Systematic review and meta‐analysis of internet‐delivered interventions providing personalized feedback for weight loss in overweight and obese adults

Summary Background Obesity levels continue to rise annually. Face‐to‐face weight loss consultations have previously identified mixed effectiveness and face high demand with limited resources. Therefore, alternative interventions, such as internet‐delivered interventions, warrant further investigation. The aim was to assess whether internet‐delivered weight loss interventions providing personalized feedback were more effective for weight loss in overweight and obese adults in comparison with control groups receiving no personalized feedback. Method Nine databases were searched, and 12 studies were identified that met all inclusion criteria. Results Meta‐analysis, identified participants receiving personalized feedback via internet‐delivered interventions, had 2.13 kg mean difference (SMD) greater weight loss (and BMI change, waist circumference change and 5% weight loss) in comparison with control groups providing no personalized feedback. This was also true for results at 3 and 6‐month time points but not for studies where interventions lasted ≥12 months. Conclusion This suggests that personalized feedback may be an important behaviour change technique (BCT) to incorporate within internet‐delivered weight loss interventions. However, meta‐analysis results revealed no differences between internet‐delivered weight loss interventions with personalized feedback and control interventions ≥12 months. Further investigation into longer term internet‐delivered interventions is required to examine how weight loss could be maintained. Future research examining which BCTs are most effective for internet‐delivered weight loss interventions is suggested.


Introduction
Obesity is of growing concern owing to the rise in prevalence with levels reaching 26% in men and 24% in women within the UK (1). In 2013, 83% of UK households had access to the internet, the vast majority through broadband connections, with over half of users able to connect to the internet via their mobile phones (2). Globally, the internet is accessed by over three billion people, over 40% of the world population (3).
Traditional weight loss interventions, such as in-person consultations, have reported mixed findings for effectiveness in terms of weight loss and its sustainability (4,5), which may be related to poor adherence rates. Reasons for non-adherence within in-person consultations include personal reasons, cost of travel, limited availability and lack of parking at venues (6). Internet-delivered weight loss interventions could minimize these problems by increasing the convenience and control for the user and health professional and reducing the cost of an intervention (7,8). The number of studies incorporating internet-delivered weight loss interventions has increased over recent years (9). Previous reviews have demonstrated that internet-delivered weight loss interventions can be effective in promoting weight loss and changes in physical activity and diet (10)(11)(12)(13). However, several reviews have shown heterogeneity in results between studies, with several reporting no consistent benefits of internet-delivered weight loss interventions in comparison with control groups (10,11,(14)(15)(16). Furthermore, many studies have demonstrated high attrition rates for both intervention and control groups (10,13,17).
Previous reviews have identified the need to identify which components of internet interventions contribute to weight loss and the effectiveness of an intervention. Taxonomies have been developed to provide definitions of active ingredients within interventions based on pre-established descriptions of behaviour change techniques (BCTs) and how these relate to theories (18). Using these taxonomies allows researchers to identify the presence of BCTs within an intervention and promotes consistent reporting whilst enabling comparison and replication of intervention features (19).
Feedback has been identified as an important and effective component within technology-based weight loss interventions (20)(21)(22). Feedback delivered by a person as part of an internet-delivered intervention can encourage, motivate and assist patients in successfully completing a weight loss program (23). Control theory (24) incorporates the BCT of 'providing feedback'. The theory's basic construct is known as the discrepancy-reducing feedback loop. This process is considered to be key to self-regulation. Self-regulation processes can be used to reduce the intention-behaviour gap and facilitate the understanding of the progression from intention to action. Self-regulation-based interventions have been identified as twice as effective as interventions without self-regulation strategies (25). The use of internet-delivered interventions can enhance weight loss effectiveness when individualized feedback and email counselling are integrated (21). Personalized feedback is generally delivered via specific tailored contacts, either web-based messaging, emails, short message service or in-person (26). It is important to identify and evaluate the types of feedback, which can be delivered via the internet.
The aim of the current study was to assess whether internet-delivered weight loss interventions providing personalized feedback (IWLPF) were more effective for weight loss in overweight and obese adults in comparison with control groups either placed on a wait list, receiving a minimal face-to-face intervention or receiving internet-delivered weight loss interventions without personalized feedback. In addition, it aims to describe how feedback is provided and to identify the BCTs incorporated within internet-delivered weight loss interventions.

Method
Guidelines set out in the Cochrane handbook for systematic reviews of interventions were followed (27), and reporting is in accordance with the PRISMA statement checklist (28). The review proposal was accepted onto PROSPERO (international prospective register for systematic reviews) on 17 May 2012, registration number: CRD42012002115.

Inclusion criteria
Criteria for considering studies are outlined in Table 1. The definition used to code for the BCT feedback was taken from the CALO-RE taxonomy definition of "Provide feedback on performance -This involves providing the participant with data about their own recorded behaviour or commenting on a person's behavioural performance (e.g. identifying a discrepancy between behavioural performance and a set goal or a discrepancy between one's own performance in relation to others)." pg. 9 (29). This definition was used throughout to guide the selection and inclusion process, coding and analysis. Reference lists of identified studies and citation indexes of papers citing the identified studies were searched. Relevant authors in the field were contacted and asked if aware of any other studies relevant to the review.

Data collection
Selection of studies All studies generated from the previously defined search strategies were evaluated against the pre-defined inclusion 542 Internet-based weight loss interventions: impact of feedback A. Sherrington et.al obesity reviews criteria by two reviewers. Any disparities were addressed by involving a third reviewer and reaching an agreement. The studies that qualified for inclusion into the review were assessed with regards to their methodological quality by two reviewers. Studies were assigned a quality rating of low, high or unclear risk of bias for each criterion based on the Cochrane Collaboration's tool for assessing risk of bias (27). Studies were scored in relation to randomisation, allocation concealment, reporting of blinding, incomplete outcome data, selective outcome reporting and any other sources of bias (Supplementary materials Table S1). The two reviewers showed high inter-rater reliability, and a third reviewer was not required (kappa = 0.89).

Data extraction, synthesis and analysis
Primary outcome analysis Weight loss was analysed at 3, 6 and 12 (or more) month data collection points as well as for the end of each study intervention.

Secondary outcome analysis
Outcomes of 5% weight loss, BMI change and waist circumference change were analysed at 3, 6 and 12 (or more) month data collection points as well as the end of each study intervention.
Retention rates are number of participants remaining and adhering to the randomized arm and also number of participants remaining in study for data collection (comparison with rates in the control group).
Coding of the BCTs was conducted for each of the studies, with 20% independently checked by the second reviewer. These were coded based on CALO-RE taxonomy of BCTs to help people change their eating and physical activity behaviours (29). When coding for the presence of BCTs within an intervention, no assumptions were made. The standardized vocabulary within the BCT taxonomy was adhered to in order to state the presence of any BCT, explicitly or implicitly, within the interventions reported in each included paper, thus promoting consistent reporting and coding between researchers (19).

Analysis
Statistical analysis of the data was carried out using Review Manager 5. Data were analysed using mean (SD) change for each IWLPF and control group receiving no personalized feedback and compared whether significant differences were present between the different arms for each outcome measure: weight loss, BMI, waist circumference and 5% weight loss. Meta-analysis was conducted to examine the studies at the end of each study intervention. As intervention length varied between the studies, time points were examined separately, including 3, 6 and 12-month analysis in addition to the end of intervention. Meta-analysis was conducted, with intention-to-treat analysis data if available from the published data, along with tests for heterogeneity. All study data included in the meta-analysis used results measured at the end of the intervention. One of the included studies, by van Wier (30), conducted a follow-up at 24 months (after a 6-month intervention). Therefore, only post intervention data was used within the meta-analysis. The follow-up data of this study, 24 month, was not included to avoid the conflation of active loss and maintenance stage results. As a variety of control groups were included in the review, e.g. wait list, face-to-face and internet-delivered, subgroup analyses were performed to separate the effect of feedback from that of delivery mode. Control groups were categorized into 'waiting list or minimal face-to-face interventions' and 'control internet-delivered interventions without personalized feedback', refer to Table 2.

Results
Fourteen articles reporting on 12 separate studies were included in the review (Fig. 1).
Study quality assessment identified that only two of the studies assessed received low risk of bias for all criteria. All quality assessments can be found in Supplementary material Table S1. Selective reporting was the only criterion to receive high risk of bias scores for four of the studies (22,(30)(31)(32). Three studies provided monetary incentives for the completion of assessments that may have acted as a co-intervention in respect of retention rates (22,33,34). Interventions Targeting diet and/or physical activity for weight loss Delivered at least in part via the internet Incorporating any form of individualized feedback to the participants either human-delivered (provided by a health care professional or researcher) or computer-generated personalized feedback (using algorithms that sent pre-programmed responses based on participant input or choices) delivered via web-based messages or email Definition of feedback used to guide process "Provide feedback on performance -This involves providing the participant with data about their own recorded behaviour or commenting on a person's behavioural performance (e.g. identifying a discrepancy between behavioural performance and a set goal or a discrepancy between one's own performance in relation to others)." pg. 9 (29) Comparator Arms comprising no individualized feedback, e.g. wait list, treatment-as-usual, intervention without feedback Outcome Primary: body weight change Secondary: body fat, waist circumference or BMI change, retention rates Study design Randomized controlled trials (including pilot studies)

Description of included studies
The characteristics of the included studies are summarized in Tables 2 and 3. All studies took place between 2001 and 2012. The majority (seven) were conducted in the USA, three in Australia, one in the Netherlands and one in the UK. The total number of participants was 3547 with 1816 females (51.2%). All 12 studies targeted changes to physical activity and diet. The length of the active interventions ranged from 3 to 24 months (21-month range, mean 8.4, SD 5.7). Seven studies included two arms, and five studies included three arms. The studies varied in terms of the features of control/comparison arms ( Table 2).

Provision of individualized feedback
Across the 12 studies, 8 incorporated human-delivered internet feedback and 5 provided computer-generated internet feedback (Table 2). One study provided the personalized feedback using both formats as the study contained two internetdelivered intervention groups (33). These two terms have been used to distinguish between interventions using personalized feedback provided by a health care professional or researcher (human-delivered) in contrast to personalized feedback created using algorithms to send pre-programmed responses based on participant input or choices (computergenerated). All 12 studies used personalized feedback to target information received on participant's weight loss progress or individual behaviour change, such as diet or physical activity level. Participant access to the internet-delivered personalized feedback was via the website (four studies) or via emails containing the feedback (six studies), with two studies remaining unclear in how it was administered. Frequency of feedback varied, the majority of studies (seven) providing it on a weekly basis. In addition to personalized feedback, two studies sent computer-generated messages when participants logged into the website (35,36). One study provided computer-generated messages to participants on completion of lesson modules or assessments (30).

Meta-analysis/synthesis of results
Internet weight loss interventions providing personalized feedback versus control groups receiving no personalized feedback The primary outcome, weight loss, is shown in Fig. 2 illustrating the meta-analysis forest plot for the 12 studies. Metaanalysis identified that provision of feedback resulted in 2.13 kg (mean difference [MD]) (p < 0.00001) greater weight loss for the IWLPF in comparison with control groups receiving no personalized feedback. Heterogeneity levels showed considerable and significant heterogeneity (I 2 = 99%, p < 0.001) between control groups not receiving personalized feedback and the IWLPF. All outcomes were found to be statistically and clinically (≥5% body weight loss) significant for study end of intervention results (Table 4). This was also true for results from data collection conducted at 3 and 6 months. In contrast, studies with duration 12 months or   Figures S1-S4). Retention rates were calculated by the number of participants who provided follow-up data at the last assessment point (varying between studies). In total, intervention groups retained
Meta-analysis showed a statistically significantly greater weight loss (2.14 kg MD, p < 0.001) for those in the IWLPF in comparison with the wait list or minimal interventions. Heterogeneity was considerable and significant between the wait list or minimal control groups and intervention groups (I 2 = 100%, p < 0.00001) (Supplementary Figure S5). Meta-analysis was performed for the three studies using control internet-delivered interventions without personal feedback (22,33,34). Results showed 2.05 kg (p < 0.0001) greater weight loss for the IWLPF in comparison with the control internet-delivered interventions receiving no personal feedback. Heterogeneity was not important and non-significant between the control internet-delivered interventions receiving no personal feedback and the IWLPF (I 2 = 0%, p > 0.05) (Supplementary Figure S6).
The BCTs incorporated most frequently are represented in Table 5 along with mean weight loss for each study's intervention and control group. The most prevalent BCT was 'providing information on consequences in general'. This was the only BCT that was present in the majority of the control groups receiving no personalized feedback. Common techniques within the IWLPF, aside from 'provide feedback on performance', were 'planning social support/social change', 'prompting self-monitoring of behaviour/behavioural outcome' and 'goal setting (behaviour and outcome)'. These most commonly used BCTs tended to be clustered within the studies.

Summary of key findings
Findings from this systematic review suggest that incorporating personalized feedback may be an important BCT for effective weight loss interventions delivered via the internet. Participants within the IWLPF were identified as twice more likely to achieve 5% weight loss than those in control groups. Shorter term data collection, 3 or 6 months, produced significant differences between the IWLPF and the control groups receiving no personalized feedback for all outcomes (weight loss, 5% weight loss, BMI and waist circumference change). In contrast, interventions lasting 12 months or longer did not produce significant differences between IWLPF and control groups receiving no personalized feedback for weight loss or 5% weight loss outcomes. Subgroup analysis identified significantly greater weight loss for the IWLPF irrespective of the comparator used, whether wait list/minimal face-to-face interventions or control internet-delivered interventions receiving no personalized feedback.

Comparison with previous literature
As in previous reviews, internet-delivered weight loss interventions appeared to be more effective than comparison groups (13,15). However, previously, in terms of significant differences between groups or clinical effectiveness of internet interventions, results were mixed (10,11,14,16). The study by van Wier (30) conducting longer term follow-up once the intervention had ended found similar findings to the results identified in this review. The significant difference between intervention and control groups identified after the intervention was delivered was lost by the 2-year follow-up. Heterogeneity between included studies was evident and is a finding common in earlier reviews (41)(42)(43). Control group type appeared to impact on heterogeneity levels. Significant heterogeneity was identified between wait list/minimal face-to-face interventions and IWLPF. In contrast, heterogeneity levels between the control internetdelivered interventions receiving no personalized feedback and the IWLPF were not significant, suggesting that the addition of feedback alone did not increase heterogeneity. Low heterogeneity suggests that feedback does not explain a great deal of the variability in interventions. The results from the BCT coding of study arms illustrated the variability between control groups receiving no personal feedback and IWLPF, with interventions containing more BCTs than the control groups. However, variability was also evident between the 12 IWLPF. The variability in included BCTs and weight loss achieved made it difficult to identify why particular studies were more effective. BCT coding identified that feedback was not the sole component that was commonly incorporated within the IWLPF. Instead, it appeared that the IWLPF used similar clusters of BCTs. However, one of these was self-monitoring that is inherent to feedback in that participants would need to monitor their weight in order to gain feedback on it.
Attrition rates from previous reviews ranged from 20 to 43% (10,13,17,44). Attrition rates in this review ranged from 12 to 47% and therefore are similar to previous findings. The review identified studies not reporting on several quality assessment criteria, with only two studies perceived low risk of bias for all criteria. Previous reviews also found mixed standards for reporting of quality criteria (11). This review identified the need for further improvement on the reporting of allocation concealment and blinding.

Strengths and limitations of review
This review focused on personalized feedback in an attempt to explain differences in findings across the studies. It has illustrated how complex and variable internet weight loss interventions can be. A limitation of the review is the inability to control for all differences emerging from the different features, often leading to high heterogeneity levels identified and therefore makes comparison of internet-delivered weight loss effectiveness very difficult to investigate. As a result, the influence of personalized feedback cannot be completely isolated from other intervention components. The BCTs used within the intervention groups were not consistent. Even within the most effective studies (in terms of weight loss), BCTs were incorporated differently. However, this approach highlights the need for researchers to both describe and investigate the exact content of interventions, to both improve replicability and to help isolate the effective components of interventions. The need to try to deconstruct complex interventions into their component elements to see what are the most effective 'active ingredients' is emphasized (45).
All the studies provided personalized feedback for weight loss or behaviour change (diet/physical activity). However, two studies generated messages when participants logged onto the websites (35,36) and, one study (30), on the completion of modules or assessments. Both these participant interactions could have a potential effect on the intervention outcome; however, this does not appear to be the case with the three studies being placed in the four least effective studies when comparing mean weight loss difference between the intervention group and control group.
The lack of a set description when defining internetdelivered weight loss intervention groups was a limitation with intervention names varied greatly, e.g. remote support, enhanced group or behavioural internet therapy. This was also a problem within the control groups, e.g. variability in the use of the term usual care. Following frameworks, such as TIDieR (46), may help to maintain a minimum standard when reporting intervention descriptions. Control groups tended to be wait list or usual care. Usual care allows real-world practices to be examined in comparison to internet-delivered weight loss interventions, but these were often what could be classified as minimal face-to-face interventions.
The majority of studies had high percentages of white, female participants, which could impact on the generalisability of the findings. Three studies provided monetary incentives for the completion of assessments, which may have biased the findings in terms of retention rates and thus outcome results (22,33,34).

Implications for policy, practice and further research
Meta-analysis results identified no significant weight loss for the IWLPF at longer term follow-up (≥12 months). Long-term maintenance is essential for health benefits, and therefore, more investigation is required to examine how weight loss could be maintained across time and how internet-delivered interventions could be refined to better support weight maintenance. Further investigation into all BCTs used in each IWLPF and the relationship to effectiveness would be an important path to explore. Owing to small sample sizes within the included studies, analysing the relationship between effectiveness and BCTs could not be conducted in this review. This would be useful to examine in future research and would enable not only individual BCT impact to be investigated but also exploration of synergistic effects between clusters of BCTs and weight loss.
Human-delivered internet feedback took the form of health care professionals or researchers producing individually created responses (emails/web-based messages) to each participant, although the use of pre-scripted responses for common queries/topics could be used. This causes potential limitations of scaling up an intervention as greater resources, labour and therefore costs would be incurred. This is especially true when compared against computer-generated options available, which are less labour-intensive after initial set-up. However, human-delivered internet feedback could still be more efficient in comparison with traditional faceto-face methods as there are wider issues such as the ability to provide health care advice quicker and easier because of greater flexibility, convenience and time efficiency for both health care professional and the patient. In addition, consultants have more readily accessible patient outcome data. Therefore, human-delivered internet feedback is an important research area to investigate. One study (33) within this review compared internet feedback examining humandelivered versus computer-generated (with results favouring human-delivered feedback), but research remains limited. Further research could highlight the advantages and disadvantages both options provide. Implications for practice relate to the use of IWLPF as alternative ways to provide weight management services. Further research is needed to establish whether internet-delivered weight loss interventions provide additional benefit than in-person services in current health care practice and to identify the most effective ways of providing personalized feedback.

Supporting information
Additional Supporting Information may be found in the online version of this article, http://dx.doi.org/10.1111/ obr.12396 Table S1: Study quality assessment. Figure S1: Internet feedback versus no feedback weight loss (kg). Figure S2: Internet feedback versus no feedback 5% weight loss. Figure S3: Internet feedback versus no feedback mean waist circumference change. Figure S4: Internet feedback versus no feedback mean BMI change. Figure S5: Wait list control/minimal intervention versus internet feedback interventions mean weight loss (kg).