NOTES / HEALTHCARE AI

Clinical decision support

What clinical decision support is, why alerts fail, how the US FDA separates device from non-device CDS in broad terms, and how human factors decide whether a tool helps.

UPDATED 18 NOV 2026#healthcare#cds#fda#human-factors

Concepts only. Not medical, legal or regulatory advice. Rules change and differ by country; read the current guidance.

What it is

Clinical decision support (CDS) software gives clinicians information to inform a decision: a drug-interaction warning, a risk score, a retrieved guideline. Most AI in clinical settings is of this kind, a tool used alongside a person rather than an autonomous agent.2 That matches the levels of autonomy idea: keep a human responsible for the decision.

Is it a device?

Diagram: first question is whether software is a medical device Diagram: my own sketch; intended use drives the answer.

In the US, whether software is a regulated device depends on its intended use, not on whether it uses AI. The FDA’s CDS guidance explains when decision support falls outside the device definition.1

Diagram: simplified four-question CDS test Diagram: my simplified paraphrase of the criteria; the guidance has detail and exceptions, and it has been revised, so check the current version.

In broad terms the criteria ask whether the software avoids analysing images or signals, displays or analyses medical information, only supports a clinician’s decision, and lets the clinician independently review the basis for its output.1 A tool that gives a time-critical directive, or whose reasoning a clinician cannot inspect, is more likely to be treated as a device. The FDA also keeps a public list of AI-enabled devices it has authorised.3

Why good models still fail in practice

  • Alert fatigue: too many low-value alerts get ignored.
  • Workflow fit: a tool outside the normal screen is not used.
  • Automation bias: people over-trust a confident suggestion.
  • Explanation: a clinician who cannot see the basis cannot judge it.

This is why DECIDE-AI focuses early studies on human factors and safety, not just accuracy.2 For LLM-based CDS, grounding on retrieved sources helps with review; see RAG overview and human approval.

Sources

  1. [1]Clinical Decision Support Software, guidance for industry and FDA staff — US FDA · 2026PARAPHRASED
  2. [2]Reporting guideline for the early-stage clinical evaluation of decision support systems driven by AI: DECIDE-AI — Vasey et al. · 2022PARAPHRASED
  3. [3]Artificial Intelligence-Enabled Medical Devices — US FDA · 2026PARAPHRASED

© Ashish Kumar. All rights reserved. Quoted and cited material belongs to its owners. How I attribute sources.

Product, company and framework names are trademarks of their respective owners. Mentioning them does not imply affiliation or endorsement.