The reference population in this study consisted of patients with fibromyalgia syndrome (FMS). According to the American College of Rheumatology [57], FMS is a condition characterized by chronic widespread pain and tenderness in 11 or more of the 18 specific tender point sites. Although the medical evidence is still lacking for precise diagnostic criteria for FMS, there are three major symptoms that are usually associated with the disease: pain, sleep disorders, and fatigue [58-60]. Alongside these somatic factors are many other psychological dimensions observed in FMS patients, such as anxiety, stress, and depression [61-63]. People with FMS usually have other concurrent conditions, including diabetes, high blood pressure, and back pain. FMS outcomes are generally measured with the Fibromyalgia Impact Questionnaire (FIQ) [64-66] or the Fibromyalgia Assessment Status (FAS) [67]. The FIQ includes items covering all the disease-specific domains, accounting for functional disability, job ability, pain intensity, sleep function, stiffness, anxiety, depression, and the overall sense of wellbeing. The FAS focuses on the three main characteristics of FMS: fatigue, sleep disturbances, and pain. Both of these measures have good psychometric properties, in terms of reliability and validity. For this study, we used the FIQ, as the FAS was still under development and testing. In other words, our health outcome variable was patients’ assessment of fibromyalgia impact.

To test our hypotheses, we developed a Web-based eHealth intervention for patients with FMS. This application, called ONESELF, was developed in collaboration with health professionals (rheumatologists, physiotherapists, and general practitioners), and it is fully compliant with the Health On the Net Foundation guidelines (HONcode). The HONcode prescribes guidelines on the quality of the contents and the overall usability of an application. Indeed, these intrinsic factors can play a decisive role in the ultimate effects of an online intervention, and compliance with these guidelines helps to temper—and to some extent rule out—potential biases caused by usability issues.

The application enabled asynchronous and synchronous interactions with health professionals and laypeople. The system also included static informative sections in a virtual library that provided users with relevant information on the disease. First aid and frequently asked questions sections provided brief and practical information on syndrome management. A virtual gymnasium provided patients with tailored multimedia contents on several physical exercises that constitute the wider part of the nonpharmacological treatment of FMS. Later, a section with testimonies, where patients could post their stories and read stories of other people with the same health condition, enhanced the dimension of social support. We also designed and implemented synchronous interaction via a chat room and asynchronous interaction in an online forum. Patients used these tools to communicate with the physicians and among themselves. A more detailed description of the design of the application is presented elsewhere [68,69], together with qualitative insights on the user experience with the system, which was generally considered useful, usable, and comprehensible. Since the first release of ONESELF in June 2008, more than 600 FMS patients, mostly from Switzerland and Italy (the site language being Italian), have registered.

This study used a pretest–posttest experimental design. Patients were contacted by two means: a list of patients who were members of the Associazione Fibromialgici Svizzeri Sezione Ticino (Ticino Fibromyalgia Patients Association) and patients visiting health professionals (rheumatologists and physiotherapists). Health professionals were involved in the recruitment to assure that patients received a diagnosis of FMS from a doctor. To be eligible for the study, patients had to meet a set of inclusion criteria: (1) be fluent in Italian, (2) have access to the Internet, (3) be confident in using a computer, and (4) have received a diagnosis of FMS from a doctor. Every patient who matched these criteria was given a letter briefly describing the aims of the study, together with a contact form. If interested in the study, patients had to send the form back to the research team, filled in with contact details. The research team contacted the patients by phone and email. Patients were introduced to the study and asked to register on the ONESELF website. After registration, they had to accept an informed consent statement and, finally, complete the first questionnaire. After completing the questionnaire, patients could access the website and start the navigation. After 5 months of access to the site, a second questionnaire was presented and completed. A maximum of three reminders were used to maximize response rate. By the end of the recruitment process, a total of 165 patients had agreed to participate in the study (Figure 2).

To investigate the effect of functional interactivity, we created three different versions of the ONESELF application, each implementing different enabling functions. Patients were randomly assigned to one of the three versions and blinded to the others, using a computer utility that assigned them to a randomly selected experimental condition until the conditions were equally filled. Patients in group 1 (n = 55) were given a static version of ONESELF, including only the library, the virtual gymnasium, the testimonials, and the generic sections such as the first aid, the frequently asked questions, and other common contents (eg, contacts, legal notices, and ownership disclosure). No interactive enabling tools such as the Web forum or the chat room were present in this version. This group was considered as the reference or control group. Patients in group 2 (n = 55) were given an interactive-only version of ONESELF, including the Web forum, the chat room, and the generic sections. Static sections were not implemented in this version. Patients in group 3 (n = 55) were given the full version of ONESELF, including both static and interactive components.

The rationale behind this choice goes back to the functional approach to interactivity. While the library, the virtual gymnasium, and the testimonials did not allow any input from the user other than traditional hypertextual navigation, the Web forum and the chat room enabled synchronous and asynchronous interactions. Although this kind of approach has been criticized for being too generic [15], it is useful to capture the contribution of different enabling functions [70], and it was adopted in other studies on interactivity as well (eg, [53,71-73]).

Each construct of the eHealth intervention effect model was translated into operational measures. The assessments were conducted using standardized online questionnaires. The questionnaires were pretested for face and content validity with two focus groups, with 4 health professionals and 4 patients respectively. The measures used in the study were as follows.

The main analyses were conducted using structural equation modeling techniques [77]. SPSS AMOS 18 (IBM Corporation, Somers, NY, USA) was used for the analyses. Specifically, we examined the effect of interactivity on knowledge, meaning, competence, self-determination, and impact. Because of the relatively small sample size, it was not advisable to run a test on a model including the five focal dependent variables at the same time. Thus, a structural equation model reflecting an analysis of covariance for the analysis of change was implemented and tested five times, varying the five focal dependent variables. Additionally, to determine whether a mediating effect existed, we ran an analysis of variance with the fibromyalgia impact as dependent variable and study condition as factor.

The model accounts for autoregressive effects of the focal dependent variables (knowledge and the four empowerment dimensions) at time 1 on the same variable at time 2, as well as outcomes at time 1 on outcomes at time 2. This strategy allows controlling for pretest scores and adjustment for measurement errors. To reduce sampling error, we included age and years since first diagnosis in the model as additional covariates, as well as the initial score on health outcomes. The main endogenous variable was the patients’ assessment of fibromyalgia impact.

The sample included 165 patients with FMS. Age ranged from 27 to 72 (mean 49.93, SD 9.93) years. Most patients were female (157, 95.2% women; 8, 4.8% men). The modal educational level was high school (63.6%); 43 (26.1%) had a lower educational level and 17 (10.3%), a higher. The mean time since first diagnosis was 5.7 (SD 4.8) years. At baseline, the average FIQ score was 5.98 (SD 1.82). Table 1 presents descriptive sample characteristics.

Table 2 reports the mean values of the relevant constructs of the model, as measured at pretest (T1) and posttest (T2).

Regarding usage of the application, the mean number of visits as recorded by a log file analyzer was 13.27 (SD 6.68). The mean time spend on the website was 4.8 (SD 3.2) minutes per visit. Table 3 reports some usage data related to the main features implemented in the experimental conditions. Given that these figures suggest that the application was not used very often, in a preliminary phase of the analysis, we introduced the number of visits and the time spent on the application as potential moderators of the relationships implied by the theoretical model. For the sake of brevity, we cannot report the whole preliminary evaluation, but the main result was that, at the observed level of usage, we found no significant moderating effect. Although this result does not exclude the presence of a trend (ie, that running the study for a longer period would generate significant interaction effects), it suggests that usage of the application had no strong influence on the relationships investigated in the present analysis. It must be acknowledged, however, that the scant usage of the system may understate the results of the present analysis. Under the assumption that more frequent usage should lead to larger effects, the differential impact of some features that were seldom used (eg, the chat room) might stay covered. We accounted for this in the power and effect size considerations that are detailed in Multimedia Appendix 1 [78-82].

Our hypotheses detailed the role of knowledge and patient empowerment as mediating variables of a possible effect of the availability of interactive elements in an eHealth intervention. Results on the differences between the three study groups indicate no significant difference in fibromyalgia impact (F _2,148 = 0.824, P = .44). This means that the two groups who were offered interactive elements did not benefit more from the website than the group offered only static elements. The absence of a difference renders the question of mediating variables obsolete. However, we did compute the model (and report it here) to determine at what stage the hypothesized relationships failed to emerge.

The following five steps address one of the five focal dependent variables (knowledge, and the four dimensions of empowerment). Each section first reports the results of the effects of one of the focal dependent variables on the endogenous variable, health outcomes (operationalized as patient assessment of fibromyalgia impact), and then the effects of the main independent variable on the focal dependent variable. This means that hypotheses 1 and 6 are treated in one step, as are hypotheses 2 and 7, 3 and 8, an so on. To make the presentation clearer, we summarize the results in Figure 3. All five models are available in Multimedia Appendix 2, Multimedia Appendix 3, Multimedia Appendix 4, Multimedia Appendix 5, and Multimedia Appendix 6. The five steps do not mention the autoregressive effects of the focal dependent variables at time 1 on time 2 and the endogenous variable at time 1 on time 2. All these effects were highly significant.

As predicted by the hypothesized model (hypothesis 6), knowledge at posttest significantly predicted health outcomes at posttest. On average, for every 1 unit that knowledge increased, negative health outcomes were predicted to decrease by 1.3 units, everything else being equal. We further hypothesized that interactivity enabled in the application increases patients’ knowledge (hypothesis 1). This result should reflect a significant difference between groups 2 and 3 (higher score) and group 1 (lower score). Despite the significant effect of knowledge on patient assessment of fibromyalgia impact, interactivity did not have any impact on knowledge, as the absence of significant differences between the study groups testifies. Given this result, hypothesis 1 was not confirmed.

As predicted by the hypothesized model (hypothesis 7), meaning at posttest significantly predicted assessment of impact at posttest. On average, for every 1 unit that meaning increased, negative assessments were predicted to decrease by 0.49 units, everything else being equal. We further hypothesized that interactivity enabled in the application should increase the patients’ meaning score (hypothesis 2). There should be a significant difference between the groups with (higher score) and without (lower score) interactivity. The only direct effect of functional interactivity on meaning can be observed in the significant difference between group 1 (ie, patients provided with the static version of the tool) and group 3 (ie, patients provided with the full-fledged application). However, the direction of the effect is not in as expected. Patients who had access to both static and interactive elements on the site scored significantly lower in meaning than did people with the static version only. Therefore, hypothesis 2 was not supported.

Contrary to the eHealth effect model prediction (hypothesis 8), competence at posttest did not significantly predict patient assessment of fibromyalgia impact at posttest. We also hypothesized that the interactivity enabled in the application should increase the patients’ competence score (hypothesis 3). However, interactivity did not have any such impact on competence, as testified by the absence of significant differences between the study groups. Given this result, hypothesis 3 was not confirmed.

Contrary to the eHealth effect model prediction, self-determination at posttest did not significantly predict patient assessment of fibromyalgia impact at posttest. Thus, hypothesis 9 was not supported. We also hypothesized that the interactivity enabled in the application should increase the patients’ self-determination score (hypothesis 4). No significant direct effect of functional interactivity was observed, as indicated by the absence of significant differences between the study groups. Therefore, hypothesis 4 was not confirmed.

In accordance with the eHealth effect model prediction (hypothesis 10), impact at posttest significantly predicted health outcomes at posttest. The impact of the syndrome as perceived by the patient appears to make the largest difference in health outcomes. From this experimental data, we observed that for every 1 unit impact score increased, negative health outcomes were predicted to decrease on average by 3.4 units (or a quarter of a standard deviation), which is indeed a relevant finding. Given this result, hypothesis 10 was confirmed. We also hypothesized that the interactivity enabled in the application should increase the patients’ impact score (hypothesis 5). This result should reflect a significant difference between groups 2 and 3 (higher scores) and group 1 (lower score). The only significant difference, though, was found between groups 2 and 3. Patients provided with the interactive-only application (group 2) scored significantly higher on impact than did patients who had access to both the interactive and the static elements (group 3). Hypothesis 5, consequently, was not confirmed.

Although not of primary relevance in this analysis, it is also interesting that the number of years that patients had FMS influenced their impact score. On average, for every 1 additional year of illness experience, the patients’ impact score was predicted to decrease by 0.04 units. The total effect of years of illness experience on health outcomes was 0.07, meaning that an additional year of FMS on average increased negative health outcomes by 0.07 units, and did so by decreasing the impact that patients perceived the syndrome to have on them. In other words, the lower a patient’s experienced control over the syndrome (impact score), the worse the health outcomes score will be.

Some limitations of this study should be acknowledged. In respect to interactivity, the preferred approach mostly relied on the functional perspective [18]. This choice allowed testing for the different enabling functions implemented in the eHealth intervention. However, the dimension of perceived interactivity [15,17] may play a major role as well. In this study, we assumed that the chosen functions had a different degree of interactivity without considering whether the patients perceived this difference. Ideally, both functional and perceived interactivity must be considered in an integrated and holistic model, as shown in a recent contribution to the study of this construct [83].

The construct of knowledge was largely based on the critical and integrative perspective on health literacy. However, the aggregate average score provided by the knowledge test may not fully capture its theoretical complexity. For example, in Schulz and Nakamoto’s conception, individual judgment skills should ideally be included in a full measure of integrative health literacy [26].

Regarding our methods, an objection can be advanced for the lack of a pure control group in the study phase. Patients provided with the static version of the application were used as a baseline, but we did not include patients who were not exposed to the application. This limitation was mainly linked to feasibility issues in reaching a sufficiently large number of FMS patients. In addition, it is not uncommon to provide people in the control group with some kind of neutral intervention, and this principle was applied in the study.

Despite these limitations, this study provided interesting insights into the differential effect of enabling functions implemented in eHealth interventions, providing some evidence of the impact of interactivity in the health context.

First, the combination of interactive and static features is not necessarily a turnkey solution to enhance patients’ knowledge and empowerment. For example, the presence of interactive features did not significantly improve the individual level of knowledge. When tested as stand-alone feature, however [84], the library of information as implemented in ONESELF had an impact on patients’ knowledge. Given that we measured knowledge in relation to FMS, however, this result is hardly generalizable to other health conditions. More holistic and comprehensive measures of individual knowledge would help in comparing results on this construct.

A second consideration relates to other studies [56,70,73], which found that interactivity can enhance individual efficacy and self-determination. The present study rejected this claim by testing the effects of this construct on a more holistic conceptualization of empowerment.

A third, more methodological observation is that testing individual features separately or grouped into clusters decreases the effect size of the overall application. The present analysis is not suggesting that one should limit an eHealth intervention to some functions, but rather that the choice of these features should be theory or goal driven. This is even more important considering that our results showed that the effect of the full-fledged application was generally weaker than a more focused version. It is possible that patients provided with the more complex application needed more time before mastering the system and the impact was somehow delayed. A longitudinal evaluation may provide further insights into this hypothesis. Indeed, the stability of a certain effect is rarely tested in eHealth studies [85]. This is partly due to the costs associated with longitudinal evaluations and partly linked to the high level of attrition [82] between pretest and posttest.

At the time of writing this paper, research on health communication and the Internet is rapidly evolving. While the debate on the theories and constructs that must be considered in the design and evaluation of eHealth applications is lively, several efforts are being made to define standard procedures to conduct and report empirical studies on eHealth interventions [86]. In line with the scientific debate around eHealth, an underlying theme reappeared throughout this work: the need to integrate a theory-driven approach into eHealth research. The model of eHealth effects defined and tested in this study is a first step on this route, but is far from complete, especially as the model did not hold. Future directions of this endeavor should focus on adapting, complementing, and refining the model and its evaluation. The constructs included in the model of eHealth effects posited some limitations at the theoretical level. For example, the construct of interactivity should be refined to include both functional and perceived perspectives. Although Rafaeli and Ariel [15] argued that interactivity is primarily a subjective construct, exclusion of interactivity at the functional level would not provide any insight into the mechanisms that occur within eHealth applications. A more integrative perspective such as the one recently proposed by Yoo [83], who combined medium and audience interactivity, is preferable and should be pursued to refine the model.

Along the same lines, while the general validity of the model of empowerment as proposed by Thomas and Velthouse [40] and Spreitzer [75] was assessed in the health care setting, other models of empowerment (eg, [87]) might be considered and integrated to develop an improved measure of this construct.

In general terms, a model should be as predictive and explanatory as it is parsimonious. Without overexpanding the eHealth effects model, there are some constructs that should be considered for its improvement.

The first two of these constructs relate to the literacy and knowledge dimensions. According to Schulz and Nakamoto [26], the construct of judgment skills as the ability to integrate procedural and declarative knowledge into one’s own experience should be included in the model to complement the construct of knowledge. Before integrating the judgment dimension, however, further effort should be put into translating it into operational measures. Additionally, the construct of media literacy [88] can further improve the cognitive component of the model. Rosenbaum and colleagues [89] defined media literacy as the understanding of the influences that occur between the media, producers, and users. Thus, media literacy can be considered a mediator between exposure to a certain medium and its effects on opinions and behavior. As such, together with health literacy, it can make a significant difference in the effects of eHealth interventions.

Eventually, an important construct that should be included in the model is computer self-efficacy, or one’s confidence in using a computer and gaining benefits from it [90,91]. Studies on eHealth often have generalizability issues because the reference population is composed of people who are implicitly considered skilled in using a computer. The CONSORT-EHEALTH guidelines for conducting research on eHealth stress the need to avoid this assumption and urge systematically testing individuals’ ability and confidence in using technology [86]. The construct of computer self-efficacy and its measurement can serve to control for this aspect, while providing further insights into the mechanisms that determine the effectiveness of eHealth interventions.