The following approaches can be considered to reduce the estimated sample size in clinical trials:

• Objective
  • Choose another objective, e.g.
    • Reduce objectives of study, e.g. phase 2 instead of phase 3, fewer primary/key secondary endpoints, descriptive instead of inferential
    • Change target concept, e.g. physical instead of psychological QOL
  • Operating characteristics:
    • Relax type 1 error
    • Relax type 2 error
    • Perform literature review and if expected treatment is uncertain use most optimistic parameter values (risky!)
• Design
  • Select a more homogeneous target population (ie lower measurement error)
  • Increase treatment effect, e.g.
    • higher dose of test drug
    • control with less effect (e.g., placebo instead of active control)
  • Extend observation period (assuming increasing effect over time)
  • Reuse participants, e.g. cross-over design
  • Add repeated measurements, ie add more visits
  • Apply group-sequential designs
  • Apply sample size reestimation
  • Relax definition of protocol deviations
  • Select endpoint with lower measurement error
    • Replacing measure with another with higher reliability/sensitivity to change, e.g. introduce more levels on response scale, VAS instead of rating scale
    • Aggregating multiple measure into an endpoint, e.g. multi-item summary score, composite endpoint, multivariate endpoint (cf MANOVA, Principal Component)
    • Continuous endpoint instead of binary (responder) endpoint
• Statistical analysis
  • Use parametric instead of non-parametric model
  • Use most distribution with best goodness-of-fit
  • Reduce parameters (df) in statistical model
  • Impute missing values (ie assume all patients are completers after imputation in sample size estimation, ie no dropout)
  • Repeated measures models (MMRM)
  • Include covariates (unexplained variation -> explained variation), e.g.
    • Standard (demographics, patient characteristics, disease characteristics)
    • Endpoint at baseline
    • Prognostic variables of placebo effect (cf Cognivia)
    • Prognostic variables of missingness
  • Select multiple testing procedure to optimize power
  • Bayesian analysis using external data (RWE) or expert elicitation