Online Program

Combination of survival curves from orthopedic registries

*Samprit Banerjee, Weill Medical College, Cornell University 
Guy Cafri, Kaiser Permanente 
Liz Paxton, Kaiser Permanente 
Art Sedrakyan, Weill Medical College, Cornell University 

Keywords: meta-analysis, survival analysis, random-effects model, orthopedics

Combining data from various distributed data sources has received a lot of attention in the past few decades particularly in the health care sector where individual data sources are smaller in size and it is useful to combine information from related but independent data sources to conduct inference. Combining summary information from independent studies provides increased power to detect overall treatment effects, estimates the degree of benefit from a particular treatment and assesses the amount of variability between studies. Our motivating example is a study of comparative effectiveness of medical devices used in hip-replacement and knee-replacement surgeries. The FDA initiative Medical Devices Epidemiology Network (MDEpiNet) including the International Consortium of Orthopedic Registries (ICOR) is being developed by world-wide registries and teams of researchers from the Science and Infrastructure Center (Cornell WCMC in collaboration with Kaiser Permanente). In this study, time to revision surgery, post hip-replacement surgery, is being examined as a survival outcome and data from 8+ US and international registries is being aggregated to conduct inference. Weighted average of hazard ratios is the most common method for meta-analysis of survival outcomes. However, such methods are unable to combine survival curves from multiple registries/studies and hence unable to conduct inference on risk progression over time. Here, we present a multivariate random-effects model to combine reported survival probabilities at multiple time-points for various covariates.