Only randomized controlled trials (RCT), quasi-randomized controlled trials, before-and-after studies, and interrupted time series analyses were included.

Studies in which the intervention was aimed at patients or clients with a medical problem were included. Studies that included children and adolescents less than 18 years of age were excluded to create more homogeneity in the study population.

Studies in which the treatment consisted of a Web-based intervention were included. Web-based interventions were defined as all interactive Web applications accessed via the Internet or an intranet. Furthermore, we excluded studies if the intervention did not contain any aspects of health education or intention to change health-related behavior.

Studies were included only if the control intervention consisted of either usual care or a face-to-face intervention.

Studies that measured empowerment or an empowerment-related component were included. Given the absence of a generally accepted definition of empowerment and conflicting views on how to measure empowerment, we decided to initially include concepts that are often linked to empowerment. Examples of these are self-efficacy, mastery, self-control, self-esteem, perceived control, perceived competence, or involvement in the decision-making process [33].

No language restriction was applied.

Of the 14 included studies, 13 were RCTs, while the study that was reported in two publications [44,45] was a quasi-RCT. Most studies compared two study groups: an intervention group in which the treatment was a Web-based intervention and a control group receiving usual care or no care. Exceptions were the studies of Nguyen et al [59] and Gollings et al [52] in which comparisons were made between a Web-based intervention and a face-to-face intervention. Furthermore, the study by Man et al [45] consisted of four intervention arms: a control group, a computer-assisted training program, an online training program, and a therapist-administered training program. For this review, we compared the online training program with the control group. Also, the study by Warmerdam et al [50] included three interventions arms: a Web-based cognitive behavior program, a web-based problem-solving program, and a control group.

The duration of the follow-up measurements varied from 8 weeks [54] up to one year [56,57]. Also, many differences were found in the exact content of the Web-based interventions.

The number of participants varied from 15 [62] to 958 [57]. Studies differed regarding patient groups. For example, various studies included infertility patients, patients with post-traumatic stress disorder, patients with diabetes, or back pain patients. The mean dropout rate was 23% (SD 11%) after an average follow-up duration of 19 weeks (SD 15 weeks). Participants’ compliance with the intervention was not clearly described in many of the included studies.

Empowerment was explicitly measured in only two studies. Both of these used the Diabetes Empowerment Scale (DES) [63]. Although the DES is meant to measure diabetes-related empowerment, the study of Hill et al [53] showed that the DES can be adapted to other diseases. All other included studies (one or more) measured empowerment-related outcomes: 9 studies measured disease-specific self-efficacy, 3 measured general self-efficacy, 2 measured mastery, and 2 measured self-esteem.

Empowerment was measured in 2 studies with the Diabetes Empowerment Scale (DES). Homko et al [55] examined the effectiveness of an Internet-based telemedicine system that was aimed at self-management of underserved women with diabetes mellitus. Empowerment was assessed at baseline and at 37 weeks of gestation. The control group received paper logbooks, which served as a sham intervention, and in which women could monitor their blood glucose levels. In the study of Hill et al [53], the influence of an online intervention containing self-help support groups and Web-based educational tools on empowerment was examined among chronically ill women.

Because the study of Hill et al included only the 10-item “Setting and Achieving Goals” subscale from the DES, our comparison of the two studies was based on the results of this subscale alone.

Based on the GRADE approach, we downgraded the level of evidence two levels, that is, from high to low, on basis of the studies’ limitations and imprecision of the results (Table 2).

Therefore, based on 2 RCTs (combined n = 157) our results, shown in Figure 2, indicate low quality evidence that Web-based interventions had a significant positive effect on empowerment measured with the DES scale. Figure 2 shows that the SMD for these studies was 0.61 (95% CI 0.29 - 0.94). In Figure 2, the green squares indicate the individual study’s effect sizes, and the black diamond represents the pooled effect of the combined studies.

Of the 14 included studies, 9 studies provided sufficient data for calculation of an SMD for disease-specific self-efficacy outcomes. Cousineau et al [51] examined the effects of an online psycho-educational support program for infertile women. In this trial, where 190 females were recruited from US fertility centers, a trend was observed for improvement of self-efficacy levels among the women exposed to the online program compared with the controls (P = .07). Hirai et al [54] found that use of an Internet-based self-change program for traumatic event-related fear, distress, and maladaptive coping increased self-efficacy significantly (P < .01). In three large and relatively high quality studies (the average Downs and Black score was 21) by Lorig et al [56-58], disease-specific self-efficacy was significantly increased after use of online interventions compared with usual care. These studies contained interventions that dealt with management of heart-lung disease and type 2 diabetes (P = .06) [57], arthritis or fibromyalgia (P = .01) [58], and back pain (P = .02) [56]. Although the duration of the interventions used in these studies was approximately 6 weeks, improvements in self-efficacy remained even after 1 year of follow-up. In a study by Man et al [45], in which people with acquired brain injury (ABI) were able to follow a tele-analogy-based problem-solving program, it was found that after 8 weeks, self-efficacy levels increased more among patients using the program compared with the controls, but this effect was not significant. Ross et al [60] found a trend in improvement of self-efficacy in patients with congestive heart failure who used a Web-based online medical record and educational guide, compared with patients who received usual care (P = .08). In a study of Tuil et al [61], patients who had to undergo in vitro fertilization (IVF) were empowered by an interactive website that contained health information, a medical record, and communication possibilities with fellow patients and a physician. No pre/post test differences in IVF-specific self-efficacy were found between the intervention and control group in this study. Finally, a pilot study conducted by Zutz et al [62] found a higher level of exercise-specific self-efficacy after use of an interactive website that was aimed at cardiac rehabilitation. Compared with changes in the control group, this intervention effect was not significant.

According to GRADE guidelines, we downgraded the level of evidence for this outcome by one level from high to moderate based on studies’ limitations (see Table 2).

Visual inspection of the funnel plot on this outcome indicated a possibility of publication bias. Figure 4 shows that small studies, as indicated by the high standard errors [SE] (y-axis), with relatively high effect sizes, as indicated by high SMDs (x-axis), were more present in this analysis than small studies with small positive or negative effects. A possible reason for this is that small and effective (pilot) studies are more likely to be published [45,54,62]. We did not, however, downgrade the quality of evidence on basis of the funnel plot because removing these small studies from the analysis did not result in a change of the pooled estimate.

Therefore, there was moderate quality evidence from 9 studies (combined n = 2402) that Web-based interventions had a significant positive effect on self-efficacy measured with disease-specific self-efficacy scales. Figure 3 shows that the SMD of these 9 studies was 0.23 (95% CI 0.12 - 0.33).

In three studies, general self-efficacy was measured using general self-efficacy (GSE) scales [64,65]. In the study of Hill et al [53], no significant effect from the computer-based intervention was found on general self-efficacy. Likewise, in the studies of Tuil et al [61] and Zutz et al [62], the Web-based interventions did not have a significant impact on general self-efficacy levels.

We downgraded the level of evidence by two levels, from high to low, based on limitations in the studies and on basis of imprecision of the results (see Table 2).

Therefore, based on the GRADE approach, there was low quality evidence from 3 studies (combined n = 293) that there was no statistically significant difference between Web-based interventions and usual care in increasing general self-efficacy. The SMD of these 3 studies was 0.05 (95% CI -0.25 to 0.35). (See Figure 5).

One study used the Pearlin Mastery Scale [66] to measure the construct mastery, an outcome that is often linked to empowerment. In this RCT by Warmerdam et al [50], two independent Web-based interventions were tested as to their effectiveness to treat adults with depressive symptoms. The first intervention, which lasted 5 weeks, was a problem-solving treatment (PST) based on the idea that by solving people’s practical problems their depressive symptoms will improve. The second intervention, which lasted 12 weeks, was a cognitive behavioral therapy (CBT), aimed at changing people’s cognitive patterns to decrease symptoms of depression. Pre- and post treatment mastery levels between the intervention groups and a control group on a waiting list were compared. Both interventions were found to have had a significant effect on mastery, when compared with the control group. In our analysis, we combined the effects of the two interventions and compared them to the studies’ control group. At 12 weeks follow-up, average mastery levels for both interventions combined was 22.32 (SD 4.17), while the waiting list control group scored 19.37 (SD 3.75). The difference between the control group and the Web-based interventions was significant.

The mean difference [MD] was 2.95 (95% CI 1.66 - 4.24).

Because there was only one study available, according to GRADE there was low quality evidence (one study, n = 263) that Web-based interventions had a significant positive effect on mastery measured with the Pearlin Mastery scale.

Only one study included a measurement of the construct self-esteem, measured with the Rosenberg Self-esteem (RSE) Scale [67]. In the study of Hill et al [53] no significant effects were found for the intervention on self-esteem. The MD was -0.38 (95% CI -2.45 to 1.69).

For this one study (n = 120), based on the GRADE approach, there was low quality evidence that there was no statistically significant difference between Web-based interventions and usual care in increasing self-esteem.

Summarized in Table 2 are the results of the GRADE approach to judging the quality of the evidence of the studies included in this review. Since all included studies were randomized trials, we started with a high quality of evidence and downgraded, if necessary, on basis of the 5 GRADE domains.

To be able to make clinical interpretations of the reported SMDs described above, we reexpressed the pooled SMDs into MDs, by using the technique of back-translation to a familiar instrument. This technique has been described in more detail elsewhere [43]. Table 3 shows the absolute effects (MDs) on the outcomes for which SMDs were calculated, by using back-translation.

Nguyen et al [59] compared an Internet-based dyspnea self-management program for chronic obstructive pulmonary disease (COPD) patients with a face-to-face delivery of the same intervention. In this pilot study, 39 COPD patients were randomized into either one of the two intervention groups. At baseline and after a six-month follow-up period, mastery was measured. Results from this study indicated a slight but not significant advantage for the Internet-based delivery compared with the face-to-face intervention (MD = 1.20, 95% CI -1.73 - 4.13).

Because we found only one study (n = 50) in this category, there is low quality evidence that there is no statistically significant difference between Web-based interventions and face-to-face interventions in increasing mastery.

Gollings et al [52] compared an Internet and face-to-face delivery of a group body image and disordered eating intervention for women. In this 8-week program, participants were able to communicate either online or face-to-face with a therapist that moderated 8 step-by-step group sessions aimed at self-evaluation, managing social pressures and problem solving around weight, and shape and eating issues, for example. Before the intervention and at 8 weeks after the intervention, self-esteem was measured with the RSE in both groups. Although self-esteem improved after both interventions, no significant differences were found between the online and face-to-face delivery (MD = -0.10, 95% CI -0.45 to 0.25).

Again, because we found only one study (n = 39) in this category, there is low quality evidence that there is no statistically significant difference between Web-based interventions and face-to-face interventions in increasing self-esteem (MD = -0.10, 95% CI -0.45 to 0.25).

Statistically significant effectiveness of Web-based interventions on disease-specific self-efficacy was found. On the other hand, no effects were found for general self-efficacy. Because general self-efficacy refers to a broad and stable sense of personal competence, it is possible that Web-based interventions aimed at a specific target patient population are too specific to influence a stable personal characteristic (ie, a trait). Therefore, it has been recommended that for the majority of applications, perceived self-efficacy should be conceptualized in a situation-specific manner [71]. In line with this recommendation, a systematic review of Murray et al [72] found that among children suffering from a chronic disease, self-efficacy was more likely to improve after use of Interactive Health Applications compared with no intervention (SMD = 0.24, 95% CI 0.00 - 0.48).

From the results of this review, little is known about the sustainability of the effects. In most cases we included data that was measured directly after participants were exposed to the intervention. The effects that are reported, therefore, reflect a direct effect of the intervention. On the other hand, in the studies of Lorig et al [56-58], which measured disease-specific self-efficacy, data was presented only at 6-month or 12-month follow-ups, while the intervention in these studies lasted 6 weeks. Because of the high number of participants included in these studies (total weight in the comparison is 67.1%) there are some signs that the effects of these interventions remain after a longer period of time.

We calculated SMDs in this review. This means that the effect sizes presented do not represent certain improvements on specific instruments. To be able to say something about the magnitude of the effect sizes we found, we used back translations of SMDs to a familiar instrument. The results of these back calculations are shown in Table 3. We should, however, realize that these transformations from SMDs to MDs are somewhat arbitrary and should be interpreted with caution. Nevertheless, we found that the statistically significant improvements in empowerment and empowerment-related outcomes were rather small effects: empowerment measured with the DES increased 6.2% after use of the Web-based interventions and self-efficacy measured with disease-specific self-efficacy scales improved 4.2% compared with usual or no care. What the direct or indirect impacts (on clinical outcomes, for example) of these improvements are remains unknown.

Based on this review, there is some evidence that the Internet can be an effective method to increase patient empowerment. The results from this review show that Web-based interventions can be effective in increasing empowerment among patients who are, for example, suffering from diabetes, depression, infertility, or arthritis. These findings are in line with the growing literature on the effectiveness of eHealth interventions in general, and on outcomes other than patient empowerment [68,72,73]. Clinicians who are interested in empowering their patients are encouraged to refer their patients to Internet empowerment websites, if available and appealing to the patient.

The outcome empowerment usually refers to achieving self-efficacy, mastery, and control. Although many researchers underline that these constructs are closely related to the concept empowerment [33], still much is unclear about how empowerment is defined and how it should be measured. For example, recent work from Aujoulat et al [74] added that empowerment also includes a process of accepting relinquishment of control instead of solely gaining control. Either way, much work lies ahead for researchers in defining and conceptualizing the term empowerment. This will enable combining more unambiguous research outcomes and lead to better insight into the conditions under which, and the individuals for whom, Web-based interventions are effective and how effectiveness can be maximized. Furthermore, future Web-based RCTs aimed at patient empowerment should be conducted on large populations, include the intention-to-treat principle in their analysis [69] and, if applicable, use “sham” website interventions to blind participants from treatment in order to increase the quality of evidence in this field of research. In this perspective, to minimize the risk of bias, researchers are encouraged to consult quality assessment lists (such as the Downs and Black list used in this review) prior to conducting a trial.