Efficacy of text messaging-based interventions for health promotion: A meta-analysis

Social Science & Medicine 97 (2013) 41e48

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Review

Efficacy of text messaging-based interventions for health promotion: A meta-analysis Katharine J. Head a, Seth M. Noar b, *, Nicholas T. Iannarino a, Nancy Grant Harrington a a

Department of Communication, University of Kentucky, United States School of Journalism and Mass Communication and Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill, 363 Carroll Hall, Campus Box 3365, Chapel Hill, NC 27599-3365, United States b

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 13 August 2013

This meta-analysis investigated the efficacy of text messaging-based health promotion interventions. Nineteen randomized controlled trials conducted in 13 countries met inclusion criteria and were coded on a variety of participant, intervention, and methodological moderators. Meta-analytic procedures were used to compute and aggregate effect sizes. The overall weighted mean effect size representing the impact of these interventions on health outcomes was d ¼ .329 (95% CI ¼ .274, .385; p < .001). This effect size was statistically heterogeneous (Q18 ¼ 55.60, p < .001, I2 ¼ 67.62), and several variables significantly moderated the effects of interventions. Smoking cessation and physical activity interventions were more successful than interventions targeting other health outcomes. Message tailoring and personalization were significantly associated with greater intervention efficacy. No significant differences were found between text-only interventions and interventions that included texting plus other components. Interventions that used an individualized or decreasing frequency of messages over the course of the intervention were more successful than interventions that used a fixed message frequency. We discuss implications of these results for health promotion interventions that use text messaging. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Text messaging Health promotion Intervention Message tailoring

Text messaging or short-message service (SMS) is a relatively simple technology for sending and receiving messages on mobile phones, and it has become a popular form of communication across the world. Messages are limited to 160 characters and can be sent from mobile phones and some Internet sites (Grinter & Eldridge, 2003). Although most commonly used for social communication, text messaging also can be applied in health promotion interventions to deliver health messages and positively influence health behaviors. In this manuscript, we describe text messaging, review previous literature on its use in health promotion interventions, and present a meta-analysis of 19 studies that report on health promotion interventions that used text messaging to change behavior. Text messaging has far reach. Mobile phones are owned by 67% of the world’s population; by 2017, estimates are that approximately eight billion mobile phones with text messaging capability will be in use (International Telecommunication Union, 2010; Portio Research, 2012). Text messaging is the most popular non-voice application on mobile phones (Lenhart,

* Corresponding author. E-mail address: [email protected] (S.M. Noar). 0277-9536/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2013.08.003

2010). In a recent survey of 21 countries, a median of 75% of cell phone owners reported regularly sending and receiving text messages (Kohut et al., 2011). Although text messaging has widespread popularity, data suggest that it differs across population segments. For example, minority groups are more likely to send/receive text messages than Whites in the United States (Lenhart, 2010). Age also makes a difference, as U.S. teens send/ receive on average 50 messages per day, whereas adults on average send/receive only 10 messages per day (Lenhart, 2010). Other research suggests text messaging is particularly popular in developing countries; in a survey of 21 countries, which included the United States, Great Britain, and China, Kohut and colleagues found text messaging was most popular in Indonesia, Kenya, and Lebanon. Far reaching capacity is not the only advantage of this technology; it is also relatively cost effective. Because the technology involved in sending a text is very simple, the cost for wireless carriers is almost nothing (despite the fact that the cost to customers can vary considerably; Stross, 2008). Additionally, new advancements in text messaging across a variety of media, such as text messaging apps on mobile phones that can send/receive messages from other media, are predicted to drive the cost of traditional text messaging plans down due to decreased demand

42

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

(Wortham, 2011). This low cost has appeal not only for users but also for researchers and practitioners wishing to employ this technology to reach target audiences. Perhaps the most interesting aspect of this technology is its omnipresence. A quick scan of most settings (e.g., a classroom, a sporting event) reveals the ubiquitous presence of mobile phones. Most U.S. cell phone owners (65%) report sleeping with their cell phone either in or right next to their bed (Lenhart, 2010). One scholar argues that our very society is changing because people are constantly on their cell phones (Katz, 2007). In other words, it seems this is not a technology that must be brought to an individual or that an individual must be “driven” to use. It is a technology that is an integral part of people’s lives. Because of text messaging’s arguably unprecedented potential to reach people with messages, researchers are becoming more interested in the role that it can play in health behavior change interventions. In the next section, we discuss the use of text messaging in health interventions. Text message-based health promotion interventions The use of text messaging in health interventions is a relatively new practice. In fact, the first formal evaluation of a health intervention using text messaging only appeared in 2002 (Neville, Greene, McLeod, Tracey, & Surie, 2002). The first randomized controlled trial (RCT) of a health promotion intervention appeared three years later (Rodgers et al., 2005). Recently, scholars have summarized the proliferation of text messaging health interventions in several systematic review articles and book chapters (Abroms, Padmanabhan, & Evans, 2012; Cole-Lewis & Kershaw, 2010; Fjeldsoe, Marshall, & Miller, 2009; Fjeldsoe, Miller, & Marshall, 2012). These reviews examine participant, intervention, and methodological characteristics of these interventions and suggest this technology is effective for delivery of health messages. Fjeldsoe et al. (2009) conducted the first systematic review when they examined 14 health interventions that used texting to deliver health messages. They found that interventions varied widely on characteristics such as research design, intervention length, nature of texting dialog (intervention- or participantinitiated, automated or sent by a person, one-way or interactive), frequency of text messages, and message type (tailored or targeted). They concluded that this technology has much potential, with 13 of the interventions showing positive behavioral changes (although not all results were statistically significant). Cole-Lewis and Kershaw (2010) also reviewed the use of text messaging in health interventions. They reviewed 12 studies, all of which used either a full experimental (RCT) or a quasiexperimental design, suggesting the growth of formal evaluations in the area. The review drew attention to three trends. First, text messaging interventions have global appeal; the 12 studies reviewed were conducted in nine different countries. Second, retention was relatively high in the text message-based interventions, with none of the studies falling below 68% retention at follow-up. Third this technology was an effective way to deliver messages and change health behaviors; eight of the 12 studies had statistically significant health behavior changes in the hypothesized direction. The authors concluded that research on text messaging interventions “should be approached with urgency” (p. 67), although they emphasized that researchers still have a long way to go to maximize the use of this technology to promote health behavior change. As the literature has developed, researchers have more incisively analyzed the application of text messaging to health interventions. Two recent book chapters draw attention to three conceptual, theoretical, and methodological trends in text

message-based health intervention research. First, researchers are increasingly recognizing conceptual distinctions between the purposes of interventions and types of health behaviors being addressed with text messaging. For example, Fjeldsoe et al. (2012) observe four purposes for which text messaging has been used: 1) enhancing health service provision, 2) distributing mass health education messages, 3) encouraging better disease selfmanagement practices, and 4) delivering personalized health promotion interventions. In terms of types of health behaviors, Abroms et al. (2012) discuss the use of text messaging in health promotion interventions (e.g., smoking cessation, weight loss), which may be argued to be conceptually distinct from interventions focused on the management of chronic health conditions (e.g., diabetes, heart disease). These conceptual distinctions are important given that interventions will be designed differently (e.g., recruitment strategy, intervention length, theory selected) depending on the population and health issue targeted and the intended outcomes. Second, there is a lack of theory-guided work in this area and particularly a lack of theory development. Research shows that some behaviors are more apt to be targeted by interventions informed by theory, such as smoking cessation and physical activity (Fjeldsoe et al., 2012). Still, many published studies of text messaging interventions are deficient in both theory application and theory testing (Fjeldsoe et al., 2012). Additionally, researchers have not engaged in theory development that is sensitive to this unique technology (Abroms et al., 2012). Third, researchers in this area are making use of more sophisticated research designs and evaluation strategies. As previously mentioned, the first evaluation of a health intervention using texting was published in 2002. By 2010, Cole-Lewis and Kershaw located 12 studies that used a full experimental (RCT) or quasiexperimental research design. Despite the methodological progress in this area, much of the research reporting on text messagebased health interventions does not employ strong research designs or evaluation strategies. As the field moves forward, Fjeldsoe et al. (2012) argue that it must continue to produce more RCTs. The current study Text messaging boasts mass public reach, accessibility, and low cost. Past reviews suggest text messaging-based interventions have the potential to be effective in changing health behavior. This new area of research, however, is still experiencing growing pains as researchers and public health practitioners work to determine the most efficient use of text messaging in health-related interventions. Also, as of yet, there has been no quantitative synthesis of outcomes of text messaging intervention studies. In an effort to provide a comprehensive yet focused look at the effectiveness of text messaging interventions, the current study presents a metaanalysis that investigates the effectiveness of health promotion interventions that incorporate text messaging. This study has two research questions. First, are text messaging-based health promotion interventions efficacious? Second, what are the moderators of efficacious text messaging-based interventions? Method Search strategy We used a comprehensive search strategy to locate studies relevant to this meta-analysis. We did not set a date limit on the search, and we considered all applicable studies located by October 1, 2011 for inclusion. The search strategy involved three steps. First, we conducted comprehensive searches of CINAHL, Communication & Mass Media

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

Complete, PsycINFO, and Medline computerized databases using various keywords applicable to this topic, including text messag*, cell* phone, prevent*, short message service, mobile phone, intervention, mHealth, health, and health behavior. Second, we examined the reference sections of the previously mentioned review articles and book chapters (Abroms et al., 2012; Cole-Lewis & Kershaw, 2010; Fjeldsoe et al., 2009, 2012). Third, we examined the reference lists of the publications identified in steps one and two for possible additional articles. Although we considered all grey literature that was revealed through our searches, none was ultimately included except for one thesis that was later published in a journal (the published form was used for the meta-analysis). In order to be considered for inclusion in this meta-analysis, a study had to: 1. Report on an intervention designed to change health behavior (and/or a health outcome) in the service of health promotion. 2. Include at least one condition that was solely mobile phone text messaging or a broader intervention that included text messaging as a component. 3. Be an RCT that randomly assigned individuals (or groups/ clusters) to study conditions and included a control/comparison condition that did not receive any text messages focused on the target behavior. 4. Report on at least one behavioral outcome. Pilot studies that measured only psychosocial outcomes (e.g., attitudes, beliefs, stage or change, behavioral intention) were not included. 5. Be published in English. The search strategy yielded hundreds of studies. After closely examining these studies and discarding ones that were not relevant, we determined that a total of 47 studies had the potential to be included in the meta-analysis. Of these: 1. Twelve (25.5%) were excluded because they did not use a true experimental design (i.e., did not randomly assign participants to a condition and/or did not have a control group). 2. Two (4.3%) were excluded because the control group received text messages focused on the target behavior. 3. Four (8.5%) were excluded because the text messaging intervention was targeted at someone other than the individual whose behavior was being changed (e.g., parents of adolescents). 4. Seven (14.9%) were excluded because they reported on data published in other reports already included in the metaanalysis. 5. One (2.1%) was excluded because it used a small computerized device and not a cell phone for text message delivery. 6. One (2.1%) was excluded because text messaging was optional in the experimental group. 7. One (2.1%) was excluded because text messaging was contingent on other behaviors (e.g., receiving a text message only if a participant did not respond to an initial phone call).

43

of variables used in tailoring, use of personalization (e.g., use of person’s name in text), communication direction (one-way or twoway), initiation of texts (automatic versus only in response to text), frequency of texts, and intervention type (e.g., text only, text plus Web, etc.); and methodological characteristics such as follow-up time to behavior measurement, comparison group type, and retention. Coding discrepancies were resolved through discussion with all four authors. We met after the first five articles were coded to compare results and discuss any questions or concerns before moving on to code the remaining articles. We calculated intercoder reliability as both 1) the percentage of agreement between coders and 2) Cohen’s kappa, which corrects for chance categorizations (Lipsey & Wilson, 2001). Percentage of agreement was calculated for each code by dividing the number of agreed instances by the total number of instances. For example, for use of theory, the two coders agreed on 18 out of the 19 studies, which equaled 95% agreement. The mean percent agreement across all coding categories and all studies was 97%, and the mean Cohen’s kappa was .94. Effect size extraction and calculation We used the standardized mean difference statistic (i.e., the difference in treatment and control means divided by the pooled standard deviation) as the effect size indicator (Lipsey & Wilson, 2001). Because this effect size index has been shown to be upwardly biased when based on small sample sizes (Hedges & Olkin, 1985), we applied the recommended statistical correction for this bias (Lipsey & Wilson, 2001). We calculated effect sizes from data reported in the article (e.g., summary statistics, proportions) using appropriate formulas (Lipsey & Wilson, 2001). In most cases, only two study conditions existed, and the contrast of those conditions provided the relevant effect size. For cases in which more than one intervention condition existed, we chose the most “potent” condition, which was the condition for which the primary study authors hypothesized the greatest effect. When more than one control condition existed, we chose the most minimal control condition in order to most accurately estimate the “true” or “absolute” effect of the text messaging-based intervention. Additionally, in most cases only one time point of data was reported. For cases in which more than one time point was reported, we used the shortest-term follow-up data in calculating effect sizes. Further, in terms of outcome measures, we computed effect sizes based on the “primary” outcome of the intervention. If a study had more than one measure of a single primary outcome, these measures were averaged. If a study had more than one primary outcome (e.g., three behaviors), these were averaged into a behavioral composite. In addition, in order to keep effect sizes consistent and interpretable, we gave a positive sign (þ) to studies in which the text messaging condition outperformed the control and a negative sign () to studies in which the control condition outperformed the text messaging condition. Meta-analytic approach

A final set of 19 (40.4%) articles met all criteria. Each of these articles reported on one study; therefore, the final set of studies included in this meta-analysis was k ¼ 19. Article coding The first and third authors coded the articles on numerous features. Coded features included participant characteristics such as age, sex, and the country where the intervention took place; intervention characteristics such as recruitment channel, target behavior, use of theory, message type (i.e., targeted, tailored), types

Effect sizes were weighted by their inverse variance and combined using fixed effects meta-analytic procedures (Lipsey & Wilson, 2001). The Q statistic and I2 were used to examine whether significant heterogeneity existed among the effect sizes. Effect sizes for hypothesized categorical moderators were calculated along with their 95% confidence intervals, and those effect sizes were statistically compared using the Qb statistic. In addition, in the case of continuous (i.e., interval level) moderator variables, correlations were calculated between particular moderator variables and the effect size. All analyses were conducted using

44

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

Comprehensive Meta-Analysis software, Version 2.2.046, and SPSS Version 19.

Results Characteristics of the individual studies are reported in Table 2 in the Appendix. The k ¼ 19 studies took place in 13 countries (Australia, Canada, Finland, France, Germany, Israel, Malaysia, New Zealand, Norway, Scotland, Thailand, United Kingdom, United States) and had a cumulative N ¼ 5958 (median N per study ¼ 174, ranging from n ¼ 58 to n ¼ 1705). Studies were published between 2005 and 2011, with a median publication year of 2009. With the exception of one study directed toward children (Shapiro et al., 2008), interventions were directed at young adult or adult populations (mean age across all studies ¼ 29.17 years). Study samples tended to include more females than males (mean proportion of females across study samples was 64%), and study participants tended to be recruited using mediated methods such as newspaper advertisements or email (63%; k ¼ 12), although a significant minority of studies used in-person recruitment methods (32%; k ¼ 6). Interventions attempted to improve the following behaviors and outcomes: smoking cessation (26%; k ¼ 5), physical activity (16%; k ¼ 3), weight loss (16%; k ¼ 3), medication for prevention (11%; k ¼ 2), primary care appointment attendance (11%; k ¼ 2), healthy pregnancy outcome (5%; k ¼ 1), safer sex/condom use and sexually transmitted infection testing (5%; k ¼ 1), and contraceptive use (5%; k ¼ 1). Also, one study intervened on three behaviors associated with childhood obesity: physical activity, sugar sweetened beverages, and television screen time. Forty-seven percent (k ¼ 9) of the studies reported text-only interventions; the remainder combined text messaging with other (single or multiple) modalities (53%; k ¼ 10). The most commonly used additional modalities were websites (50% of textplus studies; k ¼ 5), print materials (40%; k ¼ 4), and human counselors (40%; k ¼ 4). All text messaging was automated; 58% (k ¼ 11) of interventions texted participants but participants could not text back, whereas 42% (k ¼ 8) required participants to text back. Some interventions used a “fixed” message frequency; programs ranged from those that sent only one or two texts (11%; k ¼ 2) to those that sent texts once a day or more (21%; k ¼ 4). The frequency of texting in the other programs varied according to program specifications (i.e., decreasing over time, 21%, k ¼ 4; varying throughout program, 11%, k ¼ 2) or was user programmed (i.e., individualized, 16%, k ¼ 3). A majority of programs (84%; k ¼ 16) applied message tailoring in some fashion, using assessments of an individual participant’s characteristics to deliver text messages customized to that participant. Forty-seven percent (k ¼ 9) of programs used tailored messages only, 37% (k ¼ 7) used both tailored and targeted messages, and 16% (k ¼ 3) used targeted messages only (messages designed for a particular target audience but not a particular individual). In terms of tailoring, some programs specified using demographic (21%; k ¼ 4; e.g., health status, age, weight) or psychosocial (37%; k ¼ 7; e.g., self-efficacy, motivation) variables in tailoring. Finally, 21% (k ¼ 4) of studies applied a personalization strategy such as using a person’s name in the text message. Just over half of the interventions (53%; k ¼ 10) were theorybased interventions, with an explicit theory informing their development. Among those studies, the most commonly used theories were social cognitive theory (40% of theory-based studies; k ¼ 4), the transtheoretical model (20%; k ¼ 2), and implementation intentions (20%; k ¼ 2). One of the studies applying implementation intentions did so in combination with protection motivation theory. Other studies applied self-efficacy in combination with systems

contingency (10%; k ¼ 1) or the theory of planned behavior (10%; k ¼ 1). Efficacy of interventions The weighted mean effect size representing the impact of text messaging-based interventions on health behavior was d ¼ .329 (95% CI ¼ .274, .385; p < .001; N ¼ 5137). This effect size indicates that text-messaging interventions had statistically significant effects on health behavior and health-related outcomes (see Fig. 1). In order to examine the possibility of publication bias, fail-safe N values were calculated and the trim and fill procedure was applied (Lipsey & Wilson, 2001). Orwin’s method (Lipsey & Wilson, 2001) to calculate fail-safe N indicated that 107 studies with non-significant findings would need to exist to reduce the d ¼ .329 effect to a trivial effect size of d ¼ .05. Funnel plots of these effects were symmetrical, and the trim and fill analysis suggested no adjustment to these mean effect sizes (Duval & Tweedie, 2000). In sum, there appeared to be no evidence of publication bias in this literature. Heterogeneity and intervention moderators Next, we examined heterogeneity of these effect sizes. Statistical testing indicated that the weighted mean effect size was statistically heterogeneous, (Q18 ¼ 55.60, p < .001, I2 ¼ 67.62). Thus, we explored the potential impact of moderator variables on intervention efficacy. Participant and descriptive moderators were examined first. Correlations between effect size and gender (proportion female; r(18) ¼ .09, p ¼ .70) and age (r(18) ¼ .16, p ¼ .52) were not statistically significant, indicating no significant impact of these features on intervention outcomes. Differences in behaviors intervened upon were examined next (see Table 1). Results revealed

Study name

Std diff in means and 95% CI

Brendryen & Kraft (2008) Cocosila et al. (2009) Fairhurst et al. (2008) Fjeldsoe et al. (2010) Free et al. (2009) Gold et al (2011) Haapala et al. (2009) Huag et al. (2009) Jareethum et al. (2008) Leong et al. (2006) Lombard et al. (2010) Ollivier et al. (2009) Patrick et al. (2009) Prestwich et al. (2009) Prestwich et al. (2010) Rodgers et al. (2005) Shapiro et al. (2008) Tsur et al. (2008) Whittaker et al. (2011)

-1.00

-0.50 Favors Control

0.00

0.50

1.00

Favors Text

Fig. 1. Forest plot of effect sizes and 95% confidence intervals representing the impact of text messaging-based interventions on health-related outcomes.

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

45

Table 1 Weighted mean effect sizes by categorical moderating variables. Variable Behaviors Smoking cessation Physical activity Weight loss Medication Primary care appointment Other (Healthy pregnancy, contraceptive use, safer sex multiple obesity-related behaviors) Intervention type Text only Text plus Use of theory No Yes Message type Targeted only Tailored only Targeted and tailored Demographic tailoring No Yes Psychosocial tailoring No Yes Use of personalization No Yes Communication direction One-way Two-way Initiation Automatic texts Only in response to text Message frequency Low e fixed High e fixed Decreasing Individualized Varied Comparison group No treatment control Alternative intervention a b

ka

d

95% CI

p

QB

5 3 3 2 2 4

.447 .509 .255 .160 .242 .071

[.367, .526] [.236, .781] [.056, .455] [.03, .350] [.122, .362] [.104, .246]

***b *** ** NS *** NS

***

9 10

.335 .313

[.271, .400] [.205, .421]

*** ***

NS

9 10

.278 .373

[.196, .360] [.298, .447]

*** ***

NS

3 9 7

.073 .274 .442

[.071, .216] [.180, .368] [.364, .519]

NS *** ***

***

15 4

.216 .531

[.147, .285] [.439, .623]

*** ***

***

12 7

.269 .403

[.194, .343] [.321, .485]

*** ***

*

15 4

.237 .422

[.159, .315] [.344, .500]

*** ***

***

11 8

.342 .265

[.282, .403] [.130, .400]

*** ***

NS

17 2

.325 .485

[.269, .381] [.159, .812]

*** ***

NS

5 5 4 3 2

.179 .129 .523 .425 .213

[.084, .275] [.023, .282] [.438, .609] [.143, .708] [.007, .419]

*** NS *** **

11 8

.369 .226

[.304, .434] [.120, .331]

*** ***

***

*

k ¼ number of studies, d ¼ weighted mean effect size, CI ¼ Confidence Interval, NS ¼ non-significant. *p < .05; **p < .01; ***p < .001.

statistically significant differences among the behaviors intervened upon, QB ¼ 24.01, df ¼ 5, p ¼ .001. Interventions were most successful when intervening on smoking cessation and physical activity, and, to a lesser extent, weight loss and primary care appointments. Interventions targeting preventive medications and the other behaviors (as a group) were not successful, although few studies existed in these areas. Intervention moderators were examined next (see Table 1). First, we compared text-only (k ¼ 9) interventions to textmessaging interventions that also included other components (k ¼ 10). We found no significant differences between these two intervention types, QB ¼ .12, df ¼ 1, p ¼ .73. We similarly found no significant difference among text plus Web (k ¼ 5, p ¼ .45), text plus print materials (k ¼ 4, p ¼ .85), and text plus human counseling (k ¼ 4, p ¼ .24) interventions when comparing them to the other interventions. These results indicate that interventions that included components other than text messaging were not significantly more efficacious than interventions that used textmessaging alone. Use of theory also did not significantly affect the efficacy of interventions. Although theory-based interventions (d ¼ .373) had a larger effect size than those that were not theory based (d ¼ .278), this difference was not statistically significant, QB ¼ 2.81, df ¼ 1, p ¼ .09.

We also examined several message-oriented components of the interventions. We found a significant difference among interventions that used different message types, QB ¼ 21.66, df ¼ 2, p ¼ .001. The largest effect sizes exhibited were for those studies that employed both message tailoring and targeting (d ¼ .442), followed by studies employing message tailoring only (d ¼ .274). The smallest effects were observed among studies that employed only message targeting (d ¼ .073). We also examined the impact of tailoring text messages specifically on demographic and psychosocial factors in message design. Interventions were significantly more efficacious when they used demographic (QB ¼ 28.63, df ¼ 1, p ¼ .001) or psychosocial (QB ¼ 5.64, df ¼ 1, p ¼ .02) factors in tailoring than when they did not. Moreover, interventions that applied personalization were more efficacious than those that did not, QB ¼ 10.75, df ¼ 1, p ¼ .001. Next, we examined communication direction, initiation, and frequency. Interventions that employed two-way communication (intervention texts participants and participant texts back) were not significantly more efficacious than those employing one-way communication (intervention texts participants). In addition, no difference was found with regard to whether the intervention initiated a text automatically or only in response to when the participant texted the intervention (QB ¼ .91, df ¼ 1, p ¼ .34). Significant findings were observed for message frequency, however

46

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

(QB ¼ 37.68, df ¼ 4, p ¼ .001). Whereas interventions that applied low fixed (less than once a week; d ¼ .179), high-fixed (once a week or more; d ¼ .129), or varying frequencies (d ¼ .213) had lower efficacy, those interventions whose message frequency decreased over time (d ¼ .523) or was individualized by and for the user (d ¼ .425) exhibited the highest efficacy. Finally, methodological moderators were examined. The mean follow-up time period in the study was 81.26 days and mean retention at follow-up was 86.16%. Correlations between effect size and follow-up (r(18) ¼ .12, p ¼ .62) and effect size and retention (r(18) ¼ .14, p ¼ .56) were not statistically significant. Effect sizes for interventions that employed no-treatment control groups (d ¼ .369), however, were significantly larger than those that employed alternative comparisons (d ¼ .226), QB ¼ .5.16, df ¼ 1, p ¼ .02. Discussion The purpose of this meta-analysis was to examine the efficacy of text messaging-based interventions to improve health behaviors and health-related outcomes. We also sought to examine potentially important moderators of text messaging-based interventions to begin to advance our understanding of what may make such interventions efficacious. It is interesting to observe that 13 countries were represented across the studies in this meta-analysis. Such geographic and cultural diversity is highly commendable and bodes well for the potential to systematically include cultural constructs in theory and intervention development. It is also important to note that this diversity illustrates the ease with which text messaging can be used in a variety of locations around the world, further supporting the idea that this technology has a ubiquitous presence that can benefit not only health practitioners designing health promotion campaigns but also those around the world who may be left out of traditional media campaigns (i.e., television, internet). Our first research question asked whether text messaging-based health promotion interventions are efficacious. Results revealed that these interventions are generally efficacious, with a mean effect size of d ¼ .329. This represents an effect size of small to medium magnitude, using the benchmarks established by Cohen (1988). Further, this effect size compares favorably to behavioral effects reported in other meta-analyses of health promotion interventions, such as print-based tailored health behavior change interventions (Noar, Benac, & Harris, 2007; d ¼ .15), computer-delivered health interventions (Portnoy, Scott-Sheldon, Johnson, & Carey, 2008; range of d ¼ .05e.35), computer-based interventions for HIV prevention (Noar, Black, & Pierce, 2009; d ¼ .26), and message framing health interventions (Gallagher & Updegraff, 2012; d ¼ .17). Given the global reach of mobile phones and text messaging across a wide variety of audiences, the potential of text-messaging health promotion interventions to have significant impact at the population level is strong. It is important to note that this meta-analysis included only RCTs, and thus this effect size represents studies with high levels of control. Additionally, the fail-safe N publication bias analysis revealed that a large number of studies with non-significant results (k ¼ 107) would be needed to invalidate the mean effect size found in our study. Although there are likely to be text-messaging programs that remain unevaluated or for which evaluation data may not be in the public domain (Lim, Hocking, Hellard, & Aitken, 2008), it seems unlikely that enough null studies would exist to reduce the current effect size to one of trivial magnitude. In sum, our results suggest that text messaging-based health promotion interventions show considerable promise for improving a range of health behaviors and health outcomes. The second research question dealt with potential moderators of efficacious text messaging-based interventions. We coded the

studies on a variety of participant, intervention, and methodological characteristics, as informed by previous reviews on text messaging-based health promotion interventions. While no participant or methodological characteristics were significant moderators of the observed effects, several intervention characteristics proved to be. These intervention moderators, although significant in many cases, should be interpreted with caution, as in several cases the number of studies in particular cells is small. Still, several intriguing findings resulted from our analyses. First, interventions that used both targeted and tailored messages exhibited the largest effect size, followed by tailoring only, and finally by those that used targeting only. Given the wealth of literature indicating the effectiveness of tailored interventions (see Noar & Harrington, 2012), this finding is not terribly surprising. However, the combination of both tailored messages and targeted messages in the same intervention is something that has received scant attention in previous research. For example, a smoking intervention could incorporate targeted messages that appeal to all smokers (e.g., what to do when experiencing cravings) and tailored messages specified to a participant’s unique situation (e.g., messages sent at times when a participant tends to experience cravings). Text messaging represents a technology that allows us to design and implement interventions that can take advantage of both of these messaging strategies in the same intervention. Additionally, we found that interventions tailored on demographics and psychosocial variables, respectively, were significantly more efficacious than those that did not employ these tailoring techniques. These findings are fairly consistent with previous research on print tailored interventions that demonstrates that tailoring on certain demographics and psychosocial variables increases intervention efficacy (Noar et al., 2007). Moreover, personalization strategiesdsuch as using a participant’s name or the name of a significant other (e.g., a child)dwere found to increase intervention efficacy. As the literature in tailoring has advanced, substrategies within tailoring have been better defined and tested in the literature (Hawkins, Kreuter, Resnicow, Fishbein, & Dijkstra, 2008), and the current study adds to the growing evidence that personalization may be a key ingredient contributing to the efficacy of tailoring (Noar, Harrington, & Aldrich, 2009). Third, and fairly surprisingly, we found no difference in the effectiveness of interventions that were text-only versus those that included additional modalities in their design (e.g., websites, print materials, human counselors). In the realm of translational research, in which we are charged with disseminating and implementing research findings to make a positive influence on health behavior in practice, this should come as good news. Although there is great need for cost-effectiveness research in this area, it seems self-evident that a text-only intervention would be much more affordable (and far less time consuming) than an intervention requiring additional components in whatever form, especially those involving human counselors. Thus, while we cannot conclude that text messaging only will always be the most effective method in a particular health promotion context, these findings are intriguing and suggest that text messaging by itself may often be an effective channel for health promotion interventions. Fourth, also contrary to expectations, interventions that used theory were not significantly more efficacious than interventions that did not apply theory. Although the effect size was larger for theory-based interventions, the difference was not statistically significant. These findings are inconsistent with suggestions from health promotion researchers to use theory and from other metaanalyses that have shown greater effects for theory-based interventions (e.g., Noar, 2008; Noar et al., 2007). One explanation for this discrepancy may be that some interventions that did not identify an explicit theory, and thus were coded as not theory-

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

based, actually did use theory-based strategies to inform intervention design. For example, the report of Free et al.’s (2009) smoking cessation text-messaging intervention did not reference any specific theory (and thus was coded as not theory based), yet the intervention applied several techniques that it described as derived from previous systematic reviews, including setting a quit date, identifying reasons to quit, and using social support; effectively, then, one could argue that the study was more theory based than not. As theory-based techniques continue to move into the mainstream, interventions may in some cases increasingly use such techniques while not describing them as coming from theory. Future meta-analyses might consider coding protocols that go beyond theory/no theory dichotomies to examine the impact of particular theoretical concepts and behavior change techniques on intervention efficacy (Abraham & Michie, 2008). This will be increasingly possible when larger numbers of studies evaluating text-messaging interventions are published. Fifth, we found evidence that message frequency moderated intervention effectiveness. For example, compared to interventions that used a fixed frequency (e.g., once per month), interventions that texted on particular variable schedules (i.e., interventions that allowed participants to set their own schedule and interventions that used decreasing frequency over the course of the study) were most effective. Interventions that were individualized allowed users to customize their messaging schedule, which presumably led to messages being delivered at times that were most relevant to the behavior. While traditional message tailoring focuses on tailoring the content of the message itself (Noar et al., 2007), text messaging creates opportunities to tailor not only the content but also the timing and frequency of messages. This strategy is particularly appropriate given the constant presence of cell phones in some users’ lives. Such presence maximizes the opportunity to reach participants whenever they need intervention messages the most and therefore may result in messages that are better received by individuals, perhaps because the timing and frequency leads to enhanced message processing (Hawkins et al., 2008). Decreasing frequency interventions may have similarly been more sensitive to individuals’ messaging needs, particularly those individuals who had begun to change their behavior. In contrast, fixed frequency programs delivered the same message schedule week after week, regardless of the status of the person’s health behavior. Those programs may have done little to communicate social presence (the idea that an interactive program is responsive to the individual), which has been hypothesized to affect the efficacy of eHealth interventions (Hawkins et al., 2010). Limitations and future research We decided to include only RCTs in the meta-analysis. While this decision almost certainly increased the internal validity of our findings, it may have decreased the external validity of the meta-analysis. Also, while analyses of moderators can be revealing, they can also potentially be misleading given that such characteristics (e.g., use of personalization) are not randomly dispersed across studies. In addition, several of our moderator analyses had small groups of studies in some cells and thus should be interpreted with caution. In line with conceptual distinctions raised in the literature, we focused our meta-analysis on health promotion interventions and excluded chronic disease management interventions. Chronic disease interventions include those designed to improve diabetes management (e.g., Faridi et al., 2008), asthma control (e.g., Strandbygaard, Thomsen, & Backer, 2010), and adherence to HIVAIDS management therapies (e.g., Pop-Eleches et al., 2011). A future meta-analysis could address text messaging for chronic disease management. In so doing, researchers should consider conceptual factors that may differentiate behaviors designed to

47

promote health and prevent disease among healthy populations and those designed to manage an existing chronic condition. Researchers should also work to develop and extend theories of behavior change that take such differences into account. For example, we may want to consider the extent to which a behavior has to be enacted (e.g., exercise) versus avoided (e.g., alcohol consumption) or the extent to which engaging in or avoiding a behavior has immediate discernible results (e.g., not using sunscreen, which can result in immediate sunburn) versus results that are perceptible only over the longer term (e.g., skipping the treadmill one day usually does not result in weight gain but over time certainly can). An important question underlying these distinctions is how text message-based interventions should be designed to address these and other pertinent factors. A related question is how features of text messaging can be integrated with current health behavior change theories. For example, with proper programming, interventions can deliver persuasive health messages at critical decision points (e.g., at the end of the work day when the decision is whether or not to go to the gym). In theory of planned behavior terms, delivering such time sensitive messages means that the interval between the behavioral intention and the decision to perform the behavior is greatly reduced, which may increase the likelihood of subsequently performing the behavior. Exploring how characteristics unique to text messaging align with elements of health behavior change theory should hold great interest for researchers. Researchers also should be interested in going beyond behavioral theory to theory that informs message effects. For example, how do text message characteristics, such as limits on message length and format, align with theories of message design and message effects? Finally, the eHealth and mHealth fields are growing at an extremely rapid pace (see Noar & Harrington, 2012). Meta-analyses such as the current project will need to be updated frequently to reflect the growth of this field, as new RCTs are frequently being published (e.g., Kim & Glanz, 2013). Of course, mobile phone and text messaging technology is also bound to advance in the coming years, leading to new intervention modalities. Such advancement is a truism of technology these days. Thus, we must all be prepared to take advantage of such change if our interventions are to have a substantial and sustainable impact on public health. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.socscimed.2013.08.003. References1 Abraham, C., & Michie, S. (2008). A taxonomy of behavior change techniques used in interventions. Health Psychology, 27(3), 379e387. Abroms, L. C., Padmanabhan, N., & Evans, W. D. (2012). Mobile phones for health communication to promote behavior change. In S. M. Noar, & N. G. Harrington (Eds.), eHealth applications: Promising strategies for behavior change (pp. 147e 166). New York: Routledge. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates. Cole-Lewis, H., & Kershaw, T. (2010). Text messaging as a tool for behavior change in disease prevention and management. Epidemiologic Reviews, 32(1), 56e69. Duval, S., & Tweedie, R. (2000). Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics, 56(2), 455e463. Faridi, Z., Liberti, L., Shuval, K., Northrup, V., Ali, A., & Katz, D. L. (2008). Evaluating the impact of mobile telephone technology on type 2 diabetic patients’ selfmanagement: the niche pilot study. Journal of Evaluation in Clinical Practice, 14(3), 465e469.

1 References marked with an asterisk indicate studies included in the metaanalysis.

48

K.J. Head et al. / Social Science & Medicine 97 (2013) 41e48

Fjeldsoe, B. S., Marshall, A. L., & Miller, Y. D. (2009). Behavior change interventions delivered by mobile telephone short-message service. American Journal of Preventive Medicine, 36(2), 165e173. Fjeldsoe, B. S., Miller, Y. D., & Marshall, A. L. (2012). Text messaging interventions for chronic disease management and health promotion. In S. M. Noar, & N. G. Harrington (Eds.), eHealth applications: Promising strategies for behavior change (pp. 167e186). New York: Routledge. *Free, C., Whittaker, R., Knight, R., Abramsky, T., Rodgers, A., & Roberts, I. G. (2009). Txt2stop: a pilot randomised controlled trial of mobile phone-based smoking cessation support. Tobacco Control, 18(2), 88e91. Gallagher, K., & Updegraff, J. (2012). Health message framing effects on attitudes, intentions, and behavior: a meta-analytic review. Annals of Behavioral Medicine, 43(1), 101e116. Grinter, R., & Eldridge, M. (2003, April). Wan2tlk?: Everyday text messaging. Paper presented at the proceedings of the SIGCHI conference on human factors in computing systems, Ft. Lauderdale, Florida, USA. Hawkins, R. P., Han, J. Y., Pingree, S., Shaw, B. R., Baker, T. B., & Roberts, L. J. (2010). Interactivity and presence of three eHealth interventions. Computers in Human Behavior, 26(5), 1081e1088. Hawkins, R. P., Kreuter, M., Resnicow, K., Fishbein, M., & Dijkstra, A. (2008). Understanding tailoring in communicating about health. Health Education Research, 23(3), 454e466. Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando: Academic Press. International Telecommunication Union. (2010). Measuring the information society 2010. Retrieved 14.08.11 from http://www.itu.int/newsroom/press_releases/ 2010/pdf/PR08_ExecSum.pdf. Katz, J. E. (2007). Mobile media and communication: some important questions. Communication Monographs, 74(3), 389e394. Kim, B. H., & Glanz, K. (2013). Text messaging to motivate walking in older African Americans: a randomized controlled trial. American Journal of Preventive Medicine, 44(1), 71e75. Kohut, A., Wike, R., Horowitz, J. M., Simmons, K., Poushter, J., Barker, C., et al. (2011). Global digital communication: Texting, social networking popular worldwide. Pew Global Attitudes Project. Retrieved 23.03.12 from http://www.pewglobal.org/files/ 2011/12/Pew-Global-Attitudes-Technology-Report-FINAL-December-20-20111.pdf. Lenhart, A. (2010). Cell phones and American adults. Pew internet & American life project. Retrieved 23.03.12 from http://pewinternet.org/Reports/2010/CellPhones-and-American-Adults.aspx. Lim, M. S. C., Hocking, J. S., Hellard, M. E., & Aitken, C. K. (2008). SMS STI: a review of the uses of mobile phone text messaging in sexual health. International Journal of STD & AIDS, 19(5), 287e290. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage. Neville, R., Greene, A., McLeod, J., Tracey, A., & Surie, J. (2002). Mobile phone text messaging can help young people manage asthma. BMJ, 325, 600. Noar, S. M. (2008). Behavioral interventions to reduce HIV-related sexual risk behavior: review and synthesis of meta-analytic evidence. AIDS and Behavior, 12(3), 335e353. Noar, S. M., Benac, C. N., & Harris, M. S. (2007). Does tailoring matter? Meta-analytic review of tailored print health behavior change interventions. Psychological Bulletin, 133(4), 673e693. Noar, S. M., Black, H. G., & Pierce, L. B. (2009). Efficacy of computer technology-based HIV prevention interventions: a meta-analysis. AIDS, 23(1), 107e115. Noar, S. M., & Harrington, N. G. (Eds.). (2012). eHealth applications: Promising strategies for behavior change. New York: Routledge. Noar, S. M., Harrington, N. G., & Aldrich, R. S. (2009). The role of message tailoring in the development of persuasive health communication messages. Communication Yearbook, 33, 73e133. Pop-Eleches, C., Thirumurthy, H., Habyarimana, J. P., Zivin, J. G., Goldstein, M. P., de Walque, D., et al. (2011). Mobile phone technologies improve adherence to antiretroviral treatment in a resource-limited setting: a randomized controlled trial of text message reminders. AIDS, 25(6), 825e834. Portio Research. (2012). Mobile messaging futures 2012e2016dAnalysis and growth forecasts for mobile messaging markets worldwide (6th ed.). Chippenham, Wilts, UK: Portio Research Limited. Portnoy, D. B., Scott-Sheldon, L. A. J., Johnson, B. T., & Carey, M. P. (2008). Computerdelivered interventions for health promotion and behavioral risk reduction: a meta-analysis of 75 randomized controlled trials, 1988e2007. Preventive Medicine, 47(1), 3e16.

*Rodgers, A., Corbett, T., Bramley, D., Riddell, T., Wills, M., Lin, R., et al. (2005). Do u smoke after txt? Results of a randomised trial of smoking cessation using mobile phone text messaging. Tobacco Control, 14(4), 255e261. *Shapiro, J. R., Bauer, S., Hamer, R. M., Kordy, H., Ward, D., & Bulik, C. M. (2008). Use of text messaging for monitoring sugar-sweetened beverages, physical activity, and screen time in children: a pilot study. Journal of Nutrition Education & Behavior, 40(6), 385e391. Strandbygaard, U., Thomsen, S. F., & Backer, V. (2010). A daily SMS reminder increases adherence to asthma treatment: a three-month follow-up study. Respiratory Medicine, 104(2), 166e171. Stross, R. (2008, December 26). What carriers aren’t eager to tell you about texting. The New York Times. Retrieved 11.08.11 from http://www.nytimes.com/2008/ 12/28/business/28digi.html. Wortham, J. (2011, October 9). Free texts pose threats to carriers. The New York Times. Retrieved 12.03.12 from http://www.nytimes.com/2011/10/10/technol ogy/paying-to-text-is-becoming-passe-companies-fret.html?pagewanted¼all.

Further reading * Brendryen, H., & Kraft, P. (2008). Happy ending: a randomized controlled trial of a digital multi-media smoking cessation intervention. Addiction, 103(3), 478e484. * Cocosila, M., Archer, N., Haynes, R. B., & Yuan, Y. (2009). Can wireless text messaging improve adherence to preventive activities? Results of a randomised controlled trial. International Journal of Medical Informatics, 78(4), 230e238. * Fairhurst, K., & Sheikh, A. (2008). Texting appointment reminders to repeated non-attenders in primary care: randomised controlled study. Quality & Safety in Health Care, 17(5), 373e376. * Fjeldsoe, B. S., Miller, Y. D., & Marshall, A. L. (2010). Mobilemums: a randomized controlled trial of an SMS-based physical activity intervention. Annals of Behavioral Medicine, 39(2), 101e111. * Gold, J., Aitken, C. K., Dixon, H. G., Lim, M. S., Gouillou, M., Spelman, T., et al. (2011). A randomised controlled trial using mobile advertising to promote safer sex and sun safety to young people. Health Education Research, 26(5), 782e794. * Haapala, I., Barengo, N. C., Biggs, S., Surakka, L., & Manninen, P. (2009). Weight loss by mobile phone: a 1-year effectiveness study. Public Health Nutrition, 12, 2382e2391. * Haug, S., Meyer, C., Schorr, G., Bauer, S., & John, U. (2009). Continuous individual support of smoking cessation using text messaging: a pilot experimental study. Nicotine & Tobacco Research, 11(8), 915e923. * Jareethum, R., Titapant, V., Chantra, T., Sommai, V., Chuenwattana, P., & Jirawan, C. (2008). Satisfaction of healthy pregnant women receiving short message service via mobile phone for prenatal support: a randomized controlled trial. Journal of the Medical Association of Thailand, 91(4), 458e463. * Leong, K. C., Chen, W. S., Leong, K. W., Mastura, I., Mimi, O., Sheikh, M. A., et al. (2006). The use of text messaging to improve attendance in primary care: a randomized controlled trial. Family Practice, 23(6), 699e705. * Lombard, C., Deeks, A., Jolley, D., Ball, K., & Teede, H. (2010). A low intensity, community based lifestyle programme to prevent weight gain in women with young children: cluster randomised controlled trial. BMJ, 341. c3215. * Ollivier, L., Romand, O., Marimoutou, C., Michel, R., Pognant, C., Todesco, A., et al. (2009). Use of short message service (SMS) to improve malaria chemoprophylaxis compliance after returning from a malaria endemic area. Malaria Journal, 8, 236e244. * Patrick, K., Raab, F., Adams, M. A., Dillon, L., Zabinski, M., Rock, C. L., et al. (2009). A text message-based intervention for weight loss: randomized controlled trial. Journal of Medical Internet Research, 11. e1. * Prestwich, A., Perugini, M., & Hurling, R. (2009). Can the effects of implementation intentions on exercise be enhanced using text messages?. Psychology & Health, 24(6), 677e687. * Prestwich, A., Perugini, M., & Hurling, R. (2010). Can implementation intentions and text messages promote brisk walking? A randomized trial. Health Psychology, 29(1), 40e49. * Tsur, L., Kozer, E., & Berkovitch, M. (2008). The effect of drug consultation center guidance on contraceptive use among women using isotretinoin: a randomized, controlled study. Journal of Women’s Health, 17(4), 579e584. * Whittaker, R., Dorey, E., Bramley, D., Bullen, C., Denny, S., Elley, C. R., et al. (2011). A theory-based video messaging mobile phone intervention for smoking cessation: randomized controlled trial. Journal of Medical Internet Research, 13. e10.