From medical malpractice to single-payer healthcare, there is no shortage of topics to debate in healthcare. Though provision, funding and government involvement in our system continue to be central topics, there is also an emerging consensus on two key themes: the importance of aligning financial and outcome incentives through alternative payment models (APMs), and a critical focus on resolving data interoperability issues between siloed health systems that are directly limiting progress across the industry.
Fortunately, this is where emerging technology tools stand to have a significant near-term impact on the health of the American public. Unfortunately, there is also a tremendous amount of institutional inertia pushing against the adoption of solutions that could support alignment through APMs and address data interoperability issues.
This pushback comes as no surprise, as it is rooted in what are ultimately humanitarian concerns, such as loss of jobs. AI solutions deployed at scale will eliminate much of the “wasted spend” estimated at 25% of the $3.8 trillion spent on healthcare in 2019. Efficiency gains realized by AI will come, in part, at the expense of many jobs in the industry, a reality that must be considered as part of any successful deployment strategy.
Here, I will briefly outline some opportunities for AI technologies with respect to APMs and interoperability.
Alternative Payment Models
It is widely believed that some version of “value-based” contracting (payment based on health outcomes and overall "quality" of service delivered as opposed to service volume), a core component of APMs, will reduce overall costs while simultaneously improving patient outcomes.
However, this notion of “value” is most typically ill defined. This has resulted in programs with ambiguous outcomes, leaving both patients and providers wondering whether meaningful value has been achieved.
Even when well defined, as in the various quality programs administered by CMS, it can be difficult and costly to collect all the necessary information to measure value. Quality improvement initiatives require measurement, often through registries that enable providers to assess and track how their patients are doing in terms of key aspects of care and potential complications in order to identify areas for improvement. These registries require time and substantial resources to implement and coordinate.
Finally, once all the data is collected, the traditional statistical methods prevalent in the industry are inadequate for effectively modeling and dynamically managing risk in patient populations.
These tasks of measuring, predicting and even defining value are well suited to computational optimization. Emerging technology platforms will propel these efforts by integrating novel, predictive patient data streams with modern virtual care management tools. Of course, this will only be true if these platforms are able to efficiently and securely exchange information with the existing systems.
Interoperability
Health systems and payers have been relatively slow to adopt modern information management systems. Instead, they remain dependent on outdated technology (fax machines, copiers) and mostly human-managed processes. This has long been identified as a source of wasted spend in the U.S. healthcare system, and one that is ripe for remediation through existing machine learning and process automation technologies.
The major cloud hosting providers will continue using the AI tools they develop to better manage the complex flow of information, easily saving the industry tens of billions of dollars.
An obvious requirement for these efforts to be successful is establishing the trust between the tech firms, health systems and the patients they serve. This could prove to be a legitimate use case for distributed ledger technologies, as efforts there continue to mature.
Tying It Together
One of the important consequences of outdated IT practices is poor data fluidity within and between organizations. This is where APMs and interoperability are critically interrelated, as data fluidity is essential in order to train machine learning (ML) models, as well as to perform quantitative assessment of APM variants. Both of these elements are foundational for the larger-scale objective of matching individual patients to optimal care plans given all the data the system knows about them.
Payment models and interoperability may not be as exciting as deep network-based image diagnostics or virtual reality robo-surgeons. However, it is in these core domains of healthcare operations where AI will have its biggest impact — assuming, of course, that institutional inertia, regulatory agencies and other “political” forces do not interfere.
The debate over how we will spend all the money we save will have to wait for another time.
