Identifying subjects who are responsive to specific treatment is an essential requirement in advancing precision medicine. A companion diagnostic (CDx) device is often used to select recipients for drugs or medical devices that target on very specific receptive mechanisms. The task to simultaneously evaluate new therapeutic and CDx products creates additional challenges and constraints on clinical studies.
In the previous work, we reported that when presence or absence of receptor was determined using a CDx device in a therapeutic study, observed treatment effects were biased in the presence of false diagnoses and with treatment effect different in subjects with or without the receptor. When treatment effects are measured in differences, Y1-Y2 (R1) and Y3-Y4 (R0), for receptor positive and negative populations respectively, observed R1 = C(Y1-Y2)+(1-C)(Y3-Y4); where C = NPV/(NPV+PPV-1), as NPV and PPV represent positive and negative predictive values.
In this study, we investigated the biases measured in relative terms, i.e., proportions and hazard ratios, and their effects on respective hypothesis testing. Potential adjustments were experimented using simulated data.
|