September 17, 2018
Healthcare and AI: Systems of Scalable Empathy
Image Credit: WSJ & Eddie Guy
Last Thursday, I had the opportunity to participate in a panel discussion on advanced technologies (AI and Blockchain) at the Healthcare Executive Group (HCEG) event in Minneapolis. Alan Abramson, SVP and CIO of HealthPartners was leading the panel. As I was listening to the speakers in the morning and thinking through my remarks for the panel, a phrase came to mind that I believe captures the best hopes of AI in healthcare: Systems of Scalable Empathy.
We have learned from decades of system thinking that the value of software is in scalability. We have learned from design thinking that we have to put people and their behaviors at the center for successful product development. By combining these human considerations and systems of insight we can create Systems of Scalable Empathy.
As I was thinking about systems that promote this concept, I read the Saturday Essay in the WSJ by Dr. Kai-Fu Lee titled The Human Promise of the AI Revolution:
While AI is great at optimizing for a highly narrow objective, it is unable to choose its own goals or to think creatively. And while AI is superhuman in the coldblooded world of numbers and data, it lacks social skills or empathy—the ability to make another person feel understood and cared for. Analogously, in the world of robotics, AI is able to handle many crude tasks like stocking goods or driving cars, but it lacks the delicate dexterity needed to care for an elderly person or infant.
It’s true. Empathy and AI are an odd juxtaposition. The use of machines in the very human domain of healthcare would seem to be dehumanizing. In Automating Inequality, Virginia Eubanks argues that the use of systems that incorporate mathematical models can be used to instantiate broad social biases. Her book very powerfully articulates the harm that these systems are capable of, specifically when autonomy is removed from the individuals in the system (like a caseworker) or the system’s customer or client. When these simple statistical regression models are then described as AI or ML (Machine Learning), the system takes on a higher level of perceived authority.
I believe that we can use well-designed AI/ML models to create unbiased systems that support human intuition. By using models that support tailored responses to sensitive issues like discharge planning, chronic disease management, case management and inpatient adverse events, we can gain foresight into the near-term future for our patients and consumers. As a result, our capable and talented clinicians can leverage those responses or recommendations to provide care that intercepts the negative outcome before it happens. In this way, we elevate the individual and provide for more personalized care at greater scale.
The story of Ochsner Health decreasing inpatient ICU admissions with Epic and Microsoft is a great example of combining the best capability of machine learning and modeling with the bedside care of the clinical staff. Through Machine Learning, they were able to accurately predict patient deterioration hours before an adverse event. This early warning system was tightly integrated into Epic, enabling Ochsner’s Rapid Response team to intervene on patients proactively, rather than reactively. During their 90-day pilot at Ochsner Medical Center, the team successfully reduced codes outside of the ICU by 44%.
It is critical to maintain the autonomy of the person, otherwise they feel like they are simply a cog in dehumanizing machine. So how do we start creating Systems of Scalable Empathy? By giving smart people insights from the data we have at scale and coupling that with our creative imaginations, we create the space to make meaningful investments in people that make a difference.