28 August 2020 | 12:00 UTC / GMT
Time Zone Converter
Randomized controlled trials are arguably the gold standard approach for estimating causal effects of treatments or other interventions on outcomes in health and other fields. However, randomized trials are typically restricted to relatively short follow-up time and a subset of the eventual treatment population, and are infeasible for some types of intervention. Observational data which record longitudinal information on interventions and subsequent outcomes bring the opportunity to estimate effects of interventions in real world practice, in large and diverse populations, and with long-term follow-up. However, to estimate intervention effects from observational data we must tackle the challenge of confounding, especially by time-dependent covariates.
Recent years have seen great advances in the statistical methodology and thinking that enables estimation of causal effects from observational data. The developments also have implications for influencing the way that randomized trials are analysed. Applications have, however, lagged behind the methodological developments, but are increasing as more researchers are attracted by the possibilities for new understanding of causal mechanisms. This session will showcase the use of causal inference methods in practical applications. It will especially focus on survival or other time-to-event outcomes because of their importance in biometrical research, and the specific methodological challenges that they bring.
Three talks will highlight the challenges faced in applying and methods to different data sources, including electronic health records, patient registries, and in a randomized trial. Applications will include studies of work participation, organ transplantation, and breast cancer. There will be an emphasis on the suitability of different methods to address the research questions using the data at hand and the talks will cover comparison of different approaches, including marginal structural models, g-estimation, use of target or emulated trials, and sequential Cox regression.
Stijn Vansteelandt, Ghent University and London School of Hygiene & Tropical Medicine
Jon Michael Gran, University of Oslo
Ruth Keogh, London School of Hygiene & Tropical Medicine
Tomohiro Shinozaki, University of Tokyo
Jessica Young, Harvard Medical School