In short
- Clinical decision support fits at the points in the workflow where a clinician decides something: scheduling, chart review, ordering, documentation, discharge and follow-up between visits.
- Each point suits a different form, such as reminders before a visit, summaries at chart review, order sets and interaction checks at ordering, templates during documentation and worklists between visits, with interruptive alerts kept for severe, patient-specific risks.
- Timing also changes regulation, because FDA treats decision support for critical, time-sensitive decisions as a device even when similar logic used for planning is not.
Clinical decision support fits into the clinical workflow at the moments when a clinician makes a decision: before the visit, when opening the chart, while ordering, while documenting, at discharge and between visits. The form has to match the moment. Reminders work before a visit, summaries at chart review, order sets and interaction checks during ordering, templates during documentation and worklists between visits. Interruptive alerts belong only where the risk is severe and specific to the patient. Clinicians learn to ignore even a sound rule when it appears at the wrong step or in the wrong form.
This guide is for product leads and clinical informatics teams deciding where a new piece of decision support should live. For card design and rollout inside the EHR, see our separate guide to CDS Hooks adoption.
What counts as clinical decision support?
More than alerts. HealthIT.gov describes CDS as a digital tool that provides timely, person-specific information, filtered or presented at appropriate times, to improve care. Its examples include order sets tailored to conditions, patient data summaries and reports, documentation templates, diagnostic support, contextually relevant reference material, clinical guidelines, and alerts and reminders for providers and patients. FDA's January 2026 guidance quotes a similar list from the 2014 FDASIA Health IT Report.
HealthIT.gov also says that CDS information should fit into the provider's workflow so they can act quickly and confidently, and different types of CDS suit different kinds of care and settings. A pop-up alert is one tool among many, and often not the best one.
Where are the decision points in a typical encounter?
The CDS Hooks standard, currently at release 2.0.1 on FHIR R4, names many of these points as hooks that an EHR can use to call an external service. Its library includes appointment-book, encounter-start, patient-view, order-select, order-sign, problem-list-item-create, allergyintolerance-create, medication-refill and encounter-discharge. Mapped against the workflow:
- Scheduling and pre-visit: appointment-book. Good for preventive care reminders and gaps in care. FDA's examples include reminders for breast cancer screening or immunizations, and reminders to order a hemoglobin A1C test for diabetic patients based on the date of the last one.
- Visit start and chart review: encounter-start and patient-view. Good for summaries: what changed since the last visit, outstanding results, care gaps. Show these as information on the screen, without interruptive alerts.
- Ordering: order-select and order-sign. Good for order sets, interaction and allergy checks, dose adjustments, duplicate test warnings and formulary alternatives, offered as suggestions the clinician can accept.
- Documentation: problem-list-item-create and allergyintolerance-create, plus templates. Good for prompting structured capture that later decision support depends on.
- Discharge: encounter-discharge. FDA's examples include recommending assessments before discharge, such as a pain assessment, and assembling post-operative instructions, medication needs and follow-up for discharge papers.
- Between visits: outside the encounter entirely. Worklists, registries and dashboards that list patients due for follow-up, such as FDA's example of next-step options for a patient with a pre-diabetes diagnosis.
The library is not a list of what every EHR supports, so check with each EHR vendor and each customer which hooks are enabled before you design around one.
Should CDS interrupt, suggest or wait to be asked?
Default to not interrupting. AHRQ's patient safety primer on alert fatigue says clinicians override the vast majority of order-entry warnings, and recommends making only high-severity alerts interruptive, while low-severity ones become non-interruptive or are removed. It also recommends using patient data so alerts fire only for high-risk cases.
- Interrupt: severe, patient-specific risk at the moment of action, such as a dangerous interaction at order-sign. CDS Hooks cards carry an indicator of info, warning or critical, and critical should mean exactly this.
- Suggest: a proposed change the clinician can accept with one action. CDS Hooks cards can carry suggestions, each with a label and one or more actions that are processed together.
- Inform: a summary or a list placed where the clinician will look anyway, such as the chart header or a side panel.
- Wait to be asked: reference material, calculators and assistants opened on demand.
- Act in the background: CDS Hooks also defines system actions that are not shown to the user, and the EHR decides whether to apply them. Never depend on them without checking each customer's settings.
Which integration option fits which step?
- Native EHR rules: configured by the health system inside its EHR. ONC's (b)(11) criterion requires certified EHRs to let a limited set of identified users select and configure interventions. No extra service to host, but the logic is bound to one vendor's rule language.
- A CDS Hooks service: your logic runs outside the EHR and returns cards at the hook. Portable across EHRs that support the same hooks, and the main route for third-party vendors.
- An embedded app: a separate interface opened from the EHR for decisions that need more than a card, such as a dosing calculator or a care plan builder.
- Population tools outside the EHR: worklists and registries for care managers who work between visits. They run on data feeds instead of hooks.
- Patient-facing tools: reminders and education in a portal or app. FDA treats decision support aimed at patients or caregivers as meeting the device definition, so plan for that review.
How does timing change the regulatory picture?
FDA's guidance says decision support is outside the device definition only if it meets all four statutory criteria, including that it lets a health care professional independently review the basis for its recommendations. FDA states that software intended for a critical, time-sensitive task does not meet that criterion, because the clinician is unlikely to have time to review the basis. It explains this with automation bias: people over-rely on suggestions from automated systems, and the effect grows when urgent action leaves no time to consider other information.
FDA's examples show where the line falls. Preventive care reminders, discharge assessments and a 90-day and one-year postoperative risk estimate used for pre-transplant planning are non-device examples. Software that detects a life-threatening condition such as stroke or sepsis and alerts a clinician, software that analyzes hourly pulse oximetry and heart rate to flag deterioration, and software that predicts heart failure hospitalization from wearable data are device examples, partly because they support time-critical decisions. The same clinical idea can land on either side depending on the step where it fires.
What will a health system ask before turning your CDS on?
Expect questions shaped by the rules that bind the health system. ONC's (b)(11) criterion requires certified EHRs to show source attributes for evidence-based interventions, a list of 13 that runs from a bibliographic citation, the developer, the funding source and release dates to whether the intervention uses race, ethnicity, language, sexual orientation, gender identity, sex, date of birth, social determinants of health data or health status assessments. Predictive interventions require risk management practices covering validity, fairness, safety, security and privacy.
Covered entities under Section 1557 have their own duty. Under 45 CFR 92.210, they must not discriminate on the basis of race, color, national origin, sex, age or disability through their use of patient care decision support tools, must make reasonable efforts to identify tools that use those characteristics as inputs, and must make reasonable efforts to mitigate the risk of discrimination from each one. Have the inputs, sources and validation of your logic written down before the first meeting.
How do you map your own workflow before building?
- Watch the work. Sit with the clinicians who will use the tool, and write down every decision they make in the step you are targeting.
- For each decision, record who makes it, when, what data is on screen, how much time they have and what they use today.
- Choose the form to suit the step: summary, suggestion, order set, template, worklist or, rarely, an alert.
- Define the silent case. Decide when the tool should show nothing, and make that the most common outcome.
- Decide what you will measure: acceptance, overrides with reasons, and whether the target action happened afterward.
- Check the regulatory position for the step you chose, especially if it is time-critical or patient-facing.
We have shipped Epic and athenahealth integrations, and our AI assistants cite their sources, so clinicians can check the basis of what they read. We sign a BAA before anyone sees PHI, scope access so each clinician sees only their own patients and run a penetration test before launch.
Onix Systems has built software since 2000 and has 370+ engineers. If you are deciding where a piece of decision support should live in your customers' workflow, talk to one of our engineers. Our free five-day technical audit ends with a written report.
Sources
- HealthIT.gov: Clinical Decision Support
- FDA: Clinical Decision Support Software guidance (January 2026, PDF)
- HL7 CDS Hooks specification (current published release)
- ONC (ASTP): §170.315(b)(11) Decision support interventions
- AHRQ PSNet: Alert Fatigue primer
- 45 CFR 92.210: Nondiscrimination in the use of patient care decision support tools (Cornell LII)