How Will Artificial Intelligence Transform Healthcare? It May Depend on the Type of Healthcare Business
There’s a lot out there about how AI will transform the diagnosis and treatment of disease. But getting from here to there in a complex system will not be a straightforward process. Healthcare isn’t like Uber, where there are licensed workers with spare time and the means of production sitting around. Training and infrastructure impose significant barriers to entry. Healthcare won’t be so easily disrupted.
Ten years ago, an insightful book was published about the nature of change in healthcare. The Innovator’s Prescription puts healthcare in the context of other industries. Healthcare is insular: people in healthcare often have little perspective on how healthcare relates to the computer industry for example, or the automotive industry. The authors of The Innovator’s Prescription, Christiansen, Grossman and Hwang, argue that healthcare is subject to the same forces of disruption as any other industry. But how this disruption occurs may depend on the nature of the healthcare enterprise.
Healthcare organizations are really doing three jobs, and in doing all three, do none of them well. “Solution Shops” are how most people think about hospitals. A person becomes sick. Well-trained and dedicated professionals put their minds together, solve the diagnostic dilemma, and get the patient back to a productive, healthy life. But this “Dr. House” version of healthcare is the exception. True diagnostic dilemmas are rare. Healthcare is often standardized. The problem that needs fixing is often readily apparent and needs to be done efficiently with few complications. Cataract surgery, the most common surgical procedure in the United States, does not require a Solution Shop. Rather a “Value-Adding Process” needs to remove the cataract and replace it with the appropriate intraocular lens as quickly and safely as possible. Managing cataracts is not a diagnostic dilemma, and a system set up as a Solution Shop has a hard time performing Value-Adding Processes well. Finally, and with increasing importance these days, are the “Facilitated Networks” needed to manage chronic disease. Healthcare providers have traditionally functioned in a fee-for-service atmosphere. Preventing the complications of chronic illness — keeping people out of the hospital bed and operating room — is not what Solution Shops or Value-Adding Processes do best.
The difficulty arises when a single organization tries to perform all three tasks. Running a surgical center efficiently is a very different proposition than managing the health of populations or diagnosing complex processes. The Innovator’s Prescription argues that in trying to do one well, healthcare organizations are almost guaranteed to do the others poorly.
As artificial intelligence becomes more important in healthcare, each of these healthcare business models will require a separate approach and it’s worth thinking about each of them separately. As organizations contract for the care of a population, the numbers become large enough for machine learning algorithms to be increasingly effective at making predictions. Even a moderately-sized organization can care for tens of thousands of people with type II diabetes, each of whom has a risk of retinopathy, kidney failure, heart disease and peripheral neuropathy. Predicting who will develop these complications is not the same as preventing them from developing, but interventions can be modeled as well: who will respond best to frequent doctor visits and who is better off with a visit from a home nurse? As the financial responsibility for these conditions shifts to healthcare organizations, the opportunities open up. Sixty percent of diabetics don’t get their annual eye exams. Already algorithms exist to interpret retinal photographs that can be taken in a primary care office, so people at risk for vision loss can be identified earlier. No one in healthcare wants people to go blind, but with organizations starting to manage the health of populations, for the first time they will be financially incentivized to prevent it. The opportunities for predictive modeling in population health are endless. Given the amount we spend on managing chronic disease, the returns are enormous.
The approach required for Value-Adding Processes is similar. Patients have known medical issues; AI can help identify those who will not flow smoothly through the system. Who will have surgical complications? Who will develop a surgical site infection? Who will “bounce back” and need to be readmitted? Healthcare organizations are already deploying these models and we can expect interest in them to increase in the next decade. For the first time, surgeons will be able to discuss an individual patient’s risk prior to undergoing a procedure, and consent can truly be informed. The potential return on AI for Value-Adding Processes may not be as great as for population health management, but it will be more immediate since reimbursement models already penalize healthcare organizations for complications. We can expect this to be an active target for healthcare AI in the coming decade.
A lot of the talk about AI in health tends to focus on diagnosis: intelligent machines will tell doctors the answer. While there has been progress made in this direction, the Solution Shop may be the most challenging for AI. For one thing, true diagnostic dilemmas are rare. With the array of imaging and lab studies out there, the challenge is not the diagnosis itself but in reaching the diagnosis in the fastest, most cost-effective way. This is a different problem than is often envisioned, and more complicated. IBM’s Watson reached into this space for cancer care at M.D. Anderson and struggled, to say the least. This sort of work may be the last to benefit from healthcare AI.
The Innovator’s Prescription was published in 2009, an important year in healthcare. It marked the inauguration of Barack Obama and the start of some major changes in American healthcare, including and the beginning of the Affordable Care Act, as well as the HITECH act, which set the ground for AI by getting us on electronic medical records. The healthcare landscape changed just as The Innovator’s Prescription came out, and it may have not garnered the traction it otherwise would have. Rather than moving the three healthcare business models apart, which the book argues would improve healthcare innovation, we’ve seen increasing consolidation in the industry. With the enormous transformative power of AI just starting to enter healthcare, we would do well to consider that healthcare is more than one business, and AI may need to be used in different ways in each.