In short
- Medication decision support software checks a medication order against the patient's record, for interactions, allergies, drug-disease contraindications, duplicates, formulary status and relevant labs, and tells the prescriber before the order is signed.
- Certified EHRs already have to run drug-drug and drug-allergy checks during order entry, so an add-on product is worth having only if it is more specific to the patient and less noisy than those checks.
- FDA lists interaction checks, formulary guidance and lab flags tied to a prescription as examples of non-device decision support, provided the clinician can review the basis and the function is not time-critical or aimed at patients.
Medication decision support software checks a medication order against what is known about the patient, other active drugs, documented allergies, diagnoses and recent lab results, and tells the prescriber about interactions, contraindications, duplicates, formulary issues or a lab value that matters, before the order is signed. The check has to run inside the ordering workflow, use current drug knowledge and stay quiet when there is nothing useful to say. Making the checks relevant to the patient is usually harder than covering every type of check.
This guide is for product and engineering leads building medication checks, either inside an EHR or as a service that plugs into one, and it ends with where FDA draws the device line.
Which medication checks does decision support usually run?
FDA's Clinical Decision Support Software guidance, issued January 29, 2026, is a convenient catalog, because many of its examples of non-device decision support are medication checks:
- Drug-drug interactions and drug-allergy contraindications, identified from the current FDA-approved labeling or other up-to-date, peer-reviewed sources plus patient-specific information.
- A tool where a clinician enters several drug names and sees known interactions between them.
- Drug-disease interactions, such as notifying a clinician that a patient with asthma should not be prescribed a non-selective beta-blocker.
- Formulary guidance: listing formulary medications for a condition, and alerting clinicians to formulary changes with suggested alternatives.
- Duplicate therapy and medication reconciliation notifications.
- Lab values flagged in response to an order, with FDA's example of low potassium or magnesium flagged when digoxin is prescribed.
- Recommendations on using a prescription drug for the patient's documented condition, consistent with the drug's current approved labeling.
FDA adds a practical note in a footnote: the information the software relies on should be kept up to date, its source should be displayed prominently, and users should have a way to reach current information. Its example is that interaction alerts should link directly to a trusted, up-to-date source such as DailyMed.
What do certified EHRs already do?
If your customer uses a certified EHR, basic checking is already there. ONC's certification criterion 170.315(a)(4) requires that, before a medication order is completed and acted on during computerized provider order entry, the system automatically indicates drug-drug and drug-allergy contraindications to the user. The severity level of drug-drug interaction alerts must be adjustable, but only by a specific set of identified users or as a system administration function. ONC notes that adjusting severity changes what is displayed, not the clinical significance of the interaction.
The broader decision support criterion, 170.315(b)(11), requires certified EHRs to let authorized users turn on evidence-based interventions driven by problems, medications, allergies and intolerances, demographics, lab results, vital signs and procedures, in addition to the drug-drug and drug-allergy checks. Those interventions must expose source attributes such as a bibliographic citation, the developer and the release and revision dates, and users must be able to record feedback that can be exported.
For an add-on product, this means your card appears on a screen that already has alerts. Clinicians will compare yours with the alerts they already ignore, so it has to use more of the patient's data than the EHR's built-in checks do.
Where in the workflow should a medication check fire?
CDS Hooks, the HL7 standard for calling external decision support from the EHR, gives a vocabulary for this. The current published release, 2.0.1, is based on FHIR R4. Its hook library includes order-select, for when a user selects a medication during an ordering workflow, and order-sign, which carries the orders under review in a draftOrders context field. The library also lists medication-refill, allergyintolerance-create and patient-view, among others.
- Interaction, allergy and drug-disease checks: order-select, so the prescriber sees the problem while still choosing the drug, with order-sign as the final check on the full set of draft orders.
- Lab-dependent checks, such as renal or electrolyte values: order-select or order-sign, using the latest results fetched for that patient.
- Formulary alternatives: order-select, offered as a suggestion the prescriber can accept in one step.
- Refill reviews: medication-refill, where the EHR supports it.
- Medication reconciliation: patient-view or a separate review screen, because it is a review task rather than an ordering one.
Two details in the spec shape your service. First, your service can declare prefetch queries, but the EHR may answer none, some or all of them, so the service must be able to fetch missing data itself or return HTTP 412 when it cannot get what it needs. Second, every card needs a summary, an indicator of info, warning or critical, and a source. The required source field fits neatly with FDA's advice to display where the knowledge comes from.
How do you keep drug knowledge current and consistent?
Medication checks fail without any visible error when drug names do not match. The same product can arrive as a brand name, a local formulary code or free text. The National Library of Medicine's RxNorm provides normalized names for clinical drugs and links them to other drug vocabularies. NLM describes it as able to mediate messages between systems that do not use the same software and vocabulary, which is exactly the problem at the edge of a medication check.
- Map every incoming medication to RxNorm at the edge, and route anything that does not map to a review queue rather than silently skipping the check.
- Record the version of the drug knowledge base used for each alert, so you can answer the question of what the system knew at the time.
- Show the source and its date on the card, and link to the full reference, as FDA recommends.
- Treat knowledge base updates like code releases: test them against a set of known patient scenarios before they reach production.
How do you avoid alert fatigue?
AHRQ's patient safety primer on alert fatigue says clinicians generally override the vast majority of order-entry warnings, including critical ones, largely because most alerts are clinically inconsequential and teach people to click past them. It recommends increasing specificity, tailoring alerts to patient data so they fire only for high-risk cases, with the example of tying nephrotoxic drug alerts to renal function, tiering alerts by severity and making only severe alerts interruptive, while low-severity ones become non-interruptive or are removed. It also notes that developers have been reluctant to remove alerts for fear of liability.
- Use labs, weight, age and active problems to suppress alerts that do not apply to this patient.
- Do not repeat an alert the prescriber has already acknowledged for the same drug pair in the same encounter.
- Reserve interruptive alerts for severe, patient-specific risks, and show the rest as information beside the order.
- Offer override reasons, and capture them along with the decision, using the EHR's feedback tools where they exist.
- Review override rates per rule with a clinical owner, and retire or tighten rules that are almost always overridden.
When does medication decision support become a medical device?
Under section 520(o)(1)(E) of the FD&C Act, decision support falls outside the device definition only if it meets all four criteria: it does not analyze images or signals, it works with medical information, it supports a health care professional's decision, and it lets that professional independently review the basis for its recommendations. The medication examples above sit on the non-device side when they meet all four.
Several lines are easy to cross. FDA says software intended for critical, time-sensitive decisions does not meet the fourth criterion. Software that gives patients or caregivers, rather than professionals, recommendations meets the device definition. Software that analyzes signals or patterns, such as continuous monitoring data, fails the first criterion. On single recommendations, the January 2026 guidance says that when only one option is clinically appropriate and the function otherwise meets all criteria, FDA intends to exercise enforcement discretion, and gives the example of a treatment plan, including possible medications, for patients with cognitive impairment that a clinician reviews, revises and finalizes.
Where does AI help, and where should rules stay in charge?
Interaction and allergy checks against curated, versioned knowledge are well suited to deterministic rules that can be tested case by case. Language models are more useful around them: summarizing a long medication history, finding a documented reason why a flagged combination was intended, or drafting a question to pharmacy. Our AI assistants cite their sources, so every statement about a medication links back to the note or list it came from, and they run on Claude via AWS Bedrock or Anthropic directly, with provider switching. Each clinician sees only their own patients, we sign a BAA before anyone sees PHI and we run a penetration test before launch.
Onix Systems has built software since 2000 and has 370+ engineers, and we have shipped Epic and athenahealth integrations. If you are planning medication checks for an EHR or a CDS Hooks service, talk to one of our engineers. Our free five-day technical audit ends with a written report.
Sources
- FDA: Clinical Decision Support Software guidance (January 2026, PDF)
- ONC (ASTP): §170.315(a)(4) Drug-drug, drug-allergy interaction checks for CPOE
- ONC (ASTP): §170.315(b)(11) Decision support interventions
- HL7 CDS Hooks specification (current published release)
- National Library of Medicine: RxNorm
- AHRQ PSNet: Alert Fatigue primer