Digital Systems for Health: Inventing on Principle

Human Health is a complex phenomena. Software is a human designed artifact. Digital Health has struggled because we don’t adapt our software for that complexity.

The dominant design approach for modern software is to centralize decisions in algorithms. To say the least, it is efficient. Future often looks like the past for most things, and automating human decisions is convenient to solve for busy-ness. So it goes for “Digital Health”. Whether it is consumer engagement, evidence based practice, precision medicine or discovery of new cures, the transformational promise of digital systems for health is definitionally hinged on deploying data based algorithms within software.

But this approach has limits in complex real environments. Social media invented amazing algorithms for user engagement, and now we have this vexing issue of content moderation. Autonomous transport was imminent — only for us to discover that last 10% is 90% of the problem. The human genome was sequenced, and we learned that there are no easy medical miracles. And we have known for much longer that unquantifiable socio-economic complexity makes curing psychiatric and lifestyle diseases bigger than just a technology problem.

I want to advocate for a new principle for building software that is more in tune with reality. Software for complex environments should enhance, not replace, human purpose. While the principle applies generally, I will be only exploring the principle at the intersection of human health and software — where my experience lies.

To build the future of health we want, we need to start inventing on this principle.

To get healthy, CLICK HERE

We can deploy techniques like tracking, nudging, gamification and digital preemption efficiently and at scale now. These emerged and were honed in the world of consumer advertising. These techniques have proven to be effective to get and retain consumer attention. Perhaps, too effective. As information consumption is prioritized for engagement, it creates echo chambers and produces polarization of opinion that prevents effective progress on complex issues.

Now, we are deploying these techniques en masse to every aspect of human endeavor, including health. If what happened to our information consumption habits is any indication, we should worry. The problem is not the method per se. It lies more in relating software design or purpose with the messiness of human decision making. To unpack the nuance, consider the following approaches around building a software app intended to help people manage weight.

Approach A: You believe you know how people can lose weight. You hire a designer to define a software system of badges, financial rewards and instant feedback loops to influence nutrition and physical activity choices. You create a model of individual’s social, medical and economic context and create an algorithm to make data-driven interventions, manual and automated.

Approach B: You understand while the problem of obesity scales, the solution may not. You start with a virtual service that enables people to explore the interwoven nature of nutrition, socio-economic challenges and exercise with a provider or coach. You mediate transactions with an algorithm iteratively, and only to the extent, as informed by real evidence from users.

Approach A seems efficient. Gamifying behavior usually has a quick weight loss effect. Then, the person’s environment muddles the situation. People eventually lose interest in the app. Quite perversely, as people get conditioned to rewards, they lose ability to lose weight for its own sake. What seemed efficient becomes inefficient.

Approach B seems inefficient. It succeeds only when that individual becomes capable of self-deliberation for making right decisions around eating, exercising and managing their social and economic situations. It takes longer but the impact sustains. What looked inefficient turns out to be efficient

There is bigger imperative beyond the efficiency paradox. Approach A places faith in the prowess of the designer or the engineer. Approach B is designed to enable the human purpose.

Human Values vs. Economic Value

Donald Berwick once said, that words like “patient-centeredness are verbal analgesics only”, and that “they mask real pain”. Today, while we have all the tools, development of person-centered care which respects human values, preferences and autonomy remains painfully incomplete.

While building systems, we continue to be uncomfortably torn between two extremes: enabling human values, often wrongly conflated with consumerization of health, and creating economic value for the system, typically rooted in commercial interests of the health enterprise. To get past the stalemate, we need to define functional frameworks, that can be enabled with software, to close the gap between human values and economic value. Internet based Cognitive Behavioral Therapy (iCBT), typically used for psychiatric disorders, is an example of such a framework. Online delivery reduces the stigma often associated with such ailments, and enables “in the moment” need. Reduced reliance on face to face therapy eases the constrained supply of primary care and mental health providers. Unfortunately beyond iCBT, I couldn’t find many examples.

Same applies for adoption of digital interventions in general. Just because it’s easy to replace a human interaction with a software algorithm, or replace a real world experience with a digital one, doesn’t imply you should. For the swap to make sense, it needs close an existing value gap. The experiment of COVID forced adoption of virtual visits and telemedicine provides a relevant insight hard to ignore. At scale technology has been around for a considerable time with very little uptake. The emergent context of COVID created the actual demand pull. The technology closed the gap on what patients and healthcare staff valued i.e. safety and what business enterprise valued i.e. need to provide care and maintain essential services. When that value gap closed, the often cited barriers of lack of supportive payment structures or regulatory challenges got dismantled almost overnight.

I ought to clarify — I am not advocating that you shouldn’t build commercially oriented software applications in health. But if you want to make your application human-centric, then you have to close the value gap.

The Noise from the Infinite Detail

In 2010, Atul Gawande presented to President’s council of advisors, which included Eric Schmidt (then CEO of Google), on importance of checklists. Eric Schmidt was “befuddled” why medicine “hasn’t organized” around what he clearly saw as a “platform database problem”. Gawande reasoned that a database search will have a high noise to signal ratio with little actionable intelligence — “three inches of guidance, explaining what the evidence suggests and so on, needs to be turned into a useful form that tells you what you can do”.

2010 was early days of EHR implementation boom triggered by meaningful use legislation. 10 years later, we are dealing with problems like “4000 clicks a day” while “Digital Doctors” are still hard to find. Gawande revisited the issue in his New Yorker article “Why Doctors hate their Computers” in 2018. And concluded, “I’ve come to feel that a system that promised to increase my mastery over my work has, instead, increased my work’s mastery over me.”

Today there is almost a religious belief in information processing imperative — more information is always better than less. For patients, for doctors treating patients, for caregivers supporting patients, for manufacturers discovering cures from real world data. But, more information also means more noise, and increases the signal extraction overhead . So, we shouldn’t be surprised when providers are hesitant to adopt digital workflows with low signal content

Mass adoption of smartphones and ‘always carried, always on’ usage culture has created a compelling opportunity to bring real world data into clinical decisions and new product development. At the same time, we need to remind ourselves that the real world only moves forward when we create new explanatory knowledge about real world phenomena, not just find deeper associations in bigger and bigger piles of data.

Our challenge is to enhance human intuition of clinicians and researchers, with the best signals from the real world information. Not to replace the former with the latter.

Inventing on Principle

The progress in Digital Health, or lack thereof, is well documented. Investors have poured tens of billions into startups in the last decade. The Health Enterprise (collectively Providers, Payers, Bio-Pharma and Med-Tech) has been building to transform digitally during the same time. In recent years, there is excitement stemming from activity of big incumbents in the space , e.g. Amazon — PillPack, Haven , Google — Fitbit, Google — Ascension, Roche — Flatiron. Yet, we aren’t making progress towards the hallowed triple aim — enhancing patient experience, improving population health and reducing the per capita cost.

Misaligned financial incentives, lack of rapid experimentation and feedback, lack of customer focus are generally cited as reasons. These are certainly part of the problem. The real problem is that we aren’t building what is needed. I started thinking about this issue couple of years ago when I started building health and medicare software products for the enterprise. And I have been building software for Health Enterprise for much longer. Even though it may sound clichéd, I have come to believe that despite all the capabilities of modern technology platforms, it has never been more important to appreciate that technology can never be an end in itself. After all, digital is as much about health as telescopes are about astronomy.

When building solutions for health, the goal isn’t just to make people use the devices; it is also to make them think with their devices. While, human intuition can be informed with the data, but it is still the best when it comes to deciding in complex and uncertain situations. We also need to advance abstract concepts like person centered care, and create actionable frameworks to mediate access to scarce systemic resources aligned to what consumers need. And, we need to reevaluate the information processing imperative; focus more on creating signals for human agents in health systems.

Health is essential. Software that we build for health is contingent — on our goals, and purpose we want to enable. So, ultimately it is a design and an engineering problem. It is about our willingness to commit to the principle of developing software in service to the human purpose, and adapting that iteratively to the complexity of human health. It is about inventing on principle.

Other Notes: The title is inspired by Bret Victor’s talk with same title.