5O-6 USING BAYESIAN EVIDENCE SYNTHESIS TO ESTIMATE FRACTURE RISK ASSOCIATED WITH HORMONAL THERAPY IN EARLY BREAST CANCER

Wednesday, October 21, 2015: 11:15 AM
Grand Ballroom C (Hyatt Regency St. Louis at the Arch)

Ava John-Baptiste, PhD1, Taryn Becker, MD, MSc, FRCPC2, Hadas Fischer, MSc2, Kinwah Fung, MSc2, Lorraine Lipscombe, MD, MSc, FRCPC2, Peter C. Austin, PhD3 and Geoffrey Anderson, MD, MSC, PhD4, (1)Western University, London, ON, Canada, (2)Women's College Research Institute, Toronto, ON, Canada, (3)Institute for Clinical Evaluative Sciences, Toronto, ON, Canada, (4)University of Toronto, Toronto, ON, Canada
Purpose: Randomized controlled trials (RCTs) provide information on drug efficacy prior to licensing. Administrative databases serve important roles in identifying and monitoring adverse event risks post-market. For postmenopausal women with early breast cancer, aromatase inhibitors (anastrazole, letrozole or exemestane) reduce the risk of breast cancer recurrence compared to Tamoxifen, but may increase fracture risk. We incorporated prior information on the risk of hip, spine and wrist fracture derived from pre-market RCTs into post-market analyses using administrative data. 

Method: We conducted a retrospective cohort study of women age 66 or older diagnosed with early breast cancer, in Ontario, Canada from 2003 to 2010, who initiated treatment with an aromatase inhibitor (AI) or Tamoxifen (Tam). Prior information from meta-analyses of pre-market RCTs indicated an increased risk of hip, spine and wrist fracture with a probability that the relative risk (RR) exceeded 1 of 92% (RR=1.73, 95% Credible Interval (CrI) 0.81, 3.68), 98% (RR= 1.46, 95%CrI 1.02, 2.09) and 94% (RR= 1.48, 95%CrI 0.9, 2.44), respectively. Using Cox proportional hazard regression, we modeled the hazard of fracture occurrence, estimating the relative hazard ratio (HR) for AI compared to Tam, adjusting for age, fracture history, corticosteroid use, rheumatoid arthritis, dementia and diabetes. Using data augmentation, we incorporated both uninformative and informative priors into the analyses. We estimated posterior probabilities that the HR exceeded 1, assuming a normal distribution for regression coefficients.

Result: A smaller proportion of women who initiated treatment on AI (n=6,526) experienced fractures (hip: 2.4%, spine: 1.7%, wrist: 3.8%) compared to women initiated on Tam (n=3,733, hip: 3.8%, spine: 2.3%, wrist: 4.1%) but the mean time to fracture was shorter on AI. After risk adjustment, posterior probabilities that the HR exceeded 1 for AI compared to Tam were 46% (HR=0.99, 95%CrI 0.71, 1.25), 35% (HR=0.94, 95%CrI 0.78, 1.26) and 76% (HR=1.08, 95%CrI 0.88, 1.32) with an uninformative prior, and 63% (HR=1.04, 95%CrI 0.83, 1.3), 84% (HR=1.12, 95%CrI 0.89, 1.4) and 89% (HR=1.13, 95%CrI 0.93, 1.36) with an informative prior, for hip, spine and wrist fracture, respectively.

Conclusion: In our study, informative prior information derived from RCTs increased posterior probabilities of greater fracture risk for AI compared to Tam. This approach incorporated safety data from a range of sources and can be implemented in commonly used statistical software.

See more of: ORAL ABSTRACTS: APPLICATIONS IN HEALTH SERVICES RESEARCH

See more of: 37th Annual Meeting of the Society for Medical Decision Making