Abstract:
|
Most dose-finding oncology trials aim to find the highest dose yielding acceptable patient toxicity. The conventional binary toxicity endpoint treats low to moderate grade toxicities as irrelevant. The concept of composite toxicity scores that take into account all patient toxicity information have been suggested, requiring prior specification of severity weights to represent the relative toxicity burden each toxicity contributes to a patient if observed. Elicitation of weights generally rely on subjective specification through trial and error. In a statistical framework, we propose the Toxicity Score Elicitation Method (TSEM), a novel method of estimating toxicity weights based on a cumulative logit model, assuming toxicity scores to be a latent variable characterized by the entire set of observed toxicities and grades, manifesting as ordinal outcomes corresponding to intuitive dose escalation decisions. The TSEM provides a tool which can help elicit relatively accurate and cohesive toxicity severity weights with acceptability thresholds to be used when designing a dose-finding trial. The TSEM provides the first statistical based method for toxicity weight and score elicitation.
|