I am a pediatric cardiologist and have cared for children with heart disease for the past three decades. In addition, I have an educational background in business and finance as well as healthcare administration and global health – I gained a Masters Degree in Public Health from UCLA and taught Global Health there after I completed the program.
“It is imperative that clinicians have the knowledge and skills to assess and determine the appropriate application of AI outputs, for their own clinical practice and for their patients.”
Cornelius A James, clinician
Three clinicians put together this timely piece in JAMA Viewpoint, outlining the steps that would allow clinicians to be (better) engaged with AI in healthcare. The authors correctly point out that, while AI is thought by some to be hyped, it has enough substance to influence most, if not all subspecialties in the near future. The crux of its future will be predicated on how clinicians can be engaged in AI activities in healthcare, as well as be involved in its oversight.
The authors are spot on in the title of the first section of this viewpoint: welcome skepticism, avoid cynicism. The notion promulgated by AI experts that AI can replace certain clinicians, however, is dated and no longer widely discussed at meetings that focus on AI in healthcare. One can sometimes counter the excessive cynicism voiced around AI in healthcare with whether it is ethical to continue to delay deployment of AI in certain practices of healthcare. The authors go on to remind us of the lessons learned in a prior paradigm shift of electronic health records, with perhaps the most important one being the EHR’s negative sequelae on clinicians. It is hoped that AI will not repeat that oversight, with hundreds of startups and billions of dollars already in place.
There is an important difference between the adoption processes of EHR and AI: the former was aided by a federal mandate with incentives, while the latter will need to be driven more by return on investment, as well as awareness of regulatory and legal guidelines. Despite the overall lackluster performance of AI models in a real world practice setting, the estimated investments in this area continue to be on an exponential scale. There is therefore a dire need for a standardized reporting guideline for AI technologies, and this is much easier stated by authors than executed by practitioners.
Lastly, what is very much needed is clinician education, not only in AI methodologies, but also appraisal of these tools in clinical practice. Although the authors draw an analogy of this AI education to that of evidence-based medicine in previous decades, there is a sizable difference in the level of difficulty of these two knowledge domains that is not emphasized.
The proposed user guide of AI methodologies by these authors is perhaps idealistic rather than realistic. While some of the content in this viewpoint is not necessarily novel, it is nevertheless a good summary of the major issues in widespread adoption of AI in clinical medicine and healthcare. The ultimate aim of this AI adoption, of course, needs to be aligned with shared decision making and the Quintuple Aim.