Table 1. Effect of treatment-related confounding with or without predictive interaction on sample size for an adequately powered study.
| Predictive interaction | Subgroup | Treatment confounding | |||
|---|---|---|---|---|---|
| Noa | Yesb | ||||
| Biomarker + | Biomarker − | Biomarker + | Biomarker − | ||
| No | Proportion of patients | 30% | 70% | 30% | 70% |
| Proportion receiving adjuvant treatment | 80% | 80% | 90% | 60% | |
| Treatment efficacy (HR) | 0.6 | 0.6 | 0.6 | 0.6 | |
| Event rate (baseline) | 31% | 15% | 31% | 15% | |
| Event rate (post adjuvant treatment) | 21% | 10% | 20% | 12% | |
| Sample size needed | ~ 550 | ~ 1075 (2-fold) | |||
| Yes | Proportion of patients | 30% | 70% | 30% | 70% |
| Proportion receiving adjuvant treatment | 80% | 80% | 90% | 60% | |
| Treatment efficacy (HR) | 0.2 | 0.5 | 0.2 | 0.5 | |
| Event rate (baseline) | 31% | 15% | 31% | 15% | |
| Event rate (post adjuvant treatment) | 11% | 9% | 9%* | 11%* | |
| Sample size needed | ~ 11650 (21-fold) | ~ 11650 (21-fold) | |||
Modelling assumptions reported in this table are similar to UK/ANZ DCIS HER2 data and current use of adjuvant treatment in DCIS. BIOMARKER - Biomarker is expressed in 30% of samples and is associated with 2-fold increase in event risk. a No treatment confounding, for example, in a randomised trial OR when the biomarker (unlike HER2) is NOT associated with any features that influence adjuvant treatment selection (e.g. high grade, larger lesion size or necrosis which would normally lead to a greater use of adjuvant radiotherapy). b Treatment confounding present, for example, in a cohort study or single institution series when the biomarker (e.g. HER2) is associated with features that influence adjuvant treatment selection (e.g., high grade, larger lesion size or necrosis leading to a greater use of adjuvant radiotherapy in biomarker-positive subgroup even when biomarker status is not known). c Predictive interaction present (e.g. HER2), with a greater adjuvant treatment efficacy in biomarker-positive subgroup (HR = 0.2) as compared with biomarker-negative subgroup (HR = 0.5). * effect observed in the opposite direction of true effect.