2.3. SELECTION OF STUDY DESIGN
When an E-R analysis using an adjustment model concludes a positive E-R relationship it is difficult to discern whether there is a true positive E-R relationship, or there are additional hidden confounders. The only approach that allows for certainty in a positive E-R relationship is to study multiple dose levels of the drug in large, registrational trials. These trials should have randomized groups with balanced baseline characteristics. Studying multiple dose levels allows for the identification of a true E-R relationship because the dose-response relationship is not confounded by the prognostic factors that confound the E-R relationship. ncology phase 3 trials typically only study a single dose level, and the exposure range included in the E-R analysis for OS is limited. Including additional arms or performing separate trials to increase this exposure range allows for a more robust E-R analysis. In the previously described case of trastuzumab, the HELOISE trial studying high dose and standard dose trastuzumab revealed that the case-matching analysis conducted for the ToGA study was confounded, and that there appeared to be no causal relationship between exposure and response for trastuzumab in metastatic gastric cancer.2, 3 A similar scenario was observed with pembrolizumab in melanoma and NSCLC. Pembrolizumab was studied across a 5-fold dose range. Two case-matching analyses were performed for patients receiving 2 mg/kg and 10 mg/kg dose levels, respectively. In the unmatched analysis a steep E-R relationship was observed across exposure ranges within each dose level. While case-matching analysis corrected this E-R relationship to a certain degree it still suggested a positive E-R relationship. When examining hazard ratios across the two dose levels, however, the apparent E-R relationship was flat and suggested that higher exposures do not increase OS. The case-matching analysis was unable to fully account for confounders.40
Because dose-response relationships are not confounded by the prognostic factors that confound E-R relationships studying multiple dose levels is a robust approach to examine E-R relationships. Unlike statistical approaches discussed in the previous sections it requires no assumptions about covariates or the structure of the E-R relationship. The major limitation with this approach is the time and cost associated with additional trials or treatment arms. In addition, this approach may not be feasible in rare populations. The utility of studying multiple dose levels may also depend on characteristics of the drug. If a drug has a wide therapeutic window (i.e. monoclonal antibodies), and tested doses appear to be at the top of the dose-response curve studying multiple dose levels in registrational trials may not be necessary. It may be useful if a drug has a narrow therapeutic window and requires quantification of E-R relationships for optimal dose selection. Ultimately, limitations in feasibility motivate sponsors to consider alternative approaches to E-R analyses before studying multiple dose levels.