Phase I–II trial design for biologic agents using conditional auto‐regressive models for toxicity and efficacy. Issue 2 (7th October 2018)

• Record Type: Journal Article
• Authors: Muenz, Daniel G.; Taylor, Jeremy M. G.; Braun, Thomas M.
• Is Part Of: Journal of the Royal Statistical Society. Volume 68:Issue 2(2019)
• Page Start: 331
• Page End: 345
• Publication Date: 2018-10-07
• Subjects: Biologically optimal dose -- Clinical trial -- Dose finding -- Immunotherapy -- Molecularly targeted agent -- Randomization; Statistics -- Periodicals; 519.5
• DOI: 10.1111/rssc.12314
• Languages: English
• ISSNs: 0035-9254

Abstract

Summary: A traditional assumption in the design of chemotherapy phase I–II trial designs is that dose increases lead to both more toxicity as well as more efficacy. This assumption of monotonic rates of toxicity and efficacy has come into question as potential cancer treatments are less likely to be chemotherapy and are instead biologic agents. These biologic agents tend to have mechanisms of action that act as 'on–off' switches for cancer growth, so giving more of the biologic agents will not necessarily provide any more benefit (and possibly no further risk) to the patient. We propose the use of a conditional auto‐regressive (CAR) model as a way to estimate adaptively the rates of dose limiting toxicities (DLTs) and efficacy by smoothing the data collected for all doses in such a way that allows for non‐increasing rates of either outcome with dose. We present the study design for our CAR model approach and compare, via simulation, the operating characteristics of our design with two existing contemporary published approaches. We demonstrate that our CAR model approach is a viable design for an adaptive phase I–II trial that can accommodate a variety of toxicity–dose and efficacy–dose patterns.