Decision Support and 21st Century Cures
Nicholas Genes, MD, PhD
Emergency Medicine Informatics Section Chair-Elect
Icahn School of Medicine at Mount Sinai, New York, NY
In our March Newsletter, Jason Shapiro covered the 21st Century Cures Act, which passed the house and was signed by President Obama in December 2016.
While there’s plenty in the nearly 1000-page bill on EHRs, interoperability, and telehealth, a significant part of the bill also amended FDA oversight of clinical decision support, with an eye toward “big data” algorithms and AI-like guidance. Specifically, the FDA was made exempt from regulating any software that allowed physicians to “independently review” the basis of CDS recommendations – like the simple “if-then” statements behind drug-drug interactions, or more sophisticated sepsis alerts that are triggered by a combination of demographics, exam findings, and lab results. Because a doctor can easily determine the factors that went into a CDS recommendation, the FDA won’t be reviewing those tools.
But the next generation of CDS is going to be trickier to explain to doctors – weighing dozens or hundreds of variables determined by previous unsupervised “deep learning” processes, and functioning more like a black box to practitioners. Already, radiologists are forced to contend with a simple version of this scenario, when their algorithms identify one mammogram lesion as cancerous, but another one as benign. The new rules in the 21st Century Cures Act will prompt the FDA to regulate black box CDS, whenever the logic behind an alert can’t be independently verified by healthcare practitioners.
But what would this form of FDA regulation look like? Machine learning algorithms are constantly evolving and responding to new data; accordingly, the FDA would need assurance that the CDS performance is consistently improving. It’s a new way of evaluating technology that’s more akin to post-marketing surveillance for drugs, or maintenance of certification for physicians. In recent years, the FDA has declined to clarify its guidance on how it’ll approach CDS, prompting a group called the CDS Coalition to issue guidelines of its own in an attempt to self-regulate the industry. The Coalition’s draft of industry guidelines is available now for public comment. If you’re interested in the future of clinical decision support, and the role of physicians and “big data” in making medical decisions, check it out.
Return to Newsletter