Cluster randomized trial designs for modeling time-varying intervention effects.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Health Services Research

Background:

• Stepped-wedge cluster randomized trials (SW-CRTs) often assume constant intervention effects, which can lead to bias if effects vary with exposure time.
• Time-on-intervention (TOI) effect estimation in SW-CRTs accounts for changing intervention effects but faces challenges due to heteroskedastic variances.

Purpose of the Study:

• To compare the efficiency of different cluster randomized trial (CRT) designs in estimating time-on-intervention (TOI) effects.
• To evaluate the performance of SW-CRT, parallel CRT (P-CRT), parallel CRT with baseline (PB-CRT), and parallel CRT with baseline and all-exposed period (PBAE-CRT) for TOI effect modeling.

Main Methods:

• Derivation and comparison of variance estimators for TOI effects across different CRT designs.
• Proof of identical time-averaged TOI effect variance and point estimators for P-CRTs and PB-CRTs compared to constant effect models.
• Simulation study using data from a hospital disinvestment study to compare TOI effect estimates.

Main Results:

• The SW-CRT demonstrated the most efficient estimator for early TOI effects.
• The PB-CRT generally provided the most efficient estimators for long-term and time-averaged TOI effects.
• Time-averaged TOI effect estimators were equivalent to constant effect models in P-CRTs and PB-CRTs.

Conclusions:

• The PB-CRT design, when incorporating TOI effects, is a more suitable choice for modeling intervention effects that change over time.
• Choosing the appropriate CRT design is crucial for accurately estimating time-varying intervention effects.
• The findings offer guidance for optimizing CRT design selection based on the expected nature of intervention effects.