AMIA: Clinical decision apps lacking human interaction a risk

U.S. Food and Drug Administration officials are taking careful aim at clinical decision support functions in mobile medical apps, officials at the agency's public workshop on mobile medical app regulation said late last week.

Regulators clarified their definition of CDS a bit, indicating they must collect patient-specific information, analyze it in some way, and make recommendations based on that data.

Three key factors in the FDA's determination of a CDS function's risk--and therefore the risk of the mobile device they're embedded in--according to a presentation by FDA statistician Kristin Meier, include:

  • What impact the CDS could have on the patient's health;
  • How well accepted the CDS' findings are in clinical practice; and
  • How easily the system can identify errors.

A major part of the risk determination should include whether the CDS is mediated by a human being or not, according to Meryl Bloomrosen, public policy and government relations VP for the American Medical Informatics Association (AMIA).

"The FDA's most rigorous attention, should be given to applications that provide CDS in an automatic and autonomous fashion, and which intervene directly, based on patient care data," Bloomrosen said in a statement.

This generated some discussion among attendees, according to Satish Mishra, a blogger for iMedialApps. The presence of a "learned intermediary" should be a critical factor in the FDA's risk assessment, Mishra points out. "Basically the idea is that there is someone between the patient and the app [a physician, nurse, etc.] who either uses the app's output as part of a larger clinical picture, or can recognize erroneous output," she writes.

The FDA appears to take a different view, however, with Meier specifically noting that the input mechanism for patient data--human, automatic, etc.--isn't relevant to the FDA's definition. Meier did acknowledge that regulators definitely will consider what method is used to analyze the collected information: algorithms, formulae, data lookups, or rules and associations. It's not clear yet which of these regulators consider the most reliable, or least risky, method.

The FDA may need to parse out their definition of a CDS more finely, Mishra points out, adding that CDS can cover everything from an app that makes general preventive health recommendations to an app that can change the insulin deliver rate of an insulin pump.

Meier, in her presentation, notes that the agency considers some CDS capabilities, like calculating an Apgar score, to be lower risk than the ability of a system to view/analyze medical images or determine radiation therapy protocols.

To learn more:
- read through Meier's presentation
- dig into iMedicalApps' coverage
- check out Health Data Management's report
- look over the AMIA comments

Share this on Twitter

Suggested Articles

The newly launched Center for Connected Health will be largest telehealth hub in the Philadelphia region, according to Penn Medicine.

The FDA commissioner wants to use additional funding under Trump's budget to advance digital health initiatives and integrate real-world data.

The FDA's approval of an app that uses AI to notify specialists of a potential stroke offers new possibilities for triage software that uses CDS.