Healthcare AI’s slow progress from concept to clinic: an ‘inconvenient truth’

Few healthcare AI models introduced in the medical literature during the present AI research boom can be used in actual patient care.

This “inconvenient truth” deserves attention, according to the authors of a paper fleshing out the disconnect between the technology’s highly publicized potential and its paltry use in medical practice to date.

The senior author of the paper, which is running in npj Digital Medicine, is Leo Anthony Celi, MD, MPH, MSc, who is a principal research scientist at MIT, a staff physician at Beth Israel Deaconess Medical Center and an associate professor at Harvard Medical School.

Celi and colleagues identify two main reasons for AI’s lack of penetration in healthcare beyond research settings.

One, the innovations can’t do much on their own to change clinicians’ established ways of working.

“A complex web of ingrained political and economic factors as well as the proximal influence of medical practice norms and commercial interests determine the way healthcare is delivered,” the authors explain. “Simply adding AI applications to a fragmented system will not create sustainable change.”

And two, most provider organizations don’t have the kind of data infrastructure it would take to train algorithms for serving vast and varied populations.

The tools would need to both fit local practice patterns and operate consistently, the authors emphasize, even for patient subgroups whose characteristics were underrepresented in training datasets.

The hurdles don’t end there. Celi and colleagues point to widespread dissatisfaction among clinicians with EMRs, which not infrequently translates into incomplete entry of germane health data. Health data, and lots of it, is of course the lifeblood of usable healthcare AI.

Meanwhile, even if efforts to change the underlying EMR-averse culture were to succeed, data integration and interoperability among and between different providers remains more goal than process.

Other snags the authors describe include basic yet still unanswered questions about health data—who owns it, who’s responsible for it and who can use it—as well as widespread worries around data privacy and security.

Tackling these questions and concerns will require hashing them out in the public square and getting them addressed by policymakers.  

“The specific path forward,” Celi and colleagues write, “will depend on the degree of a social compact around healthcare itself as a public good, the tolerance to public-private partnership and, crucially, the public’s trust in both governments and the private sector to treat their healthcare data with due care and attention in the face of both commercial and political perverse incentives.”

On the bullish side, the growth of affordable cloud computing has been a boon to small and medium-sized tech companies as well as the giants, opening doors for innovators in many corners of healthcare, the authors point out.

“Now that the growth of cloud computing in the broader economy has bridged the computing gap, the opportunity exists to both transform population health and realize the potential of AI” in healthcare, they write.

To get there from here, Celi et al. conclude, governments must be willing to “foster a productive resolution to issues of ownership of healthcare data through a process that necessarily transcends election cycles and overcomes or co-opts the vested interests that maintain the status quo—a tall order.”

Without such a shift in the public and political will, they predict, “opportunities for AI in healthcare will remain just that—opportunities.”