Accounting for a decaying correlation structure in cluster randomized trials with continuous recruitment

Area of Science:

• Biostatistics
• Clinical Trial Design
• Epidemiology

Background:

• Sample size calculations for cluster randomized trials (CRTs) depend on within-cluster correlation structures.
• Existing correlation structures often make simplifying assumptions, such as uniform correlation.
• More complex structures exist but may not suit continuous recruitment and measurement.

Purpose of the Study:

• To propose a novel "continuous-time correlation decay" structure for within-cluster correlations in CRTs.
• To evaluate the impact of this new structure on sample size calculations for CRTs.
• To compare the proposed structure with the traditional uniform correlation assumption.

Main Methods:

• Derived the variance of the treatment effect estimator under the continuous-time correlation decay structure.
• Investigated various CRT designs, including stepped wedge and cluster randomized crossover.
• Compared the derived variances with those obtained under the uniform correlation assumption.

Main Results:

• The continuous-time correlation decay structure realistically models correlations that decrease with increasing time between measurements.
• Incorrectly assuming uniform correlation often leads to underestimation of the required sample size in CRTs.
• This underestimation is particularly likely in stepped wedge and cluster randomized crossover designs.

Conclusions:

• Accurate specification of within-cluster correlation structures is crucial for appropriate sample size calculation in CRTs.
• The proposed continuous-time correlation decay structure offers a more realistic approach for CRTs with continuous recruitment and measurement.
• Adopting realistic correlation structures ensures more reliable trial planning and adequate statistical power.