Novel methods for the analysis of stepped wedge cluster randomized trials

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Stepped wedge cluster randomized trials (SW-CRTs) are increasingly utilized due to feasibility but require accounting for time trends.
• Existing nonparametric methods for SW-CRTs lack the power of model-based approaches.
• Model-based approaches may be unreliable if their assumptions are misspecified.

Purpose of the Study:

• To develop novel analysis methods for SW-CRTs that balance robustness and statistical power.
• To reduce reliance on strong parametric modeling assumptions while retaining power.
• To offer alternatives to existing methods when standard assumptions may not hold.

Main Methods:

• Proposed novel methods include synthetic control and within-cluster crossover approaches.
• Combined methods were explored to further enhance statistical power.
• Methods were evaluated using simulated SW-CRTs to assess power and validity.

Main Results:

• The proposed methods demonstrated increased power compared to existing nonparametric approaches.
• The methods preserved nominal validity even when mixed-effects models were misspecified.
• Theoretical properties were demonstrated with less restrictive assumptions than traditional mixed-effects models.

Conclusions:

• Novel analysis methods offer a robust and powerful alternative for SW-CRTs, especially when standard modeling assumptions are questionable.
• These methods provide flexibility for identifying specific causal contrasts and can incorporate covariates to boost power.
• Investigators should consider these methods for their robustness and potential to improve SW-CRT analysis.