Meta-analysis using individual participant data (IPD) makes it possible to test additional hypotheses related to individual patient characteristics. However, analyzing IPD alone may introduce selection bias known as "data availability bias." Intergrating IPD with available study-level aggregate data (AD) can address this concern and improve efficiency. We propose a Bayesian approach that supports both fixed and random effects intergrated IPD-AD meta-analysis. Whether the data availability bias can be properly addressed depends on how study results influence the sharing of IPD and what information is available within the AD. We specify the conditions under which the resulting posterior is asymptotically normally distributed with the asymptotically consistent mean. Simulation studies showed that the proposed integrated meta-analysis led to the smaller mean squared errors than IPD-only analysis when IPD availability is not dependent on the data and smaller bias when it is.
Speaker: Mi-Ok Kim, PhD, Professor of Biostatistics, UCSF
To obtain Zoom link please email: [email protected]