Health Psychology Review

ISSN: 1743-7199 (Print) 1743-7202 (Online) Journal homepage: http://www.tandfonline.com/loi/rhpr20

Mobile phone SMS messages can enhance healthy behaviour: a meta-analysis of randomised controlled trials Jayne A. Orr & Robert J. King To cite this article: Jayne A. Orr & Robert J. King (2015) Mobile phone SMS messages can enhance healthy behaviour: a meta-analysis of randomised controlled trials, Health Psychology Review, 9:4, 397-416, DOI: 10.1080/17437199.2015.1022847 To link to this article: http://dx.doi.org/10.1080/17437199.2015.1022847

View supplementary material

Accepted author version posted online: 05 Mar 2015. Published online: 28 May 2015. Submit your article to this journal

Article views: 1101

View related articles

View Crossmark data

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=rhpr20 Download by: [38.124.40.237]

Date: 31 October 2015, At: 09:44

Health Psychology Review, 2015 Vol. 9, No. 4, 397–416, http://dx.doi.org/10.1080/17437199.2015.1022847

REVIEW Mobile phone SMS messages can enhance healthy behaviour: a meta-analysis of randomised controlled trials Jayne A. Orr* and Robert J. King School of Psychology and Counselling, Queensland University of Technology, Kelvin Grove, Brisbane, QLD 4059, Australia

Downloaded by [38.124.40.237] at 09:44 31 October 2015

(Received 6 January 2014; accepted 22 February 2015) Healthy behaviour, such as smoking cessation and adherence to prescribed medications, mitigates illness risk factors but health behaviour change can be challenging. Mobile phone short-message service (SMS) messages are increasingly used to deliver interventions designed to enhance healthy behaviour. This meta-analysis used a random-effects model to synthesise 38 randomised controlled trials that investigated the efficacy of SMS messages to enhance healthy behaviour. Participants (N = 19,641) lived in developed and developing countries and were diverse with respect to age, ethnicity, socioeconomic background and health behaviours targeted for change. SMS messages had a small, positive, significant effect (g = 0.291) on a broad range of healthy behaviour. This effect was maximised when multiple SMS messages per day were used (g = 0.395) compared to using lower frequencies (daily, multiple per week and once-off) (g = 0.244). The low heterogeneity in this meta-analysis (I2 = 38.619) supports reporting a summary effect size and implies that the effect of SMS messaging is robust, regardless of population characteristics or healthy behaviour targeted. SMS messaging is a simple, cost-effective intervention that can be automated and can reach any mobile phone owner. While the effect size is small, potential health benefits are well worth achieving. Keywords: health behaviour change; mobile phones; SMS; text messages; metaanalysis

Introduction Healthy behaviour, such as adopting healthy lifestyle choices and managing chronic disease, mitigates illness risk factors, resulting in improved health outcomes and enhanced quality of life (Swann et al., 2010). However, health behaviour change can be challenging because unhealthy behaviour may be pleasurable and even addictive, with immediate reinforcement making short-term gain more important than long-term loss (Rollnick, Miller, & Butler, 2008), and change may require effort and/or self-control which is easily depleted (Muraven, 2008). Further, behaviour change requires a complex combination of readiness (Prochaska, DiClemente, & Norcross, 1992), motivation (Miller & Rollnick, 2002) and capacity to change (Bandura, 1989; Deci & Ryan, 1985); self-efficacy and positive reinforcement of change attempts (Bandura, 1989); and feelings of relatedness to others (Deci & Ryan, 1985). *Corresponding author. Email: [email protected] Preliminary findings presented in this paper were presented at the 44th Society for Psychotherapy Research International Annual Meeting in July 2013. © 2015 Taylor & Francis

Downloaded by [38.124.40.237] at 09:44 31 October 2015

398

J.A. Orr and R.J. King

Mobile phone short-message service (SMS) messages are being used increasingly to deliver interventions designed to enhance healthy behaviour because the majority of people worldwide own a mobile phone, regardless of socioeconomic status; all mobile phones support SMS messaging; and SMS messages utilise push technology that enables outside parties to initiate contact with mobile phone users (Coffey, Sanci, Patton, Sawyer, & Haller, 2006; International Telecommunication Union, 2013). Despite behaviour change interventions grounded in theory demonstrating greater efficacy for producing health behaviour change compared to those that are not guided by theory (Michie & Johnston, 2012; NICE, 2007), very few SMS studies have reported that the SMS intervention has been based on a health behaviour theory (Abroms, Padmanabhan, & Evans, 2012; Cole-Lewis & Kershaw, 2010; Fjeldsoe, Marshall, & Miller, 2009). From a theoretical perspective, SMS interventions should be effective in changing health behaviour because SMS messages provide an ideal vehicle for delivering content based on effective behaviour change theories such as the transtheoretical model (Prochaska et al., 1992), motivational interviewing (Miller & Rollnick, 2002), social cognitive theory (Bandura, 1989), self-determination theory (Deci & Ryan, 1985), the theory of planned behavior (Ajzen, 1991, in press) and social ecological theory (Bronfenbrenner, 1977) to initiate positive short-term health behaviour change. Further, mobile phones’ compatibility with experience sampling facilitates the receipt of timely, moment-specific SMS feedback from health consumers (Morris et al., 2010). SMS messages can deliver cost-effective, automatic, real-time, brief, tailored and varied messages to initiate health behaviour change, at specified frequency when support for behaviour change is required, and in a manner that minimises time demands on health consumers while maintaining their interest (Fjeldsoe et al., 2009; Krishna, Boren, & Balas, 2009) and promoting their privacy (Lauder, Chester, & Berk, 2007; Ybarra & Eaton, 2005). Hence, SMS messages are used in health behaviour change interventions to deliver essential behaviour change constructs such as reminders or encouragement to perform certain actions, to obtain progress feedback, to enhance engagement and selfefficacy by reinforcing the adoption of healthy behaviour and to provide a source of social support so that health consumers do not feel alone (de Jongh, Gurol-Urganci, Vodopivec-Jamsek, Car, & Atun, 2012; Klasnja & Pratt, 2012). Research using SMS message interventions to enhance healthy behaviour has burgeoned over the past 10 years. We identified 40 reviews that addressed the use of SMS messages to enhance healthy behaviour, comprising systematic reviews (13%), stand-alone meta-analyses (18%), stand-alone systematic reviews (32%), narrative reviews (35%) and scoping reviews (2%). Overall, 67% of reviews concluded that SMS messages contributed to healthy behaviour, 11% concluded that it did not and 22% withheld judgement on grounds of insufficient evidence. These reviews suggest that SMS messages are most effective for relatively simple behaviour modification such as attending medical appointments (100% of studies reporting positive impact) and increasing medication adherence (85% of studies reporting positive impact). However, the impact of SMS messages on more complex health behaviour change was weaker, with only 68% of studies showing a positive impact on the adoption of healthy lifestyle choices (such as smoking cessation or healthy diet) and only 50% of studies showing positive impact on disease prevention activity (such as sunscreen use or immunisation). Most of the reviews we identified did not use meta-analytic procedures and their findings must be treated with caution because many studies reviewed were underpowered and/or included studies with design weaknesses that rendered it impossible to isolate the specific

Downloaded by [38.124.40.237] at 09:44 31 October 2015

Health Psychology Review

399

impact of SMS messages. We identified five meta-analyses that were of acceptable methodological quality and included more than one study that used SMS messages to change health behaviour (Car, Gurol-Urganci, de Jongh, Vodopivec-Jamsek, & Atun, 2012; Fanning, Mullen, & McAuley, 2012; Free, Phillips, Galli, et al., 2013; Free, Phillips, Watson, et al., 2013; Whittaker et al., 2012). All five concluded that SMS messages were helpful in changing health behaviour. Specifically, these meta-analyses concluded that SMS messages had a moderate effect on increasing physical activity (d = 0.54) (Fanning et al., 2012); a small effect on smoking cessation that equated to approximately d = 0.437 (Free, Phillips, Galli, et al., 2013) or d = 0.343 (Whittaker et al., 2012); and a small effect on increasing appointment attendance of approximately d = 0.242 (Free, Phillips, Watson, et al., 2013) or d = 0.267 (Car et al., 2012). However, these meta-analyses are subject to important limitations. Each meta-analysis focused on a specific healthy behaviour, which limits both study sample and generalisability of findings and precludes an overall estimate of the effect of using SMS messages to enhance healthy behaviour. Lack of recency is a limitation of 4 of the 5 meta-analyses because 14 new studies that meet the inclusion criteria of these meta-analyses have since been published (Car et al., 2012; Fanning et al., 2012; Free, Phillips, Galli, et al., 2013; Free, Phillips, Watson, et al., 2013). As SMS messaging is a rapidly growing area of research, meta-analyses quickly lose currency. Four meta-analyses failed to exclude studies where there were confounds in the intervention (Fanning et al., 2012; Free, Phillips, Galli, et al., 2013; Free, Phillips, Watson, et al., 2013; Whittaker et al., 2012). Two meta-analyses included studies that did not have a proper control group because the control was an alternative intervention (Free, Phillips, Galli, et al., 2013; Free, Phillips, Watson, et al., 2013). Finally, three synthesised research of mixed quality by including both randomised controlled and non-randomised trials (Fanning et al., 2012; Free, Phillips, Galli, et al., 2013; Free, Phillips, Watson, et al., 2013). Because of these shortcomings we think there remains a need for a meta-analysis that is both broader and more methodologically rigorous with respect to inclusion criteria. There is still much to learn about the role of SMS messages in health behaviour change, and it is important to understand how robust SMS message interventions are across a range of health behaviour change. The primary objective of this meta-analysis was to examine the available evidence for the efficacy of SMS messages to enhance healthy behaviour. The secondary objective was to gain a better understanding of SMS message features (dose, tailoring and directionality) that contribute to healthy behaviours of differing categories and complexities. This meta-analysis differs from others conducted to date in its inclusion of all categories of health behaviour change. Just as Smith and Glass (1977) justified synthesising a wide range of psychotherapy studies in their famous meta-analysis in order to determine the overall effectiveness of psychotherapy and to enable comparisons of the different types of therapy, the current meta-analysis justifies synthesising all eligible studies so as to obtain an overall picture of the efficacy of using SMS messages to enhance healthy behaviour. Although targeted healthy behaviours may appear heterogeneous on the surface, in so far as they have broadly similar goals we assumed that they share sufficient commonality to enable data pooling. This is because all health behaviour change requires a combination of factors such as education, motivation, memory prompts, reinforcement with corrective feedback and support to maintain motivation and self-efficacy; and outcomes are measurable. While this was a working assumption, we were mindful that the meta-analytic process itself would yield valuable

Downloaded by [38.124.40.237] at 09:44 31 October 2015

400

J.A. Orr and R.J. King

information about the validity of this assumption through both indices of heterogeneity and the results of subgroup analysis. We hypothesised that SMS messaging would have a small to medium effect on overall health behaviour change because the majority of individual studies and reviews have found such effects, especially when interventions comprised SMS messages alone. Given our more rigorous inclusion criteria, we thought that a small effect size was more likely than a medium effect. Research findings concerning the efficacy of SMS messages to enhance healthy behaviour informed our choice of moderators. The evidence for SMS dose effects is mixed, with some interventions finding benefits of high dose (e.g., Franklin, Waller, Pagliari, & Greene, 2006; Free et al., 2009, 2011; Rodgers et al., 2005) which has been refuted by others (e.g., Pop-Eleches et al., 2011; Weitzel, Bernhardt, Usdan, Mays, & Glanz, 2007). Nevertheless, we thought it possible that there would be some dose effect. SMS messages are commonly classified into three types. Personalised SMS messages include details unique to the message recipient such as his/her name and appointment details. Tailored SMS messages are those where the choice of message sent to the recipient is influenced by recipient characteristics such as his/her gender or behaviour change status. Standardised SMS messages are preformatted messages that are sent to all recipients. We anticipated that personalised SMS messages would be more effective than tailored messages, which in turn would be more effective than standardised messages. Also, we thought that two-way sender-recipient SMS message communication would be more effective than one-way communication. This may be because bidirectional communication establishes a dynamic feedback loop that facilitates ongoing and evolving sharing of information that can strengthen the likelihood of health behaviour change (Riley et al., 2011). In addition, we expected SMS messages to have a greater effect on less complex health behaviour change. If the task required to effect change is simple (e.g., appointment attendance), it should not take much to change behaviour, therefore an SMS message intervention would probably be effective. However, if the task is difficult (e.g., modifying unhealthy behaviour or managing chronic disease), then more than an SMS message intervention may be required to change behaviour. Methods Eligibility criteria The rationale, inclusion criteria and analysis methods for the meta-analysis were specified in advance and were documented in an unpublished protocol that was completed on 28 November 2012 (see Supplementary Material 1). All completed randomised controlled trials (RCTs) that compared an SMS message intervention targeting health behaviour change to a non-SMS message control that did not attempt to change health behaviour were included in this meta-analysis. No language or publication status restrictions were applied. Therefore, RCTs that compared two active treatments were excluded. For an RCT to be included, the control group had to be identical to the SMS message group, except for the absence of the SMS message intervention. Furthermore, SMS messaging had to be the only health behaviour change intervention employed. RCTs that were still under way were rejected. Also, RCTs that used mobile phones solely for telephone calls, or that used SMS messages to target mental health, or for the collection and/or transmission of patient data without feedback to patients or to trigger reactions in observational studies were omitted. Finally, RCTs that surveyed SMS message usage or attitudes to receiving SMS messages, or used SMS messages to communicate with others

Health Psychology Review

401

Downloaded by [38.124.40.237] at 09:44 31 October 2015

(e.g., medical personnel) rather than health consumers targeted for health behaviour change were excluded. Search strategy The search strategy was developed by both authors and was verified by a research librarian to ensure accuracy and inclusivity of articles retrieved. Electronic searches for publications in any language were conducted on 2 December 2012. Databases searched included PsycINFO, Cumulative Index to Nursing and Allied Health (CINAHL), Pubmed, Scopus, Web of Science, ScienceDirect and Proquest Psychology. In addition, searches were conducted on the Cochrane Central Register of Controlled Trials, Quick Find and Google Scholar. The query string used was (‘mobile phone’ or ‘cell phone’ or ‘cellular phone’ or ‘text messag*’ or SMS or ‘short messag* service’) AND (support or intervention or ‘behav* change’). Search delimiters selected included species of ‘human’ and article types of ‘Randomised Control Trial’, ‘Meta-Analysis’, ‘Review’ and ‘Systematic Reviews’ for Pubmed; publication type of ‘Randomised Control Trial’, ‘Meta Analysis’, ‘Review’ and ‘Systematic Review’ for CINAHL; document types of ‘Article’ and ‘Review’ and subject area of ‘Social Sciences’, ‘Psychology’, ‘Nursing’, ‘Health Professions’ and ‘Pharmacology, Toxicology and Pharmaceutics’ for Scopus; document types of ‘Article’ and ‘Review’ for Web of Science; subject of ‘Nursing and Health Professions’, ‘Pharmacology, Toxicology and Pharmaceutical Science’, ‘Psychology’ and ‘Social Sciences’ and document type of ‘Article’ and ‘Review Article’ for ScienceDirect; and source type of ‘Scholarly Journals’ and document type of ‘Article’, ‘Dissertation/Thesis’, ‘Review’ and ‘Literature Review’ for Proquest Psychology. Search results were exported to the Endnote (X6; Thomson Reuters, 2012) citation manager, duplicates were removed and the titles of the remaining citations were examined to eliminate those that clearly did not meet the selection criteria, e.g., studies that used the SMS drug for pancreatic cancer, those that examined radiation effects of using mobile phones and those that examined the effects on driving safety of using mobile phones. Reviews were extracted for further scrutiny. To exclude citations that did not meet the meta-analysis inclusion criteria, the English abstracts of the remaining citations were examined and the full text of articles was scrutinised when abstracts did not provide sufficient information to make a decision. Relevant review articles and RCTs were hand searched to maximise the number of potentially eligible publications. In addition, the corresponding authors of all relevant RCTs were invited to share additional research that met the meta-analysis selection criteria, regardless of whether it was published, unpublished, under review or in press, or included in conference papers, posters or dissertations. Also, these authors were asked whether they knew of any other unpublished work in the field that met the selection criteria. As the primary author conducted all searches and rejected citations that did not meet the meta-analysis inclusion criteria to obtain a shortlist of articles to be coded, the process to select citations for relevancy was repeated on 3 February 2013 and was reviewed by the secondary author. Any disparity was discussed until consensus was reached, thereby negating the need for a formal reliability study of the shortlist selection process.

Downloaded by [38.124.40.237] at 09:44 31 October 2015

402

J.A. Orr and R.J. King

Coding procedures The primary author developed a coding manual and data extraction form in consultation with the secondary author. The coding procedure was piloted by both authors using five articles, and the coding manual and data extraction form were refined accordingly. Study details, participant details, ratings and details to assess risk of bias, intervention details, control details and outcome measures were coded (see Supplementary Material 2, for an example of the data extraction form). Study details included a unique numerical identifier, the complete citation, the journal impact factor, language of publication, location of the research, the theory on which the intervention was based, the targeted behaviour change and its complexity, the design of the RCT and limitations reported by researchers. Participant details were specified separately for the intervention and control arms and included age range, mean age and standard deviation, the sample size at the start and end of the study, the percentage gender split, ethnicity, socioeconomic status, health problem/s and any additional descriptive information provided as free text. Assessment of risk of bias was assessed as no concern, concern or unclear, across the domains of utilisation of random sequence generation, concealment of allocation to conditions, blinding of outcome assessors, completeness of outcome data, reporting of all planned outcomes, attrition bias and researcher-allegiance bias and concerns about other sources of bias could be provided as free text. Intervention details included a description of the intervention, its duration, the purpose of SMS messages used [reminder, social connectedness, instruction, request for information, other (e.g., feedback, motivation, distraction, suggestion)], SMS dose, the degree of SMS message tailoring (standardised, tailored or personalised), researcher-participant directionality of SMS messaging and the time/s that SMS messages were sent. Control details coded included the type of control (waitlist, alternative treatment, treatment as usual, no treatment, other) and a description of the control condition. For each outcome measure at each stage of measurement, the name of the measure, the stage of measurement, the measurement’s quality (objective, subjective using standardised measures, subjective using self-report), the sample size and the data required to calculate an effect size was coded for the intervention and control arms. The primary author and a second coder with masters’ level qualifications independently coded all shortlisted articles using the refined coding manual and data extraction form. All non-English articles had English abstracts, and the full text of these articles was translated into English using Google Translate (Google, Mountain View, California). Authors were contacted up to three times when required data were missing from articles or when clarification regarding details of studies was required. Coder inter-rater reliability was assessed by calculating kappa and intra-class coefficients for the categorical and continuous variables, respectively. Following, inconsistencies between the two coders were discussed by both authors and the second coder and resolved by consensus. Assessment of study quality Study methodological quality was assessed independently by both authors and the second coder, with discrepancies resolved by consensus. Risk of bias in the included studies was assessed across five of the six domains described in the Cochrane Collaboration’s classification scheme for bias for RCTs: utilisation of random sequence generation, the concealment of allocation to conditions, blinding of outcome assessors to condition assignment (participant blinding was not assessed because it is not possible to conceal an

Health Psychology Review

403

Downloaded by [38.124.40.237] at 09:44 31 October 2015

SMS messaging intervention from participants), completeness of outcome data and other sources of bias (Higgins & Altman, 2008). The sixth domain, reporting of all prespecified outcomes, could not be assessed as the protocols of included studies could not be accessed. Within each study, domains were rated as low risk of bias (+) when there was no concern regarding bias, as high risk of bias (–) when there was concern regarding bias, or unclear risk of bias (?) if the information was absent. Risk of bias associated with dealing with incomplete data was considered to be low for studies that used intention-totreat analysis. Final selection of studies and outcomes After coding, each shortlisted article was reviewed independently by both authors and the coder to decide whether it should be included in the meta-analysis. The initial percentage of agreement was calculated, and any disagreements were discussed until consensus was reached. Next, as many articles selected for inclusion in the meta-analysis reported multiple health behaviour change outcomes, both authors and the coder independently selected the primary outcome measure that most directly measured the targeted behaviour change, e.g., if both a biological outcome measure and a self-report measure of behaviour change were available, the self-report outcome measure was preferred because the biological measure is of a health status indicator that may or may not result from the targeted behaviour change. In instances when the choice was between a self-report outcome measure backed up with biological testing or another self-report outcome measure, the former was chosen because of its superior accuracy. All discrepancies were discussed until consensus was reached. Power Power calculations for a random effects model (Borenstein, Hedges, Higgins, & Rothstein, 2009) were used to determine the minimum number of studies that would need to be included in the meta-analysis to provide sufficient power to detect the effect of SMS messaging on healthy behaviour. Considering past research, it was predicted that SMS messages would have a small, positive effect on healthy behaviour. Assuming an effect size of d = 0.30, a random effects analysis, a moderate level of between-study variance, statistical power of 0.80 and a significance level (α) of 0.05, approximately 6 studies with a mean sample size of 50 participants per condition (or 4 studies with 75 participants per condition, or 3 studies with 100 participants per condition) would be needed for the meta-analysis to provide sufficient power to detect the significance of an overall mean effect size of d = 0.30. Statistical methods Meta-analysis software The Comprehensive Meta-Analysis (version 2.2.021; Biostat, 2011) software package was used to create the database of studies to be included in the meta-analysis, to calculate effect sizes and to conduct all statistical analyses. Effect size determination The data required to calculate an effect size for all comparisons of each outcome measure were extracted. For dichotomous data, these were the counts for each category (outcome

404

J.A. Orr and R.J. King

satisfied and outcome not satisfied) in the intervention and control groups. For continuous data, measured at end point only, these were the mean, standard deviation (SD) and number of participants in the intervention and control groups. When means and SDs were not reported, the difference in means, intervention and control sample sizes and independent groups p value were extracted. For continuous change data, the pre and post means, SDs and number of participants in the intervention and control groups were extracted; or when unavailable, the mean change, paired t for change and sample size for the intervention and control groups were extracted. It was assumed that t tests were twotailed unless stated otherwise. The extracted data were entered into the Comprehensive Meta-Analysis database.

Downloaded by [38.124.40.237] at 09:44 31 October 2015

Data synthesis The Comprehensive Meta-Analysis software calculated the standardised mean difference (Cohen’s d) for each study’s continuous outcomes and the odds ratio for each study’s dichotomous outcomes. To ensure independence of data, steps were taken to ensure that each study contributed only one effect size to the meta-analysis. Where multiple time points for a chosen outcome measure were reported, the end point for the outcome measure was used. Where multiple outcomes in the same study based on the same participants existed, a combined effect across outcomes was computed: for dichotomous data, the multiple intervention groups were collapsed and compared to the control group, while for continuous data, the two intervention groups were collapsed to obtain a combined mean and SD which was then compared to the control group (Borenstein et al., 2009; Higgins & Green, 2009). Each effect size was weighted by its inverse variance weight in calculating mean effect sizes. To combine outcomes with continuous and dichotomous formats, the odds ratio was transformed to a standardised mean difference. Hedges g, an effect size that is a derivation of the standardised mean difference (Cohen’s d) and uses a pooled variance component and a correction factor to reduce bias for small sample sizes (Borenstein et al., 2009) was used as the estimate of effect size for each study in the meta-analysis. Effect sizes were interpreted as large (> 0.80), medium (0.50) or small (

Mobile phone SMS messages can enhance healthy behaviour: a meta-analysis of randomised controlled trials.

Healthy behaviour, such as smoking cessation and adherence to prescribed medications, mitigates illness risk factors but health behaviour change can b...
500KB Sizes 0 Downloads 5 Views