The number of patients living with chronic conditions continues to increase worldwide [1], and empowering these patients to manage their diseases is vital. Mobile health (mHealth) provides digital solutions to patients to help them track and manage their diseases. With the increased number of available mHealth apps to download and use [2], it is expected that the number of consumers, whether they are members of the general public, patients, or health care providers (HCPs), will continue to grow. The purpose of different types of mHealth apps vary from well-being, prevention, management, and monitoring to follow up with HCPs. Some of these apps may be potentially suitable to be prescribed to patients for the diagnosis or treatment of medical conditions. The concept of “prescribable” health apps is recently introduced to refer to health apps that are currently available and have demonstrated effectiveness [3].

Studies evaluating the effectiveness of mobile apps on health outcomes are increasing in number. Several systematic reviews have concluded that mobile apps have the potential to improve patients’ health conditions, such as diabetes [4], mental health [5], and cardiovascular diseases [6]. Governments in several countries have acknowledged the benefits of mHealth and have endeavored to meet the urgent need to accelerate the adoption of mHealth apps. Germany became the first country in the world to prescribe mobile apps. HCPs can prescribe mHealth apps, which can be reimbursed by health insurance companies [7]. In the United Kingdom, the National Institute for Health and Care Excellence has published guidance about “Sleepio,” a digital therapeutic to treat insomnia, and has recommended the use of Sleepio as a cost-saving option in comparison with sleep hygiene or sleeping pills [8]. On the basis of the results of 28 studies, it has been concluded that Sleepio is more effective than the usual treatment in reducing symptoms of insomnia in adults [9].

In a survey study, HCPs from the United Kingdom were more likely to prescribe apps if they were tagged with National Health Service approval or recommended by work colleagues [10]. The same study reported that National Health Service–approved mHealth apps were more influential than evidence-based research. In Germany, a study of physicians’ attitudes toward prescribable mHealth apps found that only one-third of the physicians intended to prescribe apps, and the rate of HCPs who had already prescribed them was lower than expected, despite the existence of regulations and facilitators from the government accelerating the mHealth app adoption among HCPs [11]. The study authors suggested that a range of factors influenced app prescribing, including gender, age, the lack of intention to prescribe, and limited apps for some specialties.

These studies shed light on various barriers to mHealth app adoption in clinical care and provide opportunities to design future behavior change interventions to improve HCPs’ evidence-based app prescription behaviors. To date, there have been no systematic or comprehensive literature reviews that compile evidence of interventions for enhancing HCPs’ app prescription behaviors. Bringing together the findings from such interventions could potentially provide policy makers and stakeholders with a better understanding of valuable strategies that can be implemented to enhance HCPs’ uptake of mHealth apps. In this review, we address this gap by identifying interventions that aim to redirect HCPs’ behavior toward prescribing or recommending apps to patients.

Several approaches are available to guide behavior change intervention designs. Among these approaches are the person-based approach [12], the British Medical Research Council’s framework on the development and evaluation of complex interventions [13], and intervention mapping [14]. Although each of these approaches offers considerable value to researchers, each concentrates on a different component of intervention development or has been criticized for lacking comprehensiveness and coherence [15].

The behavior change wheel (BCW) is used to characterize and evaluate behavior change interventions [15]. This framework provides a comprehensive approach to identifying sources of behavior and classifying them into the capability, opportunity, motivation, and behavior (COM-B) model, which represents the wheel’s hub (Figure 1). These components interact with each other to produce a change in behavior. Surrounding this is a layer of 9 intervention functions that can be selected depending on the behavioral analysis reached with the COM-B. The final layer contains 7 types of policies that one can use to deliver these intervention functions. The intervention functions are connected to behavior change techniques, which are the smallest active elements of an intervention (eg, self-monitoring, goal setting, action planning etc) [16]. Behavior change techniques used in interventions can be categorized using a taxonomy comprising 93 different techniques.

Understanding the target behavior is essential before designing an intervention. However, the BCW can also be applied retrospectively to intervention studies to identify and describe the behavior change strategies that have been used. It can also be used to improve current interventions or to introduce and evaluate an intervention that looks promising. Therefore, the interventions included in this review are classified into intervention functions of the BCW.

The study objectives were (1) to summarize and evaluate interventions aimed at encouraging HCPs to prescribe mHealth apps to patients and (2) to classify and map intervention strategies with the intervention functions of the BCW.

This study was based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines [18]. The research question was created based on the PICO framework (population, intervention, comparator, and outcome) and was defined as follows: “How do interventions designed to encourage mobile health app prescription change the practices, knowledge or self-confidence of healthcare providers?”

The population of interest was HCPs or trainees (eg, health personnel, general practitioners [GPs], physicians, clinicians, dietitians, and students of health specialties). Intervention studies to encourage mHealth app prescriptions, regardless of the design, were considered eligible. The primary outcome of interest reflected any measures of practice or behavior changes such as number of app prescriptions or self-reported or objectively measured changes in knowledge or confidence. The included studies had to be conducted in primary care settings. Studies were excluded if they were about patients’ adoption of mHealth apps or interventions to improve patients’ health outcomes. Other mHealth technologies such as wearables, mobile phone messaging, video consultations, or electronic health records (EHRs) were excluded. Studies of apps for HCPs’ medical education and training or decision support systems via mobile devices were also excluded.

A search of 4 electronic databases (MEDLINE, Scopus, CINAHL, and PsycINFO) was conducted to identify studies published between January 1, 2008, and August 5, 2022. The official start of mobile apps was chosen as 2008, as the iPhone App Store was launched that year [19]. Studies published in English and peer-reviewed papers were included. A manual search of the reference lists of eligible studies was conducted. Medical Subject Headings terms were used wherever possible to locate the relevant studies. The Boolean operators AND and OR were used to enhance the search strategy. The search string used in 2 databases is shown in Multimedia Appendix 1.

The search results were exported to the EndNote Web software (Clarivate Analytics) for screening and removing duplicates. After eliminating duplicates, screening of all titles and abstracts was independently conducted by 2 reviewers (OA and NM). The same 2 researchers reviewed the full texts of the papers identified as relevant to the objectives. In cases of disagreement, the research team discussed them and made the final decision.

The risk of bias was assessed using 2 quality appraisal tools based on the study design. The National Institute of Health’s quality assessment tool for before-and-after (pretest-posttest design) studies with no control group was used for uncontrolled before-and-after studies [20]. This tool is composed of 12 items with response options “yes,” “no,” “not reported,” and “not applicable,” and the overall quality of each study can be classified as “good,” “fair,” or “poor.” The grading was decided by the total score: 0 to 4 (poor), 5 to 9 (fair), and 10 to 12 (good).

The mixed methods appraisal tool (MMAT) was used to assess the quality of the remaining studies. The MMAT is a comprehensive tool for assessing the quality of quantitative, qualitative, and mixed methods study designs [21]. This tool begins with 2 screening questions to determine whether a research objective is clear and whether the collected data allow a research question to be answered. The remaining 5 questions assess the methodologies. A score of 0- 2 was considered low; a score of 3-4 was considered moderate; and a score of 5 was considered high.

The search strategy retrieved 3464 records. Of these, 466 (13.6%) studies were duplicates and were removed, leaving 2998 (86.5%) studies for screening. The screening of titles and abstracts excluded 2959 (98.7%) studies. Therefore, 39 (1.3%) studies were eligible for full-text screening. Another 10 records were identified through citation searching, and 6 (60%) of these were included for full-text screening. Full-text screening of the 45 studies yielded a total of 11 (24%) studies in the final review. The study selection process and the reasons for exclusion are shown in Figure 2.

The studies were published between 2015 and 2020 and were performed in the United States (6/11, 54%), Australia (2/11, 18%), the Netherlands (1/11, 9%), Catalonia (1/11, 9%), and the United Kingdom (1/11, 9%). All the included studies contained an interventional component. Of the total 11 studies, 6 (54%) were pretest-posttest design studies with no control group; 3 (27%) were mixed methods studies; 1 (9%) was a usability study; and 1 (9%) was a qualitative description study.

The studies varied regarding study participants; some were focused on HCPs in primary care settings [23-25], whereas others targeted specific specialties, such as dietitians [26,27], behavioral health providers [28,29], providers in weight management clinics [30], clinical nutrition and physician assistant students [31], or interdisciplinary groups [32,33]. Most studies (8/11, 73%) had a sample size ranging from 5 to 40 HCPs, apart from 3 (27.3%) studies that reported the results of interventions conducted over multiple years or many training sessions in which the sample size ranged from 78 to 760 [28,29,31]. The functions of the mHealth apps used in these studies included weight management [27,30] or diet and activity tracking [26,31]. A total of 45% (5/11) of the studies used a list of approved apps for a range of health conditions [23,25,28,29,32], and 9% (1/11) used an app for chronic obstructive pulmonary disease (COPD) [33]. An overview of the characteristics of the individual studies included in the systematic review is provided in Multimedia Appendix 2 [23-33].

The most common changes resulting from the intervention were self-reported changes in knowledge, confidence, or self-efficacy [28,30-33]. Only 1 (1/11, 9%) study reported HCPs’ intention to use or recommend the app [33], and 1 (1/11, 9%) study evaluated app acceptance and use in dietetic care [27]. However, the outcomes were objectively estimated using the change in the number of apps prescribed in 36% (4/11) studies [23-25,32].

Using the National Institute of Health quality assessment tool, the studies by Armstrong et al [28,29], Byambasuren et al [25], Chen et al [26], and Rodder et al [31] were assessed as having a moderate risk of bias. The study by Al-Lami et al [30] was assessed as having a high risk of bias owing to a lack of clarity around several categories that were not reported, particularly regarding the selection and eligibility of participants. In all studies (6/11, 54%), it was difficult to determine whether the researchers were blinded to the intervention. An additional limitation of many of the included studies was their small sample size. Of the 11 studies, this was particularly the case for 2 studies (18%): one with 5 participants [26] and the other with 6 [30]. Most studies (5/6, 83%) lacked power calculations to accommodate the consequences of participants dropping out, thus resulting in missing values in the postintervention measurements, with the exception of Byambasuren et al [25]. The studies by Armstrong et al [28,29] and Byambasuren et al [25] did not present P values to compare outcome measures at pretest and posttest. The details of the quality assessment are presented in Multimedia Appendix 3 [23-33].

Using the MMAT, the qualitative description study was deemed high quality [27]. Out of the 3 mixed methods studies, 1 (33%) was judged to be of low quality [27] and 2 (67%) were of moderate quality [24,33]. The independent assessment of their quantitative and qualitative components lowered the methodological quality of all mixed methods studies. The quantitative study was deemed to be of low quality [32] owing to concerns regarding the poor description of patient selection and the high nonresponse rate. However, all studies clearly stated the objectives, interventions, and outcomes. The follow-up periods were generally short (immediately after the intervention up to 4 months after the intervention). More details on the quality assessments for each study are shown in Multimedia Appendix 3.

The results reported by the interventions included were generally positive, with improvements being seen in several outcomes such as changes in current practices, increased knowledge, or improved self-efficacy. Although the levels of significance for the included outcomes varied, only 1 (9.1%) study found that the interventions had nonsignificant results [27].

The included studies were analyzed using the BCW framework. By doing so, the framework allowed an examination of which intervention functions are most commonly applied in the context of mHealth app prescriptions. Given that some strategies can be classified as falling into >1 category, that is, multiple intervention functions, it was difficult to link outcomes to a single intervention function. Therefore, the following sections report how the intervention strategies fit within the BCW’s 9 intervention functions. The intervention functions, outcomes, and main findings are reported separately in Table 1.

We looked at the most often used intervention functions in the studies with successful outcomes to gain a better understanding of why the included intervention functions provided substantial changes. As seen in Table 1, successful interventions used a variety of intervention functions (environmental restructuring, education, and training). These intervention functions cover most components of the COM-B model, which suggests that interventions are more likely to produce effective results and make changes in current behavior if they target a wide spectrum of the COM-B model’s components. For example, the study by Chen et al [26] used all COM-B components to increase dietitians’ app use behaviors through dietitians’ education and skill training as well as environmental restructuring by providing physical app-based infrastructure. Modeling, coaching, and improving self-efficacy were also addressed. Therefore, the reported change in ratings of dietitians’ self-efficacy when using mHealth apps was significant (P=.01). The Tukey post hoc test revealed significantly higher post–workshop mHealth app self-efficacy ratings compared with the baseline (P=.02), and the ratings were sustained at 12 weeks (P=.01).

In addition, the intervention functions used in the ineffective study were linked to only 2 COM-B model components (physical capability and physical opportunity), suggesting a need to understand the target behavior by collecting data from multiple sources to ensure successful behavior change.

This review included 11 studies investigating the effects of interventions aimed at encouraging app prescription behavior among HCPs. Most studies demonstrated positive findings for outcomes such as self-confidence, knowledge of mHealth apps, and number of app prescriptions. The studies included differed widely in terms of interventions, measures of practice change, types of mHealth apps used, modes of delivery, and study settings.

A broad range of interventions, all related to methods for enhancing mHealth app adoption, specifically mHealth prescribing, was covered in this review, including education (workshops, class lectures, individual sessions with a dietitian or research team, videos, or toolkits), persuasion (reminders or verbally about capabilities), incentivization (expectation to pass the course), training (case studies or scenarios or app appraisal tools), modeling (site champions or observing peers), enablement (support), and environmental restructuring (lists of apps, technological systems, time, and resources).

More than half of the interventions included in this review had training (8/11, 72.7%) or educational (9/11, 81.8%) functions that targeted HCPs’ capabilities (physical or psychological). A lack of knowledge and awareness of available apps are major barriers reported in the literature [34-36]. A study reported that HCPs consider the lack of knowledge of available apps that have proven their effectiveness in improving patient health outcomes an important barrier to prescribing [35]. This confirms the urgent need to provide training programs or educational sessions regarding the available mHealth apps that support patient self-management of long-term conditions.

However, the terms training and education are often used interchangeably. Some studies reported training as part of the intervention, but these studies included only elements of imparting knowledge and understanding, not skill development [24,28]. One possible explanation is the lack of attention to different types of training in behavior change interventions. Hence, there is difficulty in perceiving what training might entail. Therefore, it is vital to distinguish between the 2 terms. This overlap has also been reported between other intervention functions. Another study reported the existence of some overlap when providing education to participants because it serves a persuasion function at the same time [37]. This is mainly because education may induce positive feelings toward app prescription.

The lack of coercion (defined as “creating expectation of punishment or cost”) may be because it was not deemed a suitable approach for improving self-efficacy or confidence in app prescriptions or it may have been deemed counterproductive when trying to create positive attitudes or encourage app prescriptions. Furthermore, it is impractical to penalize HCPs for not prescribing mHealth apps.

None of the interventions adopted restrictions such as rules to increase HCP app prescriptions or decrease any competing behavior. The absence of regulations that ensure apps’ highest quality and accuracy and the lack of data validity and reliability of existing apps keep HCPs from prescribing apps to their patients [38,39]. To compensate, some interventions have provided study participants with a list of trustworthy apps to use when making app recommendations to patients. Other interventions adopted digital solutions such as building software or electronic systems containing approved apps. Removing such barriers by providing practical resources is a form of environmental restructuring.

Minimizing the disruption of HCPs’ time is a form of enablement reported in only one intervention [32]. HCPs’ concerns about time to discuss and instruct patients on how to use apps were reported as barriers to app prescription. In a pilot study, participants were concerned that recommending apps to patients would lengthen the duration of consultations [23]. In France, a qualitative study presented a theme after interviews with GPs about “Doctor Protection,” which mainly introduced concerns about increased workload and prescriptions of apps as an additional task [40]. One significant distinction between apps and medications is that many drugs can be prescribed with simple directions, whereas an app may require more specific instructions.

One way to minimize physicians’ workloads is to integrate and synchronize health information produced from mHealth apps to patients’ EHRs. By doing this, the physician’s ability to access patient data is centralized. In an acceptability and feasibility study that examined integrating patient data generated from smartphones into EHRs [41], clinicians reported that by using the graph feature, they could evaluate longitudinal data during consultations, which was quick and easy. When compared with retrieving information by recording histories, this was thought to be a possible time saver. Furthermore, this approach provided an accurate reflection of disease changes and treatment responses.

This review has several strengths and limitations. This is the first review that addresses interventions to improve HCPs’ confidence and capabilities to prescribe or recommend mHealth apps to patients. We consider the included studies to be a complete set of studies from 2008 to 2022. The studies were sourced from a variety of electronic databases, with the reference lists of the included papers checked for potentially relevant studies. This systematic review used a robust methodology that included screening all the studies for relevance by 2 independent reviewers.

However, the number of studies included in this review was limited, and the findings depended on the quality of the included studies. Of the 11 studies, 3 (27.3%) studies were found to be at high risk of bias, 7 (63.6%) at moderate risk of bias, and only 1 (9.1%) at low risk of bias (see Appendix 3). A total of 6 (54.5%) studies were pretest-posttest design interventions; these are known for their methodological issues such as selection bias and short durations, which do not make it possible to determine whether the intervention is effective and sustainable [42]. Furthermore, 7 (63.6%) studies used self-reported data to reflect possible changes in behaviors. With self-reported outcomes, it is impossible to tell whether the reported change in knowledge or practice is owing to response-shift bias or an actual adoption of the targeted behaviors. This emphasizes the importance of using objective measures instead. Objective measures such as the number of app prescriptions and OSCE scores were reported in 4 (36.4%) studies. However, the assessment of behavior changes or intentions to change app prescription behaviors cannot always be performed using objective measures. This is the case, in particular with interventions that lack technical systems to track the changes in the number of prescriptions before and after the intervention. Combining objective and self-reported measures could bring more insight and different perspectives to the study findings.

Several gaps in the research were identified in this review. Coercion and restriction were not reported in any of the interventions included; however, this may be because they were not deemed appropriate approaches for changing app prescription behaviors in this population. The impact of other forms of intervention functions not used in any of the interventions reviewed could be explored, such as the use of incentives (financial or nonfinancial) that, if appropriately applied and supported with other intervention functions, could potentially make an impact and encourage HCPs to prescribe apps [43]. Evidence on the acceptability and impact of such programs in the context of mHealth is lacking. This can be answered with future studies.

High-quality studies with adequate sample sizes and longer study periods are now essential for detecting differences in app prescription behaviors. Future interventions could adopt theoretical frameworks and behavior change frameworks to systematically understand HCPs behavior toward the prescription of mHealth apps.

The COM-B model, as part of the BCW, is a useful tool to make behavioral diagnoses and identify what needs to change [22]. Evidence from successful interventions, interviews, or surveys with HCPs about what motivates or limits their mHealth app prescription behaviors can provide sufficient information to understand the sources of those behaviors. Therefore, future interventions can address the target behavior (app prescriptions) and its influencing factors.

This study identified interventions aimed at improving HCPs’ app-prescribing behaviors. On the basis of the BCW, environmental restructuring and education were the most frequently used intervention functions in the included studies, followed by training. The findings of this study provide evidence that combining elements of training, education, and environmental restructuring is more likely to produce effective changes in HCPs’ behavior toward app prescribing.