Chronic diseases are long-lasting conditions that do not improve or cure completely over time. Chronic diseases are the leading cause of death worldwide. According to the World Health Organization, ischemic heart disease, stroke, and chronic obstructive pulmonary disease (COPD) are the top 3 causes, whereas diabetes mellitus (DM) is the ninth leading cause of death globally [1]. In the Global Burden of Disease study (2016), disability-adjusted life-years (DALYs) because of ischemic heart disease, cerebrovascular disease, and lower respiratory infections accounted for 16.1% of all DALYs [2]. Approximately 10% of the adult population (≥40 years) had COPD [3]. In recent decades, the disease burden has shifted sharply toward noncommunicable diseases (NCDs) and injuries [4]. Between 1999 and 2019, ischemic heart disease, diabetes, stroke, chronic kidney disease, lung cancer, and age-related hearing loss showed the most substantial absolute increase in the number of DALYs, giving rise to the largest burden of disease in older age groups. Although there are several chronic diseases, this study focused on 4 major NCDs: cardiovascular diseases (CVDs), cerebrovascular diseases, COPD, and DM.

These chronic diseases share several risk factors, including tobacco use, unhealthy diet, physical inactivity, and excessive alcohol consumption [5]. The World Health Organization also highlighted that high systolic blood pressure (BP), tobacco use, dietary risks (eg, low intake of fruits and vegetables and high salt intake), air pollution, high fasting plasma glucose, high BMI, and high low-density lipoprotein cholesterol are the major risk factors responsible for millions of deaths worldwide [6]. Over the past decades, global exposure to several highly preventable risks has risen by >0.5% annually (obesity, high blood sugar, alcohol use, and drug use); these factors contribute not only to the growing burden of NCDs but also to the risk factors for a growing number of fatalities and highlight the necessity for investments in public health [7].

In addition to having direct consequences for persons with chronic diseases, chronic physical illnesses may also distort the lives of their families. A study in the Netherlands showed that chronic diseases negatively impact their partners in good health in 4 main areas: personal life, social relations, finance, and intrinsic rewards [8].

Today, smartphone use and internet access have increased significantly, providing the potential to improve health through the use of information technology. The term digital health intervention refers to interventions delivered using digital technologies such as smartphones, websites, and SMS text messages to provide effective, cost-effective, safe, and scalable interventions to enhance health and health care and promote healthy behaviors [9]. Developing complex health service interventions involves the use of behavior change techniques (BCTs). A BCT is “an observable, replicable, and irreducible component of an intervention designed to alter or redirect causal processes that regulate behavior, that is, a technique is proposed to be an active ingredient” [10]. In the National Institute for Health and Clinical Excellence (NICE) guidelines, interventions for changing unhealthy individual behaviors, such as unhealthy diet, physical inactivity, alcohol overconsumption, unsafe sexual practices, and smoking, are recommended to use evidence-based BCTs strategies such as goal-setting, feedback, and social support [11]. A previous systematic review and meta-analysis concluded that digital health interventions using smartphones, PCs, and wearable devices combined with technologies such as software, mobile apps, and the internet improve healthy behavioral factors such as physical activity (PA), diet, and medication compliance [12].

Despite the well-established evidence of behavioral lifestyle interventions on chronic disease–related morbidity and mortality, particularly when implemented at a population level or in high-risk groups [13], evidence on the cost-effectiveness of digital-based health interventions for NCD prevention and control is inconclusive. Available studies on economic analyses of digital health interventions have shown conflicting evidence and inconsistent findings. One systematic review published in 2002 argued that telemedicine is not a cost-effective method of delivering health care [14], whereas the systematic review by Rojas and Gagnon in 2008 confirmed that telemedicine is cost-effective in general, as it reduces hospital use and improves patient compliance, satisfaction, and quality of life [15]. To the best of our knowledge, there is no cost-effectiveness study combining digital tools and behavioral changes for chronic diseases. This study aimed to determine whether digital tools are cost-effective for lifestyle behavior interventions.

This study was part of the Sustainable Behavior Change for Health Supported by Person-Tailored, Adaptive, Risk-Aware Digital Coaching in a Social Context (STAR-C) project. It is an interdisciplinary research program aimed at developing and assessing a technical platform that can be used for behavior change interventions targeting CVD prevention through digital coaching. A team of researchers from complementary fields, such as public health, social science, computer science, cardiology, and health economics, designed and implemented this project [16]. The project will run in two phases from 2019 to 2024: (1) a formative intervention design and development phase and (2) an intervention evaluation phase. STAR-C will use gender and equity lenses in all phases of the program [17].

This study assessed the cost-effectiveness of digital health interventions for risk-reduction behavior and provided evidence-based recommendations regarding the health economic evaluation for the STAR-C project.

We conducted this review following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. We used the Population, Intervention, Comparator, and Outcomes (PICO) framework to develop the review question to ensure that the relevant components of the question are well defined [18]. This review considered (1) studies that included adults with one or more of the 4 chronic diseases (CVDs, DM, COPD, and cerebrovascular diseases); (2) studies with economic evaluations using digital tools (telemedicine, mobile health [mHealth] apps, web-based, SMS text messaging, telephone consultations in combination with other digital support); and (3) studies that included behavior change interventions (quitting smoking, exercising optimally, taking a healthy diet, and reducing alcohol consumption). The comparators were no intervention, usual care, current practice, counselor-based counseling, or pharmacologic therapy. The following 4 major risk factors for chronic diseases were considered in this study: smoking or tobacco, overconsumption of alcohol, physical inactivity, and unhealthy diet (low intake of fruits and vegetables and excessive salt intake).

Studies were excluded if they were (1) systematic reviews or meta-analyses; (2) irrelevant publication types (editorials, letters, conference papers, commentary, case reports, study protocols, pilot studies, descriptive studies, and ecologic studies); (3) wrong study design (animal and in vitro trials and guidelines); (4) not published in English; (5) no information on outcomes (eg, pure economic studies without clinical or behavioral outcomes) or intervention costs (eg, those with only gross economic benefits were estimated); and (6) interventions using mass media, in addition to any deviation from PICO criteria.

This review considers both partial and full health economic evaluations. According to Drummond, full economic evaluation is defined as a comparative analysis of alternative courses of action in terms of both their costs (resource use) and consequences (effectiveness), such as cost-benefit analysis (CBA), cost-effectiveness analysis (CEA), and cost-utility analysis (CUA) [19]. Partial economic evaluations either focus solely on costs or resource use without considering costs related to outcomes or focus on both costs and outcomes without comparing alternative interventions such as cost comparison or cost analysis, cost consequence analysis, cost description, outcome description, and cost of illness study [20].

We developed search strategies based on the PICO framework to retrieve the relevant publications. Accordingly, we created 4 separate search blocks, each based on one of the 4 topics: cost-effectiveness, behavior change, digital health intervention, and chronic conditions under study. Controlled vocabulary, including Medical Subject Headings and keywords, was also used in the search to ensure that as many relevant articles as possible were identified using synonyms and truncations in every search block. We used a Boolean operator to expand, exclude, or join keywords, using “AND” and “OR.” We searched the following 4 main bibliographic databases: PubMed, CINAHL, Scopus, and Web of Science. In addition to the web-based search, we manually conducted an extensive literature search using references from retrieved articles or recent results of ongoing studies identified from the database searches. Interested readers can find the detailed search blocks and terms in Multimedia Appendix 1.

Initially, retrieved articles from the 4 databases were imported into Endnote on the web, a citation manager, where we removed duplicates before exporting the search results to Rayyan [21], a web-based platform to facilitate collaborative systematic review processes. First, we screened the titles and abstracts of all the search results, guided by our inclusion and exclusion criteria. If a paper was rejected, we recorded the reasons for exclusion. We downloaded all included articles for full-text reviews after the first screening. The full-text papers were again reviewed against the eligibility criteria (Multimedia Appendix 2). Two independent reviewers thoroughly scanned the titles, abstracts, and full texts. Reviewers then compared their independent decisions for inclusion, and disagreements during the review processes were resolved by discussion between the reviewers. The “blind on” option on Rayyan made it impossible to see the decision of another reviewer on a particular abstract, which helped reduce the risk of selection bias during screening.

We extracted data from each selected paper using a data-extraction form. These data included author, setting (country and year), inclusion and exclusion criteria, intervention and follow-up length of the study, number of participants in the intervention and control groups, economic perspective, uncertainty consideration (discounting and sensitivity analysis), outcomes, results, and type of behavioral interventions (Multimedia Appendix 3).

We appraised the quality of all included papers using the Joanna Briggs Institute criteria for economic evaluation and randomized controlled trials (RCTs; Multimedia Appendices 4 and 5). The economic quality criteria considered were the type of economic study, appropriate valuation of economic and clinical outcomes, uncertainty consideration (discounting), appropriate conclusions, and conflicts of interest. For the RCT criteria, this study considered the similarity of both groups at baseline, the same outcome in both groups, and the appropriate analysis. In terms of economic study design, this study rated full economic evaluation (CEA, CUA, and CBA) as high quality and others as low quality. A study was rated high quality if it used actual costs rather than estimated costs. The economic outcome of the study should be feasible for full economic analysis (eg, cost per quality-adjusted life-years [QALY] or DALY, cost per life-year saved, cost per clinical outcome, etc) to produce good quality. If the study period was >1 year, discounting should be included. This study used the NICE scale from the lowest to highest risk of bias to provide a qualitative appraisal [22]. The review used 10 criteria (a combination of economic and RCT criteria) for quality appraisal. Studies with ≥3 unfavorable responses (eg, no) were considered a high risk of bias. In comparison, we considered studies with 2 unfavorable responses a medium risk, and studies with 1 or no unfavorable responses were considered a low risk.

We assessed the cost-effectiveness of each study based on the cost-effectiveness threshold (CET) determined per country. Because it was impossible to determine the cost-effectiveness for partial economic studies, this study used the term cost saving or not cost saving, as stated in the respective study. The cost-effectiveness appraisal in this review was based entirely on conclusions of the respective studies.

A total of 675 papers appeared in the initial search results, of which 44 (6.5%) papers were eligible for full-text review, and 20 (3%) papers were included. In general, studies were excluded if they had no cost data, had no digital tools, had no lifestyle or behavior outcomes, had an inappropriate study design, or were study protocols (Figure 1).

Of the studies (9/20, 45%) with full economic evaluation, 7 (78%) studies concluded that using digital tools for behavior modification was cost-effective when the comparators were no intervention, usual care, counselor-delivered counseling, or pharmacologic therapies [25,29,30,35,36,39,40]; 6 (86%) studies concluded their cost-effectiveness from the health care payer perspective and 1 (14%) from the societal perspective [29]. Of the studies (11/20, 55%) with partial economic evaluations, 55% (6/11) of studies were cost-saving; 18% (2/11) of studies were inconclusive [26,41]; and 27% (3/11) of studies were not cost-saving [24,33,34].

The studies in this review used telephone, SMS text messaging, websites and software, mobile apps, and web-based video consultations as digital tools. The most cost-effective interventions (5/20, 25%) used telephone coaching, SMS text messaging, or health apps on mobile phones. Most studies (9/20, 45%) used telephones with other digital support as the tool for behavior change communication [24-26,28,33,34,36-38]. Typically, telephone interventions were provided by experienced nurses trained in motivational interviewing, but 22% (2/9) of these studies [36,38] used trained counselors and medical doctors.

Using a website to provide consultation or counseling for healthy behavior was the second most commonly used method in some studies (6/20, 30%) [23,29,32,34,40,41]. The studies involved a wide variety of health care professionals in web-based counseling. In addition, one study used email reminders to encourage exercise and incentives (key chains, athletic socks, book markers, and refrigerator magnets) to encourage active participation [32]; one study conducted in Italy used a website [45] for diabetes teleconsultation [41].

Another study used SMS text messaging for behavior change communication. Experts created automated messages that encouraged PA and a healthy diet for respective diseases and are typically sent out 3 to 5 times weekly [39]. In addition to behavior-related messages, they also reminded the patient about self-monitoring (eg, “time to check blood sugar”) [31].

The use of mobile apps, such as Moves, Vire, and Beddit, to encourage healthy behavior has been observed in 10% (2/20) of studies [30,42]. These apps were designed to integrate input from all monitoring devices, including pedometers that count steps, and the HORUS app collected pictures of the patients’ meals to provide dietary recommendations. These apps provided information to patients and create alerts for exercise [42]. Overall, 5% (1/20) of studies used video calls for internet-based visits and encouraged patients to exercise [27].

Table 2 presents the risk of bias across the selected studies. Four studies were deemed high risk [24,26,27,41], 6 medium risk [28,31-34,37], and 10 had a low risk of bias.

Nearly half of all studies (9/20, 45%) involved in this review had a potential conflict of interest because of stakeholder involvement in the analysis processes and unclear disclaimers [23,24,26,27,31-34,37]. Of these studies, 22% (2/9) had a serious risk of conflict of interest, as 1 author is the cofounder of mHealth Solutions company [31], and the other authors received consultation funds from pharmaceutical companies [34]. The remaining 55% (11/20) of studies were deemed to have no conflicts of interest. In total, 40% (8/20) of studies in this review showed cost-effective results without any conflicts of interest.

In this review, 20% (4/20) of studies [24,26,28,32] had unclear research questions regarding economic evaluation; 10% (2/20) of studies [27,41] had imprecise valuations, as they did not use actual costs in at least one of the cost categories.

In general, digital health interventions for healthy behavior in people with chronic diseases are cost-effective, as all studies with cost-effective results have a low risk of bias. Previous studies have shown that digital interventions positively affect smoking, alcohol consumption, diet, and PA [46]. However, it is impossible to know how this effect will sustain for many years, as many studies had considerably short follow-up periods. Studies on digital interventions for reducing behavioral risks of CVD in nonclinical adult populations revealed that they were effective 6 months after the end of the intervention, and the interventions lost their effectiveness after 12 months, according to a scoping review [46]. It also concluded that the shorter duration of effect was due to a shorter follow-up period and intention-to-treat analysis.

In most cases, studies in this review used <2 years as a follow-up period, and only 10% (2/20) used lifelong time horizons for economic evaluation [25,35]. Except for 10% (2/20) of studies [27,37] that used >2 years as an intervention period, most studies used parameters from the short-term effects of interventions to construct cost-effectiveness estimates and extrapolation. The results could be misleading because some behaviors could relapse, such as smoking, PA, and eating habits, which could diminish the effectiveness of the intervention, and hence extrapolation could overestimate the effects. This problem is particularly prevalent in mathematical modeling that predicts the outcomes of interventions over a person’s lifetime because their parameters of economic impacts are based on a model of behavioral changes beyond the intervention period.

Most studies (6/20, 30%) with cost-effective or cost-saving results were published after 2010 [25,29,30,35,39,40]. With technology costs likely to have decreased in the recent years, digital health intervention costs could have been higher in the studies published before 2010 than in more recent ones; therefore, the cost-effectiveness of digital health interventions could be confounded by the year of publication. In Australia, 2 studies used telephone coaching as the intervention method. One study conducted in 2009 showed that the intervention cost was Aus $570 (US $460) per participant [36], whereas the other study showed that it was Aus $33 (US $25) per participant in 2017 [35].

Only half of studies (10/20, 50%) used specific clinical indicators, such as hemoglobin A_1c level, low-density lipoprotein cholesterol, and BP in mmHg, to measure clinical outcomes concerning the program’s effectiveness. Other studies (10/20, 50%) used more general indicators, such as hospital admission rates, readmission rates, length of hospital stay, mortality rates, morbidity rates, and health-related quality of life, and interpreted reductions in these indicators as well as reduced health care resource use as evidence of improved clinical outcomes. For instance, decreased hospital admission rates or reduced outpatient visits could be due to reasons other than the effectiveness of the program. In addition, except for 15% (3/20) of studies [29,36,39] that used specific behavioral indicators, improvement in behavior or lifestyle was usually measured by clinical outcomes in most studies. These findings could be problematic in interpreting the program’s effectiveness, as the improvement in clinical outcomes may be due to pharmacologic effects (antihypertensive medication, for instance) rather than adoption of healthy behaviors.

Although UK NICE guidelines strongly recommends a societal perspective for economic evaluations, it was implemented in only 15% (3/20) of studies [29,41,42], whereas the others (17/20, 85%) used health care payer perspectives. The results of an economic evaluation could be more cost-effective when conducted from a societal perspective, partly because the inclusion of homecare costs and productivity loss owing to illness significantly impact economic benefits. Furthermore, nonhealth outcomes, such as waiting time, time to diagnosis, and improved education and reassurance, should also be considered when assessing the cost-effectiveness of an intervention program.

Some behavior change interventions are embedded in telemonitoring, tele-education, or teleconsultation services that act as internet-based visits and enhance patient self-monitoring [27,38,40,47]. As a result, physical access to health care services would be reduced, but this does not necessarily mean reduced demand because of a healthy lifestyle. Therefore, researchers should be aware of this pitfall and use more specific indicators to measure the outcomes of healthy behaviors.

Although 25% (5/20) of studies [28,35,37,40,42] concluded that their results could be generalized to other settings, this is only possible for populations with high chronic disease prevalence because none of these interventions were aimed at the entire population. Because of the need for more information from low- and middle-income countries (LMICs), evidence-based recommendations are challenging to develop; however, digital health interventions also have potential. Although all studies were conducted in high-income settings, scaling up the digital health intervention in LMICs is feasible because of the high NCD burden and high population in these countries. Labrique et al [48] discussed that scaling up the digital health interventions in LMICs is possible under 5 conditions: involvement of end user inputs, engagement of all stakeholders in the developmental process, a good technical profile (simplicity, interoperability, and adaptability), well-established policy, and availability of appropriate infrastructure for digital health. The mHealth platforms will be more effective than other eHealth platforms because mobile phone use is on the rise, and smartphone adoption and use is ubiquitous not only in high-income countries but also in LMICs [49]. In addition, a systematic review found that mHealth can significantly modify health behavior as smartphones become more accessible to underserved and minority communities [50].

Owing to the demand for remote health services resulting from COVID-19, health care systems have implemented digital health and telemedicine solutions. Although telemedicine and digital solutions cannot replace all components of the health care experience, they offer certain advantages, such as the convenience of care, technology-assisted remote interaction, and increased accessibility to care, which can be crucial in managing chronic diseases [51]. Cost-effectiveness, accessibility to specialty services, and the ability to assist in alleviating physician shortages are key benefits of telemedicine, especially during COVID-19 [52]. Although health care professionals’ attitudes toward telemedicine were influenced by factors such as self-efficacy, performance expectations, and facilitating conditions, mHealth emerged as the most preferred mode of telemedicine, enabling health care systems to be integrated into telemedicine systems during pandemics in low-income countries [53].

On the basis of the findings of this review, the following recommendations are suggested:

First, because most studies in this review sought to determine the effectiveness of digital health interventions based on clinical outcomes, economic evaluations were embedded in RCTs. Thus, most studies have many weaknesses in economic evaluations, such as not using QALY or DALY, no discounting, and no sensitivity analysis, which lead to uncertainty in decision making regarding cost-effectiveness. Moreover, this review contains no studies on LMICs, making it difficult to generalize the findings to broader regions because many LMICs have a poor infrastructure for digital health, such as an unstable internet connection. Finally, this review has limited conclusions owing to the heterogeneity of the interventions and diseases examined and the short follow-up periods. Furthermore, the heterogeneity of the results makes a meta-analysis difficult.

Digital health interventions for behavioral change among people with chronic diseases are cost-effective in high-income settings and can therefore be scaled up. Similar evidence from LMICs based on properly designed studies for cost-effectiveness evaluation is urgently needed. A full economic evaluation is required to provide robust evidence of the cost-effectiveness of digital health interventions and their potential for scaling up in the broader population.