The FDA has reissued a 2017 draft guidance for clinical decision support (CDS) applications as part of a reexamination of software regulations, and one difference between the 2017 draft and the latest iteration that is sure to cheer developers is that the agency now provides much more clarity as to the types of functions that will and will not fall under regulatory scrutiny going forward.

The 2017 draft was released concurrently with the draft guidance for software as a medical device, and both were accompanied by a statement by then-FDA Commissioner Scott Gottlieb, a move intended to ensure stakeholders the agency was bringing its resources to bear on the digital health predicament. Those who commented to the docket for the previous draft exhibited little affinity for that draft, however, arguing that it would force a number of such products off the market. (See BioWorld MedTech, March 1, 2018.)

The new draft runs to 27 pages, double the 13 pages in the 2017 version, and one significant difference is that the new version cites the September 2014 final guidance for risk categorization of software as a medical device (SaMD) by the International Medical Device Regulators Forum (IMDRF). The FDA noted that the IMDRF document cites the significance of any health care decisions derived in part from SaMD as a risk factor, as well as the seriousness of the patient's condition. The citation of the IMDRF guidance might be seen as part of a larger effort to harmonize the FDA regulatory approach with international standards, which will at some point be applicable to the agency's approach to quality systems management as well. (See BioWorld MedTech, Sept. 27, 2019.)

The omission of any mention of the 21st Century Cures Act in the 2017 draft might seem curious given that the Cures Act was passed prior to the publication of that draft, but the FDA noted in the latest draft that the legislation imposed restrictions on regulatory oversight. Among the four entries that are exempt from regulatory oversight are functions that reference medical information appearing in the peer-reviewed literature, and any function that aids in the effort to diagnose a patient so long as the function is not presented as the primary source of diagnostic expertise. These are characterized as non-device CDS.

PACs exempt if no processing of image

The draft said that any CDS that will be exempt from oversight must meet all four of the described conditions, however. A non-CDS cannot be designed to acquire, process or analyze a medical image or signal from an in vitro diagnostic, and cannot process any pattern or signal from a signal acquisition system. The description of image processing would seem to exempt picture archiving systems (PACs) for image storage only, and the agency notes that a CDS for measurement of physiological parameters would be a non-CDS so long as that product is not marketed or intended for any use that would qualify as a device per se.

The exemption of a CDS that is used to aid in the making of a diagnosis carries the stipulation that the developer describe the basis for arriving at the diagnosis. Another requirement is that the developer provide a function that would allow the user to examine the evidence used at arriving at a diagnosis with a sufficient amount of information to evaluate that diagnosis.

The draft describes the risk management framework found in the 2104 IMDRF guidance, and while the FDA draft provides some leeway for artificial intelligence and machine learning not found in the 2017 draft, the use of a machine learning algorithm is very much restricted where a regulated CDS for medical professional use is concerned. The only such example given is a diagnostic that suggests a need for cholesterol management, and this comes with the proviso that this allowance is based on the agency's current understanding of the risks associated with such uses. Also on the list of "at the present time" exemptions are software for making recommendations regarding diagnoses based on allergens and common cold symptoms, as well as the selection of over-the-counter remedies for seasonal allergies.

In the category of CDS items intended for patient/consumer use, the FDA said it will exempt software that "provides information" regarding the use of a prescription drug based on the patient's prescription, so long as the software does not recommend changes to dose or a cessation of use.

The draft spells out a number of conditions in which a CDS for health care professional (HCP) use would be subject to regulatory scrutiny, such as a machine learning algorithm that makes recommendations regarding seasonal allergies when the underlying data and logic are not transparent to the HCP. Any software used to identify patients who may exhibit signs of opioid addiction on the basis of prescription patterns and electronic health record data would also be subject to regulation, as are CDS for patient/consumer use that provide information related to glucose management for those with type 2 diabetes. Several functions that would ordinarily not fall into the category of a CDS as such, but which the agency would nonetheless regulate, are those used to create treatment plans for patients undergoing external beam radiotherapy or brachytherapy.

The agency said that upon finalization of the CDS draft, it will make conforming changes to the just-released draft guidance titled, "Policy for Medical Device Software Functions and Mobile Medical Applications." The comment period for the new CDS draft closes Dec. 26, 2019, and appears under the existing docket number, FDA-2017-D-6569.

No Comments