When Residual Exposure Shapes the Design
A crossover study compares treatments within the same participants across periods. That comparison is easiest to interpret when the previous period’s treatment no longer influences the outcome in the current period. With a formulation that has prolonged residual exposure, this condition may be difficult to achieve: simply inserting the usual washout interval does not establish that the next period begins free of pharmacologically relevant drug.
The design response is not automatically “use a replicate.” A longer washout can reduce residual concentrations, while replication can provide repeated observations on a treatment and may support estimation of within-subject variability. Those choices address different issues. Replication does not itself remove residual drug, and a longer washout does not guarantee that every other crossover assumption holds.
A useful design question is therefore: what washout duration makes residual influence acceptably unlikely, and what schedule provides the needed treatment comparisons without introducing avoidable burden or ambiguous transitions? The answer depends on the drug, active metabolites, outcome, assay sensitivity, and study objective—not on a universal number of half-lives.
Translate Residual Exposure Into a Washout Target
Under a simplifying first-order elimination model, concentration after treatment stops declines exponentially. If \(C_0\) is the concentration at the start of washout and \(t_{1/2}\) is the terminal half-life, the concentration after washout duration \(W\) is:
This equation follows because each elapsed half-life multiplies the concentration by one-half. After \(W/t_{1/2}\) half-lives, repeated multiplication gives \(C_0(1/2)^{W/t_{1/2}}\), which is the stated expression. If the design specifies a target concentration \(C_{\text{crit}}\), rearranging gives:
This is a planning calculation, not proof that the next period is unaffected. The target should be justified in relation to pharmacological activity and the study outcome. A concentration below an assay’s quantification limit may still be active; conversely, measurable residual concentration may not meaningfully affect the outcome. Active metabolites, absorption delays, nonlinear elimination, and long terminal phases can also make a single-half-life model inadequate.
Worked Example: Calculating a Candidate Washout
Suppose a hypothetical formulation has a terminal half-life of 36 hours. For planning, the team sets a provisional target that the residual concentration be no more than 1% of the concentration at washout start. Assume first-order elimination and a relevant single terminal phase. What washout duration meets this target?
Here \(C_{\text{crit}}/C_0=0.01\), so \(C_0/C_{\text{crit}}=100\). Substituting into the washout equation:
Converting to days gives \(239.2/24=9.967\) days, or approximately 10 days. A 7-day interval would leave, under the same model, \(2^{-168/36}=0.0394\), or 3.94% of the starting concentration. A 10-day interval would leave \(2^{-240/36}=0.00984\), or 0.984%. Thus 10 days meets the provisional 1% target under the stated assumptions; it does not establish that 1% is clinically negligible or that the assumed elimination model is adequate.
In practice, washout planning should consider the terminal behavior of the parent drug and any active metabolites, the timing of predose samples, and a biologically defensible residual threshold. When information is uncertain, a conservative interval may be appropriate, but it should be weighed against the additional time, visits, and attrition risk it creates.
What Replication Can—and Cannot—Change
A replicate schedule gives participants more than one exposure to at least one treatment. In a three-period partial-replicate schedule, for example, sequences TRR and RTT give each participant one exposure to T and two to R. A four-period full-replicate schedule such as TRTR and RTRT gives each participant two exposures to each treatment. These schedules can support repeated within-participant observations, but they also create additional treatment transitions and periods in which residual exposure could matter.
The appropriate comparison is not “washout or replication” as though only one can be used. A study may extend washout in either a two-period or replicate design. Instead, compare the consequences of each design choice: washout duration affects the likelihood of residual exposure; replication affects the number and pattern of treatment exposures, participant burden, and the information available from repeated observations.
| Design response | Potential benefit | What it does not guarantee |
|---|---|---|
| Extend washout | More time for drug and active metabolites to decline before the next period | No guarantee that residual influence is absent or that all other crossover assumptions hold |
| Use a partial replicate | Repeated observations on one treatment, with fewer periods than a four-period full replicate | No automatic removal or identification of carryover; participants still have repeated transitions |
| Use a full replicate | Repeated observations on both treatments | No automatic protection against carryover, period effects, or attrition |
Worked Example: Comparing Burden and Exposure Counts
A hypothetical study is considering a two-period TR/RT crossover with an extended washout, a three-period partial replicate (TRR/RTT), or a four-period full replicate (TRTR/RTRT). Assume equal allocation to the two sequences and that every participant completes the assigned periods. Compare the treatment exposures for 24 participants.
Two-period design. Each participant receives one T and one R, for 24 T exposures and 24 R exposures overall. There are 48 treatment periods per participant cohort in total.
Three-period partial replicate. Each participant receives one T and two R. Across 24 participants, this gives \(24\times1=24\) T exposures and \(24\times2=48\) R exposures, or 72 treatment periods overall. Each participant has three periods, one more than in the two-period option.
Four-period full replicate. Each participant receives two T and two R. Across 24 participants, this gives 48 exposures to each treatment and 96 treatment periods overall. Each participant has four periods, doubling the period count relative to the two-period option.
The replicate options increase repeated exposure information, but they also increase treatment periods and opportunities for missed visits or withdrawal. None of the three schedules, by its exposure count alone, establishes that a washout is long enough. In all options, the residual-exposure question must be addressed for each transition in the assigned schedule.
Assumptions That Remain Necessary
For a standard crossover treatment comparison, the design and analysis still require assumptions about how treatment, period, and participant effects relate to the measured outcome. For example, let \(Y_{ij}\) denote the outcome for participant \(i\) in period \(j\). A simple additive description is:
This description presumes that any relevant effect of treatment in an earlier period is absent or otherwise adequately addressed. If residual treatment influence remains and is associated with the current treatment, the current-treatment comparison may combine the current treatment effect with a prior-treatment effect. Adding replicate periods does not make that issue disappear.
The study also needs valid random assignment to sequences, a schedule that provides the intended treatment and period comparisons, and an analysis that accounts for the within-participant structure. The outcome and treatment conditions should be sufficiently stable across periods for a within-participant comparison to be meaningful, or important changes should be handled in the design and analysis. Adherence, missing outcomes, and differential withdrawal must not undermine the planned comparisons. A design can reduce one risk while leaving these other conditions unresolved.
Transition patterns deserve explicit attention. In a schedule with repeated alternating treatments, the previous treatment may be closely tied to the current treatment. In such a case, observing a difference between current-treatment outcomes does not by itself separate a current effect from a residual effect of the preceding treatment. A sequence with a different transition pattern may provide different information, but the pattern should be checked rather than assumed to solve the problem. The earlier material on treatment transitions is useful for listing which transitions actually occur.
Worked Example: Using Predose Results as a Diagnostic
Suppose a hypothetical study planned a 10-day washout based on a concentration target of 1% of the concentration at washout start. Predose concentrations are then available for 10 participants, expressed as a percentage of their own relevant starting concentrations: 0.3%, 0.6%, 0.8%, 0.9%, 1.0%, 1.1%, 1.2%, 0.7%, 1.4%, and 0.5%. Assume the 1% threshold is a justified planning target for this example and that the assay can measure concentrations around that level.
Four of the ten values exceed 1%: 1.1%, 1.2%, 1.4%, and (if equality is treated as acceptable) the 1.0% value does not exceed it. Thus \(4/10=0.40\), or 40%, are above the target. One additional participant is exactly at the target.
These observations are a warning that the planned interval may not reliably meet the target in this hypothetical group. They do not establish the size of any effect on the pharmacokinetic outcome, nor do they prove that every participant with a measurable concentration has clinically important carryover. The team should investigate the concentration profile and assay, consider whether the threshold remains appropriate, and decide whether to extend washout or revise the protocol before relying on the planned comparison. A predose value below target is a diagnostic result, not proof of complete absence of carryover.
Diagnostics, Alternatives, and Limitations
Useful diagnostics should be tied to the proposed mechanism. Predose samples can reveal residual concentration at the beginning of a period; observed concentration-time profiles can reveal a terminal phase longer than expected; and treatment-by-period patterns may flag unexpected changes. Such checks can identify a design concern, but a nonsignificant statistical test for carryover is not proof that carryover is absent. Tests may have limited ability to detect it, and an outcome-based test cannot replace pharmacological reasoning about the washout.
The main alternatives are to extend the washout, change the sequence structure, or reconsider whether a crossover is suitable. Extending washout directly targets time for elimination but increases study duration. A replicate sequence may be attractive when repeated observations are needed, but it adds periods and may make the consequences of residual exposure more complicated. A parallel-group design avoids within-participant treatment transitions, but gives up the within-participant comparison and has different precision and sample-size considerations. Detailed layout selection depends on the objectives and constraints of the study.
Common signs that the design reasoning is going wrong include treating a fixed “five half-lives” rule as universally sufficient, selecting a replicate because carryover is suspected, or concluding from a nonsignificant carryover test that the washout is adequate. The correction is to specify a relevant residual threshold, justify the elimination model and sampling plan, inspect every scheduled transition, and distinguish the carryover problem from the reason replication is being considered.
Check Your Understanding
Answer each question, then compare with the short solutions below.
- A hypothetical drug has a 24-hour half-life. Under first-order elimination, what fraction of its starting concentration remains after 120 hours?
- Why does a replicate schedule not automatically solve a prolonged-residual-exposure problem?
- A team observes predose concentrations below its assay’s quantification limit. What can it conclude, and what can it not conclude?
- Name two assumptions, other than adequate washout, that remain important for a crossover treatment comparison.
Solutions. (1) \(120/24=5\) half-lives, so the remaining fraction is \(2^{-5}=1/32=0.03125\), or 3.125%. (2) Replication adds repeated exposures but does not accelerate elimination; residual exposure can still affect later periods, and transition patterns may connect prior and current treatment. (3) The concentrations are below the assay’s quantification limit; this does not prove that no drug is present or that any remaining amount is pharmacologically irrelevant. (4) Examples include valid randomization to sequences, appropriate handling of period effects, adequate accounting for within-participant dependence, and sufficient adherence and outcome data.