Care Redesign
Man vs. Machine: Leveraging AI in Health Care (08:06)

“Thus far, the use of computer assistance to help clinicians and patients has been disproportionally focused on clinical decisions — treatment A versus treatment B,” says Arnold Milstein. But there’s an enormous opportunity to use computer assistance not just in clinical decision-making, but also in assuring critical clinical work flow.

“For example, our lab has been very active in taking assurance of intended care processes in small geographic places where getting care exactly right is enormously important to value of care,” Milstein says. They focus primarily in ICUs and in the homes of cognitively impaired seniors, and ask clinicians: What are the critical facets of work flow that, if you don’t get right, will cause the patient to have severe health and cost trouble very quickly?

“What we’re doing is essentially asking, can we use economical forms of computer vision to begin to detect omissions in intended processes and alert the clinician and medical teams that something important has been missed?” For instance, turning an unconscious patient in an ICU at a certain frequency makes a big difference in preventing decubitus ulcers. With humans being human, it’s not uncommon that someone working in the unit will document that they turned the patient, when they may, in fact, not have.

Benjamin Chu agrees that artificial intelligence and machine learning can play a big role. “We have to have a much more dynamic, iterative process for trying to evaluate effectiveness, and the only way to do that is to try to decipher the patterns from the noise. It requires that we all look at data sets, big data sets, in a way that has enough power to determine whether they’re effective or not.”

Milstein adds how valuable the artificial intelligence system was in showing predictors that made intuitive clinical sense, but that no one had ever thought about before. One example was if a patient within the course of a month had three visits to a primary care doctor without a diagnosis. Another was increasing velocity of visits to subspecialists. “Think about that — in retrospect, it makes sense, there’s something happening here, and the health system is responding with a quickening of pace and moving patients into services, but it’s not a variable that any human would have thought to test. But it turned out that the supervised machine learning approach that we applied taught us that that contributed to predicting what we call cost blooms,” Milstein says.

Namita Mohta raises the importance of preserving the sacred relationship between patients and providers despite advances with machine learning. “The thing we have to keep in mind is that we don’t repeat perceived mistakes of the EMR — that it doesn’t become a technology that somehow has the unintended consequence of burdening the relationship that we’re looking to preserve between patients and their caregivers.”

Tom Lee agrees, leveraging the flood of knowledge artificial intelligence and machine learning are providing. “There’s data showing that the doubling time of knowledge in medicine has been accelerating,” he says. “It’s gone from, when I was born in the 1950s, about every 5 years knowledge was estimated to double, to now, at the end of this decade, the prediction is that doubling time will be 73 days.” How do we deal with that? “If you’ve got four neurologists, and their neurology knowledge doubles, do you hire another four neurologists? That clearly isn’t the way to do it. We’re going to have to figure out, how do we work together, and how do we work with computers in order to have people use knowledge and not just be overwhelmed by it?”

Milstein adds that we can learn from other industries and asks that we consider the example of commercial airline pilots in the 1980s, overwhelmed by the increased flow of alerts, which interfered with their “relationship” to the plane. “Eventually what they did is they got together as a group and they said, there has to be a threshold of probability that what the computer wants to tell me is going to be useful before I want to be interrupted,” he says. That turned out to be a very successful equilibrium in allowing the human brain to do what it can do best, especially in an intimate encounter, and at the same time take advantage of this supplementary form of intelligence.

From the NEJM Catalyst event The Future of Care Delivery: Relentless Redesign at Providence St. Joseph Health, January 19, 2017.

More From Care Redesign

What Is Lean Healthcare?

Lean Healthcare is the application of “lean” ideas in healthcare to minimize waste with ongoing process improvement. Learn how to use Lean to improve patient satisfaction and care outcomes while reducing costs.

Zane05_pullquote3 hospital decentralization

Do Hospitals Still Make Sense? The Case for Decentralization of Health Care

The future is here: moving care out of the hospital and into the home and community.

Sociotechnical dimensions used to analyze EHR-related health IT safety concerns

Frustrated with Your EHR? Don’t Blame Your Vendor — Safety Is a Shared Responsibility

Two informatics experts urge individuals and organizations to work together toward safer EHR-enabled patient care.

Relative Health Care System Performance and Spending in 11 High-Income Countries

From Last to First — Could the U.S. Health Care System Become the Best in the World?

The United States could achieve the best-performing health care system in the world by undertaking coordinated efforts that address four challenges.

Addressing the Prescription Opioid Crisis: Advancing Provider Education and Collaborating with All Stakeholders

Providers have a large role to play in tackling the opioid overdose epidemic, but they can’t go it alone.

Collaboration Between Doctors and Computers - Machine Learning

Lost in Thought — The Limits of the Human Mind and the Future of Medicine

It’s ironic that just when clinicians feel that there’s no time in their daily routines for thinking, the need for deep thinking is more urgent than ever.

Root Cause Analysis: Typical Domains of Root Cause: Medical Errors

Does Every Hospital Admission Deserve a Root Cause Analysis?

Will applying the RCA rubric to hospital admissions better help define and manage care?

Uniquely Identified: The Impact of a National Health Index

What does the NHI mean to a New Zealand clinician, researcher, and health care consumer?

The Intersection of Home-Based Primary Care and Home-Based Palliative Care

My Favorite Slide: The Intersection of Home-Based Primary Care and Home-Based Palliative Care

What are the overlapping provider skill sets needed to care for homebound patients?

Simplifying Person-Centered Care with Use of the Personalized Perfect Care (PPC) Bundle

Personalized Perfect Care

The Personalized Perfect Care Bundle: Making quality metrics easier to understand and more patient-centered.


A weekly email newsletter featuring the latest actionable ideas and practical innovations from NEJM Catalyst.

Learn More »


Proactively Catching the Declining Patient

A coordinated effort by UCLA leaders to identify a high-cost population with chronic kidney disease…

Coordinated Care

145 Articles

Proactively Catching the Declining Patient

A coordinated effort by UCLA leaders to identify a high-cost population with chronic kidney disease…

Assessing and Addressing Social Needs in…

Lincoln Community Health Center improved care quality by measuring and responding to upstream social and…

Insights Council

Have a voice. Join other health care leaders effecting change, shaping tomorrow.

Apply Now