Sample size calculation for clinical trials with correlated count measurements based on the negative binomial...

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• Biomedical studies often involve correlated count measurements.
• Existing sample size calculations typically rely on the Poisson model, which may not fit real-world data due to overdispersion.
• A need exists for more robust sample size methods accommodating deviations from Poisson assumptions.

Purpose of the Study:

• To develop and investigate sample size calculation methods for clinical trials with correlated count data.
• To utilize the negative binomial distribution to address overdispersion and other complexities in count data.
• To provide practical, closed-form formulas for sample size determination.

Main Methods:

• Employed the negative binomial distribution for sample size calculations with correlated count outcomes.
• Developed closed-form formulas for comparing slopes and time-averaged responses.
• Incorporated flexibility for overdispersion, unequal intervals, randomization ratios, missing data, and correlation structures.

Main Results:

• The proposed negative binomial-based method provides flexible and accurate sample size calculations.
• Closed-form formulas simplify implementation for comparing slopes and time-averaged responses.
• Extensive simulations confirmed the method's ability to maintain nominal power and type I error rates.

Conclusions:

• The negative binomial distribution offers a more realistic and flexible approach for sample size calculations in correlated count data compared to the Poisson model.
• The derived formulas are practical for designing clinical trials with complex data structures.
• The method was successfully illustrated using a real-world epileptic trial.