
Healthcare has spent nearly three decades laying the groundwork for AI adoption, shifting from paper records to a deeply digitized ecosystem. According to Ritu Agarwal, the Wm. Polk Carey distinguished professor of information systems and health at the Johns Hopkins Carey Business School, the industry has not yet figured out how to use AI at scale.
Scaling AI means embedding the technology into clinical and administrative workflows throughout an organization. Many healthcare organizations are working to implement AI, but they are struggling to move past small pilot cohorts. Agarwal notes that AI has the potential to address deficiencies in healthcare practice and delivery, but there’s a lot of work to be done before that potential is realized.
Challenges in AI Adoption
The fact that AI in healthcare is still in its infancy contributes to the challenges. There are unresolved issues about clinical workflow integration, governance challenges, physician trust, and return on investment that will take time and research to resolve. Agarwal says that developing AI models that can guarantee adequate performance under all conditions is tricky.
A model trained with data from one hospital may not perform well in another due to differences in patient populations. This concept of model drift happens when a model built in one context is applied in another. To overcome this, additional development, testing, and more representative data sets are needed to deploy models in everyday clinical practice.
Clinician trust of AI plays a role in accelerating adoption. Factors that determine whether a new AI tool is accepted by clinicians include integration within the clinical workflow and physician involvement in the development of AI applications. Agarwal believes that physicians will resist AI if it’s just added to their processes without redesigning them.
Physician involvement in the ideation and conceptualization of AI applications is important. Agarwal notes that vendors often have a chief medical officer and a handful of clinicians on their development team, but technologists are making the key decisions. Clinicians should be front and center, perhaps leading the development of AI applications.
Measuring ROI
When it comes to measuring the return on investment of AI, Agarwal advises health system leaders to put a measurement baseline system in place and clearly identify the outcomes they expect from a given AI. This could include efficiency, effectiveness, or actual health outcomes. She recommends picking specific disease conditions to evaluate AI and going about it systematically.
The emerging evidence suggests that the upside of AI in healthcare is potentially large, but it will take time to realize. Agarwal notes that in other industries, AI has shown benefits, and there’s no reason why healthcare can’t achieve the same. She advises health system leaders not to try to address all issues at once and to start with small, systematic evaluations.
In the middle of this development, it’s clear that the healthcare industry is at a crossroads. As AI continues to advance, the industry will likely see a shift towards more integrated solutions. This could lead to a more streamlined and efficient healthcare system, but it will require significant investment and development.
Agarwal predicts that there will be a lot of consolidation in the industry, with the winners being those who have the most extensive set of relationships with health systems and the most extensive sources of data. Companies like Google and Microsoft have leadership in the market due to their size and number of customers.
These companies are well-positioned to construct massive health data sets, which will give them a competitive advantage. The electronic health record companies have curated large data sets over time, but they are less nimble and flexible than the tech companies in terms of building innovative solutions. Agarwal believes that the winners will likely get bigger, and the companies offering highly specialized point solutions will get absorbed or bought out.
Leave a Reply