OBJECTIVES: The issue of selection bias in the clinical trial has been identified and stated in the ICH Topic Choice of Control Group. However, as HTAs recommend the comparator to be standard of care (Soc), the issue of the selection is likely to play against the new product as ethically the physician is unlikely to enrol a patient who failed or did not tolerate the SoC. The objective was to estimate the impact of selection bias on efficacy results.

METHODS: We developed a simple decision tree model to estimate the impact of selection bias using the case of 2 hypothetical treatments A and B with the same average responder rate of 50% in a two-arm randomised trial. We assumed that for ethical reason previous non-responder or patients intolerant to treatment B would not be selected so only responder/patients tolerant to B will be included, and thus that previous responder will have a higher probability of response and lower probability of Drop-out (DO) due to an adverse event. We ran a different scenario with a different rate of market share (MS) for B.

RESULTS: The model shows that with previous MS of 50% for B, an additional 25% of response for previous responder and 10% less DO for B compared to A lead to a difference of 7.2% additional total responder to B (49.7% vs 42.5% p<0.05 with 400 patients per arm). When the MS is 20%, the difference remains at 2.9%, and with an MS of 70%, the difference is 10.1%.

CONCLUSIONS: This simple model shows that previous exposure to SoC may lead to substantial bias against a new product. HTAs should be made aware of this potential issue when the market share of SoC is large, several options are available and when previous response/tolerance is likely to predict future response/tolerance.