Why Participant Burden Matters
A crossover schedule is not only a sequence of treatment exposures. It is also a commitment of time and effort for each participant: attending study visits, receiving each scheduled dose, following any period-specific restrictions, completing sampling and assessments, and remaining available through washouts and follow-up. A design with more periods generally asks more of participants, but period count alone does not describe the entire burden. The length and timing of visits, the number of samples, and the practical demands between visits also matter.
Burden matters for feasibility and retention. If participants withdraw before completing a schedule, fewer will contribute observations from every planned period. That can reduce the information available for comparisons that rely on repeated exposures. Attrition may also affect interpretation if the participants who remain differ systematically from those who withdraw. Counting the opportunities for withdrawal is therefore useful during design planning, while recognizing that it does not by itself predict the actual withdrawal rate.
Counting Scheduled Burden
The counting rule is simple, but its interpretation needs care. The \(P-1\) transitions count only the gaps between the scheduled treatment periods. They do not specify the duration of each washout, the number of clinic contacts during it, or whether participants must comply with restrictions at home. Likewise, one dosing occasion could involve a long inpatient visit or a brief outpatient appointment. To compare burden meaningfully, record both the period-based counts and the study activities that are not captured by those counts.
| Periods per participant | Planned dosing occasions, if one per period | Between-period transitions | Typical implication for burden |
|---|---|---|---|
| Two | 2 | 1 | Fewer scheduled periods and transitions than three- or four-period schedules |
| Three | 3 | 2 | One more planned dosing occasion and transition than a two-period schedule |
| Four | 4 | 3 | More planned periods and transitions, with potentially greater cumulative time and effort |
These are schedule counts, not universal rankings of total effort. For example, a three-period schedule with lengthy inpatient stays could demand more time than a four-period schedule with brief outpatient dosing. A burden audit should therefore state the counting assumptions and also list sampling, travel, restrictions, and scheduled time.
Worked Example: Auditing Planned Activities
Suppose a hypothetical study compares schedules with two, three, or four periods. Each period contains one dosing occasion and one clinic visit, and every transition between periods includes one washout interval. Screening and a final follow-up visit are required for every participant, regardless of the number of periods. Count the planned doses, period visits, washout intervals, and total visits.
For a two-period schedule, the counts are 2 doses, 2 period visits, and \(2-1=1\) washout interval. Including screening and final follow-up, the total visit count is \(2+2=4\).
For a three-period schedule, the counts are 3 doses, 3 period visits, and \(3-1=2\) washout intervals. Including the two additional visits, the total is \(3+2=5\) visits.
For a four-period schedule, the counts are 4 doses, 4 period visits, and \(4-1=3\) washout intervals. Including screening and final follow-up, the total is \(4+2=6\) visits.
The comparison is conditional on the stated assumption that there is one visit per period and no extra visits during washout. If a washout requires an additional clinic visit, that visit must be added; the period count alone would not capture it.
Withdrawal Opportunities and Completion
A schedule with more periods gives participants more time and more occasions to withdraw before all planned periods are completed. To quantify this, define a transition as the interval after one period in which a participant could withdraw before starting the next. Let \(q_j\) be the probability of withdrawal during transition \(j\), conditional on having remained in the study through the preceding period. The probability of completing all \(P\) periods is the probability of surviving every one of the \(P-1\) transitions.
This product follows from the chain rule for conditional probabilities: completion requires passing each transition, and \(1-q_j\) is the conditional probability of passing transition \(j\), given survival to it. No assumption that the transitions are statistically independent is needed when each \(q_j\) is defined conditionally in this way.
A convenient simplifying assumption is a constant conditional withdrawal probability \(q\) at every transition. Then the completion probability is \((1-q)^{P-1}\). If \(N\) participants start the first period, the expected number who complete all periods is \(N(1-q)^{P-1}\). This is an expectation under the model, not a guarantee that the observed count will equal that value.
Worked Example: Equal Withdrawal Probability Across Transitions
In a hypothetical study, 120 participants begin the first period. Assume a 10% conditional probability of withdrawal before each subsequent period, the same at every transition. Estimate the number completing all periods under two-, three-, and four-period schedules.
For two periods there is one transition, so the completion probability is \(1-0.10=0.90\). The expected number completing is \(120(0.90)=108\).
For three periods there are two transitions, so the completion probability is \((0.90)^2=0.81\). The expected number completing is \(120(0.81)=97.2\), or about 97 participants.
For four periods there are three transitions, so the completion probability is \((0.90)^3=0.729\). The expected number completing is \(120(0.729)=87.48\), or about 87 participants.
Under these assumptions, the expected complete-schedule count is lower as the number of transitions increases. The calculation does not say that a particular study will lose exactly 12, 23, or 33 participants; observed withdrawals vary, and the assumed 10% probability may not match actual retention.
Worked Example: Different Withdrawal Probabilities by Transition
Consider a hypothetical four-period schedule with 80 participants starting period 1. Suppose the conditional withdrawal probabilities before periods 2, 3, and 4 are 5%, 10%, and 15%, respectively. Calculate the probability and expected number completing all four periods.
The conditional retention probabilities are \(1-0.05=0.95\), \(1-0.10=0.90\), and \(1-0.15=0.85\). Therefore,
The expected number completing is \(80(0.72675)=58.14\), or about 58 participants. Using a single average withdrawal probability without justification could give a different estimate. Here the calculation appropriately allows the conditional risk to increase across transitions.
Interpreting Attrition and Comparing Approaches
The full-completion calculation is a transparent planning summary, not a complete analysis of missing data. It focuses on participants who pass every transition. A participant who withdraws late may still have provided observations from several periods, while a participant who withdraws early may have provided fewer. Whether partial data can inform a particular analysis depends on the design, the analysis model, and the reason the data are missing. The expected number of full completers is therefore not automatically the same as the number of participants whose data can contribute to every analysis.
The main alternatives to a constant-rate calculation are to use transition-specific conditional probabilities, as in the previous example, or to describe retention under several plausible scenarios. The first is useful when different stages plausibly have different retention. Scenario analysis is useful when rates are uncertain: for instance, a study team can compare expected completion under lower, central, and higher withdrawal assumptions. None of these approaches removes uncertainty about future retention; they make the assumptions visible.
A simple comparison of schedules can support a design discussion, but it cannot determine which design is best on its own. Additional periods may support repeated observations that a shorter schedule cannot provide. Reducing periods can ease the participant commitment, but it may also remove information that motivated a replicate design. The design choice should weigh that scientific value against the added time, procedures, and risk of incomplete participation, rather than treating fewer periods as automatically preferable.
Diagnostics for burden and attrition planning include checking whether every scheduled dose, clinic visit, washout contact, and follow-up has been counted; confirming that the denominator for each withdrawal probability is participants who reached that transition; and comparing assumed rates with relevant operational experience when available. During a study, review retention by transition and investigate whether observed withdrawals cluster around particular visits or procedures. A marked departure from the assumed pattern signals that the planning calculation may no longer describe expected completion.
Common Pitfalls and Limitations
- Equating periods with all visits. A period count does not include screening, unscheduled contacts, or extra washout visits unless the accounting explicitly includes them.
- Counting \(P\) withdrawal intervals for \(P\) periods. Under the stated convention, withdrawal before a later period can occur at \(P-1\) between-period transitions; additional within-period withdrawal risks require a separate model.
- Multiplying unconditional percentages. Use conditional retention probabilities for participants who have reached each transition. A probability measured among all enrolled participants is not interchangeable with a conditional transition probability.
- Assuming attrition only reduces precision. If withdrawal is related to participant characteristics, tolerability, treatment response, or other outcomes, the remaining participants may not represent those who started. A completion-count calculation cannot establish that missingness is harmless.
- Treating an expected count as an observed count. The model gives an average over possible study outcomes, not a promise about the realized number of completers.
Check Your Understanding
Answer each question using the stated assumptions. Solutions are provided after the questions.
- A three-period schedule has one planned dose per period. How many dosing occasions and between-period transitions does each participant face?
- For a four-period schedule, assume a 20% conditional withdrawal probability at every transition. What proportion is expected to complete all periods?
- In a hypothetical two-period study, 150 participants start period 1 and the conditional withdrawal probability before period 2 is 8%. What is the expected number completing both periods?
- Why might the number of periods understate a schedule’s total visit burden?
- What is one reason that an accurate expected full-completion count may still be insufficient to assess the impact of attrition?
Solutions. (1) There are 3 dosing occasions and \(3-1=2\) between-period transitions. (2) There are 3 transitions, so the completion proportion is \((1-0.20)^3=(0.80)^3=0.512\), or 51.2%. (3) The completion probability is \(1-0.08=0.92\), so the expected count is \(150(0.92)=138\) participants. (4) Screening, final follow-up, extra washout contacts, or additional sampling visits may add visits beyond one per period. (5) The count does not show whether withdrawal is related to outcomes or participant characteristics; such a relationship could make the observed participants systematically different from those who withdraw.