The FTC recently posted its Closing Letter to HeHealth, Inc., which provides an AI-powered application, Calmara, for the detection of sexually transmitted infections (“STIs”). HeHealth marketed the app as having an 94%+ accuracy rate in detecting over ten STIs, including HPV, syphilis and herpes.  

In June, FTC staff issued a civil investigative demand to HeHealth identifying concerns related to HeHealth’s advertising claims and seeking, among other things, substantiation for those claims in accordance with the standard required for health claims, i.e., competent and reliable scientific evidence. FTC staff also sought information about the company’s privacy practices, in light of the sensitivity of the images and information collected by HeHealth from users and the company's claims about maintaining anonymity of users.  Significantly, FTC staff also sent a Notice of Penalty Offenses Concerning Substantiation of Product Claims to HeHealth, putting the company on notice that making unsubstantiated health claims is an unlawful act or practice.

In response, HeHealth provided information to, and engaged in discussion with, FTC staff.  And the upshot of those discussions is that the company agreed to terminate the STI detection application and remove all related advertising claims; refund all U.S. customers with unfulfilled app purchases; delete all customer personal information obtained through the app; and delete customer personal information obtained or accessed through its payment vendor.  And it agreed to do all of this by the just past-deadline of July 15^th.  The principals of the business are also prohibited from making misrepresentations about any similar medical detection product in the future. 

So, what was the problem with the information provided by HeHealth, information so deficient that it had to basically shut down its business in a month’s time?  According to the Closing Letter, the data HeHealth used to test and train the AI detection model included images uploaded by individuals who were never subjected to diagnostic tests to confirm whether the individual associated with the image actually had an STI.  Also, the performance of the AI model was assessed on a small number of images, and was only trained to detect visual symptoms like marks or lesions, limiting its reliability, particularly for asymptomatic individuals.  Finally, the reliability of HeHealth’s study was also undercut by the fact that four of the five authors of the study either worked for HeHealth or were paid consultants. 

Take-aways?  Methods for conducting studies and collecting data may evolve, particularly as AI-powered tools improve and become more ubiquitous and accepted, but the standard for substantiating a health or safety claim will likely remain the same: an advertiser must have, at the time of making its claims, competent and reliable scientific evidence that has been conducted and evaluated in an objective manner by qualified persons and that is generally accepted in the profession to yield accurate and reliable results. And the FTC is watching.