Why Treatment Definitions Need to Be Explicit
A reader cannot interpret an observational comparison from drug names alone. “Treatment A versus usual care” leaves open when treatment had to start, what counted as a qualifying dose, how long a prescription was considered active, and what happened when a person switched or stopped. Different answers can classify the same person differently and can change the comparison the analysis represents.
A useful report makes each strategy reproducible from the available records. It states the treatment and comparator, the date at which eligibility and assignment are anchored, the rules for initiation and exposure, and the handling of treatment changes. These specifications should agree with the target estimand described in the earlier material on target estimands. Clear reporting does not, by itself, remove confounding or establish a causal effect.
For a strategy definition to be interpretable, its alternatives must be described clearly enough to distinguish them, and the rules must apply at a stated time. Eligibility, assignment, and the start of follow-up should be aligned. Otherwise, a person may have to remain alive or event-free long enough to qualify for one group, while members of another group do not face the same requirement. Explicit definitions help readers detect that problem; they do not substitute for an appropriate design or analysis.
What to Report for Each Strategy
For the treatment strategy, report the intervention or exposure in clinically meaningful terms. Depending on the data and question, relevant details may include the product or treatment class, route, dose or dose range, frequency, and whether a prescription, dispensing, administration, or other record defines initiation. State the allowable interval from time zero to initiation, if there is one. If a prescription is used as evidence of exposure, explain what is known and what is not: a dispensing record documents dispensing, not necessarily ingestion.
Define the comparator with equal care. An active comparator might specify another treatment and the same initiation window. A no-treatment comparator should state what is prohibited or allowed—for example, no initiation of either study drug during a specified period—and whether other forms of usual care remain permitted. “Untreated” is not self-explanatory: it might mean no use of a particular drug, no use of any drug in a class, or no recorded treatment at all.
When a strategy includes a grace period, give its length, its start and end, and the operational rule used to assign or classify people. State how people who do not initiate within the period are handled. A grace period is not merely a data-cleaning detail: it can affect who is counted in each strategy and when exposure is considered to begin. If the analysis uses a design or method to handle assignment over that interval, report it rather than implying that future initiation was known at time zero.
Finally, report what happens after initiation. Say whether switching, adding another treatment, dose changes, gaps, and discontinuation change a person's classification, trigger censoring, or are retained under the originally assigned strategy. If a gap is used to define discontinuation, give its length and the rule for counting it. These choices should not be collapsed into the vague phrase “patients were followed according to treatment.”
Specify the treatment strategy and comparator in terms that distinguish them clinically and operationally.
State the eligibility date, time zero, and whether treatment assignment and follow-up begin at that same point.
Give the qualifying event, dose or exposure rule, and any grace period, including what happens when initiation does not occur.
State how switching, discontinuation, and other deviations affect classification or censoring.
Confirm that the written rules match the target estimand and the analysis actually performed.
Worked Example: Specifying Two Initiation Strategies
Hypothetical example. Suppose an observational study compares two oral drugs for adults meeting a stated eligibility rule at an outpatient visit. The visit date is the proposed time zero. Write strategy definitions that do not leave initiation or dose implicit.
A reproducible description could say: “The treatment strategy was initiation of Drug A, 10–20 mg once daily, with a dispensing record dated from day 0 through day 7 after the eligible visit. The comparator strategy was initiation of Drug B, 5–10 mg once daily, using the same dispensing window. Time zero was the date of the eligible visit. The dispensing date, rather than the date a prescription was written, defined observed initiation.”
This states the two drugs, dose ranges, evidence used for initiation, and the seven-day window. It also makes clear that the strategies are not simply “Drug A” and “Drug B” at an unspecified time. If the records show only prescription orders, the report must instead say that an order defined initiation; it should not describe that measure as confirmed dispensing.
The description still needs to say how the grace period was incorporated into the analysis and how people with no qualifying dispensing were handled. If people are classified using future dispensing while time zero is the visit date, the report must not suggest that treatment was known at the visit. The design and analysis must address the interval between time zero and initiation. The strategy wording identifies the issue; it does not resolve it automatically.
Grace Periods and the Time Before Initiation
A grace period allows treatment to begin within a specified interval after time zero. It may reflect ordinary clinical practice or delays in obtaining treatment, but it creates an important classification question: what treatment strategy applies before initiation? Reporting should state both the allowed delay and how the analysis handles it. Simply assigning everyone who eventually initiates to the treatment group from time zero can create a period in which those people had to remain eligible to initiate.
The amount of time before initiation that is guaranteed to be free of an event that prevents qualification depends on each person's actual delay—not on the full length of the grace period for everyone. Under a convention that counts each full day from time zero until dispensing, the total guaranteed pre-initiation time is the sum of individual delays. It reaches the maximum only if every person initiates at the end of the allowed interval. Exact counting also depends on whether partial days and the initiation day are included, so the convention should be stated when person-time is reported.
Worked Example: Calculating Pre-Initiation Time
Hypothetical example. Five people initiate within a seven-day grace period. Their delays from time zero to dispensing are 1, 2, 4, 6, and 7 full days, respectively. Assume the convention counts the full days before dispensing as pre-initiation time.
The total is the sum of the individual delays:
Thus, these five eventual initiators contribute 20 person-days of pre-initiation time under the stated convention. The maximum under this convention is \(5 \times 7 = 35\) person-days, but that maximum would occur only if all five initiated after seven full pre-initiation days. It would be incorrect to claim 35 guaranteed person-days merely because the grace period lasted seven days. If all five had initiated after one day, the corresponding total would be \(5 \times 1=5\) person-days.
This arithmetic illustrates why the initiation dates and the analytic handling of the grace period matter. It does not establish that a particular analysis is biased or unbiased; that depends on the full design and analysis, including how events and treatment assignment are handled.
Switching, Discontinuation, and Other Treatment Changes
A baseline-strategy analysis can keep people in the group defined at time zero even if they later switch or stop treatment. This is often called an intention-to-treat-like or treatment-policy approach in observational settings, though the exact terminology should match the design. The report should say that original classification is retained and should not imply continuous treatment when later use is unknown or differs.
An as-treated analysis instead classifies exposure according to treatment actually received over time. A per-protocol analysis seeks to compare adherence to specified strategies. These approaches answer different questions and require explicit rules for changes and deviations. If follow-up is censored at switching or discontinuation, report the event that triggers censoring and the timing rule. Censoring can create selection concerns; it is not automatically harmless simply because a person no longer follows the initial strategy.
A complete description may also need to define what counts as a treatment gap, how overlapping prescriptions are treated, whether dose changes remain within the strategy, and whether combination therapy is permitted. Do not invent detailed rules that the data cannot support. Instead, state what was measurable, what rule was applied, and any ambiguity or limitation.
Worked Example: Reporting Switching and Discontinuation
Hypothetical example. At time zero, 50 people are classified in the Drug A strategy and 45 in the Drug B comparator. During the analysis period, 6 people in the Drug A group switch to B, 5 discontinue A, and 39 have no recorded switch or discontinuation. In the Drug B group, 4 switch to A, 7 discontinue B, and 34 have no recorded switch or discontinuation.
For a baseline-strategy analysis, the report could state that all 50 people initially classified to A and all 45 initially classified to B remained in their original groups for analysis, regardless of later switching or discontinuation. The counts reconcile: \(6+5+39=50\) and \(4+7+34=45\). This describes retention of baseline classification; it does not claim that everyone continued their original treatment.
For an analysis that censors at the first switch or discontinuation, the numbers censored for those reasons would be \(6+5=11\) in the initial A group and \(4+7=11\) in the initial B group. The report would still need to specify when censoring occurred and how the analysis addressed the consequences of censoring. Those counts alone do not show that the resulting comparison is unbiased.
Diagnostics, Alternatives, and Common Pitfalls
Before reporting results, check whether the definition can be applied consistently to the records. Verify that the stated initiation event exists in the data, that dates can be ordered relative to time zero, and that dose and discontinuation rules are operational rather than aspirational. Tabulate how many people meet each strategy definition, initiate within the grace period, fail to initiate, switch, or discontinue. Large numbers of unclassifiable records, unexpected dates before time zero, or substantial differences in how treatment changes are recorded are reasons to investigate and disclose limitations.
The principal alternatives are not interchangeable reporting styles. An active-comparator, new-user comparison typically contrasts people beginning one eligible treatment with people beginning another near a common index date. A no-treatment comparison uses a different comparator definition and may include people with different care-seeking or treatment histories. A baseline-strategy approach retains initial groups; an as-treated or per-protocol approach updates classification or applies deviation rules. Each can be appropriate for a different question, but the report must name the chosen approach and make its rules explicit.
- Ambiguous comparator: “No treatment” does not reveal whether other drugs or usual care were allowed. Specify what the comparator excludes and permits.
- Misaligned dates: Eligibility, assignment, and time zero described as different dates can create an unclear comparison. State their relationship directly.
- Future-defined exposure: Calling people treated from time zero because they initiate later hides the grace-period issue. Explain how the interval was handled.
- Unspecified switching: “As treated” is incomplete without rules for switches, gaps, and discontinuation.
- Overclaiming from a definition: Clear exposure rules improve interpretability and reproducibility; they do not eliminate confounding or prove causation.
A useful final check is to ask whether another analyst could classify the same records using only the report's rules. If not, add the missing date, dose, grace-period, or treatment-change detail. Also check that the stated strategy is the one the analysis estimates—not merely the one the protocol originally intended.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A study defines initiation as a dispensing within 10 days after time zero. What two additional details should be reported about the grace period and classification?
- Three people initiate after delays of 2, 5, and 10 full days. Under a convention counting full pre-initiation days, calculate their total pre-initiation person-time.
- A report says “Drug A versus no treatment.” Name two ambiguities that prevent a reader from reproducing the comparator.
- In a baseline-strategy analysis, what should the report say about people who switch after time zero?
- Why is it incorrect to say that everyone who initiates within a seven-day grace period contributes seven guaranteed pre-initiation days?
Solutions. (1) State how the 10-day interval is counted, including its start and end, and explain how people who do not initiate within it are classified or handled in the analysis. (2) \(2+5+10=17\) person-days. (3) The report should clarify what “no treatment” excludes, whether other treatments or usual care are allowed, and the period over which nonuse is assessed; any two of these identify genuine ambiguities. (4) It should state whether switchers remain in their original baseline group or are reclassified or censored, and give the rule and timing. (5) Each person's pre-initiation time depends on their actual delay. Seven days is the maximum for an individual under the stated convention, not a guaranteed duration for someone who initiates earlier.