The Oversell And Undersell Of Digital Health

Digital health technologies hold great promise to solve some of the biggest problems in our healthcare system, including achieving higher quality, lower cost, and greater access to care. In the January 2019 issue of Health Affairs, we reported that scant evidence exists demonstrating the clinical impact of twenty top-funded digital health companies. These companies tended not to study the clinical effectiveness of their products in terms of key healthcare metrics like patient outcomes, cost, and access to care.

We found 104 peer-reviewed published studies on the products or services of these companies. The majority of the studies were from three companies. Nine companies had no peer-reviewed publications. Only 28% of the studies targeted patients with high-burden, high-cost conditions or risk factors. Healthy volunteers were the most commonly studied population. Further, 15% of all studies assessed the product’s “clinical effectiveness” and only eight studies assessed clinical effectiveness in a high-cost, high-burden population. The eight clinical effectiveness studies measured impact in terms of patient outcomes, while no studies measured impact in terms of cost or access to care. There were no clinical effectiveness studies in heart disease, COPD, mental health conditions, hyperglycemia, or low back pain. Studies that did not assess clinical effectiveness may have intended to validate the product against a gold standard measurement or report feasibility of use.

This is of particular interest given the incredible amount of funding, interest, and hype in digital health. Although these companies were only a small portion of total digital health companies, they were a large portion of total private funding and had the most resources to demonstrate impact. Further, since “digital health” currently encompasses myriad technology types and approaches, these findings have broad implications.

Does this mean digital health is doomed? Should investors, consumers, patients, and developers stop spending their money on digital health? Do our results describe all digital health sectors and companies?

No. No. No.

The Over-sell

When an exciting, new technology is successfully applied in one field, other fields take notice. If AI can drive cars, perhaps it can perform surgeries. If blockchain can replace secure currency transactions, why not replace secure health data transactions? If virtual reality can produce better gaming experiences, maybe it will produce better medical educational outcomes. Since mobile communications computing technologies (smart phones) are omnipresent, only imagination limits the potential of new approaches to healthcare.

Such approaches include endless means to predict, prevent, detect, and respond to varied health conditions through connected technologies like biosensors and telemedicine platforms. Thus, with intense optimism around new potential technologies coupled with the actual presence of enabling foundational platform technologies, like smartphones, the hype grows.

Once the ideas start flowing, developers and researchers get funded and get to work. In little time, companies are formed, claims of effectiveness or efficiency are made, and products are sold. Often these claims are not bound by regulation or the burden of showing proof. When these products hit unmet market needs, lots and lots of products sell.

To boost interest and investment, a product’s proponent may minimize important hurdles to product deployment and use. A basic prediction algorithm, for example, can assist stroke prevention treatment decisions in patients with atrial fibrillation who are at high risk for stroke. Proponents may demonstrate or claim that a more sophisticated AI prediction algorithm may outperform a basic prediction algorithm in a lab setting. Neither have impact, however, if the physician can’t or isn’t using them. In our opinion, perhaps the most important question to address hype is “will I actually use this?”

As another example, an AI application may outperform a physician in detecting sepsis in the emergency ward. When the physician may forget to review all the labs, the AI application does not. It sees and integrates not only the labs, but also the patient demographics, history, vitals, exam, microbiology, imaging data, fusing them together, to outperform its physician counterpart by seeing patterns the physician cannot.

Yet, even though the AI application may work in a laboratory setting, it may fail in the real world. Many study vs. real world discrepancies can hinder real-world impact. For example, in order for an AI application to succeed in detecting sepsis, it must have access to all that multimodal data right now in the emergency ward. In practice, however, lab data may be stored in one data repository while images and microbiology data are stored in others. This hurdle must be overcome if the AI application is to integrate and analyze the data in real time. Further, examination results may be part of structured fields of a note template that wholly misrepresent actual exam findings. Here, the AI application receives wrong data.

To move from unwarranted hype to justified enthusiasm, the technology must be used and proven effective in the real world. In AI health applications, like other digital health technologies, validation studies are mainly limited to retrospective studies and prospective studies, assessing impact on real patients are just beginning.

The Under-sell

Luddites and pessimists who wholesale dismiss these emerging technologies will miss great opportunities to improve healthcare. Despite the hype and dearth of high-quality studies demonstrating impact of these products, we are confident in our optimism about the promise of digital health.

The groundwork has been laid for digital health deployment, eventually. Most physicians and the majority of the US population have digital health applications or technology. The array of digital health technologies is rapidly rising, too, which allows for more selection from between meaningful and meaningless solutions. The cry for interoperability of technologies and standard approaches to data structure continually grows louder and louder, too. This may be the crucial backbone by which successful digital health solutions stand.

Although our study demonstrated a fairly limited volume of high-quality digital health studies in high burden populations, the volume of digital health studies is rapidly increasing. The volume of high-quality studies designed to assess clinical impact will follow suit, particularly as regulatory body and digital health company domains intersect. Some prescription digital therapeutics, for example, rigorously study their products to determine clinical impact potential while seeking clearance from the Food and Drug Administration. Certainly, other companies we did not assess in the Health Affairs study indeed target and study high-burden, high-cost populations, some of which also demonstrated clinical impact.

Three main reasons fuel our optimism:

Increasing integration: Technology interoperability for traditional medical devices (like cardiac pacemakers) and non-traditional medical devices (like digital activity trackers with data dashboards, repositories, and integrator platforms like electronic health records) will smooth the way to further deployment of digital health technologies.

Falling walls: The traditional concepts of healthcare are changing. Historically, patients typically travelled to providers in clinical buildings to be assessed and treated. Now, both patients and providers are increasingly trying new modes of care, from personal remote monitoring to provider-driven telemedicine encounters. Some new models may allow certain patients to be nearly or completely dependent on digital technologies, requiring minimal or no clinician intervention.

Left shift: Health care systems often don’t notice patients until they become ill and trek into the organization’s physical location (Figure 1). An ideal approach surveils and responds to patients before illness, making “primary (prevention-focused) care” the dominant and most effective form of healthcare. This is only possible with robust prediction and prevention systems, which in turn require robust monitoring. Instead of a system that only detects disease, intervenes to treat that disease (after it’s developed), and attempts to sustain or restore health, an ideal system would be surveilling its population continuously.

Exhibit 1: Allocation Of Resources In Current Versus Ideal Health Systems

Source: Authors’ creation, adapted from The Johns Hopkins University Applied Physics Lab National Health Mission Area “Life Cycle.”

We believe efforts to shift toward prediction and prevention will continue to percolate into the healthcare system. As compared to more traditional approaches to care – many of which require paying expensive experts and renting or buying real estate –  digital health solutions provide clear advantages. They can touch millions of people at scale, require relatively minimal development costs, and move rapidly from initial idea to market.

Conclusion

The idea of a telemedicine visit between doctor and patient to foster more effective care is not new. The concept was first reported in the April 1924 edition of Radio News (Exhibit 2), making the cover of that issue. Although many centers practice telemedicine today, that vision developed in the 1920s is still not yet a fully deployed reality.

Exhibit 2: Radio News, April 1924

Source: Photograph from author’s personal collection. Radio News is a defunct American magazine (1919-1971).

Today, telemedicine visits are hyped as one of the healthcare approaches that could revolutionize vast areas of patient-provider encounters. Yet, even at the most prolific centers that perform telemedicine, it represents less than one percent of total care volume. The major leaps in how healthcare is delivered won’t occur with a great new idea or a remarkable advance in technological capabilities. Instead, the transformation will occur when an innovation is effectively integrated into complex environments with complex people.

To truly make a transformative impact on health care, we encourage digital health companies to begin with the standard of evidence of impact that physicians, hospitals, insurance plans, and patients seek: high quality studies of highly-burdened populations using rigorous study design in real clinical environments with meaningful metrics of impact – outcomes, cost, and access to care.