A systematic review of tailored eHealth interventions for weight loss

eHealth approaches to treating obesity

Research suggests that by delivering interventions via the Internet (eHealth), some of the cost and scalability problems of obesity interventions are mitigated.^4,5 eHealth interventions can be deployed through websites, emails, text messages, monitoring devices, mobile applications, computer programmes, podcasts and personal digital assistants. ⁵ eHealth systems can set individualised goals for users, monitor diet and physical activity behaviours, and respond to data with personalised guidance, among other behaviour-change strategies.

To establish the viability of these systems for large-scale impact, research into the implementation and outcomes of eHealth interventions is needed. Hutchesson et al. ⁵ conducted a series of meta-analyses drawing on 88 eHealth weight loss or weight maintenance interventions. eHealth weight-loss interventions achieved modest weight-loss compared with no treatment (mean difference: –2.70 [–3.33, –2.08], p < 0.001; nine studies pooled in meta-analysis) or minimal treatment (mean difference: –1.40 [–1.98, –0.82], p < 0.001; 16 studies pooled in meta-analysis). Studies were found to be heterogeneous in both groups of pooled studies, I² = 49%, p = 0.04 and (I² = 72%, p < 0.001), respectively in terms of intervention strategy, technology usage and interaction style. eHealth interventions with extra features/behavioural components (e.g. self-monitoring, personalised feedback) or technologies (e.g. applications, texts or social networks) were more effective than standard eHealth programmes (drawing on either the Internet or podcasts only) (mean difference: 1.46 [0.80, 2.13], p < 0.001). ⁵ Arem and Irwin ⁶ also found the Internet-based weight-loss interventions included in their review to be heterogeneous. This was both in terms of the nature of the intervention (e.g. contact with participants, sample size and study duration) and their outcomes, with weight loss ranging from <1 kg to 4.9 kg. Crucially, since the interventions included in these reviews differed from each other in so many ways, it was impossible for the authors to make specific recommendations regarding which intervention components were necessary to bring about weight loss.

These effect sizes are modest in comparison with what may be achieved using traditional behavioural approaches. The modest weight losses demonstrated by the Hutchesson ⁵ review are several times less than in-person interventions. For example, an in-person behavioural weight-loss intervention of 4–6 months of weekly group-based lifestyle counselling and intensive dietary restriction showed that participants lose on average 10.7 kg. ⁷ These may be due in part to the varying intensity of behavioural-change strategies, personalisation strategies and supportive accountability provided by a human counsellor. To ascertain the effective components in these interventions, it is necessary to draw together the interventions that implement a common strategy.

A significant challenge to the wide-scale usage of eHealth interventions is that they typically suffer a steep decline in usage over the course of the intervention and have poor completion rates. ⁴ Neve et al. ⁸ highlight that use of eHealth technological components (both frequency and duration) may prove a useful metric to enable the establishment of an ‘optimal dose’ required for behaviour change. The review demonstrates that the higher the number of log-ins, self-monitoring occasions, chatroom attendances and bulletin board posts, resulted in greater weight loss, and this pattern was evident in the majority of studies included. Participant attrition is a key issue in assessing the efficacy of Internet-based interventions, as low levels of completion compromise the feasibility of implementing such interventions on a larger scale.

These findings provide promising support for the efficacy of eHealth interventions for weight loss.^5,6 They confer an additional advantage over traditional methods, for example, in-person behavioural counselling in delivering weight-loss interventions in a cost-effective, effectual manner suited equally to the individual and at a scale needed to address the global obesity epidemic. Efforts to investigate specific strategies that are effective for weight loss, which address the issues of modest effects and low levels of engagement are needed. Delineating the precise strategies within eHealth efforts that are effective in bringing about the greatest decrease in weight requires a granulated examination and specification of the various components, modes of delivery, technological features and behavioural theory that inform these eHealth approaches, as well of usage of technological components as part of these efforts. Web-based weight-loss interventions have been reported as being less effective than face-to-face interventions,^6,9 however, enhanced web-based interventions (e.g. more interactive and tailored) are more efficacious than basic, more generic interventions (e.g. purely informational). ⁹ Web-based interventions may prove to be clinically effective and feasible if they can mimic some of the tools and strategies of the in-person interventions, while reaching a larger population. ⁶ Enhancing web-based interventions through tailoring the eHealth interventions to characteristics of the user or patterns of behaviour is supported by a recent systematic review that evaluated web-based interventions for weight loss and maintenance. ⁹

Tailoring as a strategy to enhance eHealth interventions

Tailoring is a process whereby the provision of information, advice and support is individualised to the user. ¹⁰ Noar et al. ¹¹ explain that the process of tailoring involves an assessment of individual-level characteristics gathered by a person or self-administered (input: the basis for tailoring). This information then gets processed by either a human (human tailoring), or an expert system (computer tailoring) that uses algorithms to select content (i.e. text, images, recommendations and intervention messages) from an expert-developed database for the individual (tailoring process). Tailored material (output) is adapted using a variety of strategies to be delivered to the individual via multiple delivery modes.

Compared with non-tailored, generic materials, tailored health-messages command greater attention and are more likely to be read, elaborated upon, recalled and understood. ¹² As such, tailored health-messages are considered to be processed more deeply, contain less redundant information, and are perceived more positively by health consumers. ¹³ Tailoring strategies can range from relatively simple, for example, employing the user’s name, to the more complex, for instance adapting content to personally relevant variables. ¹⁴ Work has also been undertaken to draw together concrete definitions of tailoring strategies.^13,15,16 Krebs et al. ¹⁷ add to this by defining ‘dynamic tailoring’ as the assessment of intervention variables prior to each feedback and ‘static tailoring’ where one baseline assessment provides the basis of all subsequent feedback.

Impact of tailored approaches on health outcomes

While there is little evidence for the impact of tailored eHealth interventions for weight loss, evidence has accrued for the effectiveness of tailored web-based approaches to health interventions. For example, Lustria et al. ¹⁰ conducted a meta-analysis to assess the impact of tailored web-based interventions targeting physical activity, nutrition, smoking/tobacco use, drinking, medication adherence (asthma management), stress management and faecal soiling (encopresis). Forty experimental and quasi-experimental studies were analysed and web-based tailored interventions effected significantly greater improvement in health outcomes compared with controls both at post-testing, d = 0.139 (95% CI = 0.111, 0.166, p < 0.001, k = 40) and at follow-up, d = 0.158 (95% CI = 0.124, 0.192, p < 0.001, k = 21). While these results are encouraging, (similarly to the eHealth reviews discussed earlier^5,6) the authors identified that there was great variability in how tailoring was carried out among their included articles, including differences in intervention features, formats, and levels of interactivity.

A series of meta-analyses by Krebs et al. ¹⁷ also assessed the effect of 88 computer-tailored interventions, using computer, print, or telephone communication channels that focused on four health behaviours: smoking cessation, physical activity, diet, and mammography screening. A significant, small effect size (Hedges’s g = 0.17) was found for tailored interventions, taking the average of the four health behaviours. This is encouraging evidence for tailoring as an approach to health behaviour change in general, but more specific approaches, honing in on one specific outcome may allow for in-depth conclusions to be drawn, considering that the determinants of health behaviour change interventions differ by outcome.

These studies provide support for tailoring as a strategy for use in health behaviour-change interventions, but we argue that previous research on tailored health interventions has focused primarily on understanding whether tailoring works, rather than focusing on how it works, or which approaches are most useful under which circumstances.^12,15

The current study

Without meticulous description of intervention designs, the science and practice of tailoring within eHealth intervention design and implementation cannot be advanced. Previous work on tailoring has met with difficulty in differentiating the range of approaches to tailoring, differences in modality, intervention features and components. ¹⁰ In addition, Harrington and Noar ¹⁶ have called for improved reporting standards in tailored research.

The aim of the current paper was to review the evidence for tailored eHealth weight-loss interventions. We aim to describe in detail: 1. how tailoring was implemented in these studies and 2. whether these tailored approaches were effective in producing weight loss compared with generic or inactive controls. Specifically, we sought to identify what individual factors were assessed as part of the tailoring process, what tailoring strategies and eHealth tools were used, how engagement with eHealth interventions was conceptualised and whether tailoring increased this and, lastly, whether tailored approaches produced larger effect sizes than generic information and waitlist control approaches for weight loss.

Design

Systematic review of pre–post intervention trials with a control group, following Cochrane methodology and Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (see Appendix A). ¹⁸ The review was registered with PROSPERO (CRD42017072901). ¹⁹

A systematic review of tailored eHealth interventions for weight loss was carried out, with a title and abstract screening phase (Phase 1) and a full text screening phase (Phase 2).

Eligibility criteria

All papers that were deemed relevant at Phase 1 were included if they were quantitative peer-reviewed articles, published at any time, in the English language, with a pre-test–post-test intervention design (either randomised controlled trials (RCT) or pilot study) using a control comparison group (generic or inactive) approach. Interventions of any duration that aimed to bring about weight loss in overweight or obese (Body Mass Index (BMI) >25), free-living participants without specified clinical disease were included. Articles were required to have a stated tailored approach, which we operationalised as ‘if the intervention was personalised, and based on an assessment of individual characteristics’. ²⁰ The intervention was required to be behavioural in nature (e.g. aimed to change diet and physical activity), and delivered primarily via the Internet (Internet-based); in-person measurement of weight was permissible, as was briefing to randomisation group components. Articles were excluded if the intervention was not tailored at the individual level (e.g. generic in nature or tailored based on shared rather than individual characteristics, e.g. gender tailoring), not delivered online (e.g. delivered in-person), not aiming to change behaviours (e.g. pharmacological), and if the primary aim of the intervention was not specifically for weight loss; however, articles were included if there was more than one main aim, as long as weight loss was stated as one of the aims. Articles were excluded if they stated that the participants had a clinical population (e.g. type 2 Diabetes patients who may have been taking medication to control their condition), or if they were not published in the English language. Authors of studies that met the criteria but did not report the body weight of participants at baseline and follow-up (e.g. reported the BMI only) were contacted in an attempt to include them in the review.

Sources and Searching

Searches were run in December 2016 in five databases, including: EBSCO, Science Direct, Pubmed, EMBASE and Web of Science. Combinations of the concepts ‘tailoring’, ‘eHealth’ and ‘overweight’ were searched as part of the search strategy. For example, EBSCO search included the following combination: TI personalis* OR individualis* OR tailor* OR custom* AND AB (ehealth or e-health or website or web or web-based or website-based or online or Internet) AND SU (overweight or obesity or obese or weight management or weightloss or weight loss or diet* or physical activity). Searches were re-run on 15 March, 2018. An example of the search strategy is included in Supplemental material (Appendix B). Hand-searching was conducted of reference lists of included articles, systematic reviews and meta-analyses of eHealth interventions for weight loss, as well as in the Journal of Medical Internet Research.

Data extraction

Data extraction was conducted by KR. Fifty per cent of the extracted data were checked for accuracy by author SD and 50% by author CL. Two accepted articles had associated adjunctive articles, published separately but reporting details of the development of the tailored eHealth weight-loss intervention. Data from these articles were also included in the extraction and qualitative synthesis, in tandem with the original accepted article. The extracted information included participant information, intervention information and weight-loss outcomes. This included details on study setting; study population and participant demographics and baseline characteristics; details of the tailoring intervention and control conditions; study methodology; recruitment and study completion rates; outcomes and times of measurement; suggested mechanisms of intervention action; and information for assessment of the risk of bias. Specific to the process of tailoring, we identified the individual factors that were assessed as part of the tailoring process, how the assessment was conducted (e.g. what eHealth technology/modes of delivery were used), how often it occurred (e.g. static tailoring or dynamic), how the message output was devised, what specific tailoring and behavioural strategies were implemented, how engagement with the eHealth intervention was conceptualised and measured, whether tailoring increased this and, lastly, whether tailored approaches produced greater weight loss compared with generic or waitlist approaches. Authors were contacted for any missing information.

Data analysis

There were two components to data analysis: qualitative and quantitative synthesis. Qualitative synthesis involved examining how tailoring approaches were conducted within each intervention using information in each article’s intervention description and any corresponding intervention development articles.^21,22 As part of this process, terms listed in Harrington and Noar ¹⁶ were used to identify the tailoring strategies described. These included the following terms: ‘content-matching’ (adapting intervention content based on theoretical determinants), ‘feedback’ (providing messages to individuals about their psychological or behavioural states, which may involve description of objective data back to participant (‘descriptive feedback’), comparing their data to norms (‘comparative-normative feedback’), their previous states (‘comparative-progress feedback’) and providing an interpretation or judgement of their data (‘evaluative feedback’) and ‘personalisation’ where recognisable information relating to the participant is relayed to them, including the use of their name (‘identification’), telling participants that the intervention content has been created especially for them (‘raising expectation of customisation’) and framing messages in a context that is meaningful to them, e.g. through gender-, cultural- or ethnicity-related cues (‘contextualisation’).

Quantitative synthesis involved evaluating the effect of the tailored eHealth interventions on weight loss: effect sizes were calculated to estimate the difference between the intervention and control groups (both active and waitlist controls). Effect sizes for weight change at follow-up assessments were calculated based on the mean pre–post change in the treatment group minus the mean pre–post change in the control group, divided by the pooled pre-test standard deviation (SD). ²³ In one case, the post-test SD was used as a proxy where the pre-test SD was not reported. ²⁴ A pooled SD was calculated using Cohen’s ²⁵ formula, squaring the baseline SD of weight of the intervention group and adding the squared baseline SD for weight for the control group, and taking the square root of these. Effect sizes are therefore reported in SD units of change. Changes between the intervention and control were considered to be small (0.2 to <0.5); moderate (0.5 to <0.8) or large (>0.8). ²⁵

Quality appraisal

Quality of methodology was assessed by authors KR and SD using the Cochrane Collaboration’s tool for assessing risk of bias for RCTs. ²⁶ Bias was assessed as a judgement (high, low, or unclear) in the following domains of bias: selection, performance, detection, attrition, reporting and other bias. In this review, as the outcome of significance was weight, we included an assessment of measurement bias for weight as part of the quality review. Studies using self-reported weight, or using one-off (not averaged) weight measurements were considered to have an increased risk of measurement bias.

How was tailoring implemented?

There were six interventions included in the review, referring to 14 intervention arms, four articles compared computer tailoring to generic information control,^24,32–34 or to waitlist control;^24,31 one compared human tailoring (devised by human health counsellors) to generic information control; ³³ one used a combination of computer tailoring and human tailoring ³⁰ compared with a generic information control.

Computer tailoring, when feedback is delivered by a system, was implemented through a tailored website devised by Tailorbuilder ³⁴ and a tailored expert system. ³²

What was tailoring based on and how was it gathered? (Tailoring input)

Theoretical models of behaviour change are integral to tailoring and so any difference in theoretical components that provides the basis for the tailoring process may underlie any similarities or differences in intervention outcomes. Four out of six studies mentioned one or more theories that informed the development of the intervention: Social Cognitive Theory; ³¹ Cognitive Behavioural Theory; ³³ Self-Regulation Theory, the Theory of Planned Behaviour, the Precaution Adoption Process Model; ³⁴ and Behaviour Change Theory comprising Social Cognitive Theory, Control Theory, Operant Conditioning, Ecological Theory, and Social Network Theory. ³⁰ Tailoring was based on anthropometric data and personal characteristics including age, height, weight, gender,^31,32 family history and prior weight-loss experience; ³² behaviours including dietary factors, e.g. calories, level of physical activity,^{24,30–32,34} theoretical concepts such as future weight goals,^31,32,34 physical activity and diet goals, ³⁰ advantages and disadvantages of weight management, confidence and willingness to change behaviours, ³⁴ physical location, ³⁰ whether self-monitoring occurred; ²⁴ high-risk habits, ²⁴ health-related habits, barriers to physical activity; as well as a range of psychosocial characteristics around stress management, attitudes to overweight people, comfort eating, and sources of motivation to lose weight, among others ³² (see Table 2 for further details).

OPEN IN VIEWER

Table 2.

Summary table of tailoring methods.

Study author, year and title	Theory stated and behavioural principles	How tailoring conducted; when; how often	Tailoring based on …	Tailored output: type and mode of delivery	Engagement
Mouttapa et al., 2011The Personal Nutrition Planner: A 5-week, computer-tailored intervention for women	Social Cognitive TheoryEducation on diet and nutrition; goal setting (long- and short-term); self-efficacy (identifying short-term goal over next 7 days); provision of facilitators of diet change: shopping list; meal planner; recipe suggestions; advice for overcoming barriers	Static computer-tailoring via online questionnaire at baseline	Input 1: age, height, weight, sex, level of physical activity, and weight goals	Output 1: participants’ body mass index and estimated energy expenditure are calculated (evaluative feedback)Output 2: specific nutrition information using MyPyramid guidelines used to recommend amounts for all food groups and how nutrients in each food group benefit health (Content-matching); mode: website.Output 3: personalised newsletter based on their selection of long-term and short-term goals (content-matching, self-tailoring/customisation),mode: (if opt in) emailed newsletter	Not explicitly defined.Registration rate of intervention group to access intervention profile: 79%. Questionnaire for opinions of the intervention (not given to control)
Tate et al., 2006A randomised trial comparing human email counselling, computer-automated tailored counselling, and no counselling in an Internet weight-loss programme	Cognitive Behavioural TheoryEducation (principles of behavioural weight-loss: diet exercise and behaviour change); orientation to website (differed by group); instruction on meal replacement usage (440 kcals: coupons given) with unstructured third meal; increase exercise to 30 mins walking per day; self-monitor diet and exercise	Dynamic tailoring (weekly) via participants logging to a web-based diary Either computer-tailoring (CT) or human tailoring (HT), depending on intervention condition	Web-based diaries, including weekly weight, daily caloric intake, use of meal replacements and exercise	Human-tailoring group: weekly feedback – unstructured and based on clinical judgement (counsellors had behavioural weight-loss experience and degrees in nutrition, psychology, health education, exercise physiology) including answering questions, behavioural feedback on progress towards goals and weight loss (comparative-progress feedback; evaluative feedback), overcoming barriers and motivation.Mode: email.Computer tailoring: automated feedback based on behaviours from past week (comparative-progress feedback and evaluative feedback) for weight loss and calories consumed and burned compared with individualised diet and exercise goals, behavioural strategies for improving adherence to self-monitoring diet and exercise, overcoming barriers, and motivation or praise depending on logging frequency (evaluative feedback). Mode: website	Not explicitly defined. Log-in frequency and site component usage (number of weeks of online diary submission)Median no. of total log-ins to website. Control=34; CT=20; HT= 32.5 timesThose in the HT group submitted diaries for 17.2 weeks (SD=8.7) more than the CT group, 11.4 weeks (SD= 9.2), p<0.001Online diary submissions associated with greater weight loss for both active intervention groups (HT:r= –0.56; CT: r= –0.69, p <0.001)
Van Genugten et al., 2012Results from an online computer-tailored weight management intervention for overweight adults: randomised controlled trial	Self-Regulation Theory; Theory of Planned Behaviour; Precaution Adoption Process Model.Based on main steps for self-regulation for weight control: self-monitoring (weight, diet and PA), goal setting, action planning, evaluation, adaptation); aimed to support decision making/goal setting around behaviour change; comprised education about behaviour-health link; a review of current behaviours and feedback; social support and peer modelling; support intention formation; prompt cues; decisional balance; behavioural feedback; support for preparing and enacting behaviour change: action planning; analysing lapses in behaviours; relapse prevention; coping planning; behavioural contract	Dynamic computer-tailoring (asked to visit website every 2 weeks) Tailored website made using TailorBuilder software	Module 1: Input 1:assessments based on personal details and future weight goals; Input 2:assessment of personal advantages and disadvantages of weight-gain prevention; Input 3: assessment of confidence and willingness for weight-gain prevention; Input 4: assessment of diet and PA; Input 5: goal-setting tool and preparation tool.Modules 2 and 3: Input 1: assessment of weight and behaviours over past week; Input 2: assessment of high-risk situations in past or future; Input 3: when insufficient perceived behavioural control.Module 4: Input 1: assessment of weight; Input 2: asked to describe their personal rewards when goals are accomplished	Module 1:Output 1: graphs depicting weight history and predicted weight (comparative-progress feedback; evaluative feedback); tailored feedback on participants’ reasons for weight management and alternate views, instruction to consider both and reweigh pros and cons.Output 2: feedback decisional balance of advantages and disadvantages (descriptive feedback;content-matching); Output 3: if not confident to change behaviour receive a peer story; if not willing, anticipated regret (content-matching); Output 4: feedback on diet and PA and recommendations for change (evaluative feedback); Output 5: overall module output: summary and tailored plan.Modules 2 and 3: Output 1: factual and visual feedback on performance (comparative-progress feedback; evaluative feedback); if poor performance, decisional balance on old/new behaviours; if negative feelings reported about the person/their change, output was discussion of their feelings during failure and replacing them with positive feelings (descriptive feedback; content-matching); Output 2: guided, situation specific coping plan (content-matching); Output 3: option to change behaviour goal (self-tailoring/customisation); overall modules 2 and 3:Output 4: summary of coping plan.Module 4: Output 1: written and visual feedback on weight progress (comparative-progress feedback;); personalised normal weight range (evaluative feedback); Output 2: personal reward plan (personalisation); Output 3: overall module output: personalised certificate (identification).Mode: website	Not explicitly defined.Log-in data for each intervention moduleTailored vs control: Visited Module 1: 93.3% vs 81.5%; Module 2: 74.1% vs 66.7%; Module 3: 26.7% vs46.1%, Module 4: 15.2% (no generic Module 4)Process measures:(item-specific rating scale, summary of differences reported): Tailored group rated intervention as morepersonally relevant and novel than those receiving the generic information, ps<0.01, but read less of the information, p<0.001, and perceived the length of the information as less appropriate than those in the generic group p =0.01. There was no differenceamong groups in their ratings of usefulness of the information, attractiveness of the design, appreciation of the tool, whether they would recommend to others, and overall rating
Godino et al., 2016;Patrick et al., 2014Using social and mobile tools for weight loss in overweight and obese young adults (Project SMART): a 2-year, parallel-group, randomised, controlled trial	Abraham and Michie (2008) taxonomy of 26 behaviour change techniques (which they cite draws on: Social Cognitive Theory; Control Theory; and operant conditioning); Ecological Theory; Social Network Theory.Intention formation, goal setting, self-monitoring, feedback, and goal review; techniques to increase self-efficacy for diet and physical activity, relay the benefits of, and remove barriers to, healthy changes in physical activity and diet; blog posts that were educational; Patrick (2014) adds: social support and accountability (friends, participants, health coach via social network); formation of healthy social norms about health weight-related behaviour; location based support and prompts; problem solving	Dynamic tailoring (daily/weekly). Combination of computertailoring and human tailoring.Unclear specifically when and how due to flexible nature of intervention delivery: choice of six modes including Facebook, three study-designed mobile applications, text messaging, emails, a website with blog posts, and technology-mediated communication with a health coach (up to 10 brief (5–15 min) interactions)	Input 1: physical location; Input 2: physical activity and diet goals; preferred frequency/time of tracking, feedback and participation in goal review; Input 3: physical activity, diet and weight	Output 1: text relaying behaviour-change information according to where the participant was (e.g. at home or university-based exercise) (location-tailoring). Mode: text.Output 2: customisation of programme delivery according to preferences (self-tailoring/customisation).Output 3: Trend-setter app-issued automated feedback (comparative-progress); health coaches issued feedback on self-monitored physical activity, diet and weight-performance and progress towards goals and usage of intervention modes (e.g. apps and Facebook) (comparative-progress feedback; evaluative feedback; contextualisation of health messages by time of year).Mode: text, apps, Facebook, website, instant messenger, telephone call, or video call	Engagement operationalised as: sum of a participant’s recorded interactions on the study Facebook page (e.g. a post, comment, or like) and mobile apps (e.g. entry of the number of steps taken per day), text messages sent and replied to, and communication with the health coach between each study measurement.Intervention group: median (inter-quartile range) = 98 (9–265) interactions at 6 months, 76 (0–222) at 12 months, 41 (0–198) at 18 months, and 12 (0–161) at 24 months. Control: NR
Napolitano et al., 2013Using Facebook and text messaging to deliver a weight-loss programme to college students	None stated.Goal setting (weight-loss goal set with study staff member); self-monitoring (given advice on how, given tools including scales, book and pedometer, measuring utensils), and social support (via their selected non-study-related ‘buddy’ and daily text messages); educational content sent weekly via handout/ podcast, video demonstrations, one topic per week, topics of: self-monitoring and navigating campus; planning and nutrition; internal vs external hunger and eating triggers; physical activity; stress and distorted thinking; social support; special occasions, dining out and holidays; relapse prevention	Dynamic tailoring (daily/ weekly) Unclear how devised – appears to be automated/computer-tailored feedback (via daily text and weekly summary report) (‘messages were programmed at random intervals’ p. 26)	Input 1: whether they monitored (general and behaviour specific, e.g. diet and PA) on that day; Input 2: daily PA, calories, and weight; Input 3: baseline reported high-risk habits	Output 1: feedback on performance of self-monitoring (e.g. whether or not they monitored each behaviour, diet only, PA only, both, or neither) (descriptive feedback); Output 2: immediate feedback by text (e.g. acknowledgement of submitting self-monitoring data); (descriptive feedback); Output 3: advice on habits (e.g. late night snacking, liquid calories); Output 4: weekly tailored feedback summary reports (based on collated Input 2), which are compiled into personalised reports that summarised progress (included text and visual feedback (graphs) of average weekly weight, calories and physical activity), as well as feedback on progress towards reaching one’s behavioural goals and progress in the skills training for the week (evaluative feedback; comparative-progress feedback).Mode: SMS message, external link to secure page accessed via Facebook	Level of engagement was examined for the Facebook groups (intervention group and active control) by quantifying the number of times participants ‘liked’ a study-related post, posted a comment, and RSVP’d to an event. Active control: 4/17 participants 'liked’ study-related posts (M=1.25 likes each); 7/17 commented on study-related content at least once (M=3.05 posts per commenter); 15/17 responded to event invite (average RSVPs, M=6.54).Intervention group: 4/18 ‘liked’ study-related posts (M= 1 like each); 14/18 posted on study-related content at least once (M = 1.3 posts per commenter); 13/18 responded to event invitations (M=8.56 RSVPs of those who responded);79.7% of all general monitoring texts receiving a response
Rothert et al., 2006Web-based weight management programmes in an integrated health care setting: Arandomised, controlled trial	None stated.Social support (optional participant-nominated 'buddy' who received email prompts to provide support); educational content (initial guide); action plans (weeks 1, 3 and 6); positive reinforcement of dietary and physical activity improvements, address specific barriers, and provide support and self-monitoring resources	Static computer-tailoring algorithm (tailored expert system): Balance a self-help weight management programme developed by HealthMedia, Inc.) via baseline questionnaire	Input: demographic information; personal and family health history; former weight-loss experiences (including former use of specific weight-loss treatments and outcomes from weight-loss attempts); general self-care activities (including tobacco, physical examinations, flossing, seat belt use, and stress management); physical activity, ability to be physically active, and barriers to being physically active; perceived difficulty in controlling diet and physical activity; worry regarding body image; barriers to weight management; psychosocial stress and coping; general dietary preferences (e.g. consumption of alcoholic beverages, desserts, fast food, high-fat dairy products, fried food); foods typically consumed when stressed; weight-loss goals and motivation to lose weight; source of motivation (e.g. personal choice, pressure from others); a typological assessment of eating behaviour (e.g. whether the subject eats in response to certain emotions, restricts food intake, then eats because of hunger, etc.); attitudes regarding overweight individuals (e.g. that they lack willpower, are unattractive, cannot be physically fit, etc.); weight-related self-efficacy; weight-loss expectations (e.g. looking and feeling younger, reducing risk of disease, having clothes fit better, reassuring others, getting people to stop nagging them to lose weight, etc.); and perceived social support	Output 1: an individually tailored weight management plan in line with behavioural strategies, e.g.participants who reported greater ability to change diet than physical activity received more dietary advice; specifically cited barriers and lack of efficacy were addressed with messages tailored to those issues (content-matching).Output 2: tailored action plans delivered at 1, 3 and 6 weeks into the programme to reinforce initial plan (content-matching). Mode: website	Not explicitly defined. Process measures (% positive): tailored vs control Read information completely 82% vs 67%; materials helpful 75% vs 57%; information easy to understand 93% vs 82%; materials were personally relevant 78% vs 61%; would recommend programme to others 75% vs 59%, all ps =0.0001

Note: CT = computer tailoring; HT= human tailoring; PA = physical activity; M = mean; SD = standard deviation; SMS= short message service (text); NR= not reported.

Tailoring input, which provided the basis for the process of tailoring was gathered using online questionnaires at baseline,^31,32 web-based diaries, ³³ via purpose-built tailored website, ³³ study-designed mobile applications, ³⁰ SMS messaging, ²⁹ Facebook, and emails from a remote health coach. ³⁰

Self-tailoring or ‘customisation’ ¹⁵ reflected personal preferences, and was based on desired features, e.g. whether they wanted a newsletter, ³¹ their desired intervention mode of delivery, ³¹ and how often users wanted feedback, tracking and goal reviews. ³⁰

How was tailoring conducted? Computer tailoring

Dynamic tailoring (based on dietary or physical activity behaviours or weight, self-monitoring status or location) was conducted through daily and/or weekly input^24,30,33 or less regular input and based on theory, e.g. recommended every two weeks ³⁴ or a self-determined pattern of tracking and feedback. ³⁰ Static tailoring involved a process of distributing an online questionnaire at baseline.^31,32

Two tailoring approaches were explicitly named and described: a tailored website ³⁴ and a tailored expert system. ³²

One intervention delivered computer-tailored feedback and human tailoring via health coaches who also delivered feedback, ³⁰ and whether or not participants received the intervention depended on the usage of these components. In the other instance, daily feedback appeared to be automated as messages were programmed and feedback was immediate, however it was not explicitly stated who or what devised the weekly personalised messages; an automated system appeared to have gathered daily behavioural data and this was provided to participants through a secure link delivered by Facebook message. ²⁴ The computer-tailored feedback in Tate and colleagues’ study ³³ was based on the participants’ responses and matched with a response from a bank of messages previously devised by experts; a series of decision rules ensured that automatically generated feedback was provided on the webpage.

As an example of an in-depth level of computer tailoring, Rothert et al. ³² described the action of the tailored expert system in their intervention:

participants reporting a family history of a particular disease received information regarding the connection of obesity to this class of diseases; participants who reported greater ability to change diet than physical activity received more dietary advice; specifically cited barriers and lack of efficacy were addressed with messages tailored to those issues; psychosocial stress was compared with reported coping abilities and accompanied with tailored stress management advice; and participants who reported that overweight individuals lacked willpower were given messages attempting to change this perception to a more controllable, external attribution (p. 268).

Human tailoring

Tate et al. ³³ drew on human tailoring by human counsellors who had behavioural weight-loss experience and degrees in nutrition, psychology, health education, and exercise physiology. Feedback was based on their clinical judgement in interpreting the participants’ self-reported weekly weight, daily caloric intake, use of meal replacements and exercise, in their electronic, web-based diaries on the study website. The content of feedback for approaches involved comparative-progress feedback for weight, comparative-progress feedback on calorie goals consumed and burned compared with individualised diet and exercise goals, behavioural strategies for improving adherence to self-monitoring diet and exercise, overcoming barriers, and motivation or praise depending on logging frequency. This detail was not provided in the other study.^21,30 Based on the description provided we determined that feedback was based on app usage, behaviours and progress to goals.

How tailored interventions were delivered

A range of modes were used in the delivery of tailored interventions. Websites were used in all studies.^24,30–34 Two used SMS messages as well as Facebook.^24,30 Two studies used humans to tailor information: one used health counsellors who contacted participants via email ³³ and one used health coaches who could instant message, telephone or video-call participants. ³⁰

Type of Tailored Strategies and Output

Tailoring strategies were identified from the descriptions in the articles. Content-matching,^31,32,34 evaluative feedback,^{24,30,31,33,34} descriptive feedback,^24,34 and comparative-progress feedback,^24,30,33,34 identification, ³⁴ contextualisation, ³⁰ as well as self-tailoring (customisation)^30,31,34 and personalisation ³⁴ were implemented by a range of eHealth systems. Location-tailoring was also used. ³⁰

Output typically consisted of feedback on weight loss, physical activity and diet (e.g. calories burned and consumed) compared with individualised diet and exercise goals.^{24,30,31,33,34} Visual graphs of progress, behavioural feedback, differential behaviour-change strategies, personalised certificates and personal reward plans were also delivered to participants. ³⁴ Output also included a personalised nutrition plan and newsletter around their goals. ³¹ Feedback on behaviours such as daily self-monitoring also featured, ²⁴ as did a weight management plan with behavioural strategies (e.g. action plans) matched to the individual. ³²

Did tailoring work to increase engagement?

Four studies did not explicitly define engagement within the context of their intervention. As such, there was no consistent method of conceptualising and measuring engagement with eHealth interventions. Some studies reported usage metrics such as number of log-ins or use of eHealth features e.g. web-based behavioural diary, ³³ registration to a website, ³¹ while others used process evaluations using subjective participant reports of amount of provided material read. ³² One study drew on both online module visits and subjective process measures. ³⁴

Engagement was explicitly defined in Godino et al. ²⁷ and also in Napolitano et al. ²⁴ Godino et al. ³⁰ defined it as the sum of a participant’s recorded interactions on the study Facebook page (e.g. a post, comment, or like) and mobile apps (e.g. entry of the number of steps taken per day), text messages sent and replied to, and communication with the health coach between each study measurement. Napolitano et al. defined engagement as the number of times participants ‘liked’ a study-related post, posted a comment, and RSVP’d to an event. ²⁴ Tate and colleagues ³⁰ examined log-in frequency and number and use of online diary submissions, counting the number of weeks where behavioural diaries were submitted.

Two articles conducted a process evaluation, capturing participants’ subjective views of the intervention through self-report questionnaires.^32,34 Rothert et al. ³² reported that the tailored group responded more favourably than the information-only group for each item, including reading the intervention information completely; rating the materials as more helpful, more personally relevant, and easier to understand; and reporting that they would recommend programme to others, all ps = 0.0001. In research by van Genugten et al. ³⁴ the tailored intervention group also perceived the material to be more personally relevant, and found content to be more novel than those in the generic information group, ps < 0.01, but contrastingly reported that they read less of the information, p < 0.001. They perceived the length of the information as less appropriate than those in the generic group, p = 0.01. There was no difference among groups in their ratings of usefulness of the information, attractiveness of the design, appreciation of the tool, whether they would recommend it to others, and overall rating.

Did tailoring support weight loss compared with control conditions?

A summary of the intervention groups compared with the controls is provided in Table 3.

OPEN IN VIEWER

Table 3.

Summary table of results of included studies on weight loss.

	Baseline body weight (kg)				Follow-up(months unless otherwise stated)	Follow-up body weight (change in kg)
	Int		Control			Int		Control
	M	SD	M	SD		M	SD	M	SD	Effect size (d)
Mouttapa et al., 2011	71.7	17.43	74.91	21.42	6 weeks	↓0.59	2.12	↓0.29	5.02	–0.02
Tate et al., 2006	CT = 89; HT = 89	CT = 13.2; HT = 13	88.3	13.9	3, 6	3 months: CT = ↓5.3; HT = ↓6.16 months: CT = ↓4.9; HT = ↓7.3	3 months: CT = 4.2; HT = 3.9 months: CT = 5.9; HT = 6.2	3 months: ↓2.86 months: ↓2.6	3 months: 3.5; 6 months: 5.7	3 months:CT vs C = –0.19**; HT vs C = –0.26**; (no dif CT vs HT)6 months: CT vs C = –0.17; HT vs C = –.35***(HT>CT)
van Genugten et al., 2012	83.39	11.49	83.34	10.61	1 (NR); 6	↓0.34	12.36	↓0.6	10.81	0.02
Godino et al., 2016	76.5	12.7^	76.5	13.2^	6, 12, 18, 24	6 months = ↓1.1; 12 months = ↓1; 18 months = ↓0.3; 24 months = ↑0.3	NR	6 months = ↑0.3; 12 months = ↑0.3; 18 months = ↑0.4; 24 months = ↑1.1	NR	6 months: –0.11*12 months: –0.1**18 months: –0.05 24 months: –0.06
Napolitano et al., 2013	NR	NR	NR	NR	1; 2	1 month = ↓1.72 months = ↓2.4	1 month = 1.6; 2 months = 2.5	1 month: AC = ↓0.46; C = ↑0.28; 2 months: AC = ↓0.63; C = ↓0.24	1 month:AC = 1.4; C = 1.72 months: AC = 2.4; C = 2.6	^^1 month: T vs AC = –0.5*; T vs C = –0.79***;2 months:T vs AC = –0.71*; T vs C = –0.86*
Rothert et al., 2006	92.2	14.4	92.5	14.3	3; 6	3 months = ↓2.6; 6 months = ↓2.8	NR	3 months = ↓1.4; 6 months = ↓1.1	NR	3 months = –0.1***; 6 months = –0.12***

Note: ^raw standard deviation reported; ^^ follow-up SD used to calculate effect size; *p < 0.05 **p < 0.01, ***p < 0.001. CT = computer tailored; C = control; HT = human tailoring; T = tailored intervention; AC = active control; NR = not reported.

Four out of six articles found positive effects for tailoring on weight loss when compared with generic or waitlist controls.^24,30,32,33 Two articles found no difference between the intervention and control (waitlist ³¹ and generic information ³⁴ ). Follow-up time points ranged widely from 1 month to 24 months.

The effect sizes denoting the mean difference in weight between intervention and the control groups ranged from −0.02 to −0.86 but in general are very small to small. ²⁵ In the only article that compared tailoring types, there was no difference between human tailoring and computer tailoring at 3-month follow-up; both resulted in statistically significant weight loss compared with the control group. However, at 6 months, human tailoring was still effective in producing weight loss compared with the control, but this was no longer the case for computer tailoring. ³³ Napolitano et al. ²⁴ found the largest effect between the tailored vs waitlist control conditions, d = −0.86, followed by the difference between the tailored and generic information, d = −0.71.

Godino et al. ³⁰ found a significant difference between groups at 6 month and 12 month follow-ups, with the tailored intervention resulting in 1.33 kg more weight loss than the general information control at both 6 and 12 months (p < 0.05), but there was no significant difference at 18 or 24 months, where ultimately the effect size at 24 months was −0.06. Rothert et al. ³² found a significant difference between groups at 3 and 6 months, (mean weight loss for the tailored groups = 2.6 and 2.8 kg, respectively, ps < 0.001) favouring the intervention group, which used a tailored expert system.

Summary of key findings

A systematic review of tailored eHealth weight-loss interventions was conducted on six interventions, outlined in eight published articles, to describe how tailoring was implemented, and whether it was effective in producing weight-loss. Tailored interventions were found to be more effective in supporting weight loss than generic or waitlist controls in four of the six articles. Effect sizes were very small to moderate, with evidence of fluctuations in effect sizes and differences of effect between tailored and non-tailored interventions, and between tailoring types, over time.

We note the diversity of approaches, input and technology used in tailored interventions; tailoring included a range of approaches including computer tailoring (using purpose-built tailored websites e.g. using TailorBuilder ³⁴ and tailored expert systems ³² ), human tailoring (by remote trained health counsellors ³³ ) and a combination of the two techniques. ³⁰ Tailoring was based on a range of inputs, including anthropometric data (e.g. weight) and personal characteristics (e.g. age, prior weight-loss experience), behaviours (e.g. dietary intake, physical activity), goals (e.g. weight goal), theoretical determinants (e.g. confidence/willingness to change behaviours) psychosocial factors (e.g. stress management, social support) and location. A range of tailoring strategies was implemented, including evaluative- and comparative-progress feedback and content-matching, as outlined by Harrington and Noar. ¹⁶

Tailoring input was gathered using a variety of technological modes and included static processes of data collection from online questionnaires at baseline.^31,32 Dynamic processes also occurred where rolling input was gathered from web-based diaries, ³³ via a purpose-built tailored website, ³⁴ study-designed mobile applications ³⁰ and SMS messaging. ²⁴ Conceptualisation of engagement with tailored interventions was not uniformly defined. Our findings suggest that tailored materials were rated by participants as more personally relevant, more helpful, easier to understand, more likely to be recommended to others, and more novel than those in the generic information group.

The first goal of this review was to describe how tailoring was implemented in eHealth interventions. We have described a range of methods, inputs and technology used as part of tailored interventions. Previous reviews describing eHealth interventions have also found heterogeneity in approaches, so this was no surprise given the ubiquity of the Internet and the range of delivery options it affords.^5,6 With this advance in delivery methods comes new opportunities for tailoring health information, and we have found that tailoring is being increasingly used in innovative ways. However, if we are to build the science of tailoring it is imperative that the strategies reported in existing tailoring specification efforts^15,16 are used. In addition, the input or basis for tailoring should be specified as well as who/what devises tailoring output.

In this vein, our review has attempted to describe the range of tailoring approaches delivered by eHealth means. In Kreuter’s ³⁵ much cited original definition, tailoring is defined as the provision of individualised information, advice and support based on the participant’s known characteristics, behaviours or scores on relevant theoretical constructs. To aid the conceptualisation of tailoring in this context, we devised a model of tailoring depth, which we refer to as deep- and surface levels of tailoring (see Figure 3). This model is by no means exhaustive but we feel it is a step in the right direction in terms of differentiating between types of tailoring input to provide the basis for tailoring. It incorporates concepts discussed by Resnicow et al. in developing culturally sensitive public health interventions. ³⁶ These concepts include surface structures, in other words relating to observable or behavioural characteristics of a population, and deep structures, relating to psychosocial, cultural and historical factors of a target population. This model indicates the distinctions made when the process of tailoring is based on behaviours or observable factors and when it is based on theoretical constructs, and allows for a combination of these. This builds upon Morrison’s assertion that tailoring strategies can range from relatively simple e.g. employing the user’s name, to the more complex, e.g. adapting content to personally relevant variables. ¹⁴ Deep-level tailoring refers to tailoring message content based on theoretical determinants of a specific health outcome, and the content that is delivered aims to modify these through strategies, for example like content-matching and evaluative feedback. In our review, input for deep-level tailoring was captured using a questionnaire at baseline (static tailoring). ³² For example, participants who reported a greater ability to change their diet at baseline compared with those who reported greater ability to change their physical activity received more dietary advice. ³² Deep-level tailoring was also conducted in a dynamic manner, gathering theory-based input via modules, using more than one assessment point, in order to ensure the intervention remained suitable to the needs of participants. ³⁴ This is referred to as ‘re-tailoring’ where new information from participants is obtained and feedback given at follow-up time points. ³⁷ It was recently found that participants disengage from eHealth interventions because they believe they are no longer relevant to them and their needs. ³⁸ As such, a more continuous process of adapting and delivering content to the individual may be a promising route.

OPEN IN VIEWER

Figure 3.

Model of tailoring depth.

Surface-level tailoring is a more concrete but simpler process, when tailoring is based on participants’ (ongoing) self-reported behavioural data, e.g. amount of physical exercise or dietary intake, and evaluative or comparative-progress feedback is delivered on this basis with the aim of keeping participants on-track with their behavioural goals. We suggest that this type of tailoring is no less important, but is conducted on the basis of behaviours or observable factors, typically relying on more frequent input (e.g. daily or weekly). This process aligns with self-regulation theory in that it enables the participant to learn to self-monitor and regulate their behaviour in line with their goals. ³⁹ There may be a fundamental difference in the mechanisms at play behind this tailoring approach compared with deep-level tailoring, but this will need to be tested empirically in future research.

Tailoring may initiate processes of behaviour change, and it is possible that these processes in turn influence the efficacy of, and engagement with tailored interventions. Dynamic processes of tailoring, including movement between surface and deep level, or by tailoring type (e.g. human or computer), must be examined in relation to the efficacy of the intervention. For example, in the only study to examine human vs computer tailoring, both of these groups were effective at 3-month follow-up, however at 6 months, the human-tailoring group had significantly greater weight loss than the control. ³³ This indicates the possibility that over time, the effects of tailoring will vary, and this variation may be related to an interaction on intervention-type by intervention-time. For example, at some intervention phases it may be more effective for a computer-tailored vs a human-tailored intervention to be implemented. This is linked to the suggestion that interactivity should be considered in developing effective eHealth interventions.⁹ This also highlights the concept of engagement, an under-considered factor in the research design or assessment of factors influencing intervention efficacy. We note that engagement was not explicitly defined in the majority of included studies (N = 4), and among those that did, it was conceptualised in a variety of ways mostly in relation to interaction with study-related content but also through process evaluations of the interventions.

This is in line with how engagement has been conceptualised in eHealth interventions – in terms of both participant experience and behaviours. ⁴⁰ Future efforts to examine tailoring in eHealth interventions should measure engagement both in terms of participants’ objective usage of intervention components (e.g. behavioural- through log-in metrics, usage data) and subjective experience (e.g. through process evaluations, participant interviews or surveys). This will enable an assessment of the extent to which tailoring (and indeed other intervention features and components) works the way it is theorised to (e.g. via mechanisms of increasing attention, increasing self-referential thinking ¹⁵ ) and will provide a platform for future efforts to develop enhanced tailoring approaches addressing these mechanisms.

The second aim of the review was to establish whether tailored eHealth interventions result in weight loss compared with a generic control. Effect sizes were modest, comparable to other reviews in health behaviour change.^10,17 However, it has been shown that modest weight loss (3–5% of pre-test body weight) is enough to have a beneficial health impact, e.g. reducing blood pressure, HbA1c, triglycerides and LDL cholesterol ⁴¹ and so, even modest effects are desirable.

Four out of the six studies found beneficial effects for the tailored arm of eHealth weight-loss interventions. The two studies that found no effect were less intensively focused on weight loss, as both had a further aim in addition to weight loss, with Mouttapa et al. aiming to additionally provide a dietary intervention ³¹ and van Genugten et al. to also prevent weight-gain. ³⁴ While they met the criteria for the study, these findings suggest that the intensity of the intervention plays a key role, for example, in the study by van Genugten et al., even though the intervention was rigorously developed using behaviour change theory, the intervention could be finished within 90 minutes. ³⁴ This may suggest that though tailoring was not found to be successful in that case, it may have been that the behavioural ‘dose’ of the intervention was not high enough. The other studies suggest that the effect of tailored eHealth interventions for weight loss is effective up until 1 year with no evidence at follow-up after this point.

Limitations

The number of completed studies using tailored strategies and the Internet to support weight loss in our results belies the number of studies available for review. Despite significant effort, we were unable to secure data on weight for four studies, as weight was not reported in the published record or provided on request, and our results must be considered in this context.

Our review featured participants predominantly from the general population (82.5% of participants) and this suggests that effective tailored approaches can be delivered via the Internet in a scalable manner for weight loss. However, we did not include interventions addressing weight-loss maintenance or weight-gain prevention as part of this review, which are also important foci for population-wide obesity strategies. We note that these are important areas deserving of research attention, however interventions that address these aims differ in intensity to those addressing weight loss and so we advise a delineation of behaviour type to explicate if and how tailoring is effective. The use of tailored eHealth tools in weight-gain prevention and weight-loss maintenance should be examined in future work as we attempt to support both those at risk of becoming overweight and those who need to maintain weight loss in a cost-effective, scalable and effective manner.

Conclusions

eHealth interventions for weight loss are promising in terms of being scalable and effective, ⁷ but it is necessary to find strategies that help eHealth interventions channel the effectiveness of traditional in-person behavioural counselling approaches.^2,7,42 Tailoring is one approach that shows promise. We have found tailoring to have a small but beneficial effect on weight loss in four out of six studies in this review, yet tailoring approaches varied considerably even for one clearly defined health outcome (weight loss) and mode of delivery (Internet). To ameliorate this going forward and to build on previous research, we propose a model of tailoring depth, suggesting that there may be conceptual differences in tailoring approaches. Concepts of deep- vs surface-level tailoring could be tested experimentally to assess the mechanisms or effects of tailoring on the basis of input type, be it theoretical determinants, behavioural input or other tailoring input type. In order to develop the science of tailoring within the domain of eHealth interventions for behaviour change, it is vital to improve research designs (e.g. use tailoring vs generic control group) and methodologies (e.g. use the tailoring strategies and guidelines recommended in this field) ¹⁶ and in the wider domain of health behaviour change (e.g. by reporting behaviour-change techniques). ⁴³