C. Wu,1 C. D. Jin,1 L. Roskos,2 B. Wang1; 1AstraZenica/MedImmune, Mountain View, CA, 2AstraZenica/MedImmune, Gaithersburg, MD

BACKGROUND: Mavrilimumab, a human monoclonal antibody targeting the granulocyte-macrophage colony stimulating factor (GM-CSF) receptor α, is being investigated for the treatment of rheumatoid arthritis (RA). We describe the development of an informative dropout model with exposure-response analysis to account for voluntary subject dropout commonly seen in late-stage trials.
METHODS: Pharmacokinetic (PK) exposure and efficacy endpoint data from subjects with rheumatoid arthritis (RA) following multiple subcutaneous doses of placebo or mavrilimumab (30, 100 or 150 mg) once every other week were obtained from a double-blind Phase IIb study. A mechanistic PK model was first built, followed by an Emax model and a logit model to describe the observed DAS28 and ACR20 responses, respectively. A joint probability model was also developed to describe the risk of voluntary dropout at each scheduled visit. Clinical simulations using the final PK-efficacy-dropout model were subsequently performed to characterize the exposure-response relationship.
RESULTS: The hazard of patient dropout prior to the next scheduled visit was significantly higher in non-responders, and substantially elevated at Week 12 when patient rolling-over to an open-label extension study was permitted by the protocol. The simulated ACR20 response by carrying forward non-responder status following dropouts agreed well with the observed data. Clinical simulations suggested that the maximum efficacy was reached at 150 mg mavrilimumab dose.
CONCLUSION: The probabilities of treatment response and voluntary dropout were jointly modeled for a mavrilimumab Phase IIb study. The population model facilitated the interpretation of Phase IIb outcome and the appropriate design of late-stage clinical trials for mavrilimumab.