Estimating the Hazard Rate Accounting for Bias in Safety Reporting in a Long-Term Safety Follow-Up Study

• Activity Number: 240
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Prasheen K. Agarwal*+ and Jiandong Lu
• Companies: Johnson & Johnson Pharmaceutical R&D, LLC and Johnson & Johnson Pharmaceutical R&D, LLC
• Keywords: hazard ratio ; reporting bias ; missing data

Abstract

Introduction: A phase 2 trial and a long-term safety follow-up (LTSFU) study in COPD subjects were conducted. The data beyond the primary study suffers from potential reporting bias due to spontaneous reporting during the gap (period between end of phase2 and start of LTSFU) and unblinding prior to the LTSFU study. Method: Assuming the time to onset of malignancy is exponentially distributed with true rate parameters ?pl and ?act for placebo and active groups, respectively, we model the effect of reporting bias by a random variable zi (?pl = ?pl* zi , ?act = ?act* zi) where, ?pl* and ?act* are observed rates. For the placebo group it is possible that ?pl > ?pl*, therefore, zi ~?Uni[d-0.05, d+0.05], where d>1. For the active group it is more likely an investigator would report the onset of a malignancy during the gap, therefore, zi ~ Uni[0.95, 1.05]. Simulated datasets with data from subjects with observed LTSFU data (107) and events generated for subjects without LTSFU (124) are used to obtain estimates for true hazard ratio. Results: With more than 15% under-reporting of events in the placebo group the estimated hazard ratio is less than 1.