Mini Oral Australian Epidemiology Association ASM 2018

Statistical methods for estimating legacy effects: A simulation study (#62)

Lin Zhu 1 , Katy Bell 2 , Andrew Hayen 1
  1. Faculty of Health, University of Technology Sydney, Sydney, New South Wales, Australia
  2. School of Public Health, The University of Sydney, Sydney, New South Wales, Australia

Background: There is growing interest in possible legacy-effects of drugs, but methods of analysis are underexplored.

Methods: A simulation study was conducted to assess three methods of estimating legacy effects, which differed in terms of the selection of participants and period of data analyzed:

  1. All trial participants. Data from start of the RCT to end of post-RCT follow-up used.
  2. Participants surviving post-RCT and who were followed up. Post-RCT data used.
  3. Participants surviving post-RCT, who were followed up, and who took the drug post-RCT. Post-RCT data used.

Independent datasets were generated for scenarios where there was, and alternatively was not, a legacy effect for the drug. We estimated the legacy effect using the intention-to-treat principle (analysis according to randomized group) and Cox proportional hazard models. The three methods’ performances were compared in term of bias, mean square error, coverage and power.

Results: Under the condition that the legacy effect of taking the drug during the RCT was the same irrespective of whether or not the drug was taken post-RCT, estimations using post-RCT data had best performance (methods 2 and 3). When the size of the legacy effect differed according to whether or not drugs were taken post-RCT, estimations using post-RCT data for participants taking the drug post-RCT had best performance (method 3).

Conclusion: The most common method for estimating legacy effects, which combines initial trial and post-trial follow-up data had the worst performance. A better approach may be to use post-trial follow-up data and adjustment for post-trial drug use.