A crossover trial is a clinical trial in which participants receive two or more interventions in successive treatment periods, allowing treatment effects to be compared within the same individuals. In a randomized crossover trial, participants are randomly assigned to treatment sequences rather than to a single treatment throughout the study. This design can provide precise comparisons with fewer participants than a parallel-group trial, but its validity depends on the condition remaining sufficiently stable and on earlier treatments not materially affecting outcomes in later periods.
Basic design
The simplest design is a two-treatment, two-period crossover trial. Participants are assigned to one of two sequences:
| Sequence | First treatment period | Second treatment period |
|---|---|---|
| AB | Treatment A | Treatment B |
| BA | Treatment B | Treatment A |
A treatment may be a drug, device, dietary intervention, or other intervention; the comparator may be another active treatment or a placebo. A washout interval may separate treatment periods to allow effects of the previous intervention to diminish.
Unlike a parallel-group randomized controlled trial, in which different groups receive different interventions, a crossover trial obtains observations under multiple treatments from each participant. Stable individual characteristics—such as genetic background or longstanding disease severity—therefore contribute less to variation in the treatment comparison.
Randomization of sequence helps separate treatment effects from changes associated with the passage of time. Allocation concealment protects sequence assignment from foreknowledge during recruitment, while blinding can reduce bias in treatment delivery, participant expectations, and outcome assessment.
Crossover trials are distinct from trials in which participants switch treatments after disease progression or gain access to an experimental treatment after a primary comparison. Such switching does not, by itself, constitute a planned crossover design.
Conditions for appropriate use
Crossover designs are most suitable when:
- The underlying condition is relatively stable over the study period.
- Treatment effects develop within a practical observation period.
- Effects are reversible or diminish sufficiently before the next assessment.
- The same outcome can meaningfully be measured repeatedly.
- Receiving one intervention does not permanently alter the response to another.
They are used in studies of symptom control, physiological responses, and short-term drug effects, including some research involving asthma, hypertension, and pain. They also have an important role in bioavailability and bioequivalence studies, where within-person comparisons can reduce the influence of differences in drug absorption and elimination between participants.
The design is generally unsuitable for curative interventions, irreversible procedures such as many forms of surgery, or outcomes that cannot recur in a comparable way. Rapidly progressive disease also makes interpretation difficult because participants may differ substantially between treatment periods.
Carryover, period, and sequence effects
Carryover effects
A carryover effect occurs when an intervention received in an earlier period continues to influence a later period. Residual drug concentrations are one cause, but biological changes, learned behaviors, or persistent symptom improvement can also produce carryover.
The washout interval is chosen using knowledge of pharmacokinetics, pharmacodynamics, and the outcome being measured. Elimination from the bloodstream does not necessarily mean that all biological effects have ended. Conversely, a lengthy washout can increase study duration, withdrawal, or time without effective treatment.
A small or statistically nonsignificant test for carryover does not establish that carryover is absent. Its prevention depends primarily on the suitability of the intervention and the design, rather than on a preliminary statistical test.
Period effects
A period effect is a change associated with the timing of observation rather than the assigned treatment. Examples include seasonal variation, adaptation to study procedures, or a general improvement in symptoms over time. Balanced assignment to AB and BA sequences helps distinguish an overall period effect from a treatment effect.
Sequence effects
A sequence effect is a difference associated with the order in which treatments are received. Such differences may reflect carryover, chance differences between sequence groups, or other order-related influences. In a simple two-period trial, some potential explanations cannot be cleanly distinguished from one another using the observed data alone.
Statistical analysis
Crossover analysis accounts for the fact that repeated observations from one participant are related. Treating all measurements as independent observations generally gives an inappropriate estimate of uncertainty.
For a continuous outcome, a common model is:
where is the outcome for participant in period , represents participant-specific differences, indicates treatment, and indicates period. The coefficient estimates the treatment contrast under the model’s assumptions. Mixed-effects models can accommodate more complex designs and additional sources of variation.
The efficiency advantage depends largely on within-person variability. When an individual's measurements under different treatments are strongly correlated, a within-person comparison may have substantially lower variance than a comparison between independent groups. Sample-size calculations therefore require assumptions about within-person variation, the target treatment difference, and anticipated missing data.
Missing observations are particularly consequential when a participant completes only one treatment period. Analysis restricted to participants completing every period can introduce bias, especially when withdrawal is related to treatment response or adverse effects. Intention-to-treat principles remain relevant, but do not by themselves resolve missing-data problems or carryover.
Variants and reporting
Trials may include more than two treatments, additional periods, or repeated exposure to the same treatments. Replicate crossover designs are used in some bioequivalence studies to characterize within-person variability. An N-of-1 trial applies repeated treatment comparisons to a single individual; a conventional crossover trial usually estimates an average effect across multiple participants.
Reporting guidance associated with the CONSORT statement addresses randomized crossover trials. Important reporting elements include the treatment sequences, randomization methods, period lengths, washout rationale, timing of measurements, and statistical handling of within-person comparisons. Participant flow, withdrawals, and adverse events need to be described by period and treatment where appropriate.
The principal limitations are vulnerability to carryover, longer participation, and difficulty separating treatment effects from changes over time. As in other research involving human participants, ethical review must consider the burdens and risks of treatment withdrawal, washout, repeated exposure, and delayed access to effective care.