Keywords

1 Non-communicable Diseases as Health and Economic Challenge

With 73%, non-communicable diseases (NCDs) such as cardiovascular diseases , chronic respiratory diseases, cancer, and neurological or mental disorders were the leading cause of death in 2017 (Roth et al. 2018).Footnote 1 Equally alarming is that deaths caused by NCDs increased by circa 23% from 2007 to 7.6 million in 2017 (ibid.). Even worse, one out of three adults suffers from multiple NCDs (Marengoni et al. 2011) with disproportionally serious consequences for mortality, cognitive and physical functioning, as well as quality of life (Newman et al. 2020).Footnote 2

NCDs also lead to a significant economic burden. Data from 18 high-, middle- and low-income countries indicate that NCD-households spent on average a significantly higher percentage of their effective income on health care than non-NCD households, for example, 16% vs. 6% in China (Murphy et al. 2020). Moreover, NCDs led to 90% of health-care spending in the United States in 2014 (Buttorff et al. 2017), and recent findings show that four major NCDs (cardiovascular and chronic respiratory diseases, cancer, and diabetes) are responsible for a GDP loss of circa 2% in Europe (Vandenberghe and Albrecht 2020). Projections are alarming, too. For example, the total economic burden of NCDs in 2015–2050 in the United States is estimated to sum up to circa US$ 250,000 per capita with mental disorders and cardiovascular diseases being the two most contributing NCDs (Chen et al. 2018). With the increasing number of elderlies suffering from multiple NCDs, health expenditures sometimes increase exponentially with each additional condition (Hajat and Stein 2018) as the management gets more complex and requires health resources and specialists from various disciplines (Newman et al. 2020). So are 12% of Americans affected by five or more NCDs but contribute 41% of total health-care spending (Buttorff et al. 2017).

To address this problem, health interventions must target behavioral risk factors, for example, poor nutrition, physical inactivity, tobacco use, alcohol consumption, and metabolic risks such as obesity, high cholesterol, high blood pressure, or high blood glucose levels (WHO 2020). The prevention and management of NCDs require therefore an intervention paradigm that focuses on health-promoting behavior in our everyday lives (Katz et al. 2018; Kvedar et al. 2016). However, a corresponding change in lifestyle is only implemented by a fraction of those affected (Katz et al. 2018; Renders et al. 2000), and health coaching delivered by human experts is neither scalable nor financially sustainable.

To this end, the question arises whether digital health interventions are appropriate means to address the health and economic burden of NCDs. Digital health interventions rely on information and communication technologies and allow medical doctors and other caregivers to scale and tailor long-term treatments to individuals in need at sustainable costs (Fleisch et al. 2021; Kowatsch et al. 2019). Examples include interventions delivered via chatbots (Kramer et al. 2020), voice assistants (Bérubé et al. 2021), cars (Koch et al. 2020), smartphones (Shih et al. 2019), smartwatches (Maritsch et al. 2020), holographic digital coaches (Kowatsch et al. 2021), or health-care robots (Papadopoulos et al. 2020).

An increasing amount of technology-based digital health innovations led to investments in digital health startups that were almost continuously growing since 2011 (RH 2020; see Fig. 1). The global telehealth market size is projected to increase from US$ 61 billion in 2018 to US$ 560 billion in 2027 (FBI 2020) with companies such as American Well , Babylon Health , Kry , Lyra Health , Ping An Medical and Healthcare Management , or Teladoc Health . The global mobile and digital health market is estimated to increase from US$ 43.1 billion in 2019 to US$ 174 billion by 2024 (PB 2020) with companies like Curefit , Headspace , Keep , Omada , Virta Health , Welltok , or Zhangshang Tangyi . And in a third relevant market for digital health interventions, biometric wearables and devices are offered, for example, by Eight Sleep , iFit , Oura Health , Peloton , Supermonkey , or Whoop . This market is also projected to grow from US$ 22.8 billion in 2019 to US$ 70.9 billion in 2024 (ibid.). For the development, licensing, and commercialization of their interventions, many digital health companies, for example, Pear Therapeutics , Akili Interactive , Nightware , Propeller Health , Applied VR , or Happify Health , also partner with large companies from the pharmaceutical industry like Amgen , Astellas , AstraZeneca , Boehringer Ingelheim , Merck , Novartis , Roche , Sandoz , Sanofi , or Shionogi (Patel and Butte 2020).

Fig. 1
figure 1

US-based investments in digital health startups, Rock Health’s 2020 market update, author’s own illustration

Full size image

Against this background, we will provide an overview of digital health interventions and how they are linked to a connected ecosystem of various health-care actors. Key opportunities for these actors and digital health interventions are outlined in the next section. Afterward, we introduce the anatomy of digital health interventions with three key building blocks that focus on the prediction of vulnerable and receptive states as well as tailored support. Four digital health interventions are reviewed afterward to show their value for health-care systems. We then discuss specific challenges related to the design and delivery of digital health interventions and conclude with a summary and outlook.

2 Opportunities for Health-Care Actors and Digital Health Interventions

A health-care system involves actors that are connected to each other in various ways. Patients and healthy individuals consume health interventions while their family members and friends may act as emotional or physical supporting actors. Providers such as physicians, nurses, pharmacists, and private or public health organizations offer or prescribe health interventions, which are then reimbursed by payers, for example, health insurance companies, employers, or individuals. When it comes to individuals with multiple NCDs, a network and close collaboration of health care providers from different disciplines are required. Other actors develop and offer digital health interventions, for example, the pharmaceutical industry, MedTech and digital therapeutic companies, and research or patient organizations. And finally, regulatory and public health bodies oversee health interventions according to their effectiveness, costs, and safety. Against this background, digital health interventions offer various opportunities for these actors, as depicted in Fig. 2. Examples of these opportunities are provided in the following listing:

  1. 1.

    For patients to better manage their NCDs and advance digital health interventions: for example, digital health interventions may improve health literacy and self-management capabilities, reduce distress, or prevent life-threatening events such as heart attacks, exacerbations in chronic obstructive pulmonary disease, or very low blood glucose levels. Moreover, patients may donate their health-related self-report and behavioral and biometric data to help improve the effectiveness of digital health interventions by learning optimal features over time.

  2. 2.

    For healthy individuals to prevent the onset of NCDs and enhance digital health interventions: for example, digital health interventions may improve the physical constitution with high-intensity interval training and offering support to follow a sleep hygiene regime, to reduce alcohol consumption and binge drinking, to stop smoking, to implement a nutrient-rich diet, or to strengthen mental health with the help of slow-paced breathing and mindfulness exercises. Moreover, healthy individuals may donate their health-related self-report and behavioral and biometric data, too as digital health algorithms and statistical tests also require data from healthy control groups (e.g., to draw the line between benign and malignant skin cancer).

  3. 3.

    For family and friends and team mates to support patients and healthy individuals: for example, digital health interventions may explicitly incorporate family members and friends of a target person in the prevention or management of chronic and mental illness by prompting social activities that promote health behavior (e.g., cooking nutrient-rich meals or having hikes together) or prevent life-threatening events (e.g., co-monitoring the health condition and providing support with medication adherence).

  4. 4.

    For health-care providers (e.g., physicians, nurses, pharmacists, health coaches, patient organizations, digital health companies, public health organizations) to increase their reach and effectiveness: for example, chatbots, voice assistants, or health-care robots may take over the role of scalable digital assistants of health-care providers that reach out to their clients in their everyday lives. Health-related data unobtrusively collected by these digital assistants can be used to provide just-in-time feedback to those affected and make on-site consultations more efficient (e.g., by better understanding adverse health behavior of patients) or even obsolete (e.g., in case optimal health promoting behavior is observed). The latter case may open up time for clients that need more face-to-face support by human experts in addition to a digital coach only.

  5. 5.

    For payers (e.g., health insurance companies, employers offering health plans to their employees, healthy individuals, patients, or public bodies) to negotiate and/or adopt value-based reimbursement schemes: for example, the focus on the prevention, management, or treatment of specific NCDs, the degree of evidence available through representative randomized controlled trials, the cost of an intervention, an individuals’ health plan, or treatment success measured continuously in the field may trigger the most appropriate reimbursement scheme. Due to objective measurements and patient-reported outcomes in real-time, digital health interventions also enable reimbursement mechanisms that foster a healthy lifestyle.

  6. 6.

    For developers (e.g., digital health, MedTech, pharmaceutical, or insurance companies, health-care experts, hospitals, research and public health organizations) to improve their interventions on-the-fly and perform safety surveillance: for example, digital health interventions may incorporate self-learning mechanisms into their digital health interventions based on data collected on-the-fly to improve the accuracy of predicting life-threatening events (e.g., COPD exacerbations or very low blood glucose levels) or to better predict the moment an individual is more likely to receive, process, and react to health-promoting messages; the sensing capabilities of digital health interventions may be used to better understand any unintended side effects (e.g., smartphone addiction), too. This, in turn, enables direct market surveillance of both the effectiveness and drawbacks of digital health interventions.

  7. 7.

    For public health bodies (e.g., public centers for disease controls, public health promotion bodies) to promote effective interventions and assess population health: for example, public health bodies may offer an infrastructure that allows patients, healthy individuals, health-care providers, developers, and payers to efficiently identify effective digital health interventions for the prevention and management of chronic and mental illness that comply to various quality criteria. Moreover, anonymized data gathered through digital health interventions and donated by patients and healthy individuals may be used by public bodies to assess population health and identify shortcomings of health care. This population-wide data will then help the health-care industry and funding bodies in steering their resource allocation more efficiently.

Fig. 2
figure 2

Opportunities for various health-care actors and digital health interventions, author’s own illustration

Full size image

3 Anatomy of Digital Health Interventions

Digital health interventions may continuously use data sources to predict vulnerable states of patients or healthy individuals, select the most appropriate intervention option and dose, and deliver it at an opportune moment when the target person is in a receptive state (Nahum-Shani et al. 2018). A feedback loop informs whether and to which degree the intervention had a positive impact on the vulnerable state and, thus, may trigger optimization algorithms that learn—with every iteration of intervention delivery—which intervention option and dose works best for a particular individual and context (Hekler et al. 2018). For all of this to work seamlessly, digital health interventions require various Internet of Things services that provide health-related and contextual data and appropriate, easy-to-use user interfaces to unfold their full potential.

An overview of this closed-loop system, we call it the anatomy of digital health interventions, is depicted in Fig. 3. The three key building blocks of the anatomy, i.e., predicting states of vulnerability, predicting states of receptivity, and delivering tailored support, are described in more detail in the following sections.

Fig. 3
figure 3

Anatomy of digital health interventions, author’s own illustration

Full size image

3.1 Predicting States of Vulnerability

A vulnerable state is a “person’s transient tendency to experience adverse health outcomes or to engage in maladaptive behaviors” (Nahum-Shani et al. 2015). A very first step in the development of effective digital health interventions must be, therefore, to predict adverse health outcomes or maladaptive behaviors by using relevant sensor and/or self-report data. For example, it has been shown that stress at the workplace can be predicted via computer mouse movements (Banholzer et al. 2021) or that the breathing frequency of individuals can be measured with a smartphone’s microphone (Shih et al. 2019). Also, nocturnal cough frequency and sleep quality measured with a smartphone’s sensor data has the potential as a prognostic marker for asthma control and attacks to avoid painful and costly hospitalizations in patients with asthma (Barata et al. 2020; Rassouli et al. 2020; Tinschert et al. 2020). In another investigation, smartphone data was used to predict health-related and potentially modifiable personality states such as conscientiousness and neuroticism (Rüegger et al. 2020). Moreover, voice was used to predict emotions in individuals (Boateng and Kowatsch 2020; Boateng et al. 2020), and both driving behavior derived from a vehicle’s data bus (Kraus et al. 2018) and physiological data from wearables (Maritsch et al. 2020) were used to predict very low blood glucose levels in type-1 diabetes patients. And it was also shown how tapping patterns derived from smartphone interactions could indicate Parkinson’s disease severity (Zhan et al. 2018).

Digitally and objectively measured indicators that reflect either an organic process occurring as a consequence of disease (e.g., decrease in blood oxygen saturation during the onset of an asthma attack) or an organic response to tailored support (e.g., increase in blood oxygen saturation after inhalation of asthma medication) are defined as digital biomarkers (Coravos et al. 2019). Digital biomarkers are, therefore, an essential building block of digital health interventions and depending on their purpose can be categorized into risk, diagnostic, monitoring, prognostic, predictive, or response biomarkers (ibid.).

With the increasing amount and quality of sensor data derived from wearables, smartphones, smart speakers, TVs, and even cars (Koch et al. 2020; Sim 2019), which are pre-processed and fed into state-of-the-art machine learning algorithms (Kakarmath et al. 2020; Rajkomar et al. 2019), there is a huge potential for innovators of digital health interventions to develop scalable and personalized markers that indicate states of vulnerability.

3.2 Predicting States of Receptivity

States of receptivity are “conditions in which the person can receive, process, and use the support provided” (Nahum-Shani et al. 2015). Not only have digital health interventions to predict states of vulnerability but also states of receptivity to make the delivery of tailored support useful at all and more efficient. For example, a smartwatch-based digital health intervention may not be able to reach the vulnerable person with tailored support if that person does not wear the smartwatch.

Similar to predicting states of vulnerability, various data sources such as geographic location (being at work vs at home), acoustic samples (e.g., talking to someone else or not), Bluetooth or WIFI signals (e.g., being close to others or in specific locations), interactions with the user interface (e.g., unlocking a smartphone), communication patterns (e.g., having phone calls), user-defined events (e.g., reminder for medication intake), time (e.g., every evening at 10 pm), or accelerometer data from smartphones (e.g., walking or running) can be used for this purpose.

Indeed, there is first evidence that predicting states of receptivity is feasible (Künzler et al. 2017) and investigations are ongoing to collect various sensor data streams in multiple studies to understand better and predict states of receptivity (Kramer et al. 2019; Mishra et al. 2021). Results show that age, personality traits, day/time, physical activity, geographic location, a smartphone’s battery status, or physiological signals indicating a state of relaxation may be useful indicators for the prediction of receptive states (Chan et al. 2020; Choi et al. 2019; Künzler et al. 2019). Recently, a dynamic state of receptivity module was developed for smartphone-based and chatbot-delivered digital health interventions. This module learns over time and optimizes the prediction with incoming data to successfully predict states of receptivity in a several-week longitudinal field study (Mishra et al. 2021). However, more studies are required to improve the prediction models for different (patient) populations.

3.3 Delivering Tailored Support

If states of vulnerability and states of receptivity are predicted, the last important step of a digital health intervention concerns the delivery of tailored support. Examples of support include motivational messages (e.g., to reach a daily physical activity goal), reminders (e.g., for medication intake), self-monitoring prompts (e.g., to take blood glucose readings), educational content (e.g., health literacy video clips about asthma), or biofeedback (e.g., breathing training with heart rate variability feedback).

Support is often informed by behavioral change techniques (Knittle et al. 2020) and underlying theories about human behavior such as Social Cognitive Theory (Bandura 1991), Self-determination Theory (Ryan and Deci 2017), or the Health Action Process Approach (Zhang et al. 2019). Tailoring of support can depend on various static and time-varying characteristics of the target person, for example, gender, age, self-efficacy or mood, and the current context, for example, time of the day or the weather condition (Hekler et al. 2018; Nahum-Shani et al. 2018).

A novel way to deliver tailored support, and analog to how human health coaching, is the use of text- or voice-based conversational agents such as chatbots on smartphones, voice assistants via smart speakers, smart TVs or in cars, holographic digital coaches via virtual or augmented reality glasses, or health-care robots. Conversational agents are computer programs that imitate the conversation with a human being and have been applied in various health-care settings so far (Bérubé et al. 2021; Schachner et al. 2020; Tudor Car et al. 2020). Conversational agents are perceived as social actors, and it has been shown that individuals can build a working alliance with them (Kowatsch et al. 2021a, b). Working alliance is an important relationship quality in both face-to-face and technology-mediated health-care settings and robustly linked to treatment success (Flückiger et al. 2018).

4 Examples of Digital Health Interventions

This chapter will introduce four digital health interventions to better understand their anatomy and potential to address NCDs. First, we will introduce the digital public health intervention Ally that aims to support individuals in reaching dynamic daily physical activity goals. Second, we will provide an overview of Max, a conversational agent working together with health-care experts to support young patients with asthma and their family members. Third, we oversee Alex, a smartphone-based and holographic physiotherapy coach with the overall objective to increase treatment adherence in home exercises. And finally, we will describe Breeze, a scalable and playful biofeedback breathing training that aims to support individuals in strengthening their cardiac system and stress management capabilities.

4.1 Ally, a Digital Physical Activity Coach

Ally, the assistant to lift your level of activity, is a smartphone-based conversational agent that uses the step count of the smartphone or smartwatch sensor of an individual to infer states of vulnerability, defined in this case as physical activity levels below 10,000 steps a day. With the help of these vulnerable states, Ally motivates individuals to reach a personalized daily step goal. This motivational and tailored support involved setting up specific daily step goals, constrained to an upper limit of 10,000 steps, which were dynamically calculated based on the historic step count of an individual. This coaching approach was adopted to guide individuals step-by-step to an optimal daily step count of 10,000 steps. In addition to motivational messages delivered by Ally, individuals could win small cash or charity incentives in case a daily step goal was achieved. The goal achievement dashboard of the mobile app and an exemplary motivational conversational turn with Ally is shown in Fig. 4.

Fig. 4
figure 4

Goal achievement dashboard (left) and motivational conversational turn with Ally (right), author’s own illustration

Full size image

An 8-week experiment was carried out with 274 customers of a Swiss health insurer to assess the impact of incentives on step goal achievement. This experiment showed that cash incentives boosted the achievement of the daily step goal by circa 8% compared to a non-incentive control group (Kramer et al. 2020). Moreover, the experiment was also used to collect a vast amount of state of receptivity data with the overall goal to incorporate a state of receptivity module in a future version of Ally and other smartphone-based and conversational agent-delivered health interventions (Künzler et al. 2019; Mishra et al. 2021).

4.2 Max, a Digital Coach for Health-Care Experts, Patients, and Family Members

Successful management of chronic and mental illness requires teamwork among healthcare experts, patients, as well as supporting family members and friends. Conversational agents in the role of a team player may support this collaborative effort. With asthma being an example of a severe chronic condition, the conversational agent Max was developed to extend the reach of asthma experts into the everyday lives of young patients and their family members. Particularly, Max delivered a collaborative and gamified health literacy intervention with various educational video clips and exercises via a mobile application that was used by patients. In addition, Max sent text messages to family members and motivated them to support the patients in experiential learning activities. Moreover, Max collected relevant states of vulnerability data, such as knowledge about asthma via a quiz and video data indicating potential shortcomings in the inhalation technique of young patients. Asthma experts received an email from Max when a new video recording was available. With a web-based cockpit, the asthma experts could then assess the correctness of the inhalation technique and send their feedback directly to the mobile application. The interplay between the asthma experts, Max, the young patients, and their family members and a screenshot of the mobile application are depicted in Fig. 5.

Fig. 5
figure 5

Teamwork in the Max intervention (above) and a screenshot of the mobile application (below), author’s own illustration

Full size image

A feasibility study was conducted in 2019 to assess Max in four Swiss hospitals and two outpatient offices of the Swiss lung association (Kowatsch et al. 2021a). The results of this study show that cognitive and behavioral skills of the young patients could be improved and that 75.5% of the 49 participating patients successfully completed the intervention. Moreover, family members worked closely together with the young patients in 97.8% out of 275 coaching sessions. The vast majority of interactions, i.e., 99.5% out of 15.152 conversational turns, took place among the young patients and Max. This clearly shows the scalability of this digital health intervention. To further improve the prediction of vulnerable states, future work may investigate the utility of smartphone-recorded nocturnal cough and sleep quality data to predict asthma control and attacks, which showed promising results in adults with asthma (Barata et al. 2020; Rassouli et al. 2020; Tinschert et al. 2020).

4.3 Alex, a Smartphone-Based and Holographic Physiotherapy Coach

Various activities related to the prevention and management of chronic and mental illness require home exercises. Non-adherence to these exercises is still an important problem for health-care systems because very often, it leads to prolonged treatments and, with it, increased treatment costs. The digital physiotherapy coach Alex was developed to increase adherence to home exercises in physiotherapy. Alex is a smartphone-based conversational agent and holographic digital coach. Consistent with the approach of Max described above, Alex supports health-care experts, here, physiotherapists, by extending their reach into the everyday lives of patients with chronic back pain. Delivering psychoeducation, personalized motivational messages, and reminders via a smartphone were some of the various jobs of Alex. Also, health-related outcomes such as quality of life instruments can be gathered and monitored through smartphone-based interactions with Alex. Additionally, and in the form of a holographic digital coach, Alex provides real-time exercise instructions, monitoring, and feedback with the help of augmented reality glasses. Vulnerable states, for example, an inaccurate execution of specific movements, can be measured in real-time with sensor data streams of the augmented reality glasses and can be fed back instantly to the patient through voice and augmented visual guides via the glasses. Patient-reported health outcomes and the exercises’ performance data can also be sent back to the physiotherapist for upcoming on-site or remote coaching sessions. An overview of the interplay between the physiotherapist, Alex, and the patient and a view through the augmented reality glasses are provided in Fig. 6.

Fig. 6
figure 6

Teamwork of the Alex intervention (above) and a view with the augmented reality glasses (below), author’s own illustration

Full size image

Alex was assessed in three lab experiments with a total of 50 physiotherapy patients and 11 physiotherapists. Alex was also evaluated in a 4-week field experiment with one patient. The study participants rated Alex quite positive with respect to his utility, ease of use, and they also enjoyed the interaction with Alex. The findings also indicate that the participants were able to build a working alliance with Alex. Moreover, interviews indicated that the real-time feedback and interactions with the holographic Alex resulted in a better understanding of performing exercises correctly. Physiotherapists also indicated that Alex would improve the treatment process. The experiment in the field led to an adherence rate of 92%, i.e., 11 out of 12 coaching sessions were completed by the patient with the augmented reality glasses. In addition, the accuracy of the exercises increased substantially over the course of those 4 weeks, indicating the utility of real-time monitoring and feedback. With only one specific squat training session supported by Alex as of today, future implementations must offer a variety of additional exercises to chronic care patients at home or in rehabilitation facilities.

4.4 Breeze, a Playful, Biofeedback-Guided Breathing Training

It has been shown that slow-paced and diaphragmatic breathing has a positive impact on both cardiac functioning and psychological well-being. Slow-paced breathing training can be supported by biofeedback that allows individuals to better control their breathing with the help of adequate physiological signals. Moreover, biofeedback can improve self-efficacy, an important factor one should consider in the design of health interventions (Zhang et al. 2019). Unfortunately, biofeedback is still not scalable because it requires instructions by human experts and dedicated hardware for sensing the physiological signals. In addition, slow-paced breathing training is less attractive to males and lower-educated individuals (Shih et al. 2019).

Breeze was developed to address these limitations (ibid.). Breeze delivers a smartphone-based, scalable, playful, and biofeedback-guided breathing training. For this purpose, Breeze uses the microphone of a smartphone to detect sounds of inhalation, exhalation, and silence and to derive the underlying breathing rate as a vulnerable state. The recorded sounds are then used to deliver a playful visualization. That is, a sailing boat is displayed on the smartphone screen and reaches its destination in the shortest amount of time if an individual’s breathing follows a slow-paced pattern. The breathing detection relies on approximately 2.76 million breathing sounds that were recorded by 43 individuals. Furthermore, various control sounds were used, too, to improve the accuracy of the detection algorithm. A conceptual overview and the biofeedback-triggered visualization of Breeze is depicted in Fig. 7.

Fig. 7
figure 7

A conceptual overview (left) and biofeedback-triggered visualization of Breeze (right), author’s own illustration

Full size image

First empirical results show that Breeze was able to reduce the breathing rate in 16 individuals to a pre-defined target rate, i.e., six breathing cycles per minute. Moreover, Breeze significantly increased the high-frequency heart rate variability, an indicator of cardiac functioning (Shih et al. 2019). In another study with 170 participants, perceived relaxation scores for Breeze, and a validated breathing training were comparable, too (Lukic et al. 2021). These results suggest that Breeze could be employed in both clinical and self-management activities for the prevention and management of chronic and mental illness.

5 Challenges of Effective Digital Health Intervention

The design and delivery of effective digital health interventions come not without challenges. This section will overview challenges with regard to behavioral, regulatory, reimbursement, safety, ethical, cybersecurity, privacy, and infrastructure-related aspects.

One of the biggest challenges related to the design of effective digital health interventions lies in the human nature itself, i.e., to help individuals in bringing up the effort and discipline required to adopt and maintain health-promoting behavior. Developers of digital health interventions must be therefore well aware of various aspects like the law of least effort (i.e., individuals will likely adopt a behavior that requires the least amount of cognitive or physical resistance) (Kahneman 2012), delay discounting (i.e., preferring smaller but immediate rewards over larger future rewards) (Leahey et al. 2020), and various other theories of human behavior (e.g., Bandura 1991; Zhang et al. 2019) and behavior change techniques (Knittle et al. 2020). And although the primary target behavior of digital health interventions is a health-promoting behavior (e.g., an elevated physical activity), developers must, of course, also consider factors that promote both reach and intended use of a digital health intervention. Besides, developers must be aware that digital health interventions should offer a minimum of technology-related interactions not only to reduce any side effects such as smartphone addiction (Haug et al. 2015) but first and foremost to let individuals live their lives without (too much) being dependent on technology. Reaching out to vulnerable populations, for example, those with a lower socioeconomic status, lack of motivation, or simply those that are unaware of their adverse health behavior, especially for the prevention of NCDs, represents another major challenge. Lessons learnt from epidemiology, technology marketing, and information systems research may therefore guide additional design considerations. And finally, effective interventions lead inevitably to sustained health-promoting behavior which, in turn, makes them obsolete. This “die-after-success” design must also be considered in business models underlying digital health interventions.

Another challenge with digital health interventions is the fact that regulatory frameworks and reimbursement schemes for digital health interventions are still “work-in-progress” in many countries. Although the US Food and Drug Administration cleared and approved multiple digital health interventions under regulatory pathways for medical devices (e.g. Welldoc’s Blue Star for type-2 diabetes patients in 2010 or Pear Therapeutics’ reSET for substance use disorder in 2017), and, due to COVID-19, has even relaxed access to digital health interventions, a dedicated framework for software-as-medical-device is still missing (Patel and Butte 2020). A promising example in this regard is Germany that recently implemented a dedicated regulatory framework (FIDMD 2020). Accordingly, first digital health interventions are cleared and can be prescribed by physicians, for example, Elevida by GAIA for individuals with multiple sclerosis (US$ 900, 90 days) or somnio by mementor for insomnia patients (US$ 560, 90 days). With the introduction of a public list of cleared digital health interventions and codes for reimbursement, Germany’s approach addresses a related challenge, i.e., how to find evidence-based digital health interventions.

A further challenge concerns the safety of digital health interventions, especially, when it comes to the native and promising characteristics of digital health interventions such as their connectivity to other devices and services (e.g., blood glucose pumps and other wearables or medical devices), their updateability, or self-learning capabilities via artificial intelligence (Patel and Butte 2020). With the help of automated tests, developers and regulatory bodies can continuously assess the safety of digital health interventions, in particular, when they rely on third-party software and hardware, which is usually the case when they require or run on millions of smartphones, smart speakers, or smart watches (Kowatsch et al. 2019; Shezan et al. 2020). Moreover, mobile devices differ in price and sensor quality which may not only harm patients due to wrong predictions of vulnerable states (Shih et al. 2019) but also and inevitably to an increase of the digital divide and ethical questions about the quality of care and who will benefit most from digital health interventions around the globe (Sim 2019).

Data-driven digital health interventions as we have seen with Ally, Max, Alex, and Breeze above, must also offer easy to understand end-user license agreements, terms of service, and privacy policies (Patel and Butte 2020). Only a transparent collection, monitoring, use, and distribution of data governed by ethical considerations and regular audits by independent third parties will likely lead to broad adoption of and trust into digital health interventions.

A final challenge concerns the infrastructure required for the delivery of digital health interventions. Although COVID-19 has accelerated the digitization of health care in many countries (e.g., Agrawal 2020), there are still many people around the globe that have no or only limited access to digital health interventions (Holst et al. 2020). Even in high-income countries, technical barriers exist, for example, limited interoperability of existing health-care infrastructures and availability of Internet access in clinics or rural areas (Kowatsch et al. 2021a). To this end, novel communication infrastructures may help overcome this challenge (Khaturia et al. 2020) flanked by a strong commitment of decision-makers.

6 Conclusion

In this work, we have shown that digital health interventions have the potential to improve the prevention, management, and treatment of NCDs, the topmost global health burden of the twenty-first century. Digital health interventions offer several opportunities for various health-care actors with the overall but challenging goal to help individuals bringing up the effort and discipline to adopt and maintain a healthy lifestyle. Digital health interventions that predict vulnerable states and deliver tailored support with the most appropriate intervention option and dose in opportune moments are likely to enable precision health. And although there exist still many challenges that limit the full potential of digital health interventions today, we see a significant and increasing number of investments in digital health companies and promising research that will push the borders of our traditional health-care arena.

The ongoing socio-demographic shift toward an aging population results inevitably to more and more individuals suffering from multiple NCDs. And although many countries have already started to deal with these complex and expensive integrated care scenarios, future digital health interventions must be able to orchestrate and offer services from different health disciplines while still being accessible and easy to use. All in all, we envision a future in which digital health interventions democratize health care so that everybody in need benefits from the best knowledge available.

Success Factors for Setting Up Digital Health Interventions

  • Build up a strong interdisciplinary team consisting of experts in NCDs, health psychology and behavioral medicine, machine learning, software engineering, as well as health economics and business models.

  • Co-design digital health interventions with end users and players from the very beginning on.

  • Start with one NCD but think early about relevant comorbidities.

  • Frame digital health interventions as digital social actors so that end users can build up a working alliance with them, i.e., attachment bond and a shared agreement about treatment goals and tasks.

  • Measure vulnerable and receptive states as unobtrusive as possible based on self-reported, biometric or behavioral data.

  • Design your interventions with on-going data-based improvement and assessment cycles to improve the prediction of vulnerable and receptive states as well as intervention options and their dose.

  • Provide tailored support in a way that digital health interventions are as easy as possible to use.

  • Reduce the number of interactions with your digital health intervention to a bare minimum; this will also reduce the burden of individuals using your interventions.

  • Collect field data and provide real-world evidence for the effectiveness of your digital health interventions.

  • Open up your digital health interventions for integrated care scenarios.