Rama Doddi
What they argued
Accepts Class II patient-facing triage devices directing to telehealth nurse or ED with justified escalation balance; proposes standardized OOD deferral benchmark (95%).
Themes it raises
Across the five cross-cutting questions
High-consequence work: Directs
The comment as filed
I am a Regulatory Affairs professional with experience in medical-device regulatory strategy. Please see below my responses/suggestions for a few questions.
A.6. GenAI-Enabled Care Escalation
I recommend that FDA not treat under-escalation and over-escalation as equal risks. For Class II triage devices, manufacturers should be permitted to prioritize high sensitivity and minimize under-escalation, as failure to identify a serious condition could result in significant patient harm.
A higher rate of over-escalation may be acceptable when the device provides an appropriate intermediate escalation pathway. For example, instead of directing the patient immediately to an emergency department, the device could connect the patient with a "qualified telehealth nurse or other healthcare professional" for further assessment. Manufacturers should justify the balance between under- and over-escalation based on the device’s intended use, patient population, clinical risks, and escalation pathway.
S.3. Calibration, Uncertainty Communication, and Clinical Deferral
I recommend that FDA consider establishing a standardized out-of-domain testing dataset for GenAI-enabled medical devices. This would provide a consistent method for evaluating whether a device appropriately recognizes situations outside its intended use or validated knowledge boundaries.
For premarket evaluation, manufacturers could be required to demonstrate a predefined threshold for appropriate clinical deferral when presented with adversarial or out-of-domain scenarios as applicable. For example, FDA could consider a benchmark of at least 95% appropriate clinical deferral for predefined scenarios outside the device’s intended scope, with the threshold adjusted based on the device’s intended use and risk profile.
The evaluation should also consider over-deferral, since excessive referral to a clinician for questions within the device’s validated scope could reduce clinical utility. Manufacturers should therefore demonstrate that the device can appropriately distinguish between situations requiring clinical intervention and those within its validated capabilities.