Blog Series:  PUBLIC IMPACT ANALYTICS SCIENCE (PIAS)

“Failure doesn’t exist. It’s only a change of direction.”  [Alejandro Jodorowsky]

Traditionally, healthcare delivery mechanisms have been limited to both the availability of a provider (e.g., a physician) and the location in which care can be delivered (e.g., hospitals or outpatient clinics). However, advancements in Internet of Things (IoT), wearable devices, mobile technologies, and new tools for developing user-friendly smartphone applications have resulted in significant changes in how care is delivered.  Mobile health (mHealth) and digital health interventions, in particular, are now widely used to deliver care through innovative mechanisms.

Examples of such interventions include (a) maintaining adherence to HIV medication and smoking cessation, (b) assisting caregivers in managing veteran posttraumatic stress disorder (PTSD) and providing support with healthcare related tasks within the Veterans Affairs (VA) system,  (c) encouraging physical activity and weight loss, (d) reducing excessive alcohol use, and (e)  responding to COVID-19 and future epidemics.

It is, therefore, not surprising to see that leading voices from the medical, government, financial, and technology sectors have endorsed the idea that these technologies can “transform medicine” [1].  However, in a recent article published by the National Academy of Medicine [2], Professor Susan Murphy of Harvard’s Departments of Statistics and Computer Science and I argue that despite the rapid growth and wide use of these interventions in a variety of applications, there are urgent scientific as well as regulatory challenges that need be addressed. We point out that, without addressing such challenges, digital health interventions are not only unable to reach their full potential but may also do more harm than good.

Specifically, based on our own line of work both on the theory and in the field, we highlight that without new statistical methods, innovative experimental designs, and suitable AI-driven optimization techniques—that allow understanding (a) what interventions are effective, and (b) how interventions should be made adaptive based on the user’s current context (e.g., mood, location, weather)—“transforming” medicine via mHealth and digital health interventions is illusive.

We also discuss the critical need to (a) balance objectives on proximal and distal health outcomes, and (b) understand the causal chain link between such outcomes to achieve longer term health outcomes.  Finally, we highlight that in designing digital health interventions, ensuring that the known principle “primum non nocere,” meaning “first, do no harm,” is not violated is both vital and difficult.  For example, prompting a user for engagement during driving can cause an accident. Similarly, prompting a user to stop smoking while s/he is in an unsuitable emotional state may result in immediate alcohol consumption by the user (e.g., as a replacement for smoking).
 

What Should be Done?

To ensure that the mHealth interventions can reach their potential, we argue that at the very minimum the following advancements are needed:

1. New statistical learning and experimental designs combined with techniques used in various industries in optimizing large scale systems. Specifically, based on the state-of-the-art statistical theory in learning how best to adapt the content of each treatment as well as its delivery on the user’s context, we discuss how improving the efficiency of digital health interventions require new methods of experimental design. Such new methods are needed to provide data that facilitate the use of causal methods post-intervention to study time-varying mediators and moderators, including unobservable factors (e.g., the emotional state of the user) that can be used to make interventions more effective. Furthermore, delivery of digital health interventions can benefit from experimental designs used to optimize large-scale systems—many of which are already being used in various industries (e.g., adaptive optimization of advertisements by Google, or optimizing routes and assignment of drivers to incoming ride sharing requests by Uber).
    
2. New behavioral theories and related experiments for an improved understanding of the burden and habitual consequences of digital health deliveries. If digital health interventions are delivered excessively, they might result in user disengagement, and thus, interventions can easily become ineffective. This is more pronounced in interventions that are push-based, because they can interrupt the user in their daily life. In pull-based interventions, where notifications are not “pushed” to the user, concerns of increased burden or habituation are not as prominent. However, in such interventions, scientist need to address a similar effect:  a low rate of user participation. How to keep users engaged at the right level—neither under-engaged nor over-burdened—is an exceedingly important question that cannot be overlooked by scientists.
    
3. A more systematic way of understanding the interplay between proximal and distal health outcomes. In designing digital interventions, scientists need to carefully balance objectives on proximal and distal health outcomes. Understanding the causal chain linking the proximal effects of the treatments to longer term health outcomes can significantly help to improve delivery in the digital health ecosystem.  However, measuring both proximal and distal outcomes as well as understanding their linkages is often a perplexing task. New theories are required to assist system designer and technology developers in this vein.
    
4. Besides scientific challenges, new regulatory avenues are needed to ensure that the mass adoption of digital technologies in delivering interventions does not comprise data privacy.  Considering this, the Food and Drug Administration (FDA) in a 2019 statement “encourages the development of mobile medical apps (MMAs) that improve health care” but also emphasizes its “public health responsibility to oversee the safety and effectiveness of medical devices—including mobile medical apps” [3]. The FDA’s overall oversight of digital technologies has been controversial to members of Congress and industry and has been also criticized in various articles. Beyond data privacy concerns, digital health technologies, especially those used to continuously monitor chronic conditions and to implement disease management plans, raise important legal concerns when caregivers (e.g., medical staff or family members) who might be already overburdened with other duties fail to respond to alerts in a timely manner. Mistreatments or other adverse events caused by such oversight can create various legal ambiguities (e.g., the entity who should be responsible or whether such events should be covered by insurance plans).

Achieving these advancements and addressing the underlying challenges, however, is not an easy endeavor. It requires urgent attention as well as close collaborations among scientists, policymakers, lawmakers, and other authorities. Why does it need urgent attention and close collaborations? The answer is simple: without such attention and collaborations, digital health interventions will either transform the medicine in the wrong direction or will not prove themselves as useful tools. The good news is that it is not too late to change direction. As Alejandro Jodorowsky once said “Failure doesn’t exist. It’s only a change of direction.”

References

[1] Cortez, N. G., I. G. Cohen, and A. S. Kesselheim. 2014. FDA Regulation of Mobile Health Technologies. New England Journal of Medicine 371(4): 372- 379.

[2] Saghafian, S. and S.A. Murphy. 2021. Innovative Healthcare Delivery: The Scientific and Regulatory Challenges in Designing mHealth Interventions. National Academy of Medicine (NAM) Perspectives, Commentary.

[3] U.S. Food and Drug Administration. 2019. Device Software Functions Including Mobile Medical Applications. Available at: https://www.fda.gov/ medical-devices/digital-health-center-excellence/ device-software-functions-including-mobile-medical-applications (accessed July 7, 2021).