An Approach to Analyzing Multiple Correlated Adverse Events

• Activity Number: 559
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Zhibao Mi*+ and Dengfang Tang and Xiaoli Lu and Kousick Biswas and Joseph Collins
• Companies: VA Cooperative Studies Program and VA Cooperative Studies Program and VA Cooperative Studies Program and VA Cooperative Studies Program and VA Cooperative Studies Program
• Keywords: Adverse Effect ; Multiple correlated events ; Rate ratio ; Hazard ratio ; Competing risks

Abstract

Adverse effect assessment involves many clinical symptoms, and each individual may experience multiple adverse events (AE). A common practice for AE analysis that compares proportions of AE or the times to first event is simple and easy to implement, but it may lose data information. We propose using rate ratio (RR) calculate from AE incidence rates (IR) and hazard ratio (HR) to assess adverse effects for multiple correlated AE data. For analyzing event counts, AE specific IRs and the corresponding RRs are calculated and tested for statistical significance by computing exact probabilities, and the significance level is adjusted for multiple AE strata. The overall RR is tested using a counting process method. For analyzing times to multiple events, AE specific HRs are tested using the Cox model and overall HR is estimated and tested based on the Anderson-Gill and WLW marginal competing risk models. Multiple correlated adverse event data from a VA clinical trial was used as an example. From the analysis, HR estimated from the Anderson-Gill model was close to the overall RR. When compared with the frailty and conditional risk set models, the marginal model yielded similar HR.