, 2012; Orrow, Kinmonth, Sanderson, & Sutton, 2012), using various strategies (e.g., website, counseling), targeting different specific populations (e.g., pregnant women, primary care patients), and involving PA interventions based on various psychological theories (e.g., self-determination theory: Deci & Ryan, 2000; transtheoretical model: Prochaska, Johnson, & Lee, 2009).

In parallel, health behavior researchers have increasingly adopted the evidence-based medicine standards (Davidson et al., 2003). In particular, randomized controlled trials (RCTs) and meta-analyses of RCTs are more and more systematically considered as the highest level of evidence for the causal effects of a behavioral intervention.

Nevertheless, some authors pointed out that RCTs targeting health behavior change often present diverse characteristics (e.g., mode of delivery, length of intervention) leading to a high level of heterogeneity between studies (Coyne, Thombs, & Hagedoorn, 2010). Therefore, there is cumulative evidence that sensitivity analyses should be performed to test the robustness of meta-analytic findings (Dechartres, Altman, Trinquart, Boutron, & Ravaud, 2014; Ioannidis, Patsopoulos, & Evangelou, 2007). Based on previous investigations in epidemiology and health behavior change, three categories of moderators can be emphasized: sample and intervention characteristics (Dechartres et al., 2014), methodological quality (Johnson, Low, & MacDonald, 2015; Peters, de Bruin, & Crutzen, 2014), and theoretical implementation quality (Michie et al., 2013; Prestwich et al., 2013). Previous meta-analyses showed a significant small to moderate effect of interventions promoting PA, with high indices of heterogeneity in most cases (Cleland, Granados, Crawford, Winzenberg, & Ball, 2013; Conn, Hafdahl, Brown, & Brown, 2008; Conn, Hafdahl, & Mehr, 2011).

Several potential moderators have been examined with regard to these interventions. First, in the PA context, Conn et al. (2008) and Conn et al. (2011) conducted moderation analyses to identify whether certain sample and intervention characteristics affected the efficacy of interventions. Face-to-face–delivered interventions were more effective than mediated interventions (e.g., by mail or phone). Besides, targeting exclusively PA behavior, using behavioral rather than cognitive strategies, and encouraging PA self-monitoring were identified as the most effective interventions characteristics. Sample characteristics—that is, age, gender, ethnicity, socioeconomic status—and the type of PA measures did not moderate effect sizes. However, these results were based on two meta-analyses including interventional studies with mixed design.

Second, inconsistencies were found regarding methodological quality of experimental studies conducted on health behaviors (Johnson et al., 2015). An examination of 200 meta-analyses in health promotion RCTs underlined that only 2.5% of them investigated a possible interaction between methodological quality and effect size. While methodological weaknesses have been associated with larger efficacy of structured exercise interventions (e.g., see Carayol, Delpierre, Bernard, & Ninot, 2015), to our knowledge, no data are available to date about the moderating effect of methodological criteria in studies for PA promotion.

Third, some authors recently recommended taking into account theory-based interventions (Gourlan et al., 2016; Prestwich et al., 2013; Taylor, Conner, Lawton, 2012) so as to provide assumptions about why interventions would differentially affect health behaviors. The effectiveness of an intervention may depend not only on the theory on which it has been based (e.g., transtheoretical model) but also on how the conception and implementation of the intervention fits to the theoretical constructs of the theory (e.g., stages, self-efficacy, processes of change, decisional balance). Strategies of implementation of behavior change techniques (Dusseldorp, van Genugten, van Buuren, Verheijden, & van Empelen, 2014; Michie et al., 2013) can be classified and coded based on a Theory Coding Scheme (TCS; Michie & Prestwich, 2010). To our knowledge, only one meta-analysis investigated the moderation effect of theoretical implementation quality among theory-based interventions targeting diet and PA (Prestwich et al., 2013), but it failed to observe any moderation effect of specific items, or the overall TCS score.

Taken together, these findings suggest that several sources of variability emerge from early studies, which may partly explain the heterogeneity observed in PA promotion meta-analyses. First, sample and intervention features are highly variable, especially regarding age, population, intervention format, and duration. Second, the level of methodological quality of RCTs associated with theory-based interventions to promote PA behavior was not investigated. Third, theoretical implementation strategies appear to be highly miscellaneous and their moderation effect in RCTs promoting PA needs further investigations. Therefore, the current study aims to investigate the moderation effect of (a) sample and intervention characteristics, (b) methodological quality criteria, and (c) theoretical implementation quality criteria on the efficacy of theory-based interventions devoted to PA promotion in RCTs.

Method

Inclusion Criteria

The studies included in the current article were based on a recent meta-analysis investigating the efficacy of theory-based interventions on PA (Gourlan et al., 2016). This systematic review is reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement (Liberati et al., 2009). Studies were included in the systematic review if they met the following criteria (according to Participants, Interventions, Controls, Outcomes, Studies; PICOS; Liberati et al., 2009). Participants were all adults (≥18 years). It had to be explicitly mentioned that the intervention was based on at least one theoretical framework(s) explicitly mentioned in the text. Intervention could target only PA, or PA and other outcomes. Included trials compared an intervention group benefiting from a theory-based intervention with a control group that either did not benefit from an intervention or benefited from a non–theory-based intervention (minimal intervention, attention placebo, or active comparison control condition). The included trials measured PA as a primary or secondary outcome both at pre- and postintervention times. The indicator used had to be a direct—self-reported or objective—measurement of PA behavior (e.g., duration, energy expenditure, number of steps). Trials based on physical fitness data or intentions were not included. Only RCT studies were included.

Search Strategy

Studies were identified by searching in Medline, PsycINFO, and PSYarticles until May 15, 2013, with no language restriction. MESH and text-word terms were used in Medline with limits (i.e., age >18 years, RCT design; see details in supplementary material, available online with this article). The following theories and their constructs were included successively in search equations: transtheoretical model, social cognitive theory, theory of planned behavior, self-determination theory, protection motivation theory, theory of reasoned action, health belief model, precaution adoption process model, rubicon model, model of action phases, and health action process approach. This search was completed by manually searching bibliographies of previous relevant reviews focusing on PA (Ashford, Edmunds, & French, 2010; Daddario, 2007; Floyd, Prentice-Dunn, & Rogers, 2000; Hagger & Chatzisarantis, 2008; Hutchison, Breckon, & Johnston, 2009; Manning, 2009; Marshall & Biddle, 2001; Milne, Sheeran, & Orbell, 2000; Spencer, Adams, Malone, Roy, & Yost, 2006; Wilson, Mack, & Grattan, 2008).

Coding Procedures

A standardized data extraction was performed by three duos of independent reviewers who systematically recorded the data. Any disagreements were resolved by discussion with the third author. Descriptive data were extracted regarding (a) sample/intervention characteristics, (b) methodological quality, and (c) theoretical implementation quality. For articles in which insufficient information on program outcomes was reported, repeated attempts were made to contact corresponding authors to request the required information.

Sample and Intervention Characteristics

The following information was recorded regarding sample—year, number of participants randomly assigned, age, gender, health profile (healthy adults, adults with risk factor(s), adults with chronic illness), and comparison conditions (wait-list control, contact control, minimal intervention control, alternative treatment intervention)—and intervention characteristics: mode of delivery intervention, supervised PA, frequency of session, intervention length. The health profile variable was established as follows. Students and workers were lumped together in the “Healthy adults’” category. Samples with elderly, sedentary, pregnant, or at-risk participants (e.g., metabolic syndrome, osteoarthritis) were grouped into the “adults with risk factors” category. The “Adults with chronic illness” category included participants with diagnosed chronic illness (i.e., diabetes, cancer, low back pain, and Parkinson’s disease).

Methodological Quality

An 11-point quality score based on 11 methodological criteria was calculated. Mostly derived from Cochrane collaboration’s tools for assessing risk of bias (Higgins et al., 2011), they were specifically chosen to assess risk of bias in nonpharmacologic interventional RCTs regarding (a) participants (eligibility criteria for participants were collected), (b) interventions (precise details on both the experimental treatment and comparator are given), (c) interventions standardization (details on how the interventions were standardized are presented), (d) primary outcome (a specific primary outcome is indicated), (e) sample size (the calculation technique of sample size was specified), (f) randomization (explanations on the method used to generate the random allocation sequence are provided, including details such as block or stratification), (g) blinding (whether or not participants or administering [co]interventions or/and person assessing the outcomes were blinded to group assignment), (h) participant flow (for each group, the number of participants is stated regarding randomization assignment, intended treatment received, completion of the study protocol, analysis of the primary outcome), (i) description of care providers (i.e., counsellors, medical doctors, psychologists, exercise experts, or students [MSc or PhD]; care providers performing the intervention in each group or the number of patients treated by each care provider), (j) baseline information (description of care providers—case volume, qualification, expertise—or centers—volume—in each group), and (k) intention to treat analyses. All items were coded “yes,” “no,” or “not applicable.” Criteria (i) and (j) were relevant for interventions including face-to-face and/or phoning, but in many other types of interventions (e.g., booklets) those criteria were not applicable. Consequently, depending on the studies, the calculation of the methodological quality score was based on 8 to 11 items among the abovementioned criteria. When calculated on less than 11 items, the total methodological score was thus converted to 11 points to make it comparable from one study to another.

Theoretical Implementation Quality

TCS includes 19 items (Michie & Prestwich, 2010), but 10 items were used to code the quality of theory implementation. The selected items–coded either “yes” or “no”—were the following: (a) targeted construct mentioned as predictor of behavior, (b) theory/predictors used to select recipients for the intervention, (c) theory/predictors used to select/develop intervention techniques, (d) theory/predictors used to tailor intervention techniques to recipients, (e) all intervention techniques are explicitly linked to at least one theory-relevant construct/predictor, (f) at least one, but not all, of the intervention techniques are explicitly linked to at least one theory-relevant construct/ predictor, (g) group of techniques are linked to a group of constructs/predictors, (h) all theory-relevant constructs/predictors are explicitly linked to at least one intervention technique, (i) at least one, but not all, of the theory relevant constructs/predictors are explicitly linked to at least one intervention technique, and (j) theory-relevant constructs/predictors are measured. (Details and coding procedures are presented in supplementary material, available online with this article). Nine TCS items were not used because they were either out of the scope of the present review (e.g., mediational analysis of constructs) or redundant with methodological quality or inclusion criteria (e.g., randomization, theory mentioned).

Selected Potential Moderators

According to the authors’ field experience, literature background, and distribution of characteristics in our sample (Conn et al., 2011; Johnson et al., 2015; Prestwich et al., 2013), the following characteristics were tested as potential moderators of PA intervention efficacy: (a) sample and intervention characteristics: mean age, gender, sample comparison conditions, supervised PA, frequency of sessions, intervention length, number of experimental participants; (b) methodological quality: sample size (calculation techniques of sample size), blinding, description of care providers (e.g., case volume, qualification, expertise) or centers (volume) in each group, and the global quality score; and (c) theoretical implementation quality: three items from the TCS (Items 6, 8, 10) and the global theoretical implementation quality score.

According to previous investigations and debates (da Costa, Hilfiker, & Egger, 2013; Dechartres et al., 2014; Jüni, Witschi, Bloch, & Egger, 1999), there is no consensus about methodological quality characterization (i.e., overall methodological score vs. single items) in meta-epidemiologic studies. Consequently, methodological quality was characterized with both methods in the current investigation.

Data Analysis

Effect sizes of interventions on PA behavior were calculated using Cohen’s d (Cohen, 1988) with Comprehensive Meta-Analysis software, Version 2.2064 (Borenstein, Hedges, Higgins, & Rothstein, 2009). Positive effect sizes indicate favorable changes in the theory-based intervention group in comparison to the control group. Effect sizes of .2, .5, and .8 represent small, medium, and large effects, respectively (Cohen, 1988). Effect sizes were calculated from data on mean evolution, the number of participants, and the pooled standard deviations for each trial. When summary statistics were not reported, data from the significance levels of statistical tests (e.g., t or F values) were used to make an estimation of Cohen’s d (Lipsey & Wilson, 2001). As a complement to Cohen’s d, 95% confidence intervals (CIs) were also calculated. Heterogeneity was tested with Q test and residual between-variance study was quantified through I² statistic (Higgins, Thompson, Deeks, & Altman, 2003) considering p < .10 for heterogeneity detection and I² < 25% as small, 25% to 50% as moderate, and ≥50% as large heterogeneity.

Each variable of interest was first tested as a moderator of PA intervention efficacy in random effects bivariate meta-regression models. Assumptions regarding independence of errors, homoscedasticity of variance, and normality of continuous moderators were met.

Moderators that explained a significant part of variance in PA effect sizes, with a p value <.10 in bivariate meta-regression models, were integrated into a multivariable random effects meta-regression model. All continuous variables were zero centered based on their means; categorical variables were contrast coded (−1/+1). Beta-values (β) quantify the amount of variability in standardized mean differences associated with 1-unit increase of each moderator of interest. Meta-regression analyses were conducted using Stata Version 11.2 (StatCorp, TX). The nominal level of significance was 5% in all tests except otherwise specified.

Results

Based on the previous meta-analysis of (Gourlan et al., 2016), 77 RCTs, involving a total of 82 intervention groups, were included in the present meta-regression (see the references list of included studies in the supplementary material, available online with this article).The total number of participants was 19,357 (10,574 in the experimental condition and 8,783 in the control condition). The interventions were based on the following theories: transtheoretical model, social cognitive theory, theory of planned behavior, self-determination theory, and protection motivation theory. No RCT was identified with an intervention based on the following theories: health belief model, precaution adoption process model, rubicon model, model of action phases, and health action process approach.

Sample sizes in experimental arms included on average 129 individuals (SD = 206, Mdn = 70, range: 11-1,529) with a mean age of 48.4 years (SD = 13.93). Twenty interventions focused exclusively on women. Putting aside interventions based on only one session, the mean length of interventions was 25.3 weeks (SD = 25.5, Mdn = 14.0, range: 2-104 weeks). Forty-one (50%) theory-based interventions were delivered face-to-face, of which 34 (41%) also used another delivery method (e.g., Internet). For control groups, minimal intervention, no intervention, active control, and attention placebo were identified in 36 (44%), 18 (22%), 16 (19%), and 12 (15%) of the selected trials, respectively.

Methodological global scores ranged from 2.42 to 11 (M = 6.35, SD = 2.09, Mdn = 6.00). Three items were examined separately. Calculation of sample size was found in 38 (46%) of included interventions. Care providers were described in 21 (26%) interventions where applicable, and blinding techniques were used in 7 (8%) of included interventions where applicable (see the supplementary material, available online with this article).

The mean theoretical implementation score was 6.4 (SD = 2.09, Mdn = 6), the highest score was 8, and six studies received the minimum score of 1. A link between groups of techniques and constructs (Item 8) was observed for 44 (54%) of the included interventions. All intervention techniques were based on theoretical construct(s) (Item 6) in 31 (38%) of the included interventions, and at least one of the theoretical constructs of the theory were linked to at least one intervention technique (Item 10) in 19 (23%) of included interventions. The assessment of the quality of theoretical implementation for each intervention can be seen in the supplementary material (available online with this article).

The effect size of all interventions together was d = .31, 95% CI [.24, .37] with high heterogeneity (Q = 348.52, p < .001, I² = 76.9%; see Gourlan et al., 2016).

Bivariate Analysis

Gender, sample size, intervention length, and methodological quality score and criteria significantly moderate intervention effect on PA promotion in bivariate meta-regression models (see Table 1). Intervention length was dichotomized using the median weeks as the cutoff to define interventions “<14 weeks” and “≥14 weeks.” Interventions that included only female participants showed stronger effects than studies conducted in male or both genders. Sample size, intervention length and methodological quality score were negatively associated with intervention efficacy. Interventions that did not meet quality criteria (e.g., use of sample size calculation technique, description of care providers) displayed better efficacy. Residual heterogeneity was high (I² > 75%) in bivariate meta-regression models. No indication of moderation effect was found regarding the quality of theoretical implementation criteria.

Table 1.

Moderators of Theory-Based Interventions to Promote Physical Activity.

Study dimension and level	d [95% Confidence interval]	β	p	k	I²
Age (continuous, years)		.02	.59	78	82%
Gender		.20	.03	82	80%
Women	.47 [.31, .63]			19
Mixed or men only	.27 [.19, .36]			63
Sample		.13	.22	82	81%
Healthy adults	.28 [.06, .51]			12
With risk factor(s)	.24 [.12, .36]			32
With chronic illness	.41 [.29, .52]			38
Supervised physical activity		.04	.69	82	82%
Yes	.35 [.16, .53]			15
No	.31 [.22, .39]			67
Intervention length		−.24	.004	66	81%
<14 weeks	.47 [.33, .61]			32
>14 weeks	.19 [.11, .28]			34
Frequency of sessions (continuous)		.05	.33	51	84%
Comparison conditions		−.06	.61	82	81%
Attention placebo	.31 [.12, .50]			12
Minimal intervention	.34 [.23, .46]			36
Active control	.19 [.02, .35]			18
No intervention	.37 [.16, .58]			16
No. of experimental patients (continuous)		−.10	.001	82	78%
Calculation techniques of sample size		−.19	.01	82	79%
Yes	.22 [.12, .32]			38
No	.42 [.31, .53]			44
Blinding		−.09	.28	79	82%
Yes	.24 [.13, .34]
No	.36 [.26, .46]
Balanced methodological quality score (continuous)		−.07	.063	82	81%
Description of care providers		−.19	.01	49	79%
Yes	.18 [.04, .33]			21
No	.37 [.25, .50]			28
At least one, but not all, of the theory relevant constructs/predictors are explicitly linked to at least one intervention technique		.02	.78	82	82%
Yes	.35 [.18, .51]			19
No	.31 [.22, .39]			63
All intervention techniques are explicitly linked to at least one theory-relevant construct/predictor		−.03	.68	82	82%
Yes	.29 [.14, .45]			31
No	.33 [.24, .42]			51
Group of techniques are linked to a group of constructs/predictors		.02	.53	82	82%
Yes	.35 [.23, .46]			44
No	.33 [.24, .42]			38
Implementation quality total score (continuous)		.01	.70	82	82%

Note. Items in boldface are moderators (p < .10) included in the multivariate analysis. I² = measure of heterogeneity; k = number of intervention groups.

Multivariate Analysis

The multiple meta-regression model revealed that three moderators were negatively associated with the efficacy of interventions: intervention length (≥14 vs. <14 weeks: β = −.22, 95% CI: [−.375, −.072], p = .004), the number of experimental participants (β = −.10, 95% CI: [−.162, −.040], p = .002) and the global methodological quality score (β = −.08, 95% CI: [−.167, −.002], p = .04). The final model had an adjusted R² = 31%, F(3, 62) = 7.83, p < .001. Residual heterogeneity was high (I² = 74%).

Discussion

The present study evaluated the moderators of theory-based interventions to enhance PA behavior. Overall, the results showed that studies of lower methodological quality, having smaller size sample, and involving interventions shorter than 14 weeks were related to larger size effects. These results provide some evidence that the efficacy of theory-based interventions to promote PA (Gourlan et al., 2016) could be overestimated due to methodological weaknesses of RCTs. For example, high risk of bias at the randomization step may lead to select patients with a better profile for improvement in the intervention group (Moher et al., 1998). Then, the lack of intention to treat analysis or the lack of clear selection criteria may lead to the selection of individuals with a better compliance or adherence. Furthermore, the lack of blinding to assessors may favor an overestimation of participants’ self-reported PA knowing that social desirability has been found to be associated with an overreporting of PA (Adams et al., 2005). Our findings are consistent with those of Dechartres et al. (2014), who carried out a comparative analysis of treatment outcomes among 163 meta-analyses and found that those including trials at high or unclear risk regarding sequence allocation, treatment concealment, and blinding showed significantly better outcomes, compared to those including trials at low risk of bias. Altogether, these findings support the necessity to better control for the methodological quality in meta-analyses because poor methodological quality studies have been found to increase the risk of no evidence-based conclusions in behavioral medicine and health psychology (Coyne, 2009; Coyne et al., 2010; Johnson et al., 2015).

Smaller sample sizes in RCTs were associated with higher efficacy of interventions on PA behavior. This finding is consistent with those of previous meta-analyses examining the effects of not only psychotherapy (Cuijpers, van Straten, Bohlmeijer, Hollon, & Andersson, 2010) and rehabilitation (Nüesch et al., 2010) interventions but also pharmacotherapy (Dechartres, Trinquart, Boutron, & Ravaud, 2013). Higher heterogeneity of effect sizes combined with a publication bias could explain this finding. In theory, RCTs with small sample size lead an important variability of effect sizes in both directions (negative or positive effects). Indeed, small sample size per arm results in effect size with large deviations (based on central limit theorem; Hedges & Olkin, 1985). This heterogeneity of effect size has been empirically demonstrated in pharmacology (Gibertini, Nations, & Whitaker, 2012). A publication bias suggests that investigations are more frequently published when they reveal favorable outcomes rather than negative outcomes. A publication bias has been statistically detected in this set of studies (Gourlan et al., 2016), which strengthens the hypothesis of an overestimated theory-based interventions’ efficacy on PA behavior.

A negative association of length of intervention with efficacy of intervention for PA promotion was found, independent of sample size and methodological quality. This moderator has not been studied in previous meta-analyses and may reveal a ceiling effect after 14 weeks of intervention. Put differently, it is possible that an optimal PA behavior change would be observable during the first 3 months of intervention, whereas there would be no additional effect of pursuing intervention. This counterintuitive finding could also mirror the fact that adherence tends to decrease with time, resulting in apparently lower efficacy of longer interventions. However, it was not possible to test this hypothesis in the present study. Because of the heterogeneous nature of included interventions (e.g., face-to-face combined with booster sessions by phone, booklets by mail, supervised PA plus counselling, Internet intervention), it was not possible to code the “adherence rate” with an operational definition.

Calculation of sample size, description of care provider, and exclusively female samples were identified as moderators of interventions efficacy in bivariate analyses. Yet, when they were entered simultaneously with other selected variables, these dimensions did not reach significance. It suggests that the moderation effect of these characteristics remains minor compared to moderation effect of intervention length, sample size, and methodological quality score. However, a lack of power can also be pointed out in multivariable models due to the limited number of included studies.¹

Noteworthy, theoretical implementation quality—albeit through selected items or with a global score—did not explain variation in intervention efficacy. This finding is consistent with a previous meta-analysis focusing on diet and PA (Prestwich et al., 2013). Several hypotheses may be advanced to explain this result. First, despite repeated recommendations to improve the transparency of theory-based interventions tailoring (Abraham, Johnson, de Bruin, & Luszczynska, 2014; Michie & Abraham, 2004; National Institute of Health and Care Excellence, 2014), only a small proportion of interventions make available or publish manuals linking theory and behavioral change techniques (Hoffmann, Erueti, & Glasziou 2013; McCleary, Duncan, Stewart, & Francis, 2013). Next, approximately one quarter of included studies relied on a combination of theories. TCS items are deemed less adapted for combined theories interventions, because they have generally a poor rationale (Bandura, 1998) and because multiplying theories leads to a more important number of constructs to consider when tailoring an intervention (Michie & Abraham, 2004; Peters et al., 2014).

This study has several limitations. First, although 82 interventions were included, the results should be interpreted cautiously because of important variations in the nature of interventions, outcome measures, length of interventions, and targeted populations. Second, despite the fact that several moderators were identified, heterogeneity remains elevated. Other potential moderators may explain the efficacy of theory-based interventions to promote PA, but a lack of statistical power or lack of consistent report of information across studies (e.g., adherence rate) could have impaired their identification. Third, the association of an overall methodological score with intervention outcome may not be specific enough, since the same weight is associated to each item (Dechartres et al., 2014; Jüni et al., 1999).

These findings suggest several courses of action for trialists and behavioral scientists. Clearly, it is necessary to combine a strict methodological quality designing, and strong trial management and analysis, with adequate size sample to ensure an unbiased assessment of theory-based interventions to promote PA. A reasonable approach to examine the efficacy of behavioral change intervention could be to systematically assess the robustness of meta-analyses findings by performing sensitivity analyses or to include only high-standard RCTs in meta-analyses. Recommendation panelists should be aware and report the limits of RCTs designed to promote PA change before making pragmatic decisions and establishing guidelines (Coyne et al., 2010). Taken together, these findings should incite health behavior change trialists to use the CONSORT (Consolidated Standards of Reporting Trials) statement for nonpharmacological RCTs for reporting their results (Boutron & Ravaud, 2012) and to identify specific strategies for reducing risk of bias (De Bruin, McCambridge, & Prins, 2014).

In conclusion, this is the first study to show that efficacy of theory-based interventions promoting PA could be overestimated consequently due to methodological weaknesses of RCTs. These findings emphasize the importance of improving methodological quality standards in theory-based interventions to promote PA and, more broadly, in interventional RCTs to change health behavior. They suggest that interventions shorter than 14 weeks could be advised to maximize the increase of PA behavior at short-term. However, future investigations should explore the potential dose–effect among theory-based interventions and effective strategies to promote long-term PA modification, as well as other potential moderators of interventional efficacy, such as mode of delivery combinations, medical severity of included participants, and types of PA outcome. Finally, the high degree of heterogeneity among reviewed RCTs in this meta-analysis should be taken into consideration in the establishment of evidence-based conclusions.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Paquito Bernard is supported by a postdoctoral fellowship from the Fonds de Recherche du Québec-Santé and Psychosocial Oncology Research Training program. Olivier Lareyre and Mathieu Gourlan were supported by the SIRIC Montpellier Cancer (Grant INCa-DGOS-Inserm 6045). The SIRIC Montpellier notably aims to fund research concerning health behaviors related to cancer prevention (e.g., physical activity, smoking).

Notes

Supplemental Material

Additional supporting information is available at .

References

Abraham

Johnson

B. T.

de Bruin

Luszczynska

(2014). Enhancing reporting of behavior change intervention evaluations. Journal of Acquired Immune Deficiency Syndromes,66(Suppl. 3), S293-S299. doi:10.1097/QAI.0000000000000231

Adams

S. A.

Matthews

C. E.

Ebbeling

C. B.

Moore

C. G.

Cunningham

J. E.

Fulton

Hebert

J. R.

(2005). The effect of social desirability and social approval on self-reports of physical activity. American Journal of Epidemiology, 161, 389-398. doi:doi.org/10.1093/aje/kwi054

Ashford

Edmunds

French

D. P.

(2010). What is the best way to change self-efficacy to promote lifestyle and recreational physical activity? A systematic review with meta-analysis. British Journal of Health Psychology, 15, 265-288. doi:10.1348/135910709X461752

Avery

Flynn

van Wersch

Sniehotta

F. F.

Trenell

M. I.

(2012). Changing physical activity behavior in type 2 diabetes: A systematic review and meta-analysis of behavioral interventions. Diabetes Care, 35, 2681-2689. doi:10.2337/dc11-2452

Bandura. (1998). Health promotion from the perspective of social cognitive theory. Psychology & Health, 13, 623-649. doi:10.1080/08870449808407422

Borenstein

Hedges

L. V.

Higgins

J. P. T.

Rothstein

H. R.

(2009). Introduction to meta-analysis. Chichester, England: Wiley.

Boutron

Ravaud

(2012). Reporting guidelines for nonpharmacological trials. In Boutron

Ravaud

Moher

(Eds.), Randomized clinical trials of nonpharmacological treatments (Vol. 1, pp. 199-212). Boca Raton, FL: Chapman & Hall/CRC.

Carayol

Delpierre

Bernard

Ninot

(2015). Population-, intervention- and methodology-related characteristics of clinical trials impact exercise efficacy during adjuvant therapy for breast cancer: A meta-regression analysis. Psycho-Oncology, 24, 737-747. doi:10.1002/pon.3727

Cleland

Granados

Crawford

Winzenberg

Ball

(2013). Effectiveness of interventions to promote physical activity among socioeconomically disadvantaged women: A systematic review and meta-analysis. Obesity Reviews, 14, 197-212. doi:10.1111/j.1467-789X.2012.01058.x

10.

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.

11.

Conn

V. S.

Hafdahl

A. R.

Brown

S. A.

Brown

L. M.

(2008). Meta-analysis of patient education interventions to increase physical activity among chronically ill adults. Patient Education & Counseling, 70, 157-172. doi:10.1016/j.pec.2007.10.004

12.

Conn

V. S.

Hafdahl

A. R.

Mehr

D. R.

(2011). Interventions to increase physical activity among healthy adults: Meta-analysis of outcomes. American Journal of Public Health, 101, 751-758. doi:10.2105/AJPH.2010.194381

13.

Coyne

J. C.

(2009). Are most positive findings in health psychology false . . . or at least somewhat exaggerated? The European Health Psychologist, 11, 49-51.

14.

Coyne

J. C.

Thombs

B. D.

Hagedoorn

(2010). Ain’t necessarily so: Review and critique of recent meta-analyses of behavioral medicine interventions in health psychology. Health Psychology, 29, 107-116. doi:10.1037/a0017633

15.

Cuijpers

van Straten

Bohlmeijer

Hollon

S. D.

Andersson

(2010). The effects of psychotherapy for adult depression are overestimated: A meta-analysis of study quality and effect size. Psychological Medicine, 40, 211-223. http://doi.org/10.1017/S0033291709006114

16.

da Costa

B. R.

Hilfiker

Egger

(2013). PEDro’s bias: Summary quality scores should not be used in meta-analysis. Journal of Clinical Epidemiology, 66(1), 75-77.doi:10.1016/j.jclinepi.2012.08.003

17.

Daddario

D. K.

(2007). A review of the use of the health belief model for weight management. Medicine Nursing, 16, 363-366.

18.

Davidson

K. W.

Goldstein

Kaplan

R. M.

Kaufmann

P. G.

Knatterud

G. L.

Orleans

C. T.

. . . Whitlock

E. P.

(2003). Evidence-based behavioral medicine: what is it and how do we achieve it? Annals of Behavioral Medicine, 26, 161-171.

19.

De Bruin

McCambridge

Prins

J. M.

(2014). Reducing the risk of bias in health behaviour change trials: Improving trial design, reporting or bias assessment criteria? A review and case study. Psychology & Health, 30(1), 8-34. doi:10.1080/08870446.2014.953531

20.

Dechartres

Altman

D. G.

Trinquart

Boutron

Ravaud

(2014). Association between analytic strategy and estimates of treatment outcomes in meta-analyses. Journal of the American Medical Association, 312, 623-630. doi:10.1001/jama.2014.8166

21.

Dechartres

Trinquart

Boutron

Ravaud

(2013). Influence of trial sample size on treatment effect estimates: Meta-epidemiological study. British Medical Journal, 2013, 346. doi:10.1136/bmj.f2304

22.

Deci

E. L.

Ryan

R. M.

(2000). The “what” and “why” of goal pursuits: Human needs and the self-determination of behavior. Psychological Inquiry, 11, 227-268. doi:10.1207/S15327965PLI1104_01

23.

Dusseldorp

van Genugten

van Buuren

Verheijden

M. W.

van Empelen

(2014). Combinations of techniques that effectively change health behavior: Evidence from meta-cart analysis. Health Psychology, 33, 1530-1540. doi:10.1037/hea0000018

24.

Floyd

D. L.

Prentice-Dunn

Rogers

R. W.

(2000). A meta-analysis of research on Protection Motivation Theory. Journal of Applied Social Psychology, 30, 407-429. doi:10.1111/j.1559-1816.2000.tb02323.x

25.

Gibertini

Nations

K. R.

Whitaker

J. A.

(2012). Obtained effect size as a function of sample size in approved antidepressants: A real-world illustration in support of better trial design. International Clinical Psychopharmacology, 27, 100-106. doi:10.1097/YIC.0b013e32834f504f

26.

Gourlan

Bernard

Bortolon

Romain

A. J.

Lareyre

Carayol

. . . Boiché

(2016). Efficacy of theory-based interventions to promote physical activity. A meta-analysis of randomised controlled trials. Health Psychology Review, 10(1), 50-66. http://doi.org/10.1080/17437199.2014.981777

27.

Haanstra

T. M.

Kamper

S. J.

(2012). Increasing physical activity in healthy adults: a meta-analysis. British Journal of Sports Medicine, 46, 998-999. doi:10.1136/bjsports-2012-091641

28.

Hagger

Chatzisarantis

(2008). Self-determination theory and the psychology of exercise. International Review of Sport and Exercise Psychology, 1, 79-103. doi:10.1080/17509840701827437

29.

Hedges

L. V.

Olkin

(1985). Statistical methods for meta-analysis. Cambridge, MA: Academic Press.

30.

Higgins

J. P. T.

Altman

D. G.

Gøtzsche

P. C.

Jüni

Moher

. . . Cochrane Statistical Methods Group (2011). The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. British Medical Journal, 343. doi:10.1136/bmj.d5928

31.

Higgins

J. P. T.

Thompson

S. G.

Deeks

J. J.

Altman

D. G.

(2003). Measuring inconsistency in meta-analyses. British Medical Journal, 327, 557-560. doi:10.1136/bmj.327.7414.557

32.

Hoffmann

T. C.

Erueti

Glasziou

P. P.

(2013). Poor description of non-pharmacological interventions: Analysis of consecutive sample of randomised trials. British Medical Journal, 347. doi.org/10.1136/bmj.f3755

33.

Hutchison

A. J.

Breckon

J. D.

Johnston

L. H.

(2009). Physical activity behavior change interventions based on the transtheoretical model: A systematic review. Health Education & Behavior, 36, 829-845. doi:10.1177/1090198108318491

34.

Ioannidis

J. P. A.

Patsopoulos

N. A.

Evangelou

(2007). Uncertainty in heterogeneity estimates in meta-analyses. British Medical Journal, 335, 914-916. doi:10.1136/bmj.39343.408449.80

35.

Johnson

B. T.

Low

R. E.

MacDonald

H. V.

(2015). Panning for the Gold in Health Research: Incorporating studies’ methodological quality in meta-analysis. Psychology & Health, 30, 135-152 doi:10.1080/08870446.2014.953533

36.

Jüni

Witschi

Bloch

Egger

(1999). The hazards of scoring the quality of clinical trials for meta-analysis. Journal of the American Medical Association, 282, 1054-1060.

37.

Khan

K. M.

Thompson

A. M.

Blair

S. N.

Sallis

J. F.

Powell

K. E.

Bull

F. C.

Bauman

A. E.

(2012). Sport and exercise as contributors to the health of nations. Lancet, 380, 59-64. doi:10.1016/S0140-6736(12)60865-4

38.

Kohl

H. W.

III Craig

C. L.

Lambert

E. V.

Inoue

Alkandari

J. R.

Leetongin

. . . Lancet physical activity series working group. (2012). The pandemic of physical inactivity: Global action for public health. Lancet, 380, 294-305. doi:10.1016/S0140-6736(12)60898-8

39.

Liberati

Altman

D. G.

Tetzlaff

Mulrow

Gøtzsche

P. C.

Ioannidis

J. P. A.

. . . Moher

(2009). The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: Explanation and elaboration. PLoS Med, 6, e1000100. doi:10.1371/journal.pmed.1000100

40.

Lipsey

M. W.

Wilson

D. B.

(2001). Practical meta-analysis. Thousand Oaks, CA: Sage.

41.

Manning

(2009). The effects of subjective norms on behaviour in the theory of planned behaviour: A meta-analysis. British Journal of Social Psychology, 48, 649-705. doi:10.1348/014466608X393136

42.

Marshall

S. J.

Biddle

S. J.

(2001). The transtheoretical model of behavior change: A meta-analysis of applications to physical activity and exercise. Annals of Behavioral Medicine, 23, 229-246

43.

McCleary

Duncan

E. M.

Stewart

Francis

J. J.

(2013). Active ingredients are reported more often for pharmacologic than non-pharmacologic interventions: An illustrative review of reporting practices in titles and abstracts. Trials, 14, 146. doi:10.1186/1745-6215-14-146

44.

Michie

Abraham

(2004). Interventions to change health behaviours: Evidence-based or evidence-inspired? Psychology & Health, 19(1), 29-49. doi:10.1080/0887044031000141199

45.

Michie

Prestwich

(2010). Are interventions theory-based? Development of a theory coding scheme. Health Psychology, 29(1), 1-8. doi:10.1037/a0016939

46.

Michie

Richardson

Johnston

Abraham

Francis

Hardeman

. . . Wood

C. E.

(2013). The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: Building an international consensus for the reporting of behavior change interventions. Annals of Behavioral Medicine, 46(1), 81-95. doi:10.1007/s12160-013-9486-6

47.

Milne

Sheeran

Orbell

(2000). Prediction and intervention in health-related behavior: a meta-analytic review of protection motivation theory. Journal of Applied Social Psychology, 30, 106-143. doi:10.1111/j.1559-1816.2000.tb02308.x

48.

Moher

Pham

Jones

Cook

D. J.

Jadad

A. R.

Moher

. . . Klassen

T. P.

(1998). Does quality of reports of randomised trials affect estimates of intervention efficacy reported in meta-analyses? Lancet, 352, 609-613. doi:10.1016/S0140-6736(98)01085-X

49.

National Institute of Health and Care Excellence. (2014). Public health guidance: Update of behaviour change: The principles for effective intervention. London, England: Author.

50.

L. W. C.

Mackney

Jenkins

Hill

(2012). Does exercise training change physical activity in people with COPD? A systematic review and meta-analysis. Chronic Respiratory Disease, 9(1), 17-26. doi:10.1177/1479972311430335

51.

Nüesch

Trelle

Reichenbach

Rutjes

A. W. S.

Tschannen

Altman

D. G.

. . . Jüni

(2010). Small study effects in meta-analyses of osteoarthritis trials: Meta-epidemiological study. British Medical Journal, 341, c3515. doi.org/10.1136/bmj.c3515

52.

Orrow

Kinmonth

A.-L.

Sanderson

Sutton

(2012). Effectiveness of physical activity promotion based in primary care: Systematic review and meta-analysis of randomised controlled trials. British Medical Journal, 344, e1389. doi:10.1136/bmj.c3515

53.

Peters

G.-J. Y.

de Bruin

Crutzen

(2014). Everything should be as simple as possible, but no simpler: Towards a protocol for accumulating evidence regarding the active content of health behaviour change interventions. Health Psychology Review, 9(1), 1-14. doi:10.1080/17437199.2013.848409

54.

Prestwich

Sniehotta

F. F.

Whittington

Dombrowski

S. U.

Rogers

Michie

(2013). Does theory influence the effectiveness of health behavior interventions? Meta-analysis. Health Psychology, 33, 465-474. doi:10.1037/a0032853

55.

Prochaska

J. O.

Johnson

Lee

(2009). The transtheoretical model behavior change. In Shumaker

Ockene

Rickert

(Eds.), The handbook of health behavior change (3rd ed., pp. 59-83). New York, NY: Springer.

56.

Samitz

Egger

Zwahlen

(2011). Domains of physical activity and all-cause mortality: Systematic review and dose-response meta-analysis of cohort studies. International Journal of Epidemiology, 40, 1382-1400. doi:10.1093/ije/dyr112

57.

Spencer

Adams

T. B.

Malone

Roy

Yost

(2006). Applying the transtheoretical model to exercise: A systematic and comprehensive review of the literature. Health Promotion Practice, 7, 428-443. doi:10.1177/1524839905278900

58.

Taylor

Conner

Lawton

(2012). The impact of theory on the effectiveness of worksite physical activity interventions: A meta-analysis and meta-regression. Health Psychology Review, 6(1), 33-73. doi:10.1080/17437199.2010.533441

59.

Wilson

P. M.

Mack

D. E.

Grattan

K. P.

(2008). Understanding motivation for exercise: A self-determination theory perspective. Canadian Psychology, 49, 250-256. doi:10.1037/a0012762