Why Treatment Strategies Need Precise Rules
A causal question that contrasts “treatment” with “no treatment” may sound clear but leave important choices unresolved. Which treatment, at what dose, started when, and with what provisions for later changes? If two patients described as treated receive different drugs or follow different dosing plans, the phrase “treatment” may combine interventions that are not the same. In longitudinal records, a strategy is useful only if its rules are specific enough to distinguish the relevant treatment histories.
A target-trial protocol therefore describes each strategy as an intervention plan, not merely as a label such as “treated” or “untreated.” This specification is separate from the mechanism that assigns people to strategies and from the analysis used to estimate an effect. Those are distinct protocol elements. Here the task is to say what each strategy requires and permits, in terms that can be represented using treatment information recorded over time.
For notation, let \(A_t\) denote the treatment received at time \(t\), and let \(D_t\) denote its dose when dose is relevant. The treatment history through time \(t\) is \(\bar A_t=(A_0,A_1,\ldots,A_t)\), with corresponding dose history \(\bar D_t=(D_0,D_1,\ldots,D_t)\). A strategy \(g\) describes which treatment histories count as following that plan. The notation does not imply that a person was assigned to the strategy, followed it, or had any particular outcome.
What a Complete Specification Includes
The necessary details depend on the clinical question and the available records, but a useful specification usually answers the following questions for each strategy:
- Intervention and comparator: Name the active drug, procedure, or other intervention. State what the comparator is: a named alternative, no initiation of the intervention, or a defined care plan. “Control” and “usual care” are not adequate descriptions unless their contents are specified.
- Treatment version and dose: Identify formulation, route, dose, and schedule when differences in these features matter to the intervention. A drug name alone may combine meaningfully different versions.
- Initiation: State what treatment begins under the strategy and when the initiation rule applies relative to the protocol’s specified start. Distinguish “initiate at the specified dose” from “already receiving treatment” or “initiate at some point.”
- Continuation and duration: Say whether the initial regimen is continued, for how long, and what ends the planned course. If the strategy does not impose a fixed duration, specify the conditions under which it continues.
- Allowable changes: Specify whether dose titration, interruption, restarting, switching, discontinuation, or rescue therapy is permitted. Give the rule and any limits, such as a permitted dose range or a defined safety trigger.
- Recordable distinctions: Identify the information needed to apply the rules, such as treatment dates, doses, formulations, and reasons for changes. A rule that cannot be distinguished in the available records may not define an implementable strategy for that data source.
A rule may permit a change without requiring it. For example, “the dose may be reduced if a specified adverse event occurs” allows a reduction under that condition; it does not mean every person must have a reduction. Conversely, “increase the dose when the prespecified measurement exceeds the threshold” describes a required response under a stated condition. These are different plans and should not be collapsed into an unspecified phrase such as “dose adjusted as needed.”
Worked Examples: Turning Labels Into Classifiable Plans
Worked Example: Specifying an Initiation Strategy and Its Comparator
Suppose a hypothetical observational data set contains weekly medication records for adults with newly recorded condition X. A draft protocol compares “Drug R” with “no Drug R.” The records show drug name, formulation, dose in milligrams per day, and start and stop dates. Rewrite the contrast as two strategies that could be assessed from these fields.
Intervention strategy: At the protocol’s specified start, initiate oral Drug R, immediate-release formulation, at 10 mg per day. Continue that dose through the 12-week planned treatment period. Do not switch to another drug in the same class; a documented discontinuation ends the planned Drug R course.
Comparator strategy: Do not initiate Drug R at the protocol’s specified start and do not initiate it during the 12-week period. Other care is not restricted by this definition; if the scientific question is intended to compare Drug R with a particular alternative, that alternative must instead be named and specified.
The distinction is now checkable using the recorded drug, formulation, dose, and dates. A record showing 10 mg per day of immediate-release Drug R beginning at the specified start fits the intervention’s initiation rule. A record showing extended-release Drug R does not, even though the drug name matches. A record showing Drug R first started in week 4 does not meet either strategy as written: it fails the intervention’s start rule and violates the comparator’s no-initiation rule. The protocol must decide how such histories are handled; the vague label “no Drug R” does not resolve that question.
Worked Example: Defining Permitted Dose Changes
Consider a hypothetical strategy for an injectable medication. The protocol proposes a starting dose of 20 mg once weekly and allows a reduction to 10 mg once weekly if a prespecified safety event is recorded. It permits one restart at 10 mg after a temporary hold, but does not permit increasing above 20 mg or switching to another injectable medication. Determine whether each history meets the stated treatment rules.
History A: A patient starts 20 mg weekly, has no recorded safety event, and remains at 20 mg. This follows the specified initiation and continuation plan.
History B: A patient starts 20 mg weekly, has the specified safety event recorded, and is reduced to 10 mg weekly. This is an allowed change because the trigger is documented and the reduced dose is the permitted dose.
History C: A patient starts at 20 mg weekly and is reduced to 10 mg without a recorded safety event. The observed dose is within the stated range, but the reduction does not follow the rule permitting it. The dose values alone are insufficient to classify this history: the reason or trigger must also be available in the records.
History D: A patient starts at 20 mg weekly and increases to 30 mg. This does not follow the strategy because increases above 20 mg are not permitted. Likewise, switching to a different injectable medication is not allowed. These decisions follow directly from the stated limits; they do not require comparing outcomes.
Worked Example: Resolving an Ambiguous “Usual Care” Comparator
A hypothetical protocol describes its comparator as “usual care without Drug S.” In the records, some patients receive supportive care alone, while others receive Drug T, a different medication. The phrase leaves unclear whether Drug T is permitted. Write two possible comparator definitions and show why they are not interchangeable.
Option 1: No Drug S, other care permitted. Do not initiate Drug S during the stated period. Drug T and supportive care are permitted. Under this definition, a record showing Drug T but not Drug S is compatible with the comparator.
Option 2: Supportive care alone. Do not initiate Drug S or Drug T during the stated period; supportive care is permitted. A record showing Drug T is not compatible with this comparator.
The difference can be expressed with a simple record count. Suppose the hypothetical records include 12 comparator-labelled patients: 5 receive supportive care alone, 4 receive Drug T without Drug S, and 3 receive Drug S. Under Option 1, the first two groups meet the no-Drug-S rule, giving \(5+4=9\) records compatible with the comparator. Under Option 2, only the 5 receiving supportive care alone meet the comparator definition. The 3 receiving Drug S meet neither comparator rule. The counts differ because the strategies differ, not because of a calculation convention.
Why Specificity Matters for Causal Interpretation
Potential-outcome notation makes the issue clear. Let \(Y^g\) denote the outcome that would occur under strategy \(g\). For this notation to refer to a sufficiently defined intervention, the strategy must distinguish treatment versions that matter to the outcome. If “Drug R” combines two formulations with different dosing schedules, then \(Y^g\) may not describe one well-defined plan unless the protocol explains which versions are included and how.
This is the practical connection to consistency: when a person’s observed treatment history follows the specified strategy, the observed outcome can be interpreted as the outcome under that strategy, subject to the causal assumptions required by the study. A strategy that has not distinguished relevant versions makes that correspondence unclear. Precise treatment rules do not by themselves guarantee causal identification, but they prevent the intervention being studied from changing meaning across records.
A compact way to check a specification is to ask whether two independent reviewers, given the protocol and the same complete longitudinal record, would classify the treatment history the same way. If they cannot, the definition needs more detail. If they can only classify it using information absent from the data, the strategy may be well described but not operationalizable in that data set.
Common Pitfalls and Diagnostics
- Equating a drug label with an intervention. “Drug R” can conceal differences in formulation, route, dose, and schedule. State which versions are included or excluded.
- Using “as clinically indicated” without a rule. This phrase does not identify what event permits a change, who records it, or which changes are allowed. Replace it with a trigger and an allowed response, or explicitly state that the strategy permits clinician-directed changes and what records are required to classify them.
- Leaving the comparator implicit. “No treatment” can mean no named drug, no medication of any kind, or a specified alternative. Say which meaning applies.
- Defining exposure by a future summary alone. “Ever treated” ignores when treatment began and how it changed. If initiation and subsequent changes are part of the intervention, specify them as a longitudinal plan rather than reducing the strategy to a single ever/never label.
- Confusing the plan with observed behavior. A strategy states what intervention is specified. A treatment history records what a person actually received. Whether a history follows the plan is a separate classification question; it does not change the plan’s definition.
Useful diagnostics include tabulating recorded doses and formulations, checking whether dates permit the stated initiation rule to be assessed, and listing changes whose triggers or reasons are missing. If many records cannot be classified, do not silently infer the missing rule from the observed data. Clarify the protocol definition or acknowledge that the available records cannot implement it reliably.
Comparison and Limits
| Approach | What it specifies | Main limitation |
|---|---|---|
| Broad label such as “treated” | Usually only whether some treatment was recorded | Can combine different drugs, doses, initiation times, and later changes |
| Baseline treatment category | Treatment status at one recorded time | Does not by itself define later treatment or permitted changes |
| Precise treatment strategy | Intervention, comparator, dose, initiation, continuation, and allowable changes | Requires sufficiently detailed records and may still leave causal assumptions or assignment questions unresolved |
Greater detail is not automatically better. A protocol should specify distinctions relevant to the question and feasible to observe, rather than add arbitrary restrictions. Nor does a precise rule make an observational comparison randomized, remove confounding, or guarantee that the necessary data exist. Treatment strategies are the intervention definitions; they are one essential part of a causal design, not a substitute for its other parts.
Check Your Understanding
For each question, decide what a precise strategy would need to say. Brief solutions follow the questions.
- A protocol contrasts “Drug M” with “no Drug M.” Name two unresolved details that could change which longitudinal histories fit either strategy.
- A strategy permits dose reduction after a recorded safety event. A patient’s dose is reduced, but the record does not include the reason. Can the history be classified from dose data alone? Why?
- A comparator allows Drug N but prohibits Drug M. Does a record showing Drug N without Drug M fit the comparator? State what additional specification may still be needed.
- Explain why a treatment strategy is not the same thing as the treatment a person actually received.
- A hypothetical protocol starts Drug Q at 5 mg daily and permits an increase to 10 mg only after a recorded measurement exceeds 8 units. A record shows an increase to 10 mg when the measurement was 7 units. Does this follow the strategy?
Solutions. (1) Relevant unresolved details include formulation or dose, initiation timing, and whether later initiation, switching, or dose changes are allowed. (2) No. The observed dose change is known, but whether it followed the permitted trigger cannot be determined without the event information. (3) Yes, under the stated no-Drug-M rule, but the protocol should clarify which versions or doses of Drug N are permitted and whether other treatments are restricted. (4) The strategy is the specified intervention plan; the received treatment is the observed longitudinal history, which may or may not follow that plan. (5) No. The recorded measurement was 7 units, and \(7\) does not exceed the threshold of \(8\), so the stated condition for increasing the dose was not met.