What a Dose Recommendation Display Must Make Clear
A model-informed recommendation is useful only if the clinician can tell, without reconstructing the calculation, what regimen is proposed, what information it is based on, which assumptions matter, and what action is requested. A dose alone is not a complete recommendation. For example, “350 mg” does not say whether the dose is oral or intravenous, how often it is given, when it should begin, or whether it replaces the current regimen.
The display is not a substitute for clinical review or a place to reproduce the model’s technical documentation. Its job is to make the decision-relevant information visible and internally consistent. This tutorial focuses on the arrangement and wording of that information, not on selecting a dose or communicating the full uncertainty of a model prediction.
A reader should be able to answer four questions in sequence: What is proposed? Which patient data were used? Under what conditions does the proposal apply? What should I do next? That sequence provides a useful information hierarchy: lead with the dose and schedule, then show the key inputs and assumptions, then state the action. Supporting details can be available without competing visually with the proposed regimen.
A Practical Structure for the Display
A clear display separates information into identifiable fields rather than combining it into one sentence. The specific layout may vary, but the content should let a clinician distinguish what the system proposes from what the system observed or assumed.
- Recommendation and status: label the proposal as a recommendation, show whether it is ready for review or needs attention, and identify the medication.
- Complete regimen: display the dose with units, route, frequency or interval, and start timing. Include duration or a stopping condition when applicable.
- Relation to current treatment: state whether the proposal replaces, continues, or modifies the current regimen. Do not leave the clinician to infer this from two dose values.
- Decision-relevant inputs: show the values and measurement dates that materially support the recommendation, such as weight, laboratory results, or a drug concentration.
- Assumptions and applicability: identify consequential conditions used in generating or interpreting the recommendation, such as an assumed formulation or the absence of a specified interacting treatment.
- Required action: use a direct verb to state what the clinician should review, confirm, or enter. Avoid language that could be mistaken for an order already placed.
This is a content structure, not a rule that every detail must occupy the same amount of screen space. The proposed dose should be prominent, while input provenance and assumptions should remain accessible and legible. If the display makes the dose prominent but hides a critical condition, it is not clear: prominence without context can make an inapplicable proposal look ready to use.
Why These Fields Belong Together
The rationale can be stated as a simple chain. A model result is conditional on a particular set of input values and assumptions. A clinician interprets that result as a proposal for a regimen. The clinician can judge whether to act only if the display preserves the link between proposal, conditions, and action. If a link is omitted, the reader may mistake a conditional result for a universal instruction or mistake a proposed regimen for the regimen currently being administered.
This chain also explains why the display should name the status of the result. A current, reviewable recommendation and a result that depends on outdated data are not interchangeable. If a required input is missing, stale under a stated rule, or inconsistent with another displayed value, the interface should not present the dose as though it were an ordinary current recommendation. It should make the issue and the needed next step visible.
The display should also preserve units and time references. A laboratory value without a collection date may be hard to interpret; a dose without a time interval may be incomplete. These are not decorative details. They are part of the meaning of the information being shown.
Worked Example: Presenting a Complete Proposed Regimen
Suppose a hypothetical model for the fictional medication Drug Q proposes 350 mg by mouth every 12 hours for a 70 kg adult. For this illustration only, the model output is a predicted steady-state trough concentration of 9.6 mg/L, with a stated target range of 8–12 mg/L. The model used a creatinine clearance of 82 mL/min, recorded today. Assume the local decision rule requires a clinician to review the regimen; the recommendation does not itself place an order.
First make the dose and schedule explicit. “350 mg by mouth every 12 hours” specifies amount, route, and frequency. If the first dose is intended for 08:00 and the regimen continues at 12-hour intervals, the following dose is at 20:00. The daily amount is \(350\ \text{mg/dose} \times 2\ \text{doses/day} = 700\ \text{mg/day}\). This daily total is a cross-check; it does not replace the scheduled dose.
A clear display could read:
Drug Q — proposed regimen for clinician review
350 mg by mouth every 12 hours; first dose at 08:00; then 20:00 and 08:00 daily. This proposal is intended to replace the current regimen after review. Model-predicted steady-state trough: 9.6 mg/L; stated target: 8–12 mg/L. Inputs used: weight 70 kg; creatinine clearance 82 mL/min, measured today. Assumptions: the patient is using the modeled oral formulation and is not receiving the specified interacting treatment. Action: review the inputs and assumptions, then confirm or revise the proposed regimen before use.
The prediction is labeled as a model output, not as an observed concentration. The target is given alongside it so the reader knows what the prediction is being compared with. The proposal’s status and action are also explicit: the clinician is asked to review and confirm or revise it, not to assume that the dose has already been prescribed.
Worked Examples: Rounding and Input Status
Worked Example: Showing a Rounded Dose Without Hiding the Schedule
In a second hypothetical case, the fictional Drug R model recommends 327 mg every 12 hours before rounding. The available tablet strength is 100 mg. A configured rounding rule selects 300 mg every 12 hours, and the model’s hypothetical prediction for that rounded regimen is 9.1 mg/L against a stated target of 8–12 mg/L. The current regimen is 250 mg every 12 hours.
The rounded regimen is \(300\ \text{mg/dose} \times 2\ \text{doses/day} = 600\ \text{mg/day}\). The current daily amount is \(250 \times 2 = 500\ \text{mg/day}\), so the proposed regimen increases the total by \(600 - 500 = 100\ \text{mg/day}\). These totals provide a consistency check, but the display should still give the dose at each administration and the interval.
A useful presentation is: Proposed: Drug R 300 mg by mouth every 12 hours, replacing 250 mg every 12 hours after clinician review. Supporting detail: “Unrounded model dose: 327 mg every 12 hours; rounded to the available 100 mg tablet strength under the configured rounding rule. Predicted concentration for the rounded regimen: 9.1 mg/L; stated target: 8–12 mg/L.” The requested action should ask the clinician to review and confirm or revise the proposed regimen.
This presentation does not make the clinician infer that “300 mg” means two or three tablets, or assume that the proposal is an addition to the current dose. It also distinguishes the unrounded model result from the regimen actually being proposed.
Worked Example: Making an Out-of-Date Input Visible
In a third hypothetical case, a dosing model requires a weight measured within the preceding 30 days. The only available weight is 68 kg, measured 60 days ago. A new measurement entered today is 75 kg. The difference is \(75 - 68 = 7\ \text{kg}\), which is \(7/68 \times 100 \approx 10.3\%\) of the older weight. Under the stated 30-day rule, the old value is not current enough for a recommendation, regardless of whether the change is clinically important.
The display should not put a dose generated using 68 kg in the main recommendation position as if it were current. It should identify the status and the reason, for example: No current recommendation — required weight input is outside the 30-day validity rule. It can then show the recorded input and date, the newer weight and date, and the requested action: “Confirm the current weight and rerun the calculation.”
This wording distinguishes an input problem from a dose proposal. It does not imply that the older value is necessarily wrong, and it does not display an unreviewed dose that could be mistaken for a valid regimen. The dates make the issue inspectable; the explicit action tells the clinician what is needed next.
Diagnostics: Inspect the Display as a Clinician Would
Before release, review representative displays using both the underlying data and the rendered result. The aim is to find places where the interface changes, hides, or leaves ambiguous the meaning of the recommendation.
Check that units, values, and dates agree with the inputs used for the model result. A current weight displayed beside a result calculated from an older weight is a traceability failure.
Ask whether a clinician can identify dose, route, interval, start timing, and whether the proposal replaces or supplements the current regimen without making an assumption.
For a twice-daily regimen, verify that the daily total equals two scheduled doses. Check that a predicted concentration is not labeled as measured and that units are not omitted or carried over incorrectly.
If the proposal needs review or an input must be updated, verify that the status is visible and the requested action names the relevant issue. A warning that does not identify what is needed is difficult to act on.
A useful review method is to hide the supporting detail briefly and ask what the clinician would do based on the prominent regimen alone. Then restore the details and ask whether they explain the proposal without contradicting it. If the two views suggest different actions, the display’s hierarchy or wording needs revision.
Common Mistakes and Design Alternatives
| Approach | Strength | Risk to check |
|---|---|---|
| Recommendation-first card | Makes the proposed regimen and status easy to find. | Inputs or assumptions may be visually minimized or separated too far from the dose. |
| Detail-first report | Can show extensive input provenance and supporting information. | The clinician may have to search for the actual dose and requested action. |
| Single narrative sentence | Compact and potentially easy to read aloud. | May combine the proposal, current regimen, prediction, and action in a way that obscures their distinct roles. |
A common mistake is to display only the dose and assume that the rest is obvious. The incorrect reasoning is, “The clinician knows the patient and can infer the route, interval, and whether this replaces the current regimen.” The correction is to state those details in the recommendation itself; the display should not require reconstruction from memory or another screen.
Another mistake is to put every model input and technical detail at equal visual prominence. More information is not automatically clearer. The correction is to prioritize the proposed regimen and the conditions needed to interpret it, while keeping other relevant detail available in an organized form. Conversely, excessive compression can conceal important assumptions. A concise display is successful only if it preserves the distinctions that affect interpretation.
These approaches have limits. A display cannot repair incorrect source data, establish that a model applies to a patient outside its intended use, or replace clinical judgment. It can make input provenance and applicability easier to inspect, but the underlying calculation and the workflow still require their own verification. The display should also avoid making the recommendation look like a guaranteed outcome; a predicted value remains a model prediction, not an observed result.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A display says “Drug S: 200 mg.” Name three details that may be needed to make this a complete regimen display.
- A hypothetical proposal is 150 mg every 8 hours. What is its total daily dose, and why should that total not replace the scheduled regimen wording?
- A prediction card labels 10.2 mg/L as “measured concentration,” although it came from the model. What is wrong, and how should it be labeled?
- A recommendation uses a weight recorded 45 days ago, while the stated input rule requires a value from the preceding 30 days. What should the display communicate?
Solutions. (1) At minimum, specify route, frequency or interval, and start timing when relevant; also clarify whether it replaces the current regimen. (2) There are \(24/8 = 3\) doses per day, so \(150\ \text{mg/dose} \times 3\ \text{doses/day} = 450\ \text{mg/day}\). The display must still say “150 mg every 8 hours” because that states the amount and timing of each administration. (3) The label confuses a model output with an observed measurement. Call it a “model-predicted concentration” and show its units and relevant comparator if one is used. (4) State that no current recommendation is available under the 30-day rule, show the input date, and specify that the weight must be confirmed or updated before recalculation.