Why the Integration Point Matters
A dosing recommendation can be scientifically sound and still fail to influence care if it appears at the wrong moment. A clinician reviewing a patient’s chart, entering a medication order, and responding to a drug concentration result may be making related but distinct decisions. Each moment offers different information, has different demands on attention, and permits different next steps. Embedding dosing support in an electronic health record (EHR) therefore means more than displaying a recommendation inside the application: it means locating the support within the sequence of clinical work.
This tutorial compares three common placements: in-chart support, which appears during review of the patient record; order-entry support, which appears while a clinician creates or changes a medication order; and results-review support, which appears when a relevant laboratory result is reviewed. The goal is not to declare one placement universally best. The right choice depends on when the decision is made, which inputs are available at that time, and whether an appropriate action can follow immediately.
The comparison assumes that the dosing model and its recommendation format have already been defined, that the EHR can identify the patient and relevant medication context, and that a clinician remains responsible for the treatment decision. It also assumes that input quality and missing-data handling are addressed, as covered in the earlier material on Clinical Data Quality Checks Before Recommendation and Handling Missing Inputs in Deployed Dosing Systems. Integration placement does not correct a model-input problem or make an unsuitable recommendation safe.
Three Placement Patterns
| Placement | Typical trigger | Workflow advantage | Key data-access question |
|---|---|---|---|
| In-chart | Clinician opens a patient chart or a relevant medication/therapy view | Can support broad review and planning before a specific order is started | Can the service assemble current regimen, relevant events, and patient covariates from the chart? |
| Order-entry | Clinician creates or edits an order for the target medication | Recommendation is close to the dose action and can be compared with the proposed order | Can the service read the draft order and the clinical context needed to assess it? |
| Results-review | Clinician opens or acknowledges a relevant laboratory result | Can connect a new measurement to a dose-adjustment decision | Can the result be linked to the correct patient, medication, regimen, and collection time? |
In-chart placement can be appropriate when dosing decisions are made as part of a broader review rather than at the moment of order creation. It can give the clinician a longitudinal view, but it may be less directly connected to the order that will eventually be written. The interface should make clear which regimen and data state the recommendation uses. Merely opening a chart does not establish that the clinician intends to adjust the target medication.
Order-entry placement places support near the action. The service may receive the medication, dose, route, frequency, and other fields in the draft order, then present a recommendation for comparison. This context can reduce ambiguity about the intended drug and proposed regimen. However, the order composer may not expose every relevant input, and a recommendation that repeatedly blocks routine order work can disrupt the task. A displayed suggestion should not be treated as an authorized order unless the EHR explicitly records the clinician’s selection and resulting order.
Results-review placement is useful when a new measurement is the event that prompts reconsideration of dosing. The result alone is not enough: interpretation may depend on when the sample was collected relative to dose administration, what regimen was in force, and whether the result is current for the decision. Results review can put those elements together at a natural point in the clinician’s work, but it risks presenting advice after an order has already been changed or before the result has been interpreted in context.
Deriving a Placement From the Clinical Sequence
A practical way to choose a placement is to trace the decision in order. First identify the clinical event that creates a need to consider dosing. Then determine when the clinician has the minimum relevant information, and finally locate the point where an appropriate action is available. These are not necessarily the same moment. The best integration point is generally the earliest point at which the needed context is sufficiently complete and the clinician can take a meaningful next step.
Specify whether the clinician is planning therapy, writing or revising an order, or responding to a new result. A vague trigger such as “patient chart opened” may not identify a dosing intent.
List the inputs the recommendation needs, their source in the EHR, and whether they are available at the trigger. Check that time-dependent data refer to the relevant treatment episode.
Determine whether the clinician can review, modify, or defer a dose decision in the same workflow. A recommendation without a clear next action can create extra navigation without improving the decision.
Ask whether the recommendation might be premature, duplicated, based on a draft that later changes, or presented after the relevant action has already occurred. Choose a trigger and presentation that expose these risks rather than concealing them.
This reasoning follows from a simple dependency: a useful recommendation requires both an interpretable input state and a reachable clinical action. If the inputs are not yet available, moving support earlier does not make the recommendation more timely; it makes it under-informed. If the action has already been completed, moving support later may be clinically informative but no longer supports that immediate decision. Workflow placement is therefore a coordination problem between data availability and decision opportunity, not just a choice of screen.
Worked Example: Planning a Dose During Chart Review
Hypothetical scenario. A pharmacist reviews a patient receiving a narrow-therapeutic-index medicine during a scheduled clinic visit. The clinician may decide whether a dose change is needed after reviewing the current regimen, recent laboratory results, and relevant patient characteristics. The next order will be entered later in the visit.
An in-chart panel opened from the medication review is a plausible placement because the immediate task is to assess the treatment plan, not yet to commit an order. The panel should identify the regimen and the result dates used, present the recommendation in the context of that review, and provide a clear route to the order workflow.
The key data-access check is whether chart review can supply the active regimen and relevant results, including the result timing needed for interpretation. If the medication history contains conflicting active orders or the pertinent result is unavailable, the panel should not imply that it has evaluated a complete, current regimen. This is a data-readiness issue, not a reason to infer a dose from whichever value happens to be displayed first.
The trade-off is that the recommendation is available before order entry, but the eventual order may differ from the plan discussed in the panel. If the clinician later chooses a different regimen, the order-entry workflow should evaluate the actual proposed order rather than assume the earlier plan was carried forward unchanged. In this scenario, in-chart support is a planning aid; it is not a substitute for checking the final order.
Worked Example: Checking a Draft Order at Entry
Hypothetical scenario. A clinician is entering a maintenance dose of a medicine. The EHR order composer contains a draft dose of 125 mg every 24 hours. A dosing service recommends 100 mg every 24 hours using the patient’s current context.
Order-entry placement is well matched to the immediate task: the clinician can compare the draft and recommendation before signing. The interface should show both regimens with units and frequency, make the recommendation’s relevant context accessible, and let the clinician revise the draft or continue according to the established workflow. The service must receive the draft order as it exists at the time of calculation.
Suppose the clinician then changes the draft to 100 mg every 24 hours. The recommendation and proposed order now agree. If instead the clinician changes the interval to every 12 hours, the original comparison is no longer an adequate description of the order under consideration. A robust interaction therefore recalculates or clearly marks the recommendation as tied to the earlier draft. It should not silently present an old comparison as if it evaluated the revised order.
This placement has strong action proximity, but only if the integration reads the actual order fields and can recognize edits. If a local interface exposes the medication name but not the draft dose or interval, the service cannot reliably assess the order-entry decision from that context alone. The placement must be adapted or paired with another point where those inputs are available.
Worked Example: Responding to a New Laboratory Result
Hypothetical scenario. A laboratory reports a drug concentration of 8.4 mg/L. The patient’s chart indicates a scheduled dose of 200 mg every 12 hours, but the result screen initially shows only the concentration and its report time.
A results-review trigger is sensible only if the service can connect the result to the correct patient and treatment episode and obtain the collection time, not merely the report time. It must also access the regimen and relevant dose-administration history needed to interpret the concentration. If the result was collected before a dose change but the current order reflects the new regimen, treating the result as if it came from the current regimen could lead to an inappropriate recommendation.
For this example, assume the result is confirmed to belong to the current treatment episode, the collection time and dosing history are available, and the result is reviewed before any subsequent dose change. Under those conditions, results-review placement can present a recommendation at the moment the clinician is considering what the new measurement means. If any of those links is uncertain, the interface should make that limitation visible rather than imply that the concentration alone determines the next dose.
The result-review location is especially useful when the result itself prompts action, but it may not be the only necessary location. A later order-entry check can still be useful because the clinician’s actual order may differ from the recommendation or change after the result is reviewed.
Data Access and Workflow Diagnostics
A placement should be tested against the data that the service can actually access at that point, not against an idealized record. For each required input, document its source, how it is linked to the current patient and medication, and whether its timing is meaningful for the decision. The earlier material on Aligning Doses, Samples, and Clinical Events in Time is particularly relevant when results-review support uses samples or dose events.
- Trigger mismatch: recommendations appear when the user opens a chart for a purpose unrelated to dosing, or fail to appear when the relevant order or result is handled.
- Context mismatch: the visible regimen differs from the regimen supplied to the calculation, or a result is associated with the wrong treatment episode.
- Timing mismatch: the service uses report time where collection time matters, or evaluates a draft order that has since changed.
- Action mismatch: the recommendation is shown on a screen from which the clinician cannot take the relevant next step without searching or re-entering information.
- Repeat-display burden: the same recommendation reappears on routine chart navigation even though no relevant data or decision context has changed.
These are workflow diagnostics, not judgments about a clinician’s acceptance of the recommendation. Review whether the correct trigger fired, what data the service received, what it displayed, and what task the user was performing. Do not infer from a dismissed panel alone that the recommendation was clinically inappropriate; the user may have lacked time, needed more context, or been performing a different task.
Choosing Among Placements
The main alternatives differ in when they make support available, how directly they connect it to a dose action, and what context they can expose. None removes the need to make the recommendation understandable and actionable, subjects addressed in the earlier material on Displaying a Dose Recommendation Clearly and Making Recommendations Actionable in the EHR.
| Design question | In-chart | Order-entry | Results-review |
|---|---|---|---|
| Best fit | Therapy planning during record review | Assessing a specific proposed order | Interpreting a new measurement for dosing |
| Typical strength | Broader clinical context | Close connection to the intended dose action | Natural response point for result-driven decisions |
| Typical limitation | May be detached from the eventual order | May lack context or interrupt order work | May be too late or lack dose-event context |
| Critical data check | Current regimen and relevant chart history | Current draft order and required clinical inputs | Result identity, collection timing, and regimen history |
A single placement may be sufficient when one point reliably combines the needed information with an actionable decision. Multiple placements can be justified when the clinical process has distinct planning and execution steps, or when a result initiates reconsideration but the final order must be checked separately. More placements are not automatically better: duplicating the same recommendation can create conflicting versions, extra work, or uncertainty about which display is current.
The limitations are practical as well as technical. Different EHR views may expose different data, local workflows may place the same decision at different stages, and users may need to navigate between review and order entry. A design that works in one clinical setting may not transfer unchanged to another. The placement should be chosen with the actual decision sequence and data access in view, then assessed in realistic workflow testing rather than assumed effective because the recommendation appears in the EHR.
Common Mistake
A common mistake is to assume that order-entry is always the best integration point because the dose is entered there. The faulty reasoning is: “The model recommends a dose, so showing it beside the order guarantees that the clinician has the right context and can use it.” In reality, the order composer may not expose the sample timing, administration history, or other inputs required for a valid recommendation. Conversely, an in-chart or results-review point may provide the needed context earlier, while the final order still requires a separate check.
The correction is to map the decision trigger, required data, and next action before choosing a screen. Order-entry is a strong candidate when the draft order and necessary clinical context are both accessible. If they are not, the integration needs a different placement, additional data access, or a clear indication that a reliable recommendation cannot be made from the available context.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A dosing recommendation is needed while a clinician is comparing treatment options during a scheduled review, before an order is drafted. Which placement is a reasonable starting point, and what must still be checked?
- A recommendation appears in the order composer, but the clinician changes the dose interval after it is calculated. What is the workflow risk, and how should the system handle it?
- A drug concentration is visible in results review, but its collection time and dose history are unavailable. Why is the result-review trigger alone insufficient?
- Give one reason to use both results-review and order-entry support, and one risk of doing so.
Solutions. (1) In-chart support is a reasonable starting point because the clinician is planning therapy during record review. Check that the relevant regimen, results, and patient context are available, and verify the eventual order separately. (2) The displayed comparison may refer to a draft that is no longer under consideration. Recalculate for the revised order or clearly identify the recommendation as tied to the earlier draft. (3) Without collection timing and dose history, the service cannot establish how the measurement relates to the regimen being considered; the visible result alone does not supply sufficient context. (4) Results-review support can connect a new measurement to reconsideration, while order-entry support can evaluate the actual proposed dose. A risk is duplicative or inconsistent displays, especially if one is not updated when the order or context changes.