As technological advances pervade every aspect of daily life, there has been a corresponding proliferation in the development and implementation of technological interventions for health-related purposes. Electronic health (eHealth), the broad term for information and communication technologies deployed in health settings, is a growing area of interest as the international research community attempts to address issues facing modern health care [1]. Typically, an eHealth modality is considered to be some specific form of technology that is applied in the context of health care [2-4]. Examples of eHealth modalities include internet-based (Web-based health interventions [5-9], telephone-supported (interventions with telephone support from health practitioners) [10], interactive voice response (the use of a phone’s touch-tone keypad to provide responses to automated scripts) [11,12], virtual reality (a 3-dimensional computer-generated environment that the individual can explore, interact with, and manipulate) [13,14], videoconferencing (the use of high-quality real-time video and audio connection via online internet networks) [15], and mobile phone apps (mobile-based or mobile-enhanced programs) that deliver health-related services [16,17]. More detailed definitions of the various types of eHealth interventions are available in Multimedia Appendix 1.

The core value proposition for delivering health care via an eHealth modality is that the barriers experienced by traditional in-person treatment methods are reduced or potentially removed [4,18-21]. For instance, a Web-based eHealth intervention may improve accessibility to treatment, reduce the waiting list duration, and can be delivered more cost-effectively than in-person services [22]. For these reasons, eHealth has gained considerable traction for conditions that are long-term and where there is a shift toward self-management [23-26]. In this context, where ongoing disease management is required, eHealth interventions offer a viable and important support option. Many eHealth solutions have been developed for a variety of chronic illnesses, including diabetes [27,28], breast cancer [29], hypertension [30], cardiovascular disease [16], multiple sclerosis [31], headache [8], and chronic pain [12,14,32-38].

Chronic pain refers to pain that lasts for more than 3 months [39]. Chronic pain encompasses many diverse conditions, is highly prevalent, and is a leading cause of long-term disability [39]. Much eHealth research has been conducted in the area of chronic (noncancer) pain, and eHealth interventions have shown to be efficacious in reducing pain interference [40]. However, despite the increasing variety of eHealth modalities used for chronic pain, studies typically focus on 1 modality, and as a result, direct comparisons of modalities are rare [22]. Identifying the need to investigate the relative strengths and weaknesses of modality types, Heapy et al conducted a systematic review of eHealth self-management interventions for chronic pain, in which three modality types were evaluated, namely, telephone, interactive voice response, and internet. They concluded that each modality was effective in the context of chronic pain, but no conclusive evidence points to one being more superior than the others.

Notably, Heapy et al began the necessary steps toward ascertaining the varying efficacies of each modality as the contributing factor to intervention success. However, the authors recognized certain limitations with their review, such as the breadth of their search strategy (ie, limited to three databases) and the low number (ie, 3) of included eHealth modalities. Moreover, the review included a variety of study designs, and although they reported on the between-condition effect sizes when possible, a quantitative comparison (ie, a meta-analysis) was not conducted. Therefore, one of Heapy et al’s indications for future research was to identify the relative efficacy of modality types through direct comparison.

Although there are a growing number of eHealth interventions for chronic pain, there is a stark lack of research comparing eHealth modalities in this context. Directly comparing eHealth modalities deployed in chronic pain research could potentially yield important insights into which modalities are more efficacious in what context and for what reasons (eg, treatment fidelity, resource availability, issues with target population, typical engagement levels, and cost efficiency). Thus, from the perspectives of patient well-being, health care provision, and optimizing research interventions, there is an impetus to first identify the most effective modalities for chronic pain and to then investigate why they are the most effective.

The aim of this study was to add to the literature that concerns itself with evaluating eHealth modalities in the context of reducing pain interference for chronic pain patients by directly comparing treatment outcomes across studies that have deployed an eHealth modality. Critically, this review conducted a network meta-analysis (NMA) and quantitatively compared and ranked the eHealth modalities used for interventions in chronic pain, which has not been done before. An NMA is an extension of a meta-analysis and enables multiple treatments to be compared using direct and indirect comparisons across trials using a common comparator [41-43].

The objective of this study was to conduct a systematic review and an NMA to evaluate and compare the effectiveness of the eHealth modalities used to deliver interventions (other than drugs) for adults living with chronic noncancer pain.

The systematic review and NMA were conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the PRISMA Network Meta-Analysis extension statement (see Multimedia Appendix 2) [44]. The protocol for this study is registered with the International Prospective Register of Systematic Reviews database (registration number: CRD42016035595) [45].

The eligibility requirements for included studies in this review are outlined in Table 1. All studies included in this review were required to be published in peer-reviewed journals and available in English. The criteria were influenced by a Cochrane review of internet-delivered psychological therapies for chronic pain by Eccleston et al [49].

Studies were merged to create nodes representing the primary delivery method (eg, internet). A study was not included in the network if both arms were classified as the same modality without an additional comparator.

A total of 4 databases, Cochrane Central Register of Controlled Trials (CENTRAL; Cochrane Library), Medical Literature Analysis and Retrieval System Online (MEDLINE), Excerpta Medica dataBASE (EMBASE), and PsycINFO, were searched from inception until November 22, 2017. Necessary changes were made to adapt the search terms for different interfaces. The search strategy is detailed in Textbox 1.

The reference lists of relevant systematic reviews and of included studies were screened to identify any relevant studies. The metaRegister of Controlled Trials [50], Clinicaltrials.gov [51], and the World Health Organization’s International Clinical Trials Registry Platform [52] were also searched.

Members of the research team screened titles and abstracts to search for duplicate and nonrelevant studies; 10% of the papers were assessed in duplicate. In total, 2 review authors (SH and KF) independently screened full-text papers for inclusion. Studies were included if they (1) were randomized controlled trials (RCTs); (2) had N>20 per arm at each time point; (3) had participants with noncancer-related chronic pain; (4) were delivered via eHealth modality; and (5) measured a suitable pain outcome.

Data were independently extracted by 2 authors (BS and SH) into a preprepared excel sheet. The following items were extracted: means and SDs at postintervention for pain interference, psychological distress and HRQoL, sample size, measures, mean age, percentage of females, diagnosis, mean years of pain, method of recruitment, and presence of contact with researchers or therapists. If no SDs were reported, they were calculated from the available SEs or CIs.

In line with previous research, risk of bias within individual studies was assessed using the Cochrane Risk of Bias tool. Please see the published protocol for additional details [1]. Funnel plots and Egger tests were conducted to investigate publication bias across studies.

The network includes a node for each eHealth modality. In addition, the network contains both a control node (comprised wait list control and treatment-as-usual control groups) and an enhanced control node (eg, educational booklet).

Standardized mean differences (SMDs) between groups at postintervention and measures of uncertainty are reported. Additional summary measures such as treatment rankings and the probability of each modality arm being the best are reported.

Random-effects pairwise meta-analyses of each available comparison were run as an exploratory analysis using Stata 13 (StataCorp LLC). These analyses were carried out on both the primary and secondary outcomes: pain interference, pain severity, psychological distress, and HRQoL.

An NMA random-effects model of the eHealth modalities used to deliver chronic pain interventions with the purpose of reducing pain interference was developed in WinBUGS 14 (MRC and Imperial College of Science, Technology and Medicine). This model was based on a Bayesian framework but was created with vague priors. The NMA returned pairwise comparisons between all modalities, rankings of the modalities and assessed the probability that each modality is the best. Tests of design inconsistency [53] and loop inconsistency [54] were run using Stata 13. Node splitting was conducted on comparisons with both direct and indirect evidence [55]. Additional information is provided in the protocol [1].

As outlined in the protocol [1], the purpose of adding study-level covariates was to reduce heterogeneity by allowing the NMA to take account of additional information and minimize the differences between the studies within each modality. Covariates would be added to the model based on a reduction in the deviance information criterion (DIC). In this network, the added covariates did not have a significant effect. Sensitivity analyses investigating the influence of priors, initial values, length of burn-in, and testing convergence were carried out.

The search returned 18,470 studies (Figure 1): PsycINFO (n=1913), MEDLINE (n=5286), EMBASE (n=10,479), and CENTRAL (n=792). There were 2310 studies that were excluded as duplicates and 16,039 studies excluded on the basis of title and abstract. In total, 122 potentially eligible studies were identified and then assessed on the basis of full text. Of these, 92 studies were excluded: 51 studies were not an RCT; 13 studies had less than 20 participants per arm at each time point; 7 studies had patients with cancer-related chronic pain; 12 studies did not deliver the intervention via an eHealth modality; 3 studies did not measure an appropriate pain outcome; and 6 studies consisted of 2 arms within the same node without an additional comparator. There were 30 studies that were included in the analysis.

30 studies were included in this review. Each study arm was classified by the primary delivery method (eg, internet). Although the majority of intervention arms were compared with control arms, 1 study involved the comparison of 2 active treatments [56]. The 30 studies encompassed 61 arms: 23 internet-delivered arms [5-9,15,34,36,37,46,57-69]; 2 telephone [35,70]; 1 mobile app [71]; 2 virtual reality [14,72]; 1 videoconferencing [15]; 1 interactive voice response [12]; 25 control; and 13 enhanced controls (Table 2).

A total of 5288 participants were included in the review. There were 3005 participants randomized to interventions delivered via an eHealth modality: internet (n=2509); telephone (n=305); videoconferencing (n=65); virtual reality (n=53); mobile apps (n=47); and interactive voice response (n=26).

The risk of bias summary is presented in Table 3. In total, 18 studies were considered to have been effectively randomized, 11 studies did not provide adequate information, and 1 study did not describe randomization and was judged to be at a high risk of bias. Furthermore, 10 studies used appropriate methods of allocation concealment, 18 studies did not appropriately describe their allocation methods, and 2 studies were judged as high risk, given that allocation was not blinded from research assistants. A total of 23 studies were not at risk of detection bias; the majority of these administered their assessments online. Furthermore, 7 studies were considered unclear. Although 15 studies provided clear information on their levels of attrition, 14 of them were judged to be unclear, with many failing to report differences between completers and noncompleters, and 1 study was considered to be at high risk of bias because of statistical differences between the completers and noncompleters. In total, 28 studies reported all outcomes and were free from selective reporting bias. In addition, 2 studies were judged to be of high risk because data could not be extracted. No other sources of bias were found for the 30 studies.

The included studies indicate positive effects for interventions delivered via eHealth modalities in comparison with a control/enhanced control; 80% (24/30) of studies returned a reduction in pain interference, 69% (18/26) of studies returned a reduction in pain severity, 79% (19/24) of studies showed a decrease in psychological distress, and 67% (8/12) studies indicated an improvement in HRQoL.

Exploratory analyses were carried out on the primary outcome, pain interference, secondary outcomes, pain severity, psychological distress, and HRQoL. An NMA was conducted for the primary outcome, pain interference.

The network map in Figure 2 demonstrates the available evidence for this reduction in pain interference network. For convenience, the circular nodes are eHealth modalities and the square nodes represent the control groups.

The available evidence was used to generate the network displayed in Figure 2. The number of studies behind each direct comparison is outlined in Table 5, which also includes the percentage of contribution that each comparison made to the entire network. As expected, the internet treatment versus control comparison contributes the highest percentage (17.67%) of evidence to the network. Some of the indirect comparisons required a long pathway to be generated (eg, comparing mobile apps with telephone-delivered interventions requires the direct evidence of the internet and control nodes). The comparisons based on longer paths were communicated with less precision [41].

Given that all of the studies included in this review were randomized, the assumption of transitivity is fulfilled. A test of loop inconsistency and a Lu-Ades test of design inconsistency revealed no evidence of inconsistency (P=.85 and P=.67, respectively). Node splitting returned no evidence of inconsistency when assessing differences between direct and indirect effects.

The funnel plot (see Multimedia Appendix 4) showed no indication of publication bias with the majority of studies falling within the bands. Most studies were clustered around the zero line and have relatively large SEs.

Sensitivity analyses were carried out to assess the fit of the model. A variety of different initial values were tested; the model was run with an extended burn-in of 200,000 iterations. Both a gamma and half normal prior were used to ensure that the normal prior was uninformative, and 500,000 and 700,000 iterations were run to ensure that 600,000 were adequate. In addition, the model was run with 2 chains, and history plots showed tight iterations, indicating no evidence of nonconvergence.

Additional covariates were added to the model to explore heterogeneity. The initial NMA model returned a DIC of 12.47. When the covariates were added to the model, the DIC did not significantly reduce and they were considered not to have added enough to the model to warrant inclusion (age [DIC=13.72], gender [DIC=12.79], length of intervention [DIC=12.84], attrition [DIC=13.15], measure [DIC=13.20], contact [DIC=13.03], analysis [DIC=12.99], and condition [DIC=12.99]).

The random-effects NMA returned pairwise comparisons between each of the eHealth modalities. The majority of these comparisons were not statistically significant; however, the network indicates that all eHealth modalities were significantly better than videoconferencing. On the basis of the currently available evidence, the network also promotes the use of internet to deliver interventions. This study created a ranked list of eHealth modalities used for chronic pain by conducting a systematic review with an NMA. Study findings tentatively indicated that mobile apps and virtual reality were the most effective eHealth modalities for delivering interventions for reducing pain interference. More specifically, the joint highest ranked modalities overall, according to NMA analyses, were mobile apps, with a 43% chance that this modality delivered the most effective intervention for reducing pain interference. Following this, virtual reality had a 34% chance and telephone had a 15% chance of being the most effective delivery method. Internet-delivered interventions have a 4% chance of being the most effective at reducing pain; however, there was more certainty regarding their positioning and effectiveness as they contributed the most papers [27] to the network (comparisons including internet contributed a total of 20.78% to the network). Although the analyses revealed important insights for the potential rank order of eHealth modalities for chronic pain interventions, only tentative conclusions regarding the most effective treatment types can be drawn, as there are limitations with this review.

One limitation with this review is the disproportionate representation of different eHealth modalities included within the network. For example, of the 30 papers included in the analysis, internet was represented in 23 papers and telephone was represented in 2 papers, whereas, mobile apps, interactive voice response, and videoconferencing were each represented in only 1 paper. As a result, although we can be confident of the ranking of internet relative to the other modalities in the network, we cannot be confident of the rankings of the other modalities relative to internet. To explain further, if for example, an additional internet paper was added to this network, the modality rankings would not be anticipated to change, but if a new study based on another modality was added, then there is a chance that the modality rankings would change. However, this review is bound by the available evidence, and the current synthesis provides the first steps toward ranking which eHealth methodologies are more efficacious in the context of chronic pain.

It must also be noted that a contributing factor to the limited number of included papers and, therefore, eHealth modality types may have been the restrictive inclusion and exclusion criteria used in this study. For example, 51 studies were excluded for not being RCTs and 13 studies were excluded for not having 20 participants per arm for each time point. Therefore, had the eligibility criteria been more relaxed, arguably, more studies would have been included, allowing a larger network to be produced. However, the inclusion and exclusion criteria employed in this review followed on from a previous review in the area and the exacting criteria ensured that the included papers were of high quality and had low risk of bias [49].

Finally, because of the heterogeneous nature of intervention content, it may be contentious whether the current approach was optimal to identify the effectiveness of eHealth modalities relative to one another. For example, if each study in this review administered cognitive behavioral therapy (CBT) across each modality type, through accounting for differences in extracted variables (eg, age and gender), it could be reasonably assumed that any notable differences detected were because of the effect of the modality and not the intervention content (ie, CBT). However, although this was not the case in this network, it may also be debated that the aforementioned scenario would actually yield which eHealth modality is best for a particular treatment type (eg, CBT) and not which eHealth modality is most efficacious in the context of chronic pain. In any case, the scientific and clinical purposes of this review were to identify which eHealth modality, on the basis of the available evidence, delivers the most efficacious intervention for people living with chronic pain and not which intervention type (eg, CBT) works best with which eHealth modality.

Although there are certain limitations with this review, the findings provide support for previous research, yield tentative conclusions regarding the ranked efficacy of eHealth modalities in the context of chronic pain, and offer insight into further areas for investigation. Similar to previous research [49], the results from the exploratory meta-analysis highlight that internet-delivered interventions can reduce pain interference for people living with chronic pain. Interestingly, with regard to the results from the NMA, the 2 modalities found, albeit tentatively, to be most efficacious (virtual reality and mobile apps) are relatively new eHealth modalities compared with others in the review. The reason is not clear, but perhaps these modalities offer more immersive and convenient intervention pathways that appeal to participants.

Empowering individuals to take an active role in their own health care has been identified as a crucial factor for improving the quality of care and reducing health care costs [73-76]. This is particularly important for people with long-term health issues that require prolonged lifestyle modifications and adjustments [77-79]. Self-management of chronic/long-term illnesses through education and supportive interventions can not only decrease utilization of health care services but may also lead to improvements in clinical outcomes and overall quality of life [73]. Increasing patient engagement in health care interventions has thus become a priority for health care organizations, researchers, and policy makers. eHealth modalities offer tremendous potential to engage patients as they are flexible and can be tailored to individual patient’s needs, preferences, and circumstances [78]. However, as these technologies require actions that must be initiated and sustained by the individual, it is vital that these interventions are designed in an easily accessible and engaging manner. Perhaps as the research findings of this study tentatively show, interventions delivered via virtual reality and mobile apps could yield promising results in this area by virtue of their immersive and accessible design. In particular, research should focus on conducting interventions with mobile apps for chronic pain. With 93% of Irish consumers having access to a mobile phone [80], and a myriad of mobile apps targeting people with chronic pain (a recent review found 373 mobile apps for older adults with arthritic pain alone [81]), it is concerning that only 1 study included in this review delivered an intervention via a mobile app.

In the wider context of eHealth, there are 2 areas for future work. The first would be to replicate the synthesis of this review with different chronic conditions. The second area for future work would be to create a core outcome set for eHealth interventions, a standardized set of eHealth intervention engagement outcomes measuring, for example, fidelity, participant engagement, and user experience. Often, a treatment can have an effect in person, but this effect may not transfer to an eHealth intervention. In such instances, it is quite possible that the eHealth execution and delivery was unsatisfactory and not that the intervention content cannot be adapted to an eHealth version. A core eHealth outcome set would assist in negating such issues.

In conclusion, from both a clinical and scientific perspective, previous research has outlined a need to compare eHealth modalities in the context of chronic pain. This research is the first to use a novel statistical method, namely, NMA, to quantitatively compare eHealth modalities in this context. Similar to previous research, the results suggest that internet interventions can improve pain interference, whereas more novel modalities (ie, mobile apps and virtual reality) are most likely to be effective, but more research on chronic pain eHealth is needed. Among many areas for future research, additional research examining underutilized eHealth modalities is recommended, and a core outcome set with regard to measuring engagement within eHealth interventions in general is paramount.