What is clinical decision support (CDS)?

How it works

The US Office of the National Coordinator for Health Information Technology (ONC) describes CDS as a digital tool. It says CDS delivers timely information tailored to the individual, filtered and timed to be useful, to improve outcomes and quality of care. Its examples include condition-specific order sets, alerts and reminders, clinical guidelines, patient data summaries and diagnostic support. The tools run inside electronic health record (EHR) systems or as stand-alone systems and plug-ins.

A primer on the Agency for Healthcare Research and Quality's Patient Safety Network, last reviewed in 2025, says such systems were first developed over 40 years ago. Most were built from explicit rules, such as warning a prescriber about a documented drug allergy. It says these are increasingly giving way to more sophisticated methods, including machine learning, in which computers learn from examples.

Why it matters

The primer says decision support built into computerized ordering reliably prevents prescribing errors. It says less evidence supports its use to improve diagnosis, and too many or badly targeted warnings cause alert fatigue, blunting the systems' effect.

A 2021 study in JAMA Internal Medicine covered 38,455 hospitalizations at the University of Michigan's health system. It found the proprietary Epic Sepsis Model, at the hospital's alert threshold, did not identify 67% of patients with sepsis while generating alerts for 18% of all hospitalized patients.

How the US regulates it as of 2026

Not all CDS software is a medical device. Food and Drug Administration guidance issued on January 29, 2026 interprets a four-part test added to federal law by the 21st Century Cures Act. Under that test, CDS software falls outside the device definition only if it:

• does not analyze medical images or signals such as ECG waveforms
• works from medical information
• gives recommendations to a health care professional
• lets that professional independently review the basis for them

The FDA says software making recommendations to patients or caregivers meets the device definition. It says tools meant for critical, time-sensitive decisions fail the fourth criterion. The guidance is nonbinding.

How widely predictive AI is used

A September 2025 federal data brief on ONC's website covers US non-federal acute care hospitals. It found 71% reported using predictive AI integrated with their EHR in 2024, up from 66% in 2023. Use was uneven: 86% of hospitals in multi-hospital systems, against 37% of independent hospitals.