Artificial intelligence is now making its way into the healthcare field with the potential to influence clinical care and operational efficiency. But does that mean robots will replace hospital staff and clinical care providers? Not quite.
Here, Adam Weinstein, executive director of analytics and data sciences at Watertown, Mass.-based athenahealth, discusses how artificial intelligence will affect health systems in the short and long term, including clinical applications, population health management, data gathering and operational efficiency.
Q: How will AI change the healthcare system and affect physicians and hospitals over the next 20 years? How should hospitals prepare?
Adam Weinstein: Today there's a lot of hype and experimentation on artificial intelligence in healthcare, but we aren't seeing a ton of AI used in clinical settings on a routine, ongoing basis. You have systems like Watson running in select locations, but that hasn't impacted the broader healthcare system yet. Currently, most of the energy still goes into the use of advanced statistics on large claims datasets for population health and optimizing reimbursement.
That said, if you think about the potential opportunities over the next five years, I absolutely think we will see the broad use of AI in nearly every aspect of the healthcare system, whether it's clinical delivery or operational improvement. The real question is: what will end up being the healthcare AI killer app that becomes ubiquitous in the same way that we use machine learning in Google or Bing for search; or when we want to get from point A to point B, through Waze or Uber?
Q: How do you expect AI to integrate into the healthcare system on the provider level in the future?
AW: The patient and population health management side of healthcare today runs largely on the same processes and systems it has over the past 20 years; it's still fee-for-service, people processing claims for payment and then applying analytics on top of this. You have centers of nurse care managers making calls to manage chronic disease and healthcare teams operating on data that's weeks or months old. In the future, we'll move to a proactive and automated system where the system detects a potential problem, helps determine an intervention and uses digital marketing best practices to connect with the targeted population.
The AI system will know our habits and whether people prefer text or email for healthcare information; they'll know how to approach us. We are just starting to apply that to healthcare in the same way other industries have. If you use the technology to more readily spot outliers and think about behavior management or wellness coaching before chronic diseases progress, the AI system can do the heavy lifting and data gathering, and keep healthcare providers involved appropriately.
On the operational side, a big issue is digitized data. Healthcare is the last holdout of the fax machine. There is still a huge amount of paper records in the healthcare system, and we hope that changes over the next five years. Data systems should be able to retrieve necessary information, determine who needs that information and deliver it to the right person to support the continuum of care between providers, regardless of how that information comes in. That would take a huge amount of operational costs out of the system. These are some of the areas of AI we are focused on at athenahealth.
Q: What barriers are there to AI becoming more integrated in the healthcare system?
AW: There have been and continue to be technical barriers to AI in healthcare. The systems aren't easy to set up. A typical doctor's office won't set up an AI system, but a larger healthcare system may have the resources to.
There is also a people barrier; we need to find the people who can design and operate these systems.
The third barrier is gathering clean data. Any of these systems without a good dataset underneath it is just a technical paperweight. It can't do anything without clean data, and healthcare data is notoriously fragmented and messy. That problem hasn't gone away, and you can use AI to help with data translation, but you can't get away from the garbage-in, garbage-out problem.
Q: How can health systems develop an AI strategy? What options are available?
AW: I don't think every health system needs to prepare for AI right now, but they probably will over the next five years. One step a system can take is to decide its future AI strategy, beginning with a conversation with its current IT vendors. Ask the vendors what their AI strategy is; some have one and some don't. Based on that conversation, the health system can decide whether it is willing to go along with its vendor's strategy or develop its own.
Health systems that want to be on the leading edge can consider a best-of-breed approach. There are hundreds of healthcare AI startups that systems can bring in, but then they have the pain of integration. Most startups will run into the bumps of immature products. The health system must also hire high-powered people to help manage the integration.
The third option is for health systems to launch AI by themselves. If they have the money and data analytics capabilities, they can bring the expertise in-house and use existing technologies to develop their strategy. This is the toughest route but it gives health systems the opportunity to differentiate. Overall though, I doubt most healthcare systems will be successful in trying to pull this off by themselves.
Q: What do you see as the biggest opportunity in healthcare AI going forward?
AW: In the long term, the sky's the limit. If we can get the human and incentive components right, the technology will soon exist for AI to be ubiquitous in healthcare. You can imagine systems doing population health behind the scenes, tailoring communication not only when things go bad, but getting in front of the issues with wellness and behavioral modification, while keeping humans appropriately in the loop the whole time.
This isn't physicians versus technologies; that won't be successful. It should be the computer and doctor working together to make each other smarter as they operate.
If you start thinking about genomics and proteomics and how it will disrupt the drug industry, that could affect the FDA clearance process and clinical trials could run more smoothly and quickly. As that cycle quickens and the pace of drug discovery and personalization accelerates, we will be increasingly reliant on clinical decision support systems, tailoring clinical recommendations on a patient-by-patient basis. You can imagine the virtuous cycle from wellness to disease intervention. The healthcare system of the future will in some ways look similar, but will look and feel different from the patient perspective — hopefully much more effective and empowering.