Clinical Decision Support (CDS) is software that informs a clinician's decision while leaving the decision itself to them. It spans drug-interaction alerts, risk scores, reminders about clinical guidelines, and AI-generated suggestions — all presented to a clinician who remains the decision-maker. The defining quality is that CDS advises rather than decides; the human stays in the loop and accountable.
The regulatory line here is precise and consequential. The 21st Century Cures Act carved out a non-device lane for certain CDS, and FDA's 2022 guidance interprets that exemption narrowly. A central condition is that the tool must let the clinician independently review the basis for its recommendations — the inputs, the sources, the reasoning — rather than simply trusting the output. If the clinician cannot reasonably understand why the software is recommending something and would have to rely on it, the software is more likely to fall under FDA Software as a Medical Device (SaMD) regulation.
That transparency requirement is, in effect, an interface specification in disguise. To stay on the non-device side, your CDS must show its work: surface the source guideline or data, the patient inputs it used, and the logic that led to the recommendation, in a form a clinician can actually evaluate at the point of care. The common mistake is building a black-box recommender that just emits an answer — efficient to use, but exactly the design that pulls a product into device scope and the clearance obligations that come with it.
