Crossover Schedule Review: A Design Checklist

Why Review the Schedule as a Whole?

A crossover schedule is more than a sequence of treatment letters. It specifies which treatment each participant receives in each period, how participants are allocated to sequences, what separates administrations, and how much repeated participation the protocol requires. An error in any of these elements can leave a treatment-period combination uncovered, undermine the intended replication, make a transition difficult to interpret, or create an operational burden that threatens completion.

The purpose of a schedule review is to expose such problems while changes are still feasible. It is a structured design check, not a substitute for the earlier work on selecting a layout, evaluating carryover risk, planning washout, allocating enrollment, or constructing randomization lists. Those subjects provide the detailed methods; the audit brings their implications together and asks whether the proposed schedule is coherent and implementable.

Definition: A schedule audit is a systematic review of a proposed crossover design that checks treatment coverage by period, sequence allocation, treatment transitions, washout provisions, repeated administrations, and participant burden against the study objectives and protocol.

A Compact Framework for Coverage

Write each planned sequence as an ordered list of treatments, one entry per period. For a two-treatment study, let \(T\) denote the test treatment and \(R\) the reference treatment. A treatment-by-period count table helps reveal whether the schedule covers the planned combinations. If \(n_s\) participants are assigned to sequence \(s\), and \(I(s,p,j)\) equals 1 when sequence \(s\) assigns treatment \(j\) in period \(p\) and 0 otherwise, then the number assigned treatment \(j\) in period \(p\) is:

N_{pj}=\sum_s n_s I(s,p,j).

The table does not by itself prove that a design is suitable. It makes the schedule's structure visible: a zero means that a treatment is absent from a period, while unequal counts show the degree and pattern of imbalance. Whether a particular imbalance is acceptable depends on the design, planned analysis, and practical constraints. In a replicate design, also count how many times each participant receives each treatment; the cohort-level table cannot verify that individual-level replication is correct.

Sequence allocation must also be reviewed separately from treatment-period coverage. A randomization procedure determines how participants are assigned; balance is a property of the resulting or planned allocation counts. An allocation can be randomized without having exactly equal counts, particularly when enrollment is not divisible by the number of sequences. Conversely, equal counts alone do not demonstrate that assignment was randomized or concealed appropriately.

Worked Examples

Worked Example: Auditing a Two-Period Schedule

A hypothetical study proposes two sequences, \(TR\) and \(RT\), with 12 participants planned in each. Every participant is intended to receive both treatments once, in separate periods. A seven-day washout is proposed. Review the coverage and identify what still needs checking.

Coverage by period. In period 1, the 12 participants assigned \(TR\) receive \(T\), and the 12 assigned \(RT\) receive \(R\). In period 2 the assignments reverse. Thus:

\begin{array}{c|cc} & T & R\\ \hline \text{Period 1} & 12 & 12\\ \text{Period 2} & 12 & 12 \end{array}

Each participant receives \(T\) once and \(R\) once, and both treatments occur in both periods. The planned sequence counts are equal, so the design is balanced by sequence at the planned enrollment of 24. This arithmetic checks the proposed allocation, but randomization still requires an appropriate assignment process.

Washout check. Suppose, for illustration, the relevant terminal half-life were 24 hours and elimination followed a single exponential process. Seven days is 168 hours, or seven half-lives. The fraction remaining under that simplified model would be:

\left(\frac{1}{2}\right)^{168/24} = \left(\frac{1}{2}\right)^7 = 0.0078125 \approx 0.781\%.

This calculation supports a pharmacokinetic discussion but does not establish that seven days is adequate. The reviewers must consider the relevant drug and active metabolites, the target of the assessment, the clinical and analytical context, and the protocol's predose sampling and stopping or review rules. The assumptions of a single relevant half-life and simple exponential elimination may not capture all relevant behavior.

Conclusion. The sequence and treatment-period counts are internally consistent. Before implementation, reviewers should verify that the randomization plan matches the counts, that washout has a justified rationale beyond this simplified calculation, and that administration, sampling, and visit timing can be followed in both sequences.

Worked Example: Checking Replication in a Three-Period Layout

A hypothetical partial-replicate study uses sequences \(TRR\), \(RTR\), and \(RRT\), with six participants allocated to each. The intent is for each participant to receive \(T\) once and \(R\) twice. Check both the participant-level replication and the overall period coverage.

Participant-level check. Each sequence contains one \(T\) and two \(R\) administrations: \(TRR\) has one and two, respectively; \(RTR\) has one and two; and \(RRT\) has one and two. The individual replication requirement is met.

Period counts. In period 1, sequence \(TRR\) contributes six \(T\) administrations, while the other two sequences contribute 12 \(R\) administrations. In period 2, \(RTR\) contributes six \(T\) administrations and the other sequences contribute 12 \(R\) administrations. In period 3, \(RRT\) contributes six \(T\) administrations and the other sequences contribute 12 \(R\) administrations. Therefore:

\begin{array}{c|cc} & T & R\\ \hline \text{Period 1} & 6 & 12\\ \text{Period 2} & 6 & 12\\ \text{Period 3} & 6 & 12 \end{array}

The treatment counts are not equal within a period, but the pattern is the same across periods. Each period has 18 planned administrations, with one-third \(T\) and two-thirds \(R\). Thus it would be misleading to call the table treatment-balanced within each period; it is consistent across periods, and that pattern follows from the intended one-versus-two replication.

As a transition check, each sequence has two consecutive-period transitions. Across the six participants per sequence, the total is 12 \(T\)-to-\(R\) transitions, 12 \(R\)-to-\(T\) transitions, and 12 \(R\)-to-\(R\) transitions. This confirms that the schedule includes the transitions implied by the listed sequences; it does not, on its own, establish that carryover is negligible or that every transition is appropriate.

Conclusion. The schedule satisfies the stated replication pattern and produces the same treatment-count pattern in all three periods. The team should still verify the intended analysis, transition handling, washout between each pair of administrations, and whether the additional reference administration is acceptable to participants.

Worked Example: Finding an Allocation and Burden Concern

A hypothetical four-period full-replicate proposal uses sequences \(TRTR\) and \(RTRT\). The planned enrollment is 18, with 10 participants assigned to \(TRTR\) and 8 to \(RTRT\). The schedule requires four dosing visits and three washout intervals. What does an audit show?

Replication and period coverage. Both sequences give each participant two \(T\) and two \(R\) administrations. In periods 1 and 3, the counts are 10 \(T\) and 8 \(R\); in periods 2 and 4, they are 8 \(T\) and 10 \(R\). The counts follow directly from the sequence assignments:

\begin{array}{c|cc} & T & R\\ \hline \text{Period 1} & 10 & 8\\ \text{Period 2} & 8 & 10\\ \text{Period 3} & 10 & 8\\ \text{Period 4} & 8 & 10 \end{array}

There are no treatment-period gaps, and each person has the required two administrations of each treatment. The sequence allocation differs by two participants. It is therefore not exactly equal, but that fact alone does not invalidate a randomized plan. Reviewers should check that the planned randomization method supports the intended allocation, that the imbalance is acceptable for the study objectives, and that the allocation is not altered informally during enrollment.

Burden and implementation. Four periods mean four dosing occasions and three between-period intervals, in addition to any screening, predose, postdose, and follow-up procedures. The protocol team should map the actual participant timeline, including travel, overnight stays, sampling duration, scheduling constraints, and missed-visit procedures. A schedule that is structurally sound may still be impractical if the required visits are too demanding or difficult to complete.

Conclusion. The treatment coverage and within-participant replication are correct for the stated sequences. The allocation is modestly unequal and warrants a documented randomization and analysis review; the number and timing of visits warrant a separate feasibility review. No power conclusion follows from these schedule counts.

Reviewing the Full Protocol

A reliable audit proceeds from the written schedule to the operations required to carry it out. The following checks help keep distinct issues from being confused:

Confirm the objective and layout.
Check that the sequence list matches the intended crossover or replicate design and that each participant's treatment count matches the design objective.

Tabulate coverage.
Calculate treatment-by-period counts from the sequence allocations. Look for missing combinations, inconsistent patterns, or imbalances that need explanation.

Check allocation and transitions.
Verify planned sequence counts against the randomization plan and review the transitions that the schedule actually creates. Do not treat equal counts as proof of valid randomization.

Review washout and period procedures.
Check that the rationale and duration apply to each interval and that predose sampling, dosing, meals, and sampling times are specified consistently across periods.

Test feasibility.
Count visits, procedures, and required time for each participant. Consider how delays, missed visits, withdrawals, or rescheduling would affect the schedule and sequence balance.

The washout check should use the relevant pharmacokinetic and clinical rationale rather than a universal number of half-lives. A calculation based on a terminal half-life can be useful, but its assumptions should be stated and the protocol should specify how residual concentrations or other concerns will be assessed. For a replicate schedule, examine every interval, not just the first. Refer to the earlier material on washout planning and carryover risks when a schedule raises those questions.

Repeated administrations also need to be operationally comparable. Confirm that the same treatment is given under the intended conditions each time and that protocol instructions address deviations that could make administrations differ. Replication on paper is not sufficient if visit procedures, sampling windows, or dosing conditions vary systematically between periods. A schedule review can flag this problem; it cannot replace detailed protocol and analysis specifications.

Diagnostics, Alternatives, and Limits

Warning signs include a treatment absent from a period when period coverage is intended; a participant receiving the wrong number of administrations; a sequence count that differs from the randomization plan; an unexplained transition or washout; or a visit calendar that requires an implausible number of procedures. A second, independent reconstruction of the count table is a useful basic diagnostic. If the independently calculated counts do not agree, resolve the discrepancy before producing randomization materials.

A spreadsheet count table is a transparent first-line audit for small designs. A script or validated scheduling tool is useful when there are many sequences, participants, or constraints; it can automate counts and flag impossible assignments. Neither approach replaces a protocol review by people who understand treatment administration and participant procedures. A design or statistical review asks whether the layout supports the planned objectives and analysis, whereas an operational review asks whether it can be carried out as written. Both are necessary, but they answer different questions.

The schedule audit also has limits. It cannot establish that the study is adequately powered, that the selected outcomes or analysis are appropriate, or that a washout is clinically adequate merely because the sequence counts look balanced. Those questions require their own design, pharmacokinetic, statistical, and clinical assessments. In particular, avoid using the period-count table as a proxy for power or treating equal allocation as a guarantee against bias.

Common Mistake

A common mistake is to reason, “The sequence counts are equal, so the schedule is balanced and ready.” Equal sequence counts do not prove that every treatment appears in every period, that each participant receives the intended number of administrations, or that the randomization process is valid. The correction is to check participant-level replication, treatment-by-period counts, allocation procedures, transitions, washout provisions, and feasibility as separate items. A schedule can pass one check and fail another.

Key takeaway: Review the schedule at both the participant and cohort levels. Verify what each person receives, count treatment coverage by period, and separately assess allocation, transitions, washout, repeated procedures, and burden. A clean schedule audit is necessary for implementation, but it does not replace power, analysis, or clinical review.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

In a two-period study, 8 participants are allocated to \(TR\) and 8 to \(RT\). How many participants receive each treatment in each period?
Why does equal allocation to two sequences not, by itself, prove that randomization was conducted appropriately?
For sequences \(TRR\), \(RTR\), and \(RRT\), what is the number of \(T\) and \(R\) administrations per participant?
A proposed washout is seven half-lives under a single-exponential elimination assumption. What fraction remains under that model, and why is that calculation not sufficient to establish washout adequacy?
Name two participant-burden issues that a schedule audit should examine beyond the number of treatment periods.

Solutions. (1) Period 1 has 8 \(T\) and 8 \(R\); period 2 also has 8 \(T\) and 8 \(R\), because the sequences reverse. (2) Equal counts describe allocation totals, not how assignments were generated, concealed, or implemented. (3) Each participant receives \(T\) once and \(R\) twice. (4) The remaining fraction is \((1/2)^7=0.0078125\), or about 0.781%; the calculation depends on simplified elimination assumptions and does not address all relevant metabolites, clinical context, or protocol assessments. (5) Examples include travel and overnight requirements, total sampling time, visit scheduling, and the procedures required at each visit.