Clinical processes, not AI, key to solving healthcare's problems
Partners' Sandy Aronson has a message for healthcare IT pros: If you want to have an impact, focus on making clinical processes more effective and more efficient.
Stop believing artificial intelligence is going to solve all of healthcare's problems. Samuel "Sandy" Aronson, executive director of IT at Partners HealthCare Personalized Medicine, said the real solution is better processes.
Factors such as new data sources, new algorithms and cost pressure are creating a compounding effect that's making clinical departments "far more receptive to change than ever before," he said. IT should seize the moment and find ways to not only modernize processes but to redesign them altogether and help usher in algorithmic-based care.
"AI is not in and of itself the solution for the healthcare industry," he said. "What really matters is designing processes that enable us to access, generate and refine the data that we require to perform healthcare well, and then ensuring that we have the ability to take control of clinical processes so that we can actually improve them, so that we can feedback the learning from this data into a continuously improving process."
Partners to develop open source platform
Aronson and his team did just that with a platelet allocation application for bone marrow transplants. About 15% of the time, donor and receiver human leukocyte antigen profiles don't match, which means a patient's body ultimately rejects the new platelets. But it can take as many as seven bags of platelets before a bad match is detected.
The platelet allocation app inventories what platelets are in stock, determines the patient and donor human leukocyte antigen profiles, determines how difficult it is to match the patient and then delivers an ordered list of best matches for the patient to a blood bank technician.
"We are currently collecting data on how well this works," he said during his presentation at AI World in Boston. "But we are very optimistic this will yield a positive effect."
Ideas like these for updating clinical processes aren't hard to come by, but for them to work, they need to be tested, validated, deployed and validated again. "And in a world where most of the ideas today rely on either new data types or new kinds of algorithmic-based support, you hit this problem because you need new forms of IT support in order to launch these new techniques," he said.
Despite investment in developing these capabilities, Aronson said only a small percentage of ideas can be funded and developed at Partners. "This is a major break in innovation," he said. "I would argue the potential of AI is the ability to progress through this process."
Even so, if a process flow change works, taking it to healthcare organizations globally introduces new hurdles, making the app difficult to port from one organization to another. "You hit this issue again where the standards in the IT space are currently far from perfect," he said. "There's great work that has been done, but it does not make things plug-and-play yet."
At Partners, easing that pain involves co-developing an open source platform for SMART on FHIR applications, the standard way to build EHR apps, which it plans to unveil early next year. The goal is to help developers build less expensive, more portable applications, he said.
"The vision for this platform, when we created it, was to enable us to surgically interject new types of data and new forms of algorithmic support into the care delivery process to improve clinical decision-making," he said.
Algorithms don't work without trust
Getting to a place of algorithmic-based care requires trust, which Aronson described as multidimensional. He said protecting data, securing infrastructure at both the IT and clinical process level, as well as identifying unintended consequences are critical for establishing trust, but so is openness.
"If I can stand on my soapbox, [trust] means being open with business models so these technologies can benefit as many people as possible," he said, "but also so as many people as possible can evaluate them, criticize them and find problems to lead to their improvement."
Better processes or new processes?
But the idea of inserting data or algorithms into clinical processes raises a bigger question for Aronson. "If you were to start from scratch, would you fundamentally design these processes differently?" he asked. "What we are finding is that the answer to that question is often yes."
Older, existing clinical processes, some of which have been around for a hundred years, were created without modern algorithmic capabilities. Building from scratch could introduce "radically different" processes that are both cheaper and more effective, Aronson said.
While some view this kind of algorithmic-based care as impersonal, Aronson said the reality is algorithms are helping clinicians deliver greater personalization for patients and form greater personal connections with patients. He described algorithmic-based care as a system that models disease, not billing transactions.
"And based on those models, we want to be able to derive disease and care trajectories," he said. "And we want to use that to support the evolution of care. And we want to do it in a way that takes into account all of the different people inside and outside of the provider organization that are relevant to the patient's care as the patient progresses through their life and the disease lifecycles."
He pointed to cardiology as an example. Clinicians look at specific metrics around hypertension or lipids as predictors of heart failure. And while there are guidelines known to reduce mortality, only a small percentage of patients are treated according to those guidelines, according to Aronson.
The guidelines require a closely watched drug titration process, which means iterative interactions between doctors and patients. Not only can it be difficult to align schedules, that kind of interaction can also be uncomfortable for the patient, Aronson said. But an algorithm placed into the hands of patient guides who help the patient through the process can reduce that kind of intensity, empower the patient and, ultimately, increase the rate at which drugs are titrated.
"And when you do that," Aronson said, "the [mortality] numbers just drop."