Why most responder analyses are misleading

Area of Science:

• Clinical Trials Methodology
• Osteoarthritis Research
• Pain Management Statistics

Background:

• Responder analyses are common in osteoarthritis randomized controlled trials (RCTs).
• These analyses typically rely on observed changes in self-reported pain from baseline.
• Methodological literature indicates these responder analyses can be misleading.

Purpose of the Study:

• To illustrate the extent of the problem with misleading responder analyses using simulation.
• To evaluate the impact of natural pain fluctuations on responder analysis outcomes.

Main Methods:

• Generated individual pain trajectories using realistic assumptions (normal distribution, mean VAS 45, SDs).
• Simulated data from RCTs with varying true treatment effects (0-15 VAS points) and responder proportions (0% or 100%).
• Applied standard responder analysis methods to the simulated trial data.

Main Results:

• Observed pain change from baseline does not equate to treatment response due to natural fluctuations.
• Even with a 15-point VAS treatment effect and 0% placebo response, analyses suggested 80% responders vs. 50% in placebo.
• This underestimates treatment efficacy and falsely implies heterogeneity in treatment effects.

Conclusions:

• Responder analyses based on change from baseline in Visual Analogue Scale (VAS) pain should be discontinued for parallel-group RCTs.
• Responder criteria using change from baseline for other fluctuating patient-reported outcomes (symptoms, function, global assessment) require scrutiny due to similar limitations.