9. Crossover, Non-Inferiority, and Adaptive Designs Explained
After this lesson you can
- Explain why a crossover needs fewer participants and when carryover rules it out
- State what an n-of-1 trial can and cannot conclude
- Read non-inferior, inconclusive and inferior verdicts from an interval and a margin
- Explain why conduct and per-protocol analysis matter in non-inferiority trials
- Name four adaptive designs and the pre-specification rule they share
Video coming soon
Lessons publish daily from November 17, 2026.
This lesson covers the two-period crossover and the carryover that forbids it, the n-of-1 trial, and non-inferiority and equivalence designs read against a margin. It closes with four kinds of adaptive design and the one rule they all share.
Questions and answers
Questions a learner would ask about this lesson. Think through your own answer before you open each one.
- Why does a crossover need so many fewer participants than a parallel trial?
Show answer
Each participant receives every treatment in sequence and is compared with themselves, so between-person variation drops out. In the fictional inhaler side case, a parallel trial needs 100 per arm, 200 in total, while the crossover needs 7.84 times 256 over 64, about 31.4, so 32 participants. Each person contributes a difference, not a single score.
- When is a crossover not allowed?
Show answer
When carryover is likely, meaning the first treatment's effect persists into the second period. A crossover fits a stable chronic condition and a reversible treatment. It is ruled out for a condition that changes over the trial, a treatment that cures or teaches permanently, or outcomes that happen once, like a flare or death. Carryover must be prevented by design, because the test for it after the fact has little power.
- What can an n-of-1 trial conclude?
Show answer
An n-of-1 trial repeats randomized crossovers inside one participant, so it can conclude which treatment works better for that one person. It concludes nothing about other people, about treatments that cure or act slowly, or about rare or one-time outcomes.
- How do you read a non-inferiority result?
Show answer
Compute the two-sided interval for new minus standard and compare it with the margin set before the trial. With a ten-point margin, an interval of minus 8 to plus 4 is non-inferior because its lower bound is above minus 10. Minus 12 to plus 2 crosses the margin and includes zero, so it is inconclusive; minus 14 to minus 6 lies wholly below zero, so it is inferior.
- What rule do all adaptive designs share?
Show answer
Every adaptation is specified before the trial starts, with its trigger, timing and limits written down, so the error rate is protected by the plan. The lesson names group sequential designs, sample-size re-estimation, response-adaptive randomization, and platform and basket trials. Adaptive never means changing the question or the primary endpoint after seeing results.
Practice quiz
- Question 1The inhaler crossover uses a within-person standard deviation of differences of 16 and seeks a difference of 8. With the multiplier 7.84, about how many participants does it need?
- Question 2Why does a crossover use two sequences, A B and B A, rather than one?
- Question 3Carryover means:
- Question 4Which of these could be studied by a crossover?
- Question 5With a margin of ten points, an interval of minus 8 to plus 4 for shorter minus standard means the shorter regimen is:
- Question 6An interval of minus 12 to plus 2 against a ten-point margin is:
- Question 7In a non-inferiority trial, why does sloppy conduct such as non-adherence and missing outcomes matter so much?
- Question 8A superiority trial finds "no significant difference." What can be concluded?
- Question 9The rule every adaptive design shares is:
- Question 10Why did the Oakhaven unit reject a crossover for the main trial?
Saved in your browser only: no account, no server.
Design of Clinical Studies is education only. Every lesson follows the fictional Oakhaven Adherence Trial with invented numbers; the course uses no real data and cites no real studies.