We used the rapid evidence review approach, as it provides a streamlined alternative to standard systematic reviews. In conducting this review, we adhered to the general methodological guidance provided by the Cochrane Rapid Reviews Methods Group [18,19]. For reporting, we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement, along with the guidance for systematic reviews without meta-analysis [20,21] (completed checklist in Multimedia Appendix 1).

The inclusion and exclusion criteria were defined by using the PICOS (Population, Intervention, Comparison, Outcomes, and Study type) framework [18] Publications were included if the population was patients using digital applications. The intervention includes DiHA that are used by patients independently. There was no restriction regarding the comparison group as long as one was present. The study type is limited to trial-based or model-based full health economic studies. Publications were included if they were published between 2008 and 2023, as apps came onto the market from this period onwards. Table 1 shows the inclusion and exclusion criteria of the study.

The database search strategies consisted of search terms related to the intervention and study type. The search was divided into 2 blocks: The first block contained synonyms and generic terms for DiHA. The second block contained terms for health economic study types. The blocks were strung together using Boolean operators (AND and OR), and the respective syntax of the databases was applied. The detailed search strategies can be found in Multimedia Appendix 2.

To minimize the risk of publications being incorrectly included or excluded and to reduce selection bias, 2 reviewers (VA, JS) independently screened a random 20% sample of all records based on their titles and abstracts, while 1 reviewer (VA) screened the entire dataset [18]. Any discrepancies were resolved through consensus. The principle of (at least) 2 screeners is common practice and is recommended by the Agency for Healthcare Research and Quality, the Center for Reviews and Dissemination, the Institute of Medicine, and the Cochrane Collaboration to ensure proper application and formulation of the inclusion and exclusion criteria, as well as to prevent errors [27]. Disagreements were resolved through discussion until consensus was reached. As agreement between both reviewers was sufficiently high (above 80% raw agreement), the remaining records were screened by 1 reviewer (VA). After retrieving the full-text articles for all potentially eligible studies, each full-text article was screened for eligibility by 1 reviewer (VA). A second reviewer (JS) then screened all full-text articles excluded by the first reviewer (VA). Consensus of disagreements was again reached by discussion and by consulting additional reviewers (CH, HR). Records retrieved from databases were organized and managed using the online version of Rayyan Systems, Inc.

Data on general characteristics, methods, and outcomes of the included. EEs were extracted into a standardized Excel sheet (VA), which was piloted by 2 other reviewers (CH, HR). All extracted data were verified by the 2 additional reviewers (CH, HR). Disagreements were resolved by consensus.

To evaluate the quality of the health economic studies, 3 different assessment tools were used. These tools differ in their scope, criteria, and evaluation methods [28], and their combined use allows for a more comprehensive assessment by covering both clinical validity and economic robustness. The Cochrane risk of bias tool 2 (RoB 2) considers biases that arise at different stages of a trial. The Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) assess the reporting quality of EEs, while the Consensus Health Economic Criteria assess the methodological quality of trial-based EEs. For model-based EEs, the CHEC was extended with an additional item addressing model assumptions and validation. The following assessments were conducted independently by 2 authors.

A structured narrative synthesis was performed to synthesize the evidence in a transparent and reproducible manner. Studies were described, grouped by themes, compared, and assessed for evidence strength. EEs were differentiated by design (model-based vs trials-based) and analysis type (CEA, cost-minimization analysis, CUA, cost-benefit analysis).

The initial search in the PubMed, Web of Science, and EconBiz databases was conducted in April 2022. The search strategy of the Cochrane Library was implemented in May 2022. An update and thus a new search was carried out in the same databases in April 2023. The study selection process is presented in Figure 1. In total, 7 studies were included [34-40]. A list of articles excluded during full-text screening, along with the reasons for their exclusion, can be found in Multimedia Appendix 3.

The characteristics of the 7 EEs included are shown in Table 2. The study setting was Europe for 6 out of 7 studies, and Asia for 1 study [40]. The indications in the studies range from primary prevention to treatment. The medical specialties differ and range from urology [34,36,39], geriatrics [35], orthopedics [37], psychotherapy [38] to gynecology [40], although half of the apps can be located in the field of gynecology [34,36,39,40]. Accordingly, the active components of the apps vary widely as those are indication specific. Nevertheless, frequent indication of specific functions of the apps includes delivering medical information, sending reminders, and maintaining diary functions. The most common control group in the studies included was standard care.

Sample size ranged from 122 to 1526 patients. Almost all included evaluations are exclusively study-based, and all of them are based on a randomized controlled trial. Two of the included evaluations belong to the hybrid form and are study-based as well as model-based. In total, 3 EEs were performed from a societal perspective, 3 EEs from a payer perspective, and 1 EE from a health care provider perspective.

Only one of the studies was funded by industry [40]. Conflicts of interest were disclosed in 2 publications [34,40]. Information on funding and conflicts of interest can be found in Multimedia Appendix 4.

In summary, 7 study-based EEs were included, 2 of which incorporated a model-based component, corresponding to 7 CUAs and 3 CEAs. The time horizons considered in the study-based EEs are linked to the collection dates of the studies and range from 4 months [38] to 1 year [34,36]. Most EEs used a time horizon for effects and costs of 3 months or less. The hybrid studies used time horizons ranging from 3 months to 1 year (extrapolated from a 3 month study). One EE provided information on discounted rates. Multivariate sensitivity analysis was the most common method for calculating uncertainty (Table 3).

Table 4 provides an overview of the results from the EEs. More than half of all studies used a form of the EQ-5D questionnaire, among others, to measure the benefits [35-38]. Five of the studies used validated indication-specific questionnaires. In the study by Song and Kanaoka [40], the effect was measured through modeling (event-driven simulation). In 4 of the studies, direct medical, direct non-medical, and indirect costs were included [34,36,39,40] and in 3 of the studies, the costs considered comprised only direct medical costs [35,37,38].

For the cost-benefit ratio, 5 studies used the ICER in cost per QALY as the model output [34-36,39,40]. Pelle et al [37] and Röhr et al [38], among others, used cost-effectiveness acceptability curves to summarize the impact of uncertainty on the result of their EEs.

Overall, in more than half of the studies, the authors concluded that the app under investigation is cost-effective [34,36,39,40]. The authors of 2 of the studies concluded that the apps analyzed were not or only very unlikely cost-effective [37,38], and the authors of 1 publication concluded that the cost-effectiveness was inconclusive [35]. The additional outcomes and costs that were collected are listed in Multimedia Appendix 5.

Based on the defined inclusion and exclusion criteria, a total of only 7 full health economic studies of DiHA were identified. More than half of the included studies concluded that the app evaluated was cost-effective. Because of the broad scope of this analysis, there is a risk of oversimplification, as the included studies cover a wide range of conditions that differ significantly in their underlying mechanisms, patient populations, and intervention objectives. Grouping these diverse indications under the umbrella of DiHA does not imply a common mechanism of effect. Rather, this review provides a basis for drawing conclusions about the field as a whole by offering a first cross-indication synthesis of cost-effectiveness evidence for independently usable DiHA. However, potential publication bias should be considered, as negative results of health economic studies may not have been published [41]. Notably, only one of the included studies [40] was funded by industry. Only 2 studies used a modeling approach for EE, with one of the studies primarily based on RCT data. All other studies were trial-based EEs. The quality of the studies assessed using the CHEERS 2022 checklist and the CHEC checklist was classified as good on average. However, most RCTs whose data were used for EE were at high RoB. Overall, a good trial design is required to decide whether a given trial is a suitable vehicle for an EE; for example, it is unclear if the study design is capable of providing unbiased answers to the clinical question [42]. Applying the consensus on health economic criteria (CHEC) instrument revealed several difficulties and patterns, some of which may be attributed to CHEC not being optimized for DiHA research.

No systematic review specifically focusing on health EEs of DiHA that can be used independently by patients and without additional sensors or devices could be found. However, there are reviews that have assessed either mHealth interventions, apps in connection with other therapy components, or other adaptations. For example, Iribarren et al [24] examined the economic evidence of mHealth interventions and found that 74.3% of the included interventions were cost-effective, which is a slightly higher percentage compared to our results. Jiang et al [43] examined the cost-effectiveness of digital health interventions for cardiovascular disease and found that about half of the interventions examined achieved dominant or acceptable ICERs in terms of cost per QALY, which is lower than the proportion identified in this paper. However, whether the costs per QALY represent cost-effectiveness depends on the willingness to pay for extra QALYs, which varies in different countries. A systematic review by La Torre-Díez et al [25] examined cost-effectiveness and CUAs of telemedicine, electronic, and mobile health care systems. Their results showed a heterogeneous picture of the economic evidence. However, the authors were able to show that most of the included studies were cost-effective.

In general, there is growing interest in the topic of the cost-effectiveness of DiHA. Most investors in digital health companies recognize that demonstrating value for money is crucial for the success of digital health apps. Positive evidence on health outcomes alone may be insufficient. It’s essential to demonstrate clear, tangible benefits to the health care system to secure both reimbursement for digital health apps and support for scalable adoption [44]. Most of the included studies were published in recent years, targeting a wide range of indications and medical areas. All studies included CUAs, 3 [35-37] additionally included an evaluation based on clinical parameters and therefore CEAs. This finding is only partially consistent with the literature. Iribarren et al [24] mainly found CEAs (64%, n=25). However, according to Boehler [45], the choice of CUA offers the greatest degree of freedom for economic studies of eHealth interventions, indicating its prevalence in current app evaluation practices. All studies were based on RCTs, but one was primarily model-based [40]. Accordingly, the internal validity can be classified as high, and the external validity as low due to the lack of model-based studies and the moderate quality of hybrid studies. In the literature, a naturalistic design is usually recommended as the gold standard for eHealth interventions to enhance the generalizability to patients not involved in the study [46].

The sample sizes of the study-based evaluations were relatively small. De La Torre-Díez et al [25] concluded in their systematic review that the included studies had low participant numbers, which was seen as a major limitation of the EEs. Hazel et al [47] also address the importance of sample size, particularly regarding economies of scale. Small sample sizes may lead to lower estimated cost savings compared to larger samples and, potentially, to an underestimation of costs related to utilization, as higher utilization can increase higher provision costs. Accordingly, this issue should be considered in future evaluations of DiHA.

The time horizons considered in the study-based evaluations are primarily linked to the durations of the respective RCTs and range from 4 months to 1 year. Typically, trial follow-up periods are shorter than the period needed to fully observe the differences in health effects and use of health care resources between interventions being compared. Therefore, time horizon bias presents a significant challenge in EEs based on clinical trials [42]. In general, a longer time horizon is recommended for high-quality studies, or at least the period should be appropriate to the disease in question [48,49]. However, the rapidly evolving nature of apps and the often short periods of use might render long periods unrealistic. One possible approach to address this could be to follow up on studies or to supplement clinical study data with real-world data [50]. In the reviews by Rinaldi et al [51] and Iribarren et al [24] considerably longer observation periods were reported. However, both reviews did not exclusively focus on apps, which suggests that the choice of time horizons is related to the type of intervention.

The comparison groups of the included studies predominantly consisted of standard care or care-as-usual. The frequent use of standard care as the comparator highlights the challenges in determining an appropriate control group and indicates that health care apps tend to be seen as an addition, a delay, or a first step before other therapies. In some cases, these apps may also be used to help patients in tapering off certain therapies, serve as an escalation of treatment, or even act as an alternative to more invasive interventions, such as additional medication, surgery (eg, bariatric surgery), or rehabilitation programs [52,53]. For apps in particular, it is recommended to choose a perspective that also considers non-medical effects, thus adopting the societal perspective [16], as it was done in 3 of the included studies.

The assessment of the RoB revealed that the RCTs underlying the health economic studies should be viewed with some or high concern regarding bias. The results of RoB 2 show that there is a need for improvement in outcome measurement regarding the RoB due to open label or non-blinding. RCTs in DiHA could make better use of the digital space for blinding, but the implementation of blinding in DiHA is difficult [24,54-56]. DiHA studies also have the potential to minimize bias risks due to their digital nature, in which multiple study sites or even multinational trials are easier to implement. However, these options were not used in the included studies. Despite this, the assessment of study quality showed that the overall quality can be categorized as good.

A few limitations of this study should be noted. First, due to the inclusion and exclusion criteria focusing solely on complete health economic studies, it was not possible to investigate whether novel modeling frameworks regarding health economic studies [57] are already being applied to apps or whether further approaches exist. However, since complete health economic studies are considered to be of high methodological quality and provide a foundational overview of current modeling practices, this limitation can be addressed in future research. In addition, the inclusion criteria were strict, as only studies with indication-specific apps were included. This may have resulted in the exclusion of certain apps, highlighting the need for future studies to include broader apps.

Second, literature searches have inherent limitations. Due to the limited selection of databases, not all publications may have been identified [26]. This is particularly true for apps, which are novel interventions and may lack standardized keywords. This challenge is further compounded by the absence of an internationally standardized definition for DiHA.

Third, statements regarding the current practice of modeling should be treated with caution due to the small number of studies included. However, by conducting a systematic literature search in relevant specialized databases, it can be assumed that many published studies could be identified. Given the increasing interest in it, it is expected that many more studies will emerge in the coming years, allowing for updates to this research.

Given the current lack of high-quality, condition-specific EEs, the broad, exploratory synthesis of this study is both a necessary and valuable first step, as it serves to highlight the emerging potential as well as the methodological gaps in the field of independently usable apps. It could be shown that evaluated apps cover a wide range of conditions. Although 4 out of 7 publications concluded that the app was likely to be cost-effective, this review does not provide a basis for drawing conclusions about the field of DiHA as a whole. Rather, it shows a first cross-indication synthesis of cost-effectiveness evidence for independently usable DiHA, which is an increasingly important category of interventions, particularly in the context of reimbursement and health policy. Nevertheless, health EE is not yet a standard component of evidence generation for digital health apps. Given the increasing demands on health care systems, rising costs, and labor shortages, digital health apps present promising solutions. Still, manufacturers must demonstrate value for money to ensure digital health apps’ integration in health systems. To help innovators and manufacturers conduct health EEs and provide decision-makers with the necessary evidence, more robust information on evidence requirements, including, for instance, accepted study designs or modeling practices, is crucial.