This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Although web-based interventions for promoting health and health-related behavior can be effective, poor adherence is a common issue that needs to be addressed. Technology as a means to communicate the content in web-based interventions has been neglected in research. Indeed, technology is often seen as a black-box, a mere tool that has no effect or value and serves only as a vehicle to deliver intervention content. In this paper we examine technology from a holistic perspective. We see it as a vital and inseparable aspect of web-based interventions to help explain and understand adherence.
This study aims to review the literature on web-based health interventions to investigate whether intervention characteristics and persuasive design affect adherence to a web-based intervention.
We conducted a systematic review of studies into web-based health interventions. Per intervention, intervention characteristics, persuasive technology elements and adherence were coded. We performed a multiple regression analysis to investigate whether these variables could predict adherence.
We included 101 articles on 83 interventions. The typical web-based intervention is meant to be used once a week, is modular in set-up, is updated once a week, lasts for 10 weeks, includes interaction with the system and a counselor and peers on the web, includes some persuasive technology elements, and about 50% of the participants adhere to the intervention. Regarding persuasive technology, we see that primary task support elements are most commonly employed (mean 2.9 out of a possible 7.0). Dialogue support and social support are less commonly employed (mean 1.5 and 1.2 out of a possible 7.0, respectively). When comparing the interventions of the different health care areas, we find significant differences in intended usage (p = .004), setup (p < .001), updates (p < .001), frequency of interaction with a counselor (p < .001), the system (p = .003) and peers (p = .017), duration (F = 6.068, p = .004), adherence (F = 4.833, p = .010) and the number of primary task support elements (F = 5.631, p = .005). Our final regression model explained 55% of the variance in adherence. In this model, a RCT study as opposed to an observational study, increased interaction with a counselor, more frequent intended usage, more frequent updates and more extensive employment of dialogue support significantly predicted better adherence.
Using intervention characteristics and persuasive technology elements, a substantial amount of variance in adherence can be explained. Although there are differences between health care areas on intervention characteristics, health care area per se does not predict adherence. Rather, the differences in technology and interaction predict adherence. The results of this study can be used to make an informed decision about how to design a web-based intervention to which patients are more likely to adhere.
Web-based interventions for promoting health and health-related behaviors are seen in many variations and health care areas. According to Barak et al. [
...a primarily self-guided intervention program that is executed by means of a prescriptive online program operated through a website and used by consumers seeking health- and mental health–related assistance. The intervention program itself attempts to create positive change and or improve/enhance knowledge, awareness, and understanding via the provision of sound health-related material and use of interactive web-based components.
A web-based intervention can involve therapy that lasts for a predetermined, fixed period of time. However, it can also be a continuous program with no specific end date that supports self-management among patients with a chronic condition. It is made up of different, inseparable aspects which, according to Barak et al [
Evidence exists to support the effectiveness of web-based interventions. Research has shown these interventions to be effective in different areas of health care [
When looking at literature about adherence to a therapeutic regimen [
Adherence to web-based interventions has been the subject of research for some time. Many studies focus on whether and which respondents’ characteristics can explain variations in adherence [
Recently, two systematic reviews on the influence of intervention factors on adherence to web-based interventions were published [
Furthermore, regarding the intervention factors, both studies use an ad hoc classification of these factors without a theoretical foundation, which makes it difficult to generalize and explain the results. We consider a web-based intervention as consisting of content, interaction, and technology. And, although these aspects are inseparable, they can be looked at in a structured manner. Both earlier reviews use a classification that, in our opinion, has substantial overlap in the goals to be achieved with these aspects. For example, in the review by Brouwer [
The current study attempts to overcome these shortcomings by employing a more objective and comparable measurement of adherence to web-based interventions and a classification of technology based on persuasive technology literature.
From the field of persuasive technology we learn that technology has the capacity to be persuasive through its role as a tool, a medium, and a creator of experiences [
This study investigates whether intervention characteristics and persuasive design affect adherence to a web-based intervention. Web-based interventions are applied in various health care domains and intuitively it seems that there are differences between web-based interventions aimed at people with a chronic condition, at lifestyle change, or at mental health, because of the target group, involvement with a health care professional, and duration of the interventions. However, the underlying principles may well be the same. Therefore, from an intervention perspective, there is no absolute need to see these areas as being so different from each other that they cannot be compared. Consequently, it is interesting to see whether the preconceptions about the differences can be confirmed and whether there is added value for researchers and designers in one area to look at interventions from a different area.
Our systematic review aims to answer the following research questions: (1) What are the key characteristics of web-based interventions in terms of technology and interaction? (2) Are there any differences in intervention characteristics between web-based interventions aimed at chronic conditions, lifestyle, or mental health? (3) What percentage of participants adhere to web-based interventions? (4) Which characteristics of web-based interventions related to technology and interaction are linked to better adherence? These insights can help us understand and reduce the impact of non-adherence.
We conducted a comprehensive literature search using the following bibliographic databases: Web of Knowledge, EBSCOhost, PiCarta, SciVerse Scopus, and ScienceDirect. We used a combination of the constructs “web-based,” “intervention,” “adherence,” and “health.” For each construct, we used several keywords (see
Flow diagram of study selection.
The review is limited to studies of web-based interventions in the health care domain. The criteria used for including a study were: (1) it involved a web-based intervention for promoting health through behavioral change; (2) the web-based intervention was intended to be visited and used on more than one occasion; (3) the research included an assessment of the effect of the intervention; (4) the study reported objective, quantifiable measurements of usage for the intervention; and (5) the study was published in either English or Dutch. Exclusion criteria were as follows: (1) dropout attrition and non-adherence were indistinguishable; (2) the intervention was aimed at care providers or relatives of the “patient;” (3) the description of the intervention did not include information about the applied persuasive features of the technology; and (4) the web-based intervention was not primarily intended to be used through a computer or laptop at the user’s or patient’s home. In addition, we only included peer-reviewed, published articles.
The study selection was done in three steps. First, the titles of all retrieved articles were screened for eligibility by two authors (SK and RK). Second, the abstracts of all initially relevant articles were screened for eligibility by the same authors. Finally, the full text of all remaining publications was checked for inclusion by two authors (SK and RK or SK and JvG). In cases where the suitability of a study came into question during one of the steps, it was included in the next step. Disagreements about including the full text publication were discussed until agreement was reached. To check whether any eligible publications had been overlooked during the initial search process, the reference lists of all systematic reviews that were identified in the original search were checked to find additional publications that met our inclusion criteria.
The characteristics of all of the interventions that were included were coded by two researchers (SK and RK) using a data extraction form based on a protocol for the systematic review of eHealth technologies [
The following characteristics were coded:
The name of the intervention was recorded. If the intervention had no name, the intervention was named after the first author of the primary article about the intervention.
The targeted behavior or condition of each intervention was recorded. Furthermore, we recorded the area of health care targeted by the intervention (chronic condition, lifestyle, or mental health).
For each intervention, the studies that were used to code the characteristics of the intervention were recorded. Furthermore, we also recorded whether these studies were randomized controlled trials (RCTs) or observational studies without randomized control groups.
Intended usage was defined as the extent to which the developers of the intervention felt that the intervention should be used to achieve the desired effect ([
All reported information regarding the usage of the intervention (related to its intended usage) was collected, including the number of times the user or patient logged on and the number of modules completed ([
A percentage of adherence was calculated to enable us to compare the different interventions. We did this by calculating the percentage of participants that adhered to the intervention. For example, when the intended use of an intervention was “complete 8 modules” and 60 out of 100 participants completed 8 modules, the adherence was 60%. For each intervention that was included, we calculated one overall adherence percentage. When more studies about the same intervention yielded different adherence percentages, we calculated the overall adherence percentage using a weighted average, based on the number of participants in each study. Furthermore, when the study included a waiting list and the respondents in this waiting list received access to the intervention at a later stage, the adherence was calculated based on usage data for all participants, including the waiting list group.
The frequency of content updates for the web-based intervention for a participant was recorded. This could be based on new information being uploaded for all participants or on a new lesson becoming available for a specific participant.
The duration of the intervention in weeks was recorded.
For each intervention, we created a record indicating whether the setup was modular (ie, content is delivered in a sequential order, whereby new content is made available when the user reaches a certain point) or free (ie, all the content of the intervention is available to the user from the start).
All information about the interaction with participants was recorded ([
We recorded when interaction with the system, counselor, or peers took place through a different modality than web-based (face-to-face meeting, telephone, or SMS). An exception was made when the study protocol included a face-to-face meeting or telephone intake. This was not coded as interaction through a different modality because it was not part of the actual intervention.
The applied principles of persuasive technology within the interventions were coded according to the PSD framework of Oinas-Kukkonen and Harjumaa [
All data on each intervention was entered in SPSS version 19.0 (IBM Corporation, Somers, NY, USA), and we treated each intervention as a separate case. Descriptive data of the combined data of all included interventions on all variables were calculated using SPSS. Differences in variables between health care areas were calculated using Fisher’s exact tests (because of the small expectation values) and one-way analyses of variance. To investigate whether the characteristics of the included interventions could predict the observed adherence, we performed a hierarchical multiple linear regression analysis, using a block-wise “enter” method. The first block was related to the context of the web-based intervention and included the health care area (coded as dummy variables) and the study design (RCT vs observational), which other researchers have proposed to influence adherence or the effect of web-based interventions [
PSD framework elements coding scheme.
Principle and definition according to PSD framework [ |
Coded as element included when the web-based intervention: | Example | ||
|
|
|
||
|
Reduction | A system that reduces complex behavior into simple tasks helps users perform the target behavior, and it may increase the benefit/cost ratio of a behavior. | Specifically divides the target behavior into small, simple steps | A web-based intervention for weight management includes a diary for recording daily calorie intake, thereby dividing the target behavior (reducing calorie intake) into small, simple steps of which one is recording calorie intake |
|
Tunneling | Using the system to guide users through a process or experience provides opportunities to persuade along the way. | Delivers content in a step-by-step format with a predefined order | A web-based intervention for the prevention of depression that delivers the content in sequential lessons that can only be accessed when the previous lesson is completed |
|
Tailoring | Information provided by the system will be more persuasive if it is tailored to the potential needs, interests, personality, usage context, or other factors relevant to a user group. | Provides content that is adapted to factors relevant to a user group, or when a counselor provides feedback based on information filled out by a participant | A web-based intervention for supporting self-management among patients with diabetes provides information adapted to patients based on whether they have diabetes mellitus type I or II |
|
Personalization | A system that offers personalized content or services has a greater capability for persuasion. | Provides content that is adapted to one user (ie, the name of the user is mentioned and/or the user can adapt a part of the intervention) | A web-based intervention for increasing physical activity allows users to choose whether they want to see their weekly activity score on the home page or not |
|
Self-monitoring | A system that keeps track of one’s own performance or status supports the user in achieving goals. | Provides the ability to track and view the user’s behavior, performance or status | A web-based intervention for the treatment of alcohol dependence provides a diary to track and view daily alcohol use |
|
Simulation | Systems that provide simulations can persuade by enabling users to observe immediately the link between cause and effect. | Provides the ability to observe the cause-and-effect relationship of relevant behavior | A web-based intervention for smoking cessation includes a calculator that shows how much users will save when they quit smoking |
|
Rehearsal | A system providing means with which to rehearse a behavior can enable people to change their attitudes or behavior in the real world. | Provides the ability and stimulation to rehearse a behavior or to rehearse the content of the intervention | A web-based intervention for supporting self-management in patients with epilepsy starts each lesson with the same important exercise for stress-management |
|
||||
|
Praise | By offering praise, a system can make users more open to persuasion. | Offers praise to the participant on any occasion | A web-based intervention that aims to promote healthy nutritional habits compliments participants when they have eaten 2 pieces of fruit for 5 days |
|
Rewards | Systems that reward target behaviors may have great persuasive powers. | Offers some kind of reward when the participant performs a target behavior relating to the use or goal of the intervention | A web-based intervention for the treatment of social phobia gives points to participants when they engage in exposure exercises |
|
Reminders | If a system reminds users of their target behavior, the users will more likely achieve their goals. | Provides reminders about the use of the intervention or the performance of target behavior | A web-based intervention to support self-management among patients with rheumatic arthritis sends an automatic email message to remind the participant that the new lesson may begin |
|
Suggestion | Systems offering fitting suggestions will have greater persuasive powers. | Provides a suggestion to help the participants reach the target behavior | A web-based intervention for weight management provides low-calorie recipes |
|
Similarity | People are more readily persuaded through systems that remind them of themselves in some meaningful way. | Is designed to look familiar and designed especially for the participant | A web-based intervention for the treatment of panic disorder in teenage girls explains the exercises through a teenage girl with panic problems |
|
Liking | A system that is visually attractive for its users is likely to be more persuasive. | Is visually designed to be attractive to the participants | During the design of a web-based intervention to increase physical activity in middle-aged women, a representative group is asked for feedback on the design and their feedback is subsequently incorporated in the new design |
|
Social role | If a system adopts a social role, users will more likely use it for persuasive purposes. | Acts as if it has a social role (eg, a coach, instructor, or buddy) | A web-based intervention to support self-management among patients with migraine incorporated an avatar to guide the participant through the intervention |
|
||||
|
Social learning | A person will be more motivated to perform a target behavior if (s)he can use a system to observe others performing the behavior. | Provides the opportunity and stimulates participants to see others using the intervention or performing the target behavior | A web-based intervention for weight management provides the option, and stresses the importance, of posting physical activity self-monitoring data on the discussion board and commenting on the performance of others |
|
Social comparison | System users will have a greater motivation to perform the target behavior if they can compare their performance with the performance of others. | Provides the opportunity for participants to compare their behavior to the target behavior of other participants and stimulates them to do this | A web-based intervention for drug abuse prevention for teenagers automatically compares the response of the participant to other users of the intervention |
|
Normative influence | A system can leverage normative influence or peer pressure to increase the likelihood that a person will adopt a target behavior. | Provides normative information on the target behavior or the usage of the intervention | A web-based intervention to promote self-management among patients with COPD provides feedback on the level of physical activity of the participant by comparing it to the physical activity of well-managed COPD patients |
|
Social facilitation | System users are more likely to perform target behavior if they discern via the system that others are performing the behavior along with them. | Provides the opportunity to see whether there are other participants using the intervention | A web-based intervention for smoking cessation includes a discussion board for users of the intervention |
|
Cooperation | A system can motivate users to adopt a target attitude or behavior by leveraging human beings’ natural drive to cooperate. | Stimulates participants to cooperate to achieve a target behavior | A web-based intervention for the promotion of physical activity stimulates participants to form groups and to achieve the group goal of a certain number of steps each week |
|
Competition | A system can motivate users to adopt a target attitude or behavior by leveraging human beings’ natural drive to compete. | Stimulates participants to compete with each other to achieve a target behavior | A web-based intervention for diabetes management among children includes a leaderboard in which the children who enter blood glucose levels at the right times receive the highest place |
|
Recognition | By offering public recognition for an individual or group, a system can increase the likelihood that a person/group will adopt a target behavior. | Prominently shows (former) participants who adopted the target behavior | A web-based intervention treatment of anxiety includes a testimonial page where successful users of the intervention tell their story |
The search yielded 7345 unique titles. After title, abstract, and full-text screening, 101 articles on 83 interventions were included (
The 83 interventions that were included are presented in
When comparing the interventions of the different health care areas using Fisher’s exact tests, we find significant differences on intended usage (
When examining the persuasive technology elements that are presented in
We performed a hierarchical multiple linear regression, using a block-wise “enter” method, to explore the predictors of adherence. Variables expected to predict adherence were entered in the analysis in blocks of related constructs, as specified in the methods section. The final model explained 55% of the variance in adherence. In this model, interventions studied with a RCT design (instead of an observational study), increased interaction with a counselor, more frequent intended usage, more frequent updates and more extensive employment of dialogue support significantly predicted better adherence.
Descriptive variables of the included interventions per health care area
Variable |
|
Chronic |
Lifestyle |
Mental |
Total |
Intended usage | <= 1/month | 1 (5) | 3 (19) | 1 (2) | 5 (6) |
|
1/month – 1/week | 4 (21) | 4 (25) | 2 (4) | 10 (12) |
|
1/week | 13 (68) | 6 (38) | 40 (83) | 59 (71) |
|
>1/week | 1 (5) | 3 (3) | 5 (10) | 9 (11) |
Setup | Free | 5 (26) | 10 (63) | 1 (2) | 16 (19) |
|
Modular | 14 (74) | 6 (38) | 47 (93) | 67 (81) |
Updates | None | 1 (5) | 5 (31) | 1 (1) | 7 (8) |
|
yes, FNSa | 0 (0) | 2 (13) | 0 (0) | 2 (2) |
|
<= 1/month | 2 (11) | 1 (6) | 1 (2) | 4 (5) |
|
1/month – 1/week | 3 (16) | 1 (6) | 3 (6) | 7 (8) |
|
1/week | 12 (63) | 6 (38) | 42 (88) | 60 (72) |
|
>1/week | 1 (5) | 1 (6) | 1 (2) | 3 (4) |
Duration (weeks) | mean (sd) | 18.2 (15.8) | 29.8 (33.9)b | 11.1 (18.5) | 15.8 (18.5) |
|
Median | 11 | 17 | 9 | 10 |
Interaction with counselor | None | 2 (11) | 8 (50) | 10 (21) | 20 (24) |
yes, FNS | 3 (16) | 3 (19) | 2 (4) | 8 (10) | |
|
<1/week | 5 (26) | 3 (19) | 2 (4) | 10 (12) |
|
1/week | 7 (37) | 2 (13) | 23 (48) | 32 (39) |
|
>1/week | 2 (011 | 0 (0) | 11 (23) | 13 (16) |
Interaction with system | None | 7 (37) | 1 (6) | 14 (29) | 22 (27) |
yes, FNS | 6 (32) | 1 (6) | 3 (6) | 10 (12) | |
|
<1/week | 1 (5) | 5 (31) | 2 (4) | 8 (10) |
|
1/week | 2 (11) | 6 (38) | 14 (29) | 22 (27) |
|
>1/week | 3 (16) | 3 (19) | 15 (31) | 21 (25) |
Interaction with peers | none | 5 (26) | 10 (63) | 24 (50) | 39 (47) |
|
yes, FNS | 10 (53) | 4 (25) | 10 (21) | 24 (29) |
|
<1/week | 2 (11) | 0 (0) | 1 (2) | 3 (4) |
|
1/week | 1 (5) | 2 (13) | 13 (27) | 16 (19) |
|
>1/week | 1 (5) | 0 (0) | 0 (0) | 1 (1) |
Face-to-face | included | 3 (16) | 1 (6) | 1 (2) | 5(6) |
Phone | included | 7 (37) | 5 (31) | 17 (35) | 29 (35) |
SMS | included | 0 (0) | 2 (13) | 5 (10) | 7 (8) |
Adherence | mean (sd) | 55.3 (19.8) | 32.8 (23.0) | 54.2 (27.4) | 50.3 (26.2) |
a FNS = Frequency not specified; b Based on 13 interventions. Three interventions (23, 26, and 27) did not specify duration.
Persuasive technology in web-based interventions included in this study per health care area.
Variable |
|
Chronic |
Lifestyle(N = 16), |
Mental(N = 48), |
Total(N = 83), |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Reduction |
|
10 (53) | 10 (63) | 14 (29) | 34 (41) | .033 |
|
Tunneling |
|
17 (90) | 10 (63) | 48 (100) | 75 (90) | <.001 |
|
Tailoring |
|
16 (84) | 14 (88) | 43 (90) | 73 (88) | .814 |
|
Personalization |
|
4 (21) | 2 (13) | 3 (6) | 9 (11) | .209 |
|
Self-monitoring |
|
12 (63) | 15 (94) | 12 (12) | 39 (47) | <.001 |
|
Simulation |
|
2 (11) | 3 (19) | 2 (4) | 7 (8) | .118 |
|
Rehearsal |
|
1 (5) | 1 (6) | 0 (0) | 2 (2) | .175 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Praise |
|
0 (0) | 0 (0) | 0 (0) | 0 (0) |
|
|
Rewards |
|
0 (0) | 2 (13) | 1 (2) | 3 (4) | .134 |
|
Reminders |
|
13 (68) | 11 (69) | 37 (77) | 61 (74) | .656 |
|
Suggestion |
|
11 (58) | 4 (25) | 9 (19) | 24 (29) | .008 |
|
Similarity |
|
4 (21) | 1 (6) | 16 (33) | 21 (25) | .088 |
|
Liking |
|
2 (11) | 4 (25) | 8 (17) | 14 (17) | .561 |
|
Social role |
|
1 (5) | 0 (0) | 4 (8) | 5 (6) | .819 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Social learning |
|
5 (26) | 3 (19) | 24 (50) | 31 (39) | .044 |
|
Social comparison |
|
1 (5) | 1 (6) | 12 (25) | 14 (17) | .088 |
|
Normative influence |
|
0 (0) | 0 (0) | 1 (2) | 1 (1) | 1.000 |
|
Social facilitation |
|
14 (74) | 5 (31) | 24 (50) | 43 (52) | .046 |
|
Cooperation |
|
1 (5) | 2 (13) | 0 (0) | 3 (4) | .041 |
|
Competition |
|
0 (0) | 1 (6) | 0 (0) | 1 (1) | .193 |
|
Recognition |
|
0 (0) | 1 (6) | 2 (4) | 3 (4) | .767 |
|
|
|
|
|
|
|
a Based on Fisher’s exact test. Note: results in
Predictors of adherence in a hierarchical multiple linear regression.
Step | Variable |
|
|
|
|
1 | Constant | 0.40 | .06 |
|
<.001 |
|
Chronic | 0.04 | .07 | .07 | .55 |
|
Lifestyle | -0.17 | .08 | -.25 | .025 |
|
Study design | 0.18 | .06 | .30 | .007 |
2 | Constant | 0.25 | .09 |
|
.006 |
|
Chronic | 0.07 | .07 | -.11 | .34 |
|
Lifestyle | -0.11 | .08 | -.16 | .17 |
|
Study design | 0.16 | .07 | .28 | .014 |
|
Freq. interaction with counselor | 0.04 | .02 | .28 | .055 |
|
Freq. interaction with system | 0.01 | .02 | .03 | .79 |
|
Freq. interaction with peers | 0.01 | .02 | .05 | .63 |
|
Phone | 0.09 | .06 | .16 | .17 |
|
Face-to-face | -0.08 | .12 | -.08 | .48 |
|
SMS | 0.04 | .10 | .04 | .69 |
3 | Constant | -0.04 | .21 |
|
.85 |
|
Chronic | 0.08 | .07 | .13 | .26 |
|
Lifestyle | -0.07 | .09 | -.09 | .47 |
|
Study design | 0.18 | .06 | .30 | .005 |
|
Freq. interaction with counselor | 0.02 | .02 | .12 | .31 |
|
Freq. interaction with system | -0.02 | .02 | -.09 | .42 |
|
Freq. interaction with peers | 0.01 | .02 | .05 | .60 |
|
Phone | 0.13 | .06 | .26 | .027 |
|
Face-to-face | -0.08 | .11 | -.08 | .47 |
|
SMS | 0.02 | .09 | .03 | .81 |
|
Intended usage | 0.09 | .05 | .23 | .057 |
|
Setup | -0.15 | .11 | -.22 | .18 |
|
Updates | 0.10 | .03 | .43 | .004 |
|
Duration | -0.00 | .00 | -.06 | .63 |
4 | Constant | -0.12 | .19 |
|
.51 |
|
Chronic | 0.08 | .06 | .14 | .20 |
|
Lifestyle | -0.04 | .08 | -.01 | .96 |
|
Study design | 0.15 | .06 | .26 | .008 |
|
Freq. interaction with counselor | 0.04 | .02 | .22 | .039 |
|
Freq. interaction with system | -0.04 | .02 | -.22 | .058 |
|
Freq. interaction with peers | -0.03 | .03 | -.15 | .34 |
|
Phone | 0.05 | .06 | .10 | .37 |
|
Face-to-face | -0.10 | .10 | -.10 | .31 |
|
SMS | 0.02 | .08 | .02 | .85 |
|
Intended usage | 0.11 | .04 | .27 | .014 |
|
Setup | -0.16 | .10 | -.23 | .11 |
|
Updates | 0.09 | .03 | .40 | .002 |
|
Duration | -0.00 | .00 | -.02 | .88 |
|
Primary task support | -0.02 | .03 | -.11 | .41 |
|
Dialogue support | 0.09 | .03 | .36 | .006 |
|
Social support | 0.07 | .04 | .27 | .095 |
Note
In this systematic review, we have attempted to synthesize the combined knowledge of eHealth researchers to gain insights into the factors that affect adherence to web-based interventions in the areas of chronic conditions, lifestyle, and mental health. In this study, we viewed technology from a theoretical perspective and conceived adherence as an objective measurement that allows for comparison between different interventions.
We included 101 publications describing research into 83 interventions. Mental health interventions (n = 48) constituted the largest part of these interventions. Looking at the key characteristics of web-based interventions in terms of technology and interaction, it appears that the typical web-based intervention is meant to be used once a week, is modular in setup, is updated once a week, lasts for 10 weeks, includes interaction with the system, a counselor, and peers on the web, includes some persuasive technology elements, and results in about 50% of the participants adhering to the intervention.
However, to answer our second research question, there do appear to be differences between health care areas. Overall, lifestyle interventions are longer and less strict (more employ a free setup, less frequent intended usage, fewer updates, and less interaction) than interventions aimed at chronic conditions and mental health, which seems to result in lower adherence with lifestyle interventions. Mental health interventions follow the weekly, modular format the most, with only one intervention using a free setup. This may be explained by the difference in scope of lifestyle and mental health interventions: lifestyle interventions may be more oriented towards long-term changes, while mental health interventions are often aimed at treatment that is delivered in a short, strict format. However, interventions for a chronic condition are also aimed at a long-term change or goal, but these interventions are on average more strict than lifestyle interventions. More counselor involvement is likely to be an explanation because these interventions are often offered in a health care setting and we saw a significant difference between these areas.
Regarding persuasive technology, we see that primary task support elements are most commonly employed, especially in interventions aimed at chronic conditions and lifestyle. Tunneling, which is a technological result of a modular setup, is employed most often in mental health interventions and less frequently in lifestyle interventions. This difference is a logical result of the differences in setup between interventions in these areas. This finding is not surprising, taking into account that most mental health interventions are based on regular face-to-face therapy where psycho-education and behavior modification is usually delivered step-wise (see [
Only a mean of 1.5 out of a possible 7 dialogue support elements are employed per web-based intervention. It should be noted that we have not coded the elements that may be present in email-like messages sent by a counselor because we feel that this is part of the counselor interaction and not so much a part of the dialogue support that Oinas-Kukkonen [
Social support is widely recognized as an important strategy in behavior change [
Our third research question was about the percentage of participants that adhere to web-based interventions. We found an average adherence of 50%, which confirms that non-adherence is an issue in web-based interventions. There was a wide range in the level of adherence, with 6 interventions scoring below 10% adherence and 5 interventions scoring 90% adherence or higher. Our last research question was aimed at determining which characteristics of web-based interventions relating to technology and interaction are related to better adherence. Using a hierarchical multiple linear regression, our final model explains 55% of the variance in adherence, which, in our view, is a substantial amount that provides valuable insights into the issue of adherence.
Interestingly, the first two models (including the context of the intervention and the interaction within the intervention) were not significant. It was only when aspects relating to the format of the intervention and the technology employed were entered that the model reached significance. In the final model, an RCT, as opposed to an observational study, significantly predicted better adherence. A likely explanation is that the observational studies in our review were mainly small pilot studies and large real-life studies. Pilot studies are likely to show lower adherence rates because the interventions are not fully tested and are improved after the outcomes of the pilot are known. Real-life observational studies have been shown to have lower adherence rates, which suggests that the formal structure of a trial is important for participants to adhere to an intervention [
The frequency of interaction with a counselor was a significant predictor of adherence. This finding concurs with reviews of Brouwer [
In the final model, the frequency of interaction with the system seems to negatively influence adherence, although not significantly. This surprising finding may be explained by the fact that more interaction with the system meant, in many cases, that there was no interaction with a counselor. More frequent intended usage also predicts better adherence. This might seem counterintuitive, but might also mean that when people are expected to be more active they become more engaged with the system. Moreover, more frequent intended usage will, in many cases, lead to more frequent reminders and we know that reminders can positively influence adherence [
Finally, more extensive employment of dialogue support is related to better adherence. This outcome was predicted by the persuasive system design model [
A final comment on the model for the prediction of adherence is on the different health care areas. We see that in the first model, lifestyle interventions, as opposed to mental health interventions, predict a lower adherence, but when adding the characteristics of the interventions in the model, this predictive value is negated. It seems that the health care area
Taking into account the results of this study, it seems reasonable to not only hope for adherence, but to plan for adherence when designing web-based interventions. Although 33 studies that are included in this review state that they have planned for adherence, it is remarkable that 18 state that encouraging adherence is a task for the counselor [
Moreover, it seems valuable to look much further than the health care area for which the intervention is being designed. Although each health care area has its own demands and limitations, the different areas might learn from each other’s strong points. Lifestyle interventions, although aimed at long-term goals, might benefit from incorporating segments with a more strict format and shorter duration. Mental health interventions might be extended to aim at more long-term goals like relapse prevention. They may therefore employ a less strict format, while being aware that adherence might become a larger problem. Moreover, mental health interventions might include the primary task support elements used in chronic condition and lifestyle interventions.
Furthermore, we now have evidence that certain intervention characteristics and persuasive technology can improve adherence. It seems that expecting a certain amount of engagement from the target group can actually be helpful in promoting adherence and is something that seems to be easy to implement in new and existing web-based interventions. We must keep in mind that the effect of intended usage might also be due to a bias among the participants when only those participants who agree in advance with a high level of engagement participate in such interventions. Duration seems harder to change. Cutting an intervention into shorter segments may be enough to improve adherence, but this should be investigated further. Including and possibly increasing the frequency of interaction with a counselor seems a more costly way to improve adherence and might, therefore, be a less than optimal starting point when specifically used as a strategy to increase adherence. Increasing dialogue support using persuasive technology seems to be a more cost-effective vantage point in this respect and may even be enhanced by the increasing use of mobile technology, which seems likely to, in turn, offer a valuable platform for introducing on-the-spot reminders and feedback.
Additionally, our results can be of value for blended care (ie, a combination of online and face-to-face care) by clarifying the crucial aspects for promoting adherence in web-based interventions. When it is not possible to adapt a web-based intervention to promote adherence, it may be feasible to include a face-to-face segment in the overall intervention at a crucial stage to make up for the predicted loss of adherence.
The results of this study can be used to make an informed decision about how to design a web-based intervention that has a greater likelihood of patient adherence. It must be noted, however, that we do not advocate a so-called “technology push” where technology is introduced only for the sake of the technology and the ability to create the technology. It should always be created in close collaboration with the target audience and with a clear goal to create a viable eHealth technology [
In this study, we defined adherence as being the proportion of participants who use the intervention as it is intended to be used. By doing this, we have created an adherence measurement from objective data that is comparable between interventions. We feel that the study shows that this is a promising approach and this adherence measurement can be used for a wide variety of studies. However, to date, few studies report adherence as the measurement we have chosen to use. For review studies, this means that researchers have to define the intended use, search for the usage data that corresponded to this intended use, and then calculate the adherence. This might lead to a different interpretation of the usage data than the original authors intended. However, from our experience, we can say that as long as there is enough information on the intervention and the usage, it is feasible to calculate an objective and comparable adherence measurement. For intervention studies, we would advise researchers to at least provide the information needed (ie, intended usage and usage data related to this intended usage) to calculate this adherence measurement and, preferably, to state the calculated adherence percentage for easy comparison between interventions.
In this study, we have excluded many interventions because data about usage was absent or the usage data that was presented had no direct relationship to the intended use. For example, we excluded studies that only presented mean login data per week for all respondents and had an intended usage of once a week because these data do not show us which percentage of respondents logged in each week. This strict selection based on usage data might have introduced a bias in our included studies.
We have coded the web-based interventions included in this study based on the descriptions in the published literature. Although we have made an effort to find all the information in the published literature about each intervention, our coding was limited by the description of the interventions on paper. As is noted by other authors, the description of these interventions is varied [
Lastly, a limitation of this review might be that we have only focused on the published literature. We have not included grey literature and have therefore included little real-life adherence data. As noted by Christensen [
Overall, our results confirm the conclusions of prior studies [
The data and results from this study provide numerous points of departure for future research. To increase our understanding of the characteristics of web-based interventions and their effect on adherence, it would be interesting to compare interventions that show high adherence with interventions that show low adherence using in-depth, qualitative analyses. The positive deviance approach used by Schubart [
Keywords literature search.
Included interventions, targeted behavior or conditions, and studies.
Characteristics of, and adherence to, web-based interventions included in this study.
None declared.