Why Map the Clinical Workflow?
A precision-dosing recommendation is useful only if it reaches the right clinical decision-maker at the right point in care and is acted on appropriately. A dosing service therefore sits inside a chain of clinical work: a dose or monitoring need is recognized, patient information is assembled, a recommendation is produced and reviewed, and any resulting treatment decision is entered into the clinical record or order system. The chain can fail even when the model calculation itself is correct.
A workflow map makes that chain visible. It identifies the clinical stages, the people or systems responsible for each stage, the information or decision passed between them, and the evidence that the handoff occurred. The earlier material on From Population Model to Point-of-Care Service places the model within a service; the focus here is the operational path around it. The map is not a substitute for clinical judgment, model evaluation, or local governance. It is a way to see where those activities meet and where responsibility could become unclear.
Representing the Pathway and Its Handoffs
For a basic precision-dosing pathway, distinguish at least four clinical stages. Prescribing or request initiation identifies the need for dosing support and initiates the service. Data collection assembles the information needed for a patient-specific assessment. Recommendation review presents the output to an authorized clinical reviewer, who evaluates it in context and records a decision. Order entry records and transmits a treatment order when the decision calls for one. These are functional stages, not necessarily separate people, screens, or software components.
For each boundary, specify a sender, a recipient, what is transferred, and what confirms receipt or completion. For example, a request can be sent but not become an actionable request if the receiving service does not see it. A recommendation can be generated but remain clinically unused if no reviewer is assigned. A reviewer can decide on a dose change while the corresponding order is never entered. The map should show these as distinct states rather than treating the whole sequence as one event.
| Stage | Typical responsibility | Evidence to look for | Potential handoff concern |
|---|---|---|---|
| Prescribing or request initiation | Clinician or care team identifies a need and initiates a request | A request is recorded and routed to the intended service | Request is incomplete, misrouted, or not acknowledged |
| Data collection | Clinical team or service gathers inputs for assessment | Required information is assembled and made available for review | Collection is delayed, responsibility is unclear, or information is not returned to the service |
| Recommendation review | Authorized clinician reviews the recommendation and records a decision | A documented decision, including a decision not to change treatment | Output is not reviewed, review is delayed, or the decision is not communicated |
| Order entry | Authorized prescriber or delegate enters an indicated order | The order is recorded in the intended clinical system | Decision and order differ, or an indicated order is not entered |
The evidence column is important: it gives a map an observable endpoint at each stage. “Recommendation sent” is not interchangeable with “recommendation reviewed.” Likewise, a documented decision not to change treatment is not an order-entry failure. Workflows commonly branch after review: a dose change may require an order, while a decision to continue, defer, or seek further assessment may not. The map must represent those routes explicitly.
Measuring Stage Transitions Without Losing the Denominator
A workflow audit may report the proportion of eligible pathways that reach a stage, the proportion moving from one stage to the next, or the proportion completing the whole pathway. These answer different questions. A transition proportion has as its denominator the pathways that reached the preceding stage. An end-to-end proportion has as its denominator the original eligible pathways. The denominator must be stated alongside each percentage.
For a sequence of nested events \(A_1,\ldots,A_k\), where each event means reaching the next defined stage, the chain rule gives the probability of reaching all stages as the product of conditional transition probabilities:
This identity follows by repeatedly applying \(P(B \cap C)=P(B)P(C\mid B)\). It does not require the transitions to be independent. It does require the events and denominators to refer to the same pathway cohort and to be defined consistently. A branch-specific rate cannot simply be substituted for a rate conditional on all pathways reaching that stage.
In an audit, counts are often more transparent than probabilities. If \(n_0\) eligible pathways enter, and \(n_1,n_2,\ldots\) reach successive stages, a transition proportion is \(n_j/n_{j-1}\), provided the later count refers to pathways that came from the preceding stage. An overall completion proportion is the number completing the specified endpoint divided by \(n_0\). Report both the numerator and denominator so that branch definitions and exclusions can be checked.
Worked Example: Calculating Completion With an Order Branch
Hypothetical audit. A service follows 250 eligible pathways. Of these, 225 have completed data collection, 198 reach recommendation review, and 178 have a documented review decision. Among the 178 decisions, 120 indicate that an order should be entered; 108 of those orders are entered. The other 58 decisions indicate no order is needed, and all 58 have a documented no-order disposition. Define successful completion as a documented disposition and, when an order is indicated, entry of that order.
The transition proportions before the decision branches are:
Among the 178 pathways with a documented decision, 108 complete the order-indicated branch and 58 complete the no-order branch. Thus, \(108+58=166\) pathways meet the defined endpoint. The decision-stage completion proportion is \(166/178=0.933\), and the end-to-end completion proportion is:
The order-entry rate for the order-indicated branch alone is \(108/120=0.900\), or 90.0%. That is a valid branch-specific rate, but it is not the completion rate among all 178 pathways reaching the decision stage. Multiplying the earlier transition proportions by \(108/120\) would treat every decision-stage pathway as if it were in the order-indicated branch. The branch-aware calculation instead uses \(166/178\) for completion across both defined dispositions; \(0.900 \times 0.880 \times 0.899 \times 0.933\) is approximately \(0.664\), matching \(166/250\) apart from rounding.
Locating a Handoff Problem
An overall completion percentage says how often a pathway reaches its endpoint, but not where it loses momentum or why. A stage-by-stage map supports more focused review. Counts can reveal where the largest number of pathways stops; timestamps can reveal where delay accumulates; and interviews or direct observation can help explain what happened at a particular boundary. These are complementary kinds of evidence. A lower transition proportion identifies a place to investigate, not by itself the cause of the loss.
Worked Example: Finding Where to Investigate First
Hypothetical audit. In a separate cohort of 80 eligible pathways, 76 have a request initiated, 68 complete data collection, 64 receive a recommendation, and 56 have a documented review decision. Of those 56 decisions, 42 indicate an order and 39 of those orders are entered; the remaining 14 indicate no order, with all 14 dispositions documented.
The transition proportions are \(76/80=95.0\%\) for request initiation, \(68/76=89.5\%\) for data collection, \(64/68=94.1\%\) for recommendation availability, and \(56/64=87.5\%\) for a documented review decision. The lowest of these transitions is from recommendation availability to documented review, suggesting that review assignment, timing, or documentation is a sensible first area to investigate.
For the final branch, \(39+14=53\) pathways meet the completion definition, so the end-to-end completion proportion is \(53/80=66.3\%\). The order-entry proportion is \(39/42=92.9\%\) among decisions indicating an order. These measures describe different parts of the workflow. The data do not establish why 8 of the 64 pathways with a recommendation lack a documented review decision, nor whether every pathway still in progress should be considered a failure. Follow-up review is needed to distinguish delay, non-documentation, a legitimate deferral, and a genuine handoff failure.
Useful diagnostics include comparing counts at each stage, checking whether a stated transition denominator matches the pathways eligible to make that transition, and separating unresolved or pending cases from completed failures. Where timing matters clinically, examine the time between events as well as the proportion reaching them. A high eventual completion proportion can still conceal delays that make a recommendation less useful. Conversely, a low proportion in a short observation window may partly reflect pathways that have not yet had time to finish.
Review exceptions as well as averages. For example, a recommendation may be reviewed but the decision documented in a different place; an order may be entered by a different authorized role; or an appropriate decision may be to defer action. The map should make the intended route and evidence visible enough to ask whether an apparent gap reflects missing work, work recorded elsewhere, or a legitimate alternative disposition. Determining the clinical appropriateness of an individual decision remains a review task, not a conclusion drawn from transition rates alone.
Choosing a Useful Workflow View
A stage-and-handoff map is well suited to showing the sequence, roles, branch points, and completion evidence. A simple funnel table is useful for quantifying how many pathways reach each stage. A time-based view is more informative when the main concern is delay, while a qualitative walkthrough can uncover informal workarounds or ambiguous responsibilities that aggregate counts cannot show. These approaches answer different questions; none makes the others unnecessary.
The map should be detailed enough to expose consequential boundaries but not so detailed that it becomes an inventory of every click or screen. Start with the patient-care sequence and the responsible roles. Add a branch when the clinical disposition changes what happens next. Add an operational detail when it changes responsibility, creates a possible failure, or is needed to define measurable completion. This keeps the map centered on the clinical workflow rather than on a particular interface.
Define which clinical need or eligible request starts the pathway, and what event counts as its endpoint.
Show request initiation, data collection, recommendation review, and order entry as distinct activities, including the no-order branch where relevant.
For each transition, identify who sends, who receives, what is transferred, and what evidence establishes receipt or completion.
State each numerator, denominator, time window, branch, and pending-case rule before interpreting a percentage.
Common Mistakes and Limitations
- Using a branch-specific rate for the whole cohort. An order-entry rate among decisions that indicate an order is not automatically the completion rate among all reviewed pathways. Define the branch and calculate each denominator separately.
- Calling a generated recommendation a completed clinical action. Generation, receipt, review, decision, and order entry are distinct events. Name the event actually observed.
- Treating every non-order decision as a failure. A documented decision not to change treatment may be the appropriate endpoint. Define completion in terms of the intended clinical disposition.
- Inferring a cause from a low transition proportion. A gap points to a location for investigation; it does not tell whether the cause is delay, unclear responsibility, an alternate route, or a data-recording issue.
- Ignoring time and cohort follow-up. Pathways still underway at the time of an audit are not necessarily failures. State the observation window and how pending cases are handled.
A workflow map also has limits. Clinical practice varies across settings and can include exceptions that a single diagram cannot represent well. Counts may depend on documentation quality, and rates may change when eligibility or completion definitions change. The map does not validate the dosing model, establish that a recommendation is clinically appropriate, or determine who has prescribing authority. Its value is narrower and practical: it clarifies the intended path and makes handoffs and measurable gaps easier to examine.
Check Your Understanding
Answer each question, then compare with the brief solutions below.
- Why should “recommendation generated” and “recommendation reviewed” be treated as separate workflow events?
- In 100 eligible pathways, 80 reach review. Of those 80, 50 indicate an order and 45 orders are entered; the other 30 have a documented no-order decision. What is the completion proportion among reviewed pathways under the branch-aware definition used here?
- A report gives an order-entry rate of 92% among pathways with an order-indicated decision. Can this be reported as the end-to-end completion rate for all eligible pathways? Explain.
- A workflow audit finds that only 70% of recommendations have a documented review decision within one day. What does this result establish, and what does it not establish?
Solutions. (1) Generation shows that an output exists; review shows that it reached a clinical reviewer and was considered. The handoff between them can fail or be delayed. (2) Completion is \(45+30=75\) of 80 reviewed pathways, so \(75/80=93.75\%\), or 93.8%. (3) No. The 92% denominator excludes pathways with no order indicated and does not include earlier stages. An end-to-end rate needs the original eligible cohort as denominator and a defined endpoint covering all relevant branches. (4) It establishes the observed proportion meeting that documentation-and-time definition in the audited cohort. It does not, by itself, establish why the remaining pathways lack timely documentation, whether they were reviewed elsewhere, or whether the clinical decisions were appropriate.