Why Sequence-Group Balance Matters
A four-period full-replicate schedule gives each participant two periods on the test formulation and two on the reference formulation. But the order determines which treatment appears in each period. In a TRTR sequence, test (T) is given in periods 1 and 3, and reference (R) in periods 2 and 4. In an RTRT sequence, those assignments are reversed. Assigning participants to both sequences can therefore balance T and R within each period across the study.
This balance concerns the planned treatment schedule, not the outcomes measured after dosing. Its basic design question is simple: how many participants should be assigned to each sequence so that the treatment counts in each period are as even as possible?
Counts Implied by the Two Sequences
Let \(m\) be the number assigned to TRTR and \(k\) the number assigned to RTRT. Each sequence contributes a fixed treatment assignment to every period. The table shows the counts after combining participants in both groups.
| Period | T count | R count | T minus R |
|---|---|---|---|
| 1 | \(m\) | \(k\) | \(m-k\) |
| 2 | \(k\) | \(m\) | \(k-m\) |
| 3 | \(m\) | \(k\) | \(m-k\) |
| 4 | \(k\) | \(m\) | \(k-m\) |
The difference reverses direction in alternating periods. If TRTR has more participants, T has more assignments in periods 1 and 3, while R has more in periods 2 and 4. If the sequence groups have equal sizes, both treatments have the same count in every period. This is different from merely ensuring that the study contains equal total numbers of T and R exposures: each participant already has two exposures to each treatment, so those overall totals are equal even when treatment-by-period counts are not.
Here \(d_p\) is the T count minus the R count in period \(p\). The sum of the four differences is always zero, because each sequence contributes two T assignments and two R assignments. That overall cancellation does not mean the treatments are balanced within each period: the differences can be positive in two periods and negative in the other two.
Worked Example: Equal Allocation Gives Exact Balance
Suppose 24 hypothetical participants are assigned, with 12 to TRTR and 12 to RTRT. Calculate the treatment counts in each period.
In period 1, the 12 TRTR participants receive T and the 12 RTRT participants receive R, giving T = 12 and R = 12. In period 2, the assignments reverse, giving T = 12 and R = 12. Periods 3 and 4 repeat the period 1 and period 2 patterns, respectively.
The treatment-by-period count table therefore has 12 participants on each treatment in all four periods. The difference is \(d_p=0\) for every period. This is exact balance in the assigned schedule; it does not imply that treatment responses will be equal or that all other design concerns have been resolved.
Why Equal Sequence Groups Minimize Imbalance
A useful summary of treatment-by-period imbalance is the sum of squared within-period count differences. Squaring prevents positive and negative differences from cancelling across periods, and gives greater weight to larger differences. From the table above, the summary is:
For a fixed total sample size \(N=m+k\), this quantity is smallest when \(|m-k|\) is as small as possible. If \(N\) is even, choose \(m=k=N/2\), giving zero imbalance. If \(N\) is odd, the sequence-group sizes cannot be equal integers. The smallest possible difference is one, achieved by allocating \((N+1)/2\) participants to one sequence and \((N-1)/2\) to the other. Which sequence receives the extra participant determines the direction of the one-person differences, not their size.
Worked Example: The Best Possible Allocation for an Odd Total
Consider 25 hypothetical participants. Exact equality between two integer sequence-group sizes is impossible. Compare allocations of 13 to TRTR and 12 to RTRT.
In periods 1 and 3, TRTR contributes 13 T assignments and RTRT contributes 12 R assignments. In periods 2 and 4, RTRT contributes 12 T assignments and TRTR contributes 13 R assignments. Thus the treatment counts are:
Periods 1 and 3: T = 13, R = 12, so \(d_p=1\).
Periods 2 and 4: T = 12, R = 13, so \(d_p=-1\).
The sum of squared differences is \(1^2+(-1)^2+1^2+(-1)^2=4\), the smallest possible for this two-sequence design with 25 participants. The study still has 50 T assignments and 50 R assignments overall, because each participant receives each treatment twice. Those equal overall totals coexist with a one-participant difference in every period.
Worked Example: Consequences of Unequal Sequence Allocation
Suppose 30 hypothetical participants are assigned, but an allocation decision places 18 in TRTR and 12 in RTRT. Calculate the treatment-by-period counts and compare them with an equal allocation.
For periods 1 and 3, T = 18 and R = 12, so the difference is 6. For periods 2 and 4, T = 12 and R = 18, so the difference is \(-6\). The sum of squared differences is \(6^2+(-6)^2+6^2+(-6)^2=144\).
An allocation of 15 participants to each sequence instead gives 15 T and 15 R assignments in every period, with sum of squared differences \(0\). Both allocations produce 60 T exposures and 60 R exposures overall. The unequal allocation therefore does not change the overall treatment totals, but it creates substantial treatment-by-period imbalance. If there is no design reason to favor one sequence, equal allocation is the direct way to avoid that imbalance.
Assigning Participants and Checking the Schedule
The count result tells the planner what group sizes to target; it does not by itself specify how to assign individuals. A randomized assignment to sequence helps prevent systematic differences between the groups in participant characteristics. When exact or near-exact group sizes are required, a randomization procedure can allocate participants while maintaining the target numbers. For an odd planned total, one group receives one additional participant. The protocol should specify the target allocation before assignments begin.
Unrestricted random assignment may, by chance, produce unequal sequence-group sizes, especially in a small study. With only these two sequences, any realized difference \(m-k\) directly determines the size of the alternating period-specific count differences. A planned allocation with equal group sizes controls that count imbalance more reliably. It does not guarantee that the groups will be similar on every participant characteristic; randomization addresses that concern by chance rather than by making the observed characteristics identical.
It is useful to distinguish the planned counts from the counts that remain in the analysis. If all assigned participants complete all four periods, the treatment-by-period counts in the completed records match the assignment table. If participants withdraw or have unusable period records, the counts among complete records can differ from the planned counts. A schedule can be balanced at assignment and become unbalanced in the observed records.
Choose \(m\) and \(k\) to be equal when the planned total is even, or to differ by one when it is odd, unless a study-specific design constraint justifies another allocation.
Use the planned randomized assignment procedure and preserve a record of each participant’s assigned sequence.
For each period, count assigned T and R treatments. Repeat the tabulation for the records actually available for analysis if withdrawals or missing periods occur.
Compare observed counts with the planned table and check for assignment errors, deviations from the schedule, or differential loss of records. Do not treat a count discrepancy alone as evidence of a treatment effect.
What Balance Does—and Does Not—Achieve
Treatment-by-period balance makes the scheduled numbers on T and R equal within each period. This can prevent one treatment from being systematically represented by more participants in one period than the other treatment, simply because one sequence group is larger. It is a property of the assignment pattern and can be verified from sequence and period records without looking at pharmacokinetic outcomes.
Balance is not a remedy for every design problem. Equal group sizes do not remove carryover, guarantee comparable period-specific responses, or establish that treatment effects are constant across periods. Nor does a balanced count table establish that treatment caused any observed outcome difference. Those questions require appropriate design and analysis; the count table only describes the schedule.
An alternative is to allocate participants without a strict sequence-group target and accept the realized imbalance. That approach can be reasonable when other constraints govern allocation, but its treatment-by-period count consequences should be checked rather than assumed away. At the other extreme, assigning everyone to just one of the two sequences maximizes the imbalance: in each period, only one treatment is represented. Using both sequences addresses this particular problem; it does not make the groups interchangeable for every scientific purpose.
Common Mistakes and Diagnostics
- Checking only total exposure counts. Each participant receives T twice and R twice, so total counts match even if the treatment counts differ within every period. Inspect the four periods separately.
- Assuming two sequences automatically mean balance. Balance depends on the numbers assigned to each sequence. With \(m\ne k\), each period has a difference of \(|m-k|\).
- Calling an odd-sample allocation exactly balanced. Two integer group sizes cannot be equal when their sum is odd. A difference of one is the closest possible allocation, not exact balance.
- Using planned counts as if they were completed-record counts. Withdrawals can alter the counts available for analysis. Re-tabulate the actual records by treatment and period.
- Interpreting balanced counts as evidence that carryover is absent. The table records assigned treatments, not residual drug concentrations or response mechanisms. Schedule balance does not diagnose or eliminate carryover.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- If 16 participants are assigned to TRTR and 16 to RTRT, what are the T and R counts in period 3?
- With 21 participants, what sequence-group sizes minimize the largest treatment-by-period count difference, and what is that difference?
- A study has 14 TRTR participants and 10 RTRT participants. What are the T and R counts in periods 2 and 4?
- Why can equal total numbers of T and R exposures fail to demonstrate treatment-by-period balance?
- Why should treatment-by-period counts be tabulated again after withdrawals?
Solutions. (1) Period 3 has 16 T assignments from TRTR and 16 R assignments from RTRT. (2) Use group sizes 11 and 10; the difference is one participant in every period, so the largest absolute count difference is 1. (3) In periods 2 and 4, T = 10 and R = 14, because the sequence assignments reverse relative to periods 1 and 3. (4) Each participant receives two exposures to each treatment, so totals match even when a particular treatment is more common in particular periods. (5) Withdrawals or missing period records can change the sequence-group sizes represented in the analysis, so the planned balance may no longer describe the available records.