Statistical Calculators › Means: Correlated, Paired, and Cross-Over Designs › Higher-Order Crossover Design Sample Size Calculator
← All Calculators

Means: Correlated, Paired, and Cross-Over Designs

Higher-Order Crossover Design Sample Size Calculator

Calculates the exact total sample size required for a two-treatment, higher-order crossover design to achieve a specified power when testing the difference between two means. Based on the this method methodology of Chen, Chow & Li (1997).

Study Design & Assumptions

Subjects are allocated equally across the treatment sequences. The calculation searches over the allowable per-sequence sample size.
Specify variance parameters

Sample Size Result

The calculator searches for the smallest equally allocated design satisfying the requested power.
Enter the study assumptions and click Calculate Sample Size.

Methodology

This calculator implements the sample-size and power methodology described by the software/this method for tests of the difference between two means in higher-order crossover designs. The method covers four two-treatment designs: Balaam's design, the two-sequence dual design, the four-period/two-sequence design, and the four-period/four-sequence design.

Power calculation

Let n be the average number of subjects assigned to each sequence, σw the within-subject standard deviation, and Δ = |μ1 − μ2| the absolute mean difference. The noncentrality quantity used by this method is

z = Δ / [σw √(b / n)]

For a two-sided test, the power is calculated as the probability that the corresponding t statistic falls beyond either critical value:

Power = TV(z − tV,1−α/2) + TV(−z − tV,1−α/2)

For a one-sided test, the corresponding expression is:

Power = TV(z − tV,1−&alpha)

Here TV is the cumulative Student's t distribution with V degrees of freedom. The design-specific values of V, b, and the number of sequences are:

Balaam: V = 4n − 3, b = 2, sequences = 4
Two-Sequence Dual: V = 4n − 4, b = 3/4, sequences = 2
Four-Period / Two-Sequence: V = 6n − 5, b = 11/20, sequences = 2
Four-Period / Four-Sequence: V = 12n − 5, b = 1/4, sequences = 4

The total evaluable sample size is N = n × number of sequences. The calculator increments the number of subjects per sequence until the requested power is reached, then reports the corresponding total sample size.

Estimating the within-subject standard deviation

this method permits the within-subject standard deviation to be supplied directly, or calculated from a between-subject standard deviation and within-subject correlation. When those latter quantities are selected, the calculator uses:

σw2 = σb2(1 − ρ)
σw = σb√(1 − ρ)

Dropout inflation

If an expected dropout rate is entered, the evaluable sample size is inflated according to the this method convention:

N′ = ⌈ N / (1 − DR) ⌉

where DR is the anticipated dropout proportion. The dropout-inflated number is an enrollment target; the underlying power calculation remains based on the evaluable sample size N.

Worked validation example

This worked example uses a three-period, two-sequence dual design (ABB | BAA), a two-sided test with α = 0.05, target power of 0.90, an absolute mean difference of 14, and a within-subject standard deviation of 25. This yields an exact total sample size of 52, corresponding to 26 subjects per sequence, with achieved power 0.9039.

Design = ABB | BAA
α = 0.05
Power target = 0.90
1 − μ2| = 14
σw = 25
n per sequence = 26
Total N = 52
Achieved power = 0.9039

The implementation reproduces this validation result: with n = 26, V = 4(26) − 4 = 100 and b = 3/4, the calculated two-sided power is approximately 0.90394, which rounds to the this method value of 0.9039.

References

Chen, K.W., Chow, S.C., & Li, G. (1997). A Note on Sample Size Determination for Bioequivalence Studies with Higher-Order Crossover Designs. Journal of Pharmacokinetics and Biopharmaceutics, 25(6), 753–765.

Chow, S.C., Shao, J., & Wang, H. (2003). Sample Size Calculations in Clinical Research. Marcel Dekker, New York.

Chow, S.C. & Liu, J.P. (1999). Design and Analysis of Bioavailability and Bioequivalence Studies. Marcel Dekker, New York.

the software, LLC. this method Sample Size Software: Tests for the Difference of Two Means in a Higher-Order Cross-Over Design, Chapter 527.