Top Tops - Efficient Study Designs

 

Efficiency might relate to statistical methodology (e.g. small sample size or a using single control arm for various intervention arms in a single trial) and/or to financial and organisational structure (e.g. conducting two similar studies at a time using the same management and protocol).

Cross-over, factorial and stepped-wedge designs

Trials that try to reduce bias of time effects, interactions and testing new interventions

Cross-over trials 
This is like a standard two arm trials (control : intervention) but after the treatment arm stops receiving the intervention the control arm then receives it.  So everyone gets the intervention in the end. 

• big assumption is: when treatment is withdrawn the disease state reverts to a constant level (possibly following a washout period)
• type of RCT where each participant receives each treatment
• randomisation determines ordering of treatments
• has some benefits compared to parallel type (e.g. estimates treatment effect within each patient, usually requires fewer patients, and may be more attractive to patients)
• major requirement: patients need to be in a similar state prior to starting each treatment period
• design issues to consider include period effect, carryover effect, washout periods
• pitfalls: more opportunity for placebo response, induction periods and no assessment of long term outcomes

Factorial trials 
This is like a standard two arm trial (control : intervention) but there is more than one treatment arm (multiple arm).

• big assumption is: no interaction between the interventions
• looks to assess two (or more) interventions in one trial
• sample size – separate calculation for each intervention and take the largest
• pitfalls: interactions pose a huge problem. The approach can investigate the interaction but then sample sizes can get very large. It may be better to design as multi-arm instead if interaction is a strong possibility.

Stepped-wedge design 
This is where a cluster (group of participants) are randomised to intervention (from a control state) at a different time for each cluster.

• a cross-forward design, clusters randomised to sequences
• three main types of participation: exposure and measurement
  • closed cohort (schools, trial in one school year)
  • open cohort (care home, community)
  • continuously recruited new individuals (immediate care for patients arriving at hospital emergency department)
• some assumptions are made (which might not be realistic)

Platform trial designs

Trials that try to address several research questions under one administrative trial structure

Multi-arm multi-stage trials (MAMS)
This is like the factorial design (multi arm) but after a pre-specified time period, arms can be changed to other treatments/interventions that have been found to be more effective.  

• test many relevant approaches
• use fewer resources (two trials for the price of 1½)
• are cheaper per comparison and have less central bureaucracy
• use interim “lack of benefit” analyses
• ask if there are reasons to continue to investigate an approach

Umbrella trial designs 
These use multi treatments as multi-arms, the treatments are selected based on biomarkers, where everyone who has a certain biomarker would be on the same treatment

• single trial protocol
• single disease area
• stratified subgroups – defined by predictive/ prognostic factor, e.g. Biomarker testing
• multiple new treatment options to test – potentially targeted at specific subgroups
• adaptiveness is a key part of design
• multi-stage approach as per MAMS design
• opportunities to offer patients to entry to the comparison that offers the most promising new treatment

Basket designs 
This is a single intervention but where the effects are investigated across different diseases

• single trial protocol
• single treatment to be tested
• multiple disease areas, with shared treatment pathway
• separate comparisons for each disease area
• multi-stage approach as per MAMS design
• need for a baseline before the intervention

Author: Shaun Barber      Created: March 2022    Last Updated: December 2022