Abstract
Objective
This systematic review aimed to assess the effectiveness, feasibility and acceptability of mobile health (mHealth) technology (including wearable activity monitors and smartphone applications) for promoting physical activity (PA) and reducing sedentary behaviour (SB) in workplace settings.
Methods
Systematic searches were conducted in seven electronic databases (MEDLINE, SPORTDiscus, Scopus, EMBASE, PsycINFO, Web of Science and the Cochrane library). Studies were included if mHealth was a major intervention component, PA/SB was a primary outcome, and participants were recruited and/or the intervention was delivered in the workplace. Study quality was assessed using the Effective Public Health Practice Project (EPHPP) tool. Interventions were coded for behaviour change techniques (BCTs) using the Coventry, Aberdeen and London – Refined (CALO-RE) taxonomy.
Results
Twenty-five experimental and quasi-experimental studies were included. Studies were highly heterogeneous and only one was rated as ‘strong’ methodological quality. Common BCTs included self-monitoring, feedback, goal-setting and social comparison. A total of 14/25 (56%) studies reported a significant increase in PA, and 4/10 (40%) reported a significant reduction in sedentary time; 11/16 (69%) studies reported a significant impact on secondary outcomes including reductions in weight, systolic blood pressure and total cholesterol. While overall acceptability was high, a large decline in technology use and engagement was observed over time.
Conclusions
While methodological quality was generally weak, there is reasonable evidence for mHealth in a workplace context as a feasible, acceptable and effective tool to promote PA. The impact in the longer term and on SB is less clear. Higher quality, mixed methods studies are needed to explore the reasons for decline in engagement with time and the longer-term potential of mHealth in workplace interventions.
Keywords
Introduction
Physical inactivity is considered one of the biggest public health problems of the 21st century. 1 Failing to meet the recommended guidelines is associated with an increased risk of morbidity due to cardiovascular disease, cancer and metabolic syndrome and general mortality.2–6 There is now also substantial evidence that sedentary behaviour (SB) is an independent predictor of poor health and mortality.7–9
Interventions to increase physical activity (PA) levels and reduce SB are clearly vital. The workplace is viewed as an important setting for health promotion and disease prevention. 10 Around half of weekday sitting time is work-related,11,12 and up to 71% of working hours in office workers are spent sedentary. 13 Occupational sedentary time is predicted to further increase in future with rises in automation and information technology use. 14 Promotion of PA in the workplace has many potential benefits, including improved health and wellbeing of employees and economic benefits for employers. 15
Mobile health (mHealth) technology has rapidly gained popularity in the general population. mHealth technology includes wearable PA monitors or trackers and smartphone applications (apps) designed to help people to manage their own health and wellbeing. The potential value of mHealth in health promotion lies in its widespread appeal, accessibility and ability to reach large populations at a low cost. 16 It also offers the potential for tailoring of interventions to the needs of individuals or specific groups.
Studies have investigated the use of mHealth to promote PA in a range of settings, including the workplace. 16 Whilst the results of clinical and general population studies suggest that mHealth may be a feasible and cost-effective way to promote PA, 17 the findings of existing reviews have been inconclusive. Some reviews have reported nonsignificant effects of mobile technology on PA levels, 18 and where beneficial effects are reported, effect sizes have generally been small,17,19–21 Additional limitations of previous reviews are the inclusion of studies where mHealth devices were used as a data collection tool rather than as an intervention in their own right,20,22 and a lack of a comprehensive description of interventions and study procedures. 19 Furthermore, with two recent exceptions,17,23 few reviews of mHealth interventions have assessed both PA and SB outcomes.
Identification of behaviour change techniques (BCTs) using standardised taxonomies is important for recognition of effective and acceptable components, to allow replication and comparison of interventions, and to facilitate further development and testing of theories. 24 There is also evidence that including established BCTs is associated with greater intervention effectiveness. 25 Despite this, previous reviews have concluded that many mHealth interventions lack an explicit theoretical basis,19,20 and it remains unclear which components are most effective and accepted. 16 Identification or coding of included BCTs, and identifying the theoretical basis of existing studies are therefore important gaps to address.
As mHealth is such a rapidly progressing field due to advances in technology, studies have increased exponentially in a short space of time. Early reviews predominantly comprised studies of text messaging (SMS) interventions, but the emergence of new technologies (e.g. tablets, commercial wearable activity monitors, and ‘exergaming’) means the evidence should be frequently reviewed in order to accurately reflect the current status. Furthermore, the use and effectiveness of mHealth interventions in specific population groups remains unclear. 23 It is important to consider setting or context in the evaluation of mHealth interventions as due to their complex nature, various components may produce different outcomes for different individuals in different settings. 26 Workplace mHealth interventions may differ from general interventions in terms of both intervention content and timing of effectiveness. 27 To the authors’ knowledge, there has been no previous systematic review of mHealth technology for promoting PA and reducing SB in workplace settings. A recent review of general digital health interventions in the workplace concluded that the evaluation of smartphone apps in this context is an important ‘next step’ for future research. 28
Employee populations potentially have much to gain from mHealth interventions for PA and SB, yet little is known about the impact of this technology in a workplace context. Feasibility and acceptability are important aspects to consider but remain understudied and underreported. 26 This review therefore aims to provide a comprehensive synthesis of current evidence in relation to the effectiveness, feasibility and acceptability of mHealth interventions in the promotion of PA and reduction of SB in the workplace. This includes a description of intervention content in terms of common BCTs using an established behaviour change taxonomy, and a consideration of subgroup differences and the wider impact of interventions on health and related outcomes.
Methods
Protocol and registration
The review was conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines (Additional file 1). 29 The protocol was registered with the University of York Centre for Reviews and Dissemination PROSPERO database (CRD42017058856).
Data sources and search strategy
Searches were conducted in the following databases: MEDLINE, SPORTDiscus, Scopus, EMBASE, PsycINFO, Web of Science and the Cochrane library (including the Cochrane Database of Systematic Reviews (CDSR), Cochrane Central Register of Controlled Trials (CENTRAL), Database of Abstracts of Reviews and Effect (DARE) and Health Technology Assessment (HTA)). Studies with a publication date between January 2007 (around the time smartphones were introduced) and February 2017 were included, with no language restriction. Full updated searches were later conducted to include studies to the end of February 2018, then to the end of December 2018. A master search strategy was developed (Additional file 2) and revised for each database (see Additional file 3 for example search strategy for MEDLINE). Both free text searching and controlled vocabulary were used, including key terms such as ‘mHealth’, ‘smartphone’, ‘application or app’, ‘activity monitor or tracker’, ‘physical activity’, ‘sedentary’, ‘workplace’ and ‘occupation’.
In addition, relevant studies were identified via forward and backward citation searching, including reference lists of included articles and published systematic reviews. A search of grey literature, using the same key terms and for the same time period, included dissertations and theses (ProQuest Dissertations and Theses Global), ‘mHealth Evidence’, and the ‘Fitabase’ research library (studies using the Fitbit® activity tracker).
Inclusion criteria and study selection
Both experimental (e.g. randomised controlled trials, RCTs) and quasi-experimental (e.g. pre-post uncontrolled trials) study designs were included. Studies were included if they: 1) used mHealth (including mobile phone, smartphone apps, personal digital assistants (PDAs), tablets, wearable activity monitors/trackers) as a major component of the intervention, as stated by the authors or apparent from the context of the paper, 2) included a control or comparison group (experimental studies) or pre- and post- exposure data (quasi-experimental and observational studies), 3) recruited participants in the workplace and/or the intervention was delivered in the workplace, and 4) included any measure of PA and/or SB (self-reported or objective) as a primary quantitative outcome.
Pilot and feasibility trials were included if they met the inclusion criteria. Interventions could be either standalone mHealth or multi-component (e.g. facilitated with telephone counselling). The rationale for including multi-component interventions was that many digital workplace interventions for PA and SB, as delivered in the real world, are part of multi-component health promotion programmes, 30 and we wanted to maximise the number of studies for inclusion and scope of the review. Interventions could be designed as an exclusive workplace or a wider lifestyle intervention (i.e. where the intervention was initiated or delivered in the workplace but also included activity outside of working hours). Studies using smartphone apps for PA or SB alone or with other behaviours (e.g. diet, weight) were included.
Exclusion criteria were studies reporting web-only interventions or traditional pedometers (i.e. not able to transmit data to a consumer interface), as these fall outside the realm of mHealth technology. Interventions involving basic text messaging (SMS) alone were excluded as these have been more extensively reviewed in the past, 19 and are felt to be a different type of intervention than more advanced mHealth tools such as multimedia smartphone apps and activity monitors. Studies using mobile devices for data collection only, and studies with clinical or student populations (i.e. school, college, university) were excluded. Studies reporting only qualitative data, non-human studies, review articles and editorials, and reports published only as conference abstracts or proceedings were also excluded.
All search results were imported into EndNote X7 bibliographic software (Thompson Reuters, San Francisco, CA, USA) and duplicates removed. Two independent reviewers (SAB and AJW) screened papers for eligibility by title and abstract followed by full text screening. Disagreements were resolved through discussion and consulting a third reviewer (KM).
Data extraction
Standardised data extraction forms were completed by one reviewer (SAB) and verified by a second reviewer (AJW). Any disagreements were resolved through discussion and consulting a third reviewer (KM). The following data were captured: author; year; country; setting or workplace; study design; participants (number and characteristics); intervention description (type of mHealth technology or tool, intervention components including whether standalone mHealth or multi-component, theoretical basis, key motivational strategies or BCTs, duration and frequency); control or comparator; study aim (i.e. increase PA and/or reduce SB); primary PA/SB outcome (including method of assessment); secondary outcomes; duration of follow-up; main study results including any relevant subgroup findings; details of acceptability, engagement and attrition. Key within- and between-group quantitative findings were summarised for each study; significant effects were P < 0.05.
Study quality assessment
Included studies were appraised using the Effective Public Health Practice Project (EPHPP) quality assessment tool for quantitative studies. 31 This tool was developed for health promotion interventions and was selected for its application to a wide range of study designs (e.g. RCTs, cohort trials and case-control studies). The tool has demonstrated content and construct validity and both intra- and inter-rater reliability.31,32
The EPHPP quality assessment tool assesses six domains: 1) selection bias; 2) study design; 3) confounders; 4) blinding; 5) data collection methods; and 6) withdrawals and dropouts. Each study was given a rating of ‘strong’, ‘moderate’ or ‘weak’ for each domain; based on this, a global rating was then assigned for each study – ‘strong’ (no weak ratings), ‘moderate’ (one weak rating) or ‘weak’ (two or more weak ratings). Intervention integrity (proportion of participants receiving the intended intervention), fidelity of delivery (whether studies measured consistency of intervention) and appropriateness of analysis methods were also separately considered.
Two independent raters (SAB and AJW) used the tool to assess risk of bias and study quality. KM was consulted to resolve any uncertainties.
Coding of BCTs
Interventions in included studies were coded for BCTs using definitions provided in the ‘Coventry, Aberdeen and London – Refined’ taxonomy for PA and healthy eating behaviours. 33 This 40-item evidence-based taxonomy was selected as it was specifically designed for PA and healthy eating behaviours, and is widely used including to characterise smartphone apps for PA and wearable activity monitors.34,35 Content was coded for each intervention as a whole (i.e. mHealth and any additional components) using information from relevant results and protocol papers. Coding was completed by two independent reviewers (SAB and AJW) who were trained in Michie et al.’s Behaviour Change Taxonomy v1, 36 and consensus was reached through discussion.
Results
Study selection
A flow diagram of the study selection process is shown in Figure 1. A total of 2820 publications were identified in the initial searches (2815 from databases and 5 from other sources). After removal of duplicates, 1897 publication titles and abstracts were screened. The full text was obtained for 71 publications; of these, 18 publications describing 15 studies met the criteria for inclusion.37–54 An updated search to February 2018 found an additional five publications describing four studies,55–59 and a second updated search to December 2018 found an additional seven publications for six studies,60–66 resulting in a total of 30 publications (25 studies) for inclusion in the review.
Flow diagram of study selection process.
Study and intervention characteristics
The characteristics of the 25 included studies are summarised in Table 1. Eleven studies were conducted in the USA,42–46,50,53,57,59,60,62 five in Australia,37,38,49,55,56,61,66 two in Canada,58,63 two in the Netherlands,52,54 one in Belgium,64,65 Singapore,39,40 Finland,47,48 Norway, 51 and one in multiple countries. 41 Workplace settings included academic and academic medical institutions,42–44,50,53,57,59,60,62 healthcare,63,66 health insurance,45,49 wellbeing improvement, 46 property and infrastructure,37,38 pension insurance,47,48 financial services, 61 road maintenance, 51 and haulage.55,56 Six studies targeted multiple organisations.39–41,52,54,58,64,65 Both public and private sector organisations were represented.
Characteristics of included studies.
M: male; F: female; IG: intervention group; CG: control/comparison group; FTE: full time equivalent; PA: physical activity; SB: sedentary behaviour; OB: objective; SR: self-reported; QoL: quality of life; RCT: randomised controlled trial; ± or SD: standard deviation; MVPA: moderate to vigorous physical activity; BP: blood pressure; BMI: body mass index; MET: metabolic equivalent; IPAQ: international physical activity questionnaire; QQ: Quantity and Quality questionnaire; IQR: Interquartile Range; HDL: high-density lipoprotein; LDL: low-density lipoprotein; AQuAA: Activity Questionnaire for Adolescents and Adults; CDC: Centers for Disease Control and Prevention
The number of participants ranged from 20 in a feasibility cohort study, 58 to over 69,000 in a large international cohort study. 41 The majority of studies targeted sedentary, office-based employees. Of the 25 studies, 16 had a markedly higher proportion of female (≥60%) than male participants.42–46,48,49,52,57–63,66
The most common study designs were individual RCTs (n = 10)39,40,43–48,52,54,60,62 and pre-post prospective cohort studies (n = 10).41,49–51,55–59,61,66 One study used a combination of these designs in two phases. 53 Other designs included cluster RCTs,37,38,65 a parallel group uncontrolled randomised trial, 63 and a prospective cluster trial with an asynchronous control group. 42 Study duration varied greatly, with length of follow-up ranging from 6 weeks to 12 months.
Assessing the effectiveness, feasibility and/or acceptability of mHealth technology for PA promotion was the primary aim of 16 studies.39,40,43–51,53,57–59,62,63,65 Six studies targeted both PA and SB in a single intervention41,42,52,55,56,60,66 and three studies aimed to reduce SB (but also included PA as an outcome measure).37,38,54,61 Although recruitment and/or delivery of the intervention took place in the workplace in all cases, 24 of the 25 studies used mHealth as a wider lifestyle intervention, including both workplace and non-workplace activity. Only one study, designed to reduce SB, was based exclusively in the workplace. 54
The main mHealth tools used were wearable activity monitors or trackers (n = 11),39,40,42,46–49,51–54,57,63 smartphone apps (n = 6),41,43–45,62,66 or a combination of the two (n = 8).37,38,50,55,56,58–61,65 Some studies had additional mHealth and technology intervention components, including motivational or persuasive text messaging43,46,54 or e-mails, 61 computer software or websites linked to the activity monitor,39,40,42,46–48,51–54,57–59,63 and dedicated social media groups. 66 Eleven studies assessed mHealth as a standalone intervention,41,46,49,50,52,54,59,60,63,65,66 whereas 14 studies used mHealth in the context of a multi-component workplace health or PA programme.37–40,42–45,47,48,51,53,55–58,61,62 Among the multi-component programmes in particular, interventions were diverse and additional components included educational materials on health and PA,37–40,43,47,48,55,56,58,61 managerial support,37,38 financial incentives or rewards,39,40,44,45,51,53,55–57,62 shared active workstations, 42 online or telephone counselling,47,48,58 personalised feedback on activity,55–57 wellness education delivered in the workplace,42,53,61 group-based action planning, 61 and access to personal training and nutritionists. 53 Further detail on intervention content is given in Table 2.
Summary of intervention components.
BCT: Behaviour Change Technique; SA: Standalone mHealth; MC: Multi-Component programme; Y: Yes, included in intervention; N: No, not included in intervention;?: unclear or difficult to determine whether included from available intervention description; PA: Physical Activity; SB: Sedentary Behaviour; MVPA: Moderate to Vigorous Physical Activity
Intervention duration ranged from 1 to 12 months. Frequency of delivery of the intervention components was variable but daily wear of activity monitors was encouraged in most studies. Fifteen studies reported that the intervention was based on a named behaviour change theory and/or principles of behavioural economics.39,40,42–46,54,57,58,60–66 A further two studies alluded to behaviour change techniques or theory in their discussion,49,50 and eight studies had no clear theoretical basis.37,38,41,47,48,51–53,55,56,59 The most frequently cited behaviour change theories were the Theory of Reasoned Action, 67 the Socio-Ecological Model, 68 Social Cognitive Theory and Self-Efficacy, 69 Self-Determination Theory, 70 other social influence theories such as self-presentation theory and Cialdini’s social influence strategies,71,72 and the Health Action Process Approach. 73
A control or comparator group was present in 14 of the 25 studies.37–40,42–48,52–54,60,62,65 Of these, six could not be classed as a ‘true’ control group as the participants received at least a partial mHealth intervention,43–45,54,60,62 and another three studies supplied controls with wearable activity monitors for data collection.42,46,53
Outcome measures of PA and SB were heterogeneous. The most frequently used outcome measures for PA included daily step count, daily or weekly minutes or metabolic equivalent (MET) minutes of total activity or moderate to vigorous PA (MVPA). Other outcomes included exercise frequency and proportion of participants meeting step or PA goals. Studies that assessed SB commonly reported daily or weekly sedentary time, although the one study using an exclusive workplace intervention used computer activity as a proxy for SB. 54 Objective PA/SB outcomes were used in 20 studies,37–40,42–46,49,53–63,65,66 whilst four studies relied on self-report for the primary measure of PA or SB.47,48,50–52 Ganesan and colleagues used pedometer data that was self-entered by participants. 41
Study quality
A summary of the risk of bias and quality assessment for the included studies is presented in Table 3. Using the EPHPP tool, only one study was judged as ‘strong’ quality. 53 Nine studies were assigned a ‘moderate’ quality rating,39,40,43,44,46,52,60,62,63,65 and 15 studies were given a ‘weak’ rating.37,38,41,42,45,47–51,54–59,61,66 All except two studies were judged as ‘weak’ in terms of selection bias; participants were typically self-selected employees who volunteered to take part in a wellness programme.50,53 Representativeness and level of participation were unclear in many of the included studies.
Summary of risk of bias assessment.
S: strong; M: moderate; W: weak; IG: Intervention Group; CG: Control/Comparison Group
All 25 studies used robust experimental or quasi-experimental designs. Of the 25 studies, 15 reported controlling for important confounders in their design and/or analysis. Ten studies were assigned a ‘weak’ rating in this domain due to lack of reporting or poor control of confounders in analysis.42,45,49,50,56–59,61,66 No studies received a ‘strong’ rating for blinding due to the difficulty and impracticality in blinding participants to this type of mHealth intervention. Blinding of outcome assessors was often not described, and two studies were rated as ‘weak’ in this domain as outcome assessors were reported to be unblinded.37,38,47,48
A ‘strong’ data collection method for the main PA/SB outcome was used by 14 studies; this included research-grade accelerometers (e.g. activPAL™, Actigraph™, GENEActiv™)37–40,55,56,58,61,65,66 and commercial activity monitors with high validity and reliability for the particular measure (e.g. Fitbit® used to monitor steps),42,46,49,53,59,60,74 and the International Physical Activity Questionnaire (IPAQ) with reasonable validity and reliability.47,48,75 Eight studies used ‘moderate’ data collection tools with either acceptable validity or reliability, including smartphone-integrated accelerometers,43–45,62 the Activity Questionnaire for Adolescents and Adults (AQuAA),52,76 the Tractivity® activity monitor, 63 self-entered pedometer data, 41 and step data from the Fitbit® converted to MVPA. 57 Two studies used self-reported data in non-validated questionnaires,50,51 and one study 54 used computer software and an activity monitor with unreported validity and reliability; these were therefore given a ‘weak’ data collection rating.
Withdrawals and dropouts were reported by the majority of studies (n = 24). Definitions of attrition varied between studies but it was possible to calculate attrition rates based on the number of participants failing to provide data at the final follow-up, which ranged from 0% to 74%. Only four studies38,66,41,56 were rated as ‘weak’ in this domain due to having particularly high attrition rates of greater than 40%.
Regarding intervention integrity, most studies reported the proportion of participants receiving the allocated intervention, which was most frequently in the range of 80–100%. Approximately two-thirds of studies reported measuring consistency of delivery or use of the intervention, with outcomes such as device wear time and interaction with technology. In the majority of studies, it was judged to be possible that participants had received an unintended intervention or this could not be determined from the reports.
Data analysis methods were generally deemed appropriate. Of the 13 RCTs, 8 used intention-to-treat analysis.37–40,44,45,52,53,60,62
BCTs
Due to the relatively small number of studies and BCTs identified, it was not possible to determine which techniques were associated with intervention efficacy. In many cases it was difficult to determine intervention content and specific BCTs used from the available descriptions. The most frequently identified BCTs (or categories of BCTs) are shown in Table 2. These included self-monitoring of the behaviour or outcome (n = 22, 88% of studies), provision of feedback on the behaviour or outcome (n = 21, 84%), goal-setting for the behaviour or outcome (n = 17, 68%), social comparison (n = 14, 56%), social support (n = 12, 48%), rewards and incentives contingent on progress towards or achieving the behaviour (n = 11, 44%), and provision of information on consequences of PA and SB to the individual or in general (n = 11, 44%). Prompts and cues (n = 9, 36%) were also a common intervention component. Action planning was identified in eight studies (32%), graded tasks were described by four studies (16%), information on where and when to perform the behaviour was given in four studies (16%), and barrier identification/problem solving was used in three studies (12%). Instruction on how to perform the behaviour, shaping, and prompting anticipated regret were each used in two studies (8%). Information about others’ approval and environmental restructuring were each found in only one study (4%). Individual or team competitions, and various types of gamification (such as virtual avatars and racing around a virtual landscape) were not part of the CALO-RE taxonomy but were used in several studies with smartphone apps and websites. Of the 40 BCTs listed in the CALO-RE taxonomy, 16 were not identified in any of the coded interventions.
Prompts and cues were used more frequently in interventions for SB; these were found in 5 of 9 studies (56%) that aimed to reduce SB compared with 6 of 22 (27%) aiming to promote PA. Rewards and incentives were more frequently part of interventions targeting PA (11/22 studies, 50%) compared with 3 of 9 (33%) studies that aimed to reduce SB.
Effects of interventions
Statistical methods of combining the results were not considered feasible for several reasons. There was high methodological heterogeneity with a range of different study designs, outcome measures (particularly for PA) and outcome time points. Incomplete reporting of outcome data and standard deviations precluded the calculation of reliable effect sizes. Some studies reported change in PA while others reported absolute values. In addition, several studies were either uncontrolled or did not have a ‘true’ control group (i.e. the comparison group received an mHealth intervention), which would have resulted in an underestimation of effect sizes. The data were therefore summarised narratively and visually. A summary of the main results for each included study is shown in Table 4. Impact on PA, SB and health and other related outcomes is reported separately.
Summary of main results.
IG: intervention group; CG: control/comparison group; PA: physical activity; SB: sedentary behaviour; MVPA: moderate to vigorous physical activity; MD: mean difference; BMI: body mass index; BP: blood pressure; ±: standard deviation; N/A: non-applicable; AUSDRISK: Australian Type 2 Diabetes Risk Assessment Tool; IQR: interquartile range; LDL: low-density lipoprotein; HDL: high-density lipoprotein; CRP: C-reactive protein; OR: odds ratio; CDC: Centers for Disease Control.
Impact on PA
A significant increase in one or more measures of PA, over time or relative to the control or comparison group, was reported by 14 of the 25 studies (56%).38,40,41,43–46,50,51,53,58–60,62 These outcomes included mean (or median) daily steps, frequency and/or duration of activity, and odds of meeting step goals. Schrager and colleagues reported a significant impact of the intervention only in participants with a low baseline activity level. 50 Six studies (24%) reported no significant impact on any PA outcome.42,48,54,56,61,65 Three studies (12%) reported reductions in PA; two uncontrolled studies reported reductions in daily steps, 63 and MVPA66 over time, and one RCT found a reduction in light intensity PA relative to the control group, but only in a highly educated subgroup. 52 It was not possible to determine the impact of the intervention in two studies; in one the pre- to post- change in PA was unclear, 49 and another (a feasibility study) did not report the statistical significance of changes as there was no reliable baseline measure of PA. 57 It should be noted that five of the 14 studies that found a relative increase in PA did not have a true (i.e. non-mHealth) control group (see Table 4); the results suggested that one or more mHealth or complementary components had contributed to this increase, including a smartphone app, 60 motivational text messages 43 and financial incentives.44,45,62
Of the 10 studies rated as ‘high’ or ‘moderate’ quality, 7 (70%) reported a significant impact of the intervention on PA.40,43,44,46,53,60,62 Only 4 of the 11 studies (36%) using a wearable activity monitor as a single mHealth tool reported a significant absolute or relative increase in PA, compared with 10 of the 14 studies (71%) using smartphone apps or activity monitors combined with apps. Of the 14 studies (64%) using multi-component interventions, 9 reported a significant impact on PA,38,40,43–45,51,53,58,62 compared with five of the 11 studies (45%) that assessed standalone mHealth interventions.41,46,50,59,60 There were no other discernible associations between type or length of intervention, type of workplace and impact on PA.
Significant effects on PA were reported from 1 month to 12 months after beginning the intervention, although only three studies reported a significant increase in PA at a time point of 6 months or later.38,40,50 In some cases an initial increase in PA was not sustained at later follow-up.44,45,62,63 In contrast, Brakenridge and colleagues reported a significant impact of the intervention at 12 months but not 3 months. 38
There was wide variation in effect sizes. For example, for studies reporting a significant positive impact of the intervention on mean daily step count, this ranged from a between-group difference of around 847 (95% CI 68–1625)38 to 2183 (95% CI 992–3344)60 steps per day. The large international cohort study reported the largest effect, with a mean pre–post increase of 3519 (95% CI 3484–3553) steps per day. 41
Impact on sedentary behaviour
Of the 10 studies reporting impact of their intervention on sedentary time, only 4 (40%) found a significant reduction; these were a short-term wearable activity monitor and text messaging intervention in the workplace; 54 an activity monitor and smartphone app intervention; 60 an activity monitor, app and behavioural counselling intervention; 58 and a standalone smartphone app intervention. 41 Van Dantzig and colleagues found a mean between-group difference in reduction in computer activity (a proxy for sedentary time) of 4.1 min, 30 min before and after receiving a persuasive text message. 54 Gremaud et al. reported a reduction of 26.6 min (95% CI –70.9 to –17.3) in the mean longest sedentary time in those with an activity monitor and app compared with the activity monitor only group. 60 Neil-Sztramko et al. found a mean reduction in both objective and self-reported weekly sedentary time of 405.5 and 425.3 min, respectively, from baseline to 12 weeks post-intervention. 58 Ganesan and colleagues reported a mean reduction in self-reported daily sitting duration of 0.74 h (95% CI 0.78–0.71) after 100 days of the smartphone app intervention. 41
Two studies using objective measures of sedentary time showed no significant impact of a smartphone app, pedometer and social media intervention, 66 and a multi-component programme including an activity monitor and smartphone app combined with group-based action planning and a healthy living seminar. 61 Another study found no impact of an activity monitor on self-reported sedentary time at either 3 or 8 months follow-up. 52 A further two studies using objective measures showed significantly higher daily standing time and lower daily sedentary time respectively in controls relative to the intervention group.38,42 Another study using accelerometer data reported a significant increase in the mean proportion of time spent sedentary from baseline to follow-up, but only in workday non-work time (there was a slight reduction in proportion of work time spent sedentary). 56
Impact on other outcomes
Of the 25 studies, 16 (64%) assessed the impact of the mHealth intervention on secondary outcomes including health and fitness, wellbeing and determinants of PA.37–43,47–53,55,56,58,63,65,66 Of these 16 studies, 11 (69%) found an improvement in at least one outcome over time or relative to the control or comparison group.41,43,49–53,56,58,63,66 Significant beneficial effects included weight or BMI reduction,41,58 reduced body fat percentage, 63 reduced systolic blood pressure,53,63 reduced resting pulse rate, 43 reduced total and low-density lipoprotein (LDL) cholesterol and increased high-density lipoprotein (HDL) cholesterol,51,53 improved ‘AUSDRISK’ (Australian Type 2 Diabetes Risk) score, 49 improved diet,56,66 improved aerobic fitness, 51 improved self-reported health or wellness,50,66 greater self-reported energy and emotional wellbeing, 58 reduced sleep disturbance, 58 and improved self-efficacy for walking. 43 However, the study by Skogstad and colleagues could not attribute the changes in aerobic fitness and cholesterol levels to changes in individual PA levels. 51
Slootmaker and colleagues reported a significant impact on secondary outcomes in subgroups only. This included increased awareness of PA level in overweight participants only (after 3 months) and reduced body weight in lower educated participants only (after 8 months). 52 Four studies found no impact on any secondary outcomes,38,40,42,65 and one RCT found a significant between-group difference in weight loss and percentage body fat, but in favour of the control group. 48 Only two studies assessed work-related outcomes including work productivity and sickness absence, 48 and job performance, job control and work satisfaction; 38 there was no significant effect on these outcomes in the short or long term.
Subgroup findings
The most important subgroup and sensitivity findings (where applicable) for each study are reported in Table 4. Potential effect modifiers associated with intervention effectiveness were low baseline activity level,40,46,50 lower education level, 52 African American ethnicity, 57 non-obesity, 57 and high risk of diabetes. 49
Feasibility and acceptability/additional findings
Three studies were designed primarily to assess feasibility of the intervention and/or trial methods, including measures of engagement, acceptability, attrition, demand (i.e. reach and recruitment) and implementation (i.e. delivery of the intervention).55–58,66 Many effectiveness studies also reported some of these outcomes, with engagement and attrition measured most frequently. Definitions of engagement and acceptability were variable. Engagement with interventions tended to be measured quantitatively using outcomes such as activity monitor wear time, usage time for apps, features used or proportion of text messages read. Acceptability was a broader concept incorporating both quantitative and qualitative measures such as participant satisfaction, perceived usability, perceived effectiveness and usefulness of the intervention (for PA/SB/other outcomes), preferred components, intentions to continue technology use, barriers to use/engagement, and adverse events. Only a small number of studies assessed qualitative experiences as reported by the participants as a measure of acceptability. For example, Rowe-Roberts and colleagues used focus groups to gain further insight into employee experiences of using the activity monitor, 49 while Gilson and colleagues interviewed drivers and depot managers to capture experiences, insights into perceived impact of the intervention and barriers to PA.55,56
The main findings in relation to engagement, acceptability and attrition are summarised in Table 4. A clear decline in technology usage and engagement over time was reported by all longer duration studies (i.e. more than 12 weeks) that assessed these outcomes. Schrager and colleagues reported that only 33% of participants used their activity monitor after 6 months, 50 Brakenridge and colleagues reported that activity monitor use had ceased in all participants by 12 months, 38 and Finkelstein and colleagues found that only around 10% of participants still wore their activity monitor after 12 months. 40 Common reasons for lack of engagement were broken or lost devices,49,59 forgotten devices,49,50,65 lack of interest or boredom,50,52 beliefs the device was not accurate, 50 technical issues,50,60,65 fashion, 50 privacy concerns, 56 data usage costs, 56 and usability issues such as difficulty navigating the website. 52
Overall, participant satisfaction was high, and employees perceived wearable activity monitor and smartphone app interventions to be an acceptable and useful method to improve PA. Of the studies aiming to reduce SB, only two included qualitative measures of acceptability.56,66 In one study, the activity monitor and smartphone app were perceived by drivers and depot managers as feasible, acceptable and as having a positive impact on PA and SB. 56 In contrast, a study of a smartphone app for improving diet and PA (and reducing SB) in nurses found low perceived usefulness, and interviews revealed difficulties in changing more than one behaviour at a time, and the desire for a workplace champion to implement the intervention. 66 Additional findings in relation to acceptability included individual differences in preferred motivational strategies according to levels of activity and engagement, 49 and higher compliance with activity monitor wear with team-based competition as opposed to individual monitoring. 53 It is also important to consider adverse events associated with technology use; in one study around 27% of participants reported at least one adverse event, typically related to reactions to wearing the activity monitor or accelerometer. 38 Due to the relatively small number of studies reporting measures of acceptability, and the heterogeneity of interventions and outcomes, no associations between acceptability and intervention type or length or type of workplace were evident.
Attrition rates ranged from 0% to 74%. Predictors of loss to follow-up included female gender, 40 younger age 51 and ethnicity, with lower attrition in Chinese participants. 40
Discussion
While methodological quality of many of the included studies was weak, based on this review there is reasonable evidence that mHealth interventions in workplace settings are a potentially effective and feasible method for increasing PA. There is some evidence that they may also be effective in reducing SB. However, findings are mixed and effect sizes are small, particularly for the impact on SB and in the longer term.
A significant increase in PA, either over time or relative to the control or comparison group, was observed in 56% (14/25) of studies, and in a higher proportion of studies rated as ‘high’ or ‘moderate’ quality (7/10, 70%). The findings in relation to SB were less clear, with only 40% (4/10) of studies reporting a significant reduction in sedentary time, and a further three studies reporting relative increases in certain measures of sedentary time. It may be that reducing sedentary time at work leads to corresponding increases in time spent sedentary outside of work; this demonstrates the importance of holistic interventions that take both work and non-work contexts into account. 56
It is important to assess feasibility in addition to effectiveness of complex interventions such as mHealth.77,78 Many studies included measures of engagement with the intervention, and the vast majority showed a decline in engagement over time. It is not yet clear whether this disengagement from the technology is detrimental to behaviour change or if sustained behaviour change can be achieved without continued used of the mHealth tool. Future studies could draw comparisons with, and learn from, eHealth interventions to reduce SB and increase PA in the workplace, such as the studies by Mainsbridge et al. (2014),79 Pedersen et al. (2014)80 and Irvine et al. (2011). 81 Only a small number of studies included qualitative measures of acceptability such as interviews to explore participants’ experiences, mechanisms of behaviour change and reasons for the decline in engagement. Future studies should focus on these areas. There also appears to be a need for more standardised definitions, assessment and reporting of engagement and acceptability in the field of mHealth. 26
The findings generally concur with the existing evidence for potential effectiveness and acceptability, most prominent in the short term, reported in reviews of mHealth interventions for PA and SB in non-workplace contexts.17,19–21,23 Due to considerable heterogeneity and the small number of high quality studies, it was not possible to draw any definitive conclusions on the relative effectiveness or acceptability of different types of interventions, although there was some evidence that wearable activity monitors alone, and standalone mHealth interventions with no additional ‘offline’ components, were less likely to result in increased PA. Similarly, previous reviews have suggested that multi-component interventions may be more effective than standalone mHealth interventions.23,30
This review is the first to focus on mHealth technology for the promotion of PA and reduction of SB in workplace interventions. A recent systematic review by Stephenson and colleagues that assessed the impact of computer-based, mobile and wearable technologies on SB suggested that the effects of workplace interventions may be more prominent than non-workplace interventions at medium-term follow-up. 27 While there was insufficient data to test this hypothesis in the present review, this highlights the potential importance of setting and the possibility of differential results.
There was a small amount of evidence to suggest that mHealth for PA and SB may be more effective for more sedentary employees,40,46,50 and those with lower levels of education. 52 There may also be differential effectiveness according to health status at baseline, 49 BMI and ethnicity. 57 Future studies should aim to clarify which subgroups are likely to benefit most from workplace mHealth interventions. The acceptability and impact of mHealth for underrepresented groups such as shift workers, who experience unique barriers to PA and may have an elevated risk of cardiovascular disease, 82 diabetes 83 and obesity, 84 should also be explored further.
The review found some evidence for a positive impact on health and wellbeing outcomes (physiological and psychological) of mHealth interventions for PA and SB. It is recommended that future studies investigate the wider impact on health and wellbeing in addition to measures of ‘organisational wellness’ such as productivity, sickness absence and economic analyses, which were included as outcomes in only a small minority of studies. Most studies included in this review focused on workplaces in developed countries, with many based in academic and healthcare organisations. There will be a need for more diverse samples in a greater range of workplace settings as mHealth becomes more prevalent.
A ‘weak’ quality rating was assigned to a high proportion (15/25, 60%) of studies. Selection bias and lack of blinding were the weakest areas overall, although these are common issues in workplace and mHealth interventions.21,85 Many studies lacked a true control group or did not include a reliable measure of baseline activity. Studies were highly heterogeneous in terms of methodology and outcomes, and some studies used data collection methods for the primary PA or SB outcome with below satisfactory validity and reliability. The mHealth tool itself may be an efficient method for data collection, for example most commercial activity monitors provide a real-time, objective, valid and reliable measure of step count. 74 This will be an important advance for studies that currently rely on self-reported data. There is also a need for improved reporting and consistent use of outcome measures to facilitate future synthesis of findings and meta-analyses. Combined with the relatively small number of included studies and mostly small sample sizes, these factors make it difficult to draw definitive conclusions regarding the impact of mHealth on PA and SB.
The most frequently used mHealth interventions were wearable activity monitors or trackers and/or smartphone apps. However, interventions were highly heterogeneous in terms of both mHealth and additional content, frequency, duration and mode of delivery. Similar to previous reviews of mHealth for PA and SB, the most commonly identified BCTs included self-monitoring, feedback, goal-setting and social comparison.16,23,27 Several studies incorporated rewards (virtual or real) and social support in their interventions. Prompts and cues were more frequently used to target SB; this BCT was also frequent in the workplace interventions reviewed by Stephenson et al., compared with non-workplace interventions. 27 However, descriptions of interventions and BCTs were unclear or incomplete in many cases and it was not possible to determine with confidence which specific techniques were incorporated.
Future studies should aim for more transparent reporting of intervention content and specification of embedded BCTs, to facilitate identification of the most impactful and acceptable intervention components. There may also be a need for new behaviour change taxonomies specifically for mHealth interventions, for example to include in-app competitions, various types of gamification, virtual avatars, and to distinguish between virtual and real rewards. It was also apparent that many interventions did not have a strong theoretical or evidence basis. It has been suggested that new behaviour change models and theories may be needed to account for the interactive, dynamic and adaptive nature of mHealth interventions. 86
Long-term impact and acceptability of mHealth technology is still unclear. There is a need for studies with longer duration of follow-up, further qualitative investigation of reasons for the substantial decline in engagement over time and subsequently how engagement may be maximised. Mixed methods studies will be particularly valuable to elucidate the feasibility and acceptability of mHealth to promote PA and reduce SB in a workplace setting, as well as determining the longer-term impact on outcomes.
Strengths and limitations
Strengths of this review are that it was conducted in accordance with PRISMA guidelines, 29 the robust nature of the search strategy, study selection and data extraction process, and the systematic assessment of study quality using the EPHPP tool. 31 The review comprehensively included a range of study types, with a combination of quantitative and mixed methods studies. This enabled synthesis of findings related to acceptability and engagement in addition to intervention effectiveness. The identification of BCTs using an established taxonomy will facilitate comparison of interventions and possible future replication. The review is the first to consider studies of mHealth for PA and SB that were conducted in a workplace setting.
The main limitations are that meta-analysis could not be performed due to the relatively small number of included studies, heterogeneity of methods and outcomes and incomplete reporting. The high proportion of studies rated ‘weak’ for methodological quality limits confidence in the findings. Furthermore, the possibility of publication bias should be recognised.
Summary of recommendations for future research
According to the findings of this review, it is recommended that future studies:
Use larger samples in more diverse workplace settings (outside of academia and healthcare), include underrepresented groups such as shift workers, and consider behaviour both within and outside of the workplace. Report more fully intervention components (including the identification of BCTs using established taxonomies such as CALO-RE) and outcomes. Focus on SB in addition to PA, and use objective and efficient data collection methods (including the mHealth tool itself) to capture this data. Where practicable, include a no-intervention control (experimental studies) or at least a reliable baseline measure of PA/SB (for quasi-experimental studies). Consider the wider impact on health and wellbeing, and work-related outcomes such as productivity and sickness absence. Use mixed and qualitative methods to explore short- and long-term impact, feasibility and acceptability, including participants’ experiences, reasons for the decline in engagement with mHealth technology, mechanisms of behaviour change, and the relationship between engagement and intervention effectiveness. Capture data on adverse events associated with mHealth technology use. Explore further the relative impact and feasibility of standalone mHealth and multi-component interventions, including those combined with other online and offline components. Explore subgroup differences, including which interventions and components/BCTs are most acceptable and impactful, and for whom.
Conclusion
There is reasonable evidence to support the use of mHealth in the promotion of PA in workplace interventions. Despite low methodological quality, early studies have demonstrated feasibility, acceptability and potential effectiveness of mHealth based interventions in a workplace context. The longer-term impact, and the impact on SB, are less clear. There is a clear need for new high quality, mixed methods studies with better reporting of interventions and outcomes, in order to explore the reasons for decline in engagement over time and the longer-term potential of mHealth in workplace interventions for promoting PA and reducing SB.
Supplemental Material
Supplemental Material1 - Supplemental material for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review
Supplemental material, Supplemental Material1 for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review by Sarah Ann Buckingham, Andrew James Williams, Karyn Morrissey, Lisa Price and John Harrison in Digital Health
Supplemental Material
Supplemental Material2 - Supplemental material for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review
Supplemental material, Supplemental Material2 for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review by Sarah Ann Buckingham, Andrew James Williams, Karyn Morrissey, Lisa Price and John Harrison in Digital Health
Supplemental Material
Supplemental Material3 - Supplemental material for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review
Supplemental material, Supplemental Material3 for Mobile health interventions to promote physical activity and reduce sedentary behaviour in the workplace: A systematic review by Sarah Ann Buckingham, Andrew James Williams, Karyn Morrissey, Lisa Price and John Harrison in Digital Health
Footnotes
Acknowledgements
We would like to thank the PenCLARHC Evidence Synthesis Team at University of Exeter Medical School for their support and assistance.
Contributorship
SAB conceptualised and designed the study with support from AJW, KM, LP and JH. SAB conducted the searches. SAB and AJW completed screening, data extraction, quality assessment and BCT coding. SAB drafted and wrote the manuscript; AJW, KM, LP and JH provided critical feedback. All authors read and approved the final version.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Ethical approval
Not applicable.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: SAB is supported by a PhD studentship from the University of Exeter Medical School and Devon and Cornwall Police.
Guarantor
SAB
Peer Review statement
This manuscript was reviewed by Corneel Vandelanotte, Institute for Health and Social Science Research, Central Queensland University, Australia; Ana Howarth, Cigna Global Well-being Solutions Ltd, USA; and S. Abdin, University of the West of England, United Kingdom.
Supplementary Material
Supplementary material is available for this article online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
