Clinical decision support systems: what they are and how to evaluate one

See also our free clinical calculators, each with its primary references.

Clinical medicine asks one person to hold a great deal in mind at once: the history, the exam, prior results, current guidelines, what was already done, and what the next step should be. Clinical decision support exists because that load is real. The question for any practice or health system is not whether decision support is a good idea in principle. It is whether a specific system reduces the load on your clinicians or adds to it.

What counts as a CDSS

One widely cited review defines a clinical decision support system as software intended to improve healthcare delivery by enhancing medical decisions with targeted clinical knowledge, patient information and other health information.^[1] It describes two broad families:^[1]

• Knowledge-based systems, in which rules (IF-THEN statements) are written and the system retrieves data to evaluate them. Drug-interaction checks, preventive-care reminders and most order-set logic work this way.
• Non-knowledge-based systems, which use machine learning or statistical pattern recognition instead of hand-written rules.

In practice the category is broad. A dose-range check, a sepsis screening alert, a risk calculator, a differential-diagnosis generator and a system that reads a note and proposes a workup are all decision support. They fail in different ways, so they should be evaluated differently.

What the evidence shows

The evidence is large and consistent in one direction and thin in another.

• A systematic review of 148 randomized trials found decision support effective at improving process measures, such as ordering preventive services (OR 1.42), ordering clinical studies (OR 1.72) and prescribing (OR 1.57). Evidence for clinical, economic, workload and efficiency outcomes remained sparse; only 20% of trials assessed clinical outcomes.^[2]
• A meta-analysis of controlled trials found that decision support increased the proportion of patients receiving desired care by 5.8% (95% CI 4.0% to 7.6%), with substantial heterogeneity. For clinical endpoints the median improvement was 0.3%.^[3]
• The trials that did improve practice shared features: support delivered automatically in the workflow, at the time and place of the decision, with a recommendation rather than just an assessment. Of 32 systems with all four features, 30 (94%) improved practice.^[4]

The honest summary: well-designed decision support moves what clinicians do, by a modest average amount, and the effect depends heavily on design. Claims that a system improves outcomes deserve a request for the study.

The known risks

The same review lists the pitfalls: fragmented workflows, alert fatigue and inappropriate alerts, effects on user skill, dependence on computer literacy, the burden of maintaining content, poor data quality and incorrect content, poor transportability and interoperability, and financial strain.^[1] It notes that up to 95% of CDS alerts are inconsequential.^[1] Alert fatigue alone is enough to sink an otherwise sound system.

How regulators look at it

FDA. Section 520(o)(1)(E) of the Federal Food, Drug, and Cosmetic Act excludes certain clinical decision support software from the definition of a device. FDA's current guidance, issued January 29, 2026, explains the four criteria. In short, the software must not acquire, process or analyze a medical image or a signal from an in vitro diagnostic or signal-acquisition system; it must display, analyze or print medical information; it must support or provide recommendations to a health care professional; and it must enable that professional to independently review the basis for the recommendation, so that they do not rely primarily on it.^[5] Whether a given product meets those criteria is a product-specific question. "Not FDA-cleared" and "not a device" are different statements.

ONC. For certified health IT, the HTI-1 final rule created a "decision support interventions" criterion (45 CFR 170.315(b)(11)). It requires certified systems to let users select evidence-based and predictive interventions, to support "source attributes" describing how an intervention was developed and performs, and to apply risk management to predictive interventions.^[6] A December 2025 proposed rule would remove the source-attribute and risk-management requirements; as of this writing it has not been finalized.^[7]

Interoperability. CDS Hooks, an HL7 specification, defines a standard way for an EHR to call an external decision-support service at a point in the workflow and receive cards back.^[8]

Ten questions to ask any vendor

1. What exactly does it return? A list of alerts, prose to read, or structured items (coded orders, referrals, escalation criteria) the chart can act on?
2. Where does it appear? Inside the note or order at the moment of the decision, or in a separate window?
3. Does it know what was already done? A recommendation for a test the patient had yesterday is worse than no recommendation.
4. How specific are its interruptions? Ask for override rates or false-positive rates for any interruptive alert.
5. Can the clinician see the basis? Every recommendation should show its reason and its source.
6. What evidence supports it? Published validation, a local pilot, or none yet? All three exist; know which you are buying.
7. What happens to patient data? Is there a business associate agreement, which cloud and model providers sit behind it, and is data retained?
8. How is content maintained? Who updates it when a guideline changes, and how fast?
9. How does it integrate? API, CDS Hooks, a native EHR module, or copy and paste?
10. What does it cost, and is the price published?

Where Pogosh fits

Pogosh is decision support that reads the note and returns structured next steps: labs with LOINC codes, imaging with RadLex codes, specialist routing adapted to the setting, specialty, age and sex (including what happens after the emergency department), escalation criteria, and a separate strict emergency check. It offers relevant calculators but never runs them on its own, and it can draft a note without inventing negatives. It is available as a REST API with a HIPAA business associate agreement on AWS, and as an OpenEMR module. Pricing is published: $149 per month for a clinic and $299 per month per location for a practice, a $499 monthly minimum for EHR platforms, and custom enterprise terms.

Against the questions above, two answers matter most. Everything Pogosh returns is advisory; a clinician reviews it and can override all of it. And Pogosh is not FDA-cleared and has no published independent validation yet. Judge it the way you would judge any system: on your own cases.