Estimating Causal Treatment Effects in the Sequential Parallel Comparison Design (SPCD).

Area of Science:

• Clinical Trials Methodology
• Psychiatric Research
• Biostatistics

Background:

• High placebo response rates in clinical trials, especially in psychiatry, impede treatment efficacy assessment.
• The Sequential Parallel Comparison Design (SPCD) was proposed to mitigate placebo effects by pooling treatment effects.
• Causal interpretation of the SPCD pooled treatment effect (Δ_SPCD) remains challenging.

Purpose of the Study:

• To explore the causal interpretation of the pooled SPCD treatment effect.
• To contrast Δ_SPCD with two causal estimators: average treatment effect among non-responders (Δ_NR) and a novel joint estimator (Δ_Joint).
• To evaluate bias and performance of these estimators via simulation and reanalysis of the ADAPT-A trial.

Main Methods:

• Simulation studies were conducted to compare bias and mean squared error (MSE) of Δ_SPCD, Δ_NR, and Δ_Joint.
• The G-formula approach was utilized for causal estimation.
• The ADAPT-A trial, the first using the SPCD design, was reanalyzed.

Main Results:

• Δ_SPCD tended to underestimate treatment benefits compared to both causal estimators in most simulated scenarios.
• The novel Δ_Joint estimator demonstrated statistically superior performance over Δ_NR and Δ_SPCD in terms of bias and MSE.
• Simulation findings were corroborated by the reanalysis of the ADAPT-A trial.

Conclusions:

• Interpreting Δ_SPCD as a causal effect can be misleading due to potential underestimation of treatment benefits.
• The Δ_Joint estimator offers a more accurate and statistically robust approach for analyzing SPCD trials.
• Adoption of the Δ_Joint estimator is recommended for applicable studies to improve causal inference in clinical trial reviews.