Why the Comparator Must Be a Strategy
An external control is informative only if its patients' care represents the comparator of interest. Naming a comparator “standard care” or “supportive care” is not enough: those labels can cover different start times, switches, discontinuations, and accompanying treatments. If the trial comparator permits one pattern of care but external patients received another, their outcomes may not represent the trial comparator—even when the treatments share a name.
The practical task is to translate the trial comparator into rules that can be checked against routine-care records. Those rules need to specify what treatment may be used, when it should begin, what changes are allowed, and which concomitant treatments are compatible. A rule also needs a time origin and a way to classify records that do not contain enough detail to decide whether it was followed.
The strategy should be defined from the trial comparator, not reverse-engineered from whichever external patients are easiest to find. The earlier material on the external-control estimand and emulating a target trial provides the broader comparison framework; here the focus is making the comparator's treatment rule operational.
Write the Strategy as Explicit Rules
Begin by stating the treatment regimen in clinically meaningful terms: the intervention or care, dose or intensity where relevant, route, and any allowed alternatives. Then specify four time-varying components. Each should be stated in language that can be applied consistently to trial and routine-care records.
- Initiation: Identify the time origin, the required treatment, and any allowed initiation window. “Start at baseline” and “start within 2 days after baseline” are different rules.
- Switching: State whether a switch is prohibited, permitted only for a defined reason, or allowed among specified alternatives. Identify what counts as a switch, such as stopping one drug and beginning another.
- Discontinuation: State whether stopping is allowed, under what conditions, and whether a treatment interruption or permanent cessation violates the rule. Do not treat a short interruption as equivalent to permanent discontinuation unless the rule says so.
- Concomitant care: Specify background treatments that are required, permitted, or prohibited, including rescue treatments when they could change the comparator being represented.
Rules should be anchored to the same time zero used to identify eligible external patients. For example, if time zero is the date a patient first meets the specified eligibility conditions, a medication begun before that date is not automatically an initiation after time zero. Whether prior use is allowed must be stated. Likewise, a strategy that permits a short initiation window must say how long it is and how it is measured.
Worked Example: Turning Supportive Care Into a Rule
Suppose a hypothetical randomized trial compares a study drug with supportive care for adults with a defined condition. The trial protocol specifies that comparator participants receive supportive care at randomization; they may use acetaminophen for symptoms, but may not start another disease-modifying drug during the first 30 days. Define a strategy that could be checked in external records.
Time origin and initiation. Set time zero to the date of the treatment decision corresponding to randomization. The strategy is supportive care beginning at time zero. Because the protocol specifies no later start window, an external record showing supportive care first begun two days later does not by itself establish that the initiation rule was met.
Switching and discontinuation. The stated protocol rules do not describe a switch among supportive-care options or stopping supportive care. Do not silently invent a permitted switch or assume indefinite continuation. Either use further protocol detail to define those rules or mark that part of the comparator as unspecified and assess whether it matters for the intended comparison.
Concomitant care. Acetaminophen is permitted, while starting another disease-modifying drug within 30 days is prohibited. A record of acetaminophen use is compatible with the strategy; a record of a prohibited drug start is a deviation. If medication records are incomplete, absence of a recorded prohibited drug is not proof that none was used.
This definition is useful because it turns a broad label into observable decisions. It also exposes protocol details that must be clarified before external patients can be classified.
Classifying Records Without Overclaiming
For each rule, record what is known, what conflicts with the rule, and what cannot be determined. A useful classification has three states: documented aligned, documented deviation, and alignment indeterminate. The last state matters: incomplete records should not be converted into evidence of adherence.
A record may establish one component but not the full strategy. For example, a prescription date can establish that treatment was prescribed, but not necessarily that it was dispensed or taken. A medication list may show a drug was active without establishing its start date. The classification should match the data's actual level of detail, not the level of detail desired by the analysis.
Worked Example: Applying an Explicit Initiation Window
In a hypothetical comparator protocol, supportive care must begin from time zero through the end of day 2, inclusive. No other disease-modifying drug may be started during the first 30 days. Among 120 eligible external patients, records establish that 110 began supportive care within the specified window. Of those 110, six began a prohibited disease-modifying drug during the first 30 days. For the other 104, records establish no prohibited start and no other departure from the specified strategy.
The initiation window is explicit, so all 110 patients with documented starts from day 0 through day 2 meet the initiation rule. The six patients with a prohibited drug start do not meet the full strategy. The remaining count is
Thus, 104 patients are documented as aligned with the complete stated strategy, given the information in the example. The other 10 patients did not have a documented supportive-care start in the allowed window; they are not classified as aligned. This arithmetic would not be justified if the protocol required initiation exactly at time zero, or if the records failed to establish whether the other strategy rules were followed.
The classification also depends on the rule's exact wording. If the trial required initiation precisely at randomization but an external database recorded only calendar dates, a patient listed as starting “on the same date” might have started before or after the treatment decision. Unless the data establish the required timing, the initiation status is indeterminate—not confirmed adherence.
Worked Example: When Timing Cannot Establish Alignment
Consider a hypothetical comparator rule requiring treatment to begin exactly at time zero. An external cohort contains 110 patients whose records say treatment began on the same calendar date as time zero, but the database has no times of day. Six of these patients also have a documented prohibited concomitant treatment.
The six known prohibited-treatment records establish a deviation from the full strategy. For the other 104 patients, the calendar date alone does not establish whether initiation occurred at time zero: treatment could have begun earlier or later that day. Subtracting six from 110 does not establish 104 aligned patients. The defensible conclusion is that 110 patients have a recorded same-day start, six have an additional known deviation, and exact initiation alignment for the remaining 104 is indeterminate from these data.
The correction is not to assume that same-day initiation is “close enough.” If a clinically justified initiation window is acceptable, define it in advance and apply it consistently. Otherwise, retain the uncertainty in the classification.
Why Strategy Definitions Matter
A treatment strategy must be sufficiently well defined that the care it represents has a coherent interpretation. Suppose one external patient receives supportive care alone, another receives supportive care plus a routinely used co-treatment, and a third receives a different active drug. Pooling all three under a vague “usual care” label may mix clinically distinct treatment versions. If the trial comparator did not include those versions, the pooled external group may not represent that comparator.
A short justification follows from the logic of potential outcomes. Let \(g\) denote a fully specified treatment strategy, and let \(Y^{g}\) denote the outcome that would occur for a patient under that strategy. Interpreting an observed outcome as information about \(Y^{g}\) requires that the patient's observed care actually follows \(g\). If “supportive care” leaves important treatment choices undefined, patients assigned that label may in fact follow different strategies \(g_1\), \(g_2\), and so on. Then the observed outcomes do not automatically represent one common \(Y^{g}\).
This is a consistency requirement, not a guarantee that external and trial patients are comparable in every other respect. Even perfect documentation of strategy alignment does not by itself remove differences in prognosis, eligibility, measurement, or care setting. It establishes what treatment the external observations represent; it does not establish that treatment groups are exchangeable or that a causal effect is unbiased.
Alternative Descriptions and Practical Diagnostics
The main alternative to a sustained strategy is a treatment-policy description: classify patients according to an initial treatment decision and retain them in that group regardless of later switching or discontinuation. These answer different questions. A sustained strategy concerns following specified care over time; a treatment-policy approach concerns the effect of an initial assignment or decision despite later care changes. External records should not be treated as implementing one approach while the trial comparator is interpreted as the other.
| Feature | Sustained treatment strategy | Treatment-policy description |
|---|---|---|
| Initiation | Must meet the specified initiation rule | Defines the initial treatment or decision |
| Later switching or stopping | Evaluated against explicit rules | Does not change the initial group definition |
| Concomitant care | Specified when it changes the strategy | May occur after the initial decision and remain part of the observed course |
| Interpretation | Outcomes under following the defined care rule | Outcomes following the initial treatment policy, including later changes |
Before classifying patients, check for practical signs that the definition is failing:
- The protocol's treatment instructions are too vague to decide whether a switch or add-on is allowed.
- Start or stop dates are missing, recorded only at a coarse time scale, or inconsistent across sources.
- Routine-care products, doses, or co-treatments differ materially from the trial comparator but share its broad label.
- A “no recorded treatment” value is being interpreted as proof that treatment was not received.
- Patients with indeterminate strategy status are being counted as aligned without a stated rule.
A practical response is to clarify the comparator definition, report which components are observable, and keep indeterminate cases distinct from documented alignment. If a rule cannot be measured in the external data, do not imply that the data demonstrate adherence to it. A narrower but verifiable definition may be more credible than a broad label that combines incomparable care.
Check Your Understanding
Answer each question, then compare with the brief solutions.
- A protocol requires a drug to start exactly at time zero. External records show only the calendar date, not the time. Can a same-date record establish alignment?
- A comparator allows initiation from day 0 through day 3, inclusive. A patient starts on day 3 and has no recorded switch or prohibited co-treatment; the records cover all specified rules. How should this patient be classified?
- Why is “no prohibited drug appears in the database” insufficient when medication capture is known to be incomplete?
- In a hypothetical cohort, 75 patients have a documented start within an explicitly allowed window; 5 have a known prohibited co-treatment, and records establish no other deviations for the remaining patients. How many are documented aligned?
- What is the key difference between a sustained strategy and a treatment-policy description?
Solutions. (1) No. The start could have occurred before or after time zero, so exact timing is indeterminate. (2) Documented aligned, provided the records establish the other specified rules; day 3 is within the inclusive window. (3) Incomplete capture means the drug may have been used but not recorded, so absence from the database does not establish absence of use. (4) \(75-5=70\) patients are documented aligned. (5) A sustained strategy evaluates whether patients follow specified care rules over time; a treatment-policy description retains the initial group definition despite later switching or discontinuation.