Mats Stensrud | Why reporting hazard ratios in medical research is a bad idea

Описание: Hazard ratios are routinely reported as effect measures in clinical trials and observational studies. However, many methodological works have raised concerns about the interpretation of hazard ratios as causal effects. These concerns are often related to three points: (i) depletion of susceptible individuals leads to selection bias and complicates the causal interpretation of the hazard ratio, (ii) the hazard ratio is not collapsible, and (iii) the conventional proportional hazards assumption rarely holds in medical studies. We discuss the relation between these three points. We ground our presentation on an example about effect of endocrine therapy in reducing the risk of recurrence or death in a population of patients with breast cancer. We also describe why survival curves and risk differences do not exhibit any of the undesirable properties of hazard ratios.

About the speaker: Mats Stensrud is an MD and associate professor at the Institute of Mathematics, EPFL, where he holds the Chair of Biostatistics (since 2020) and serves as director of the doctoral program in mathematics (since 2025). He develops causal and statistical methodology with two main goals: make the target of inference scientifically meaningful, and keep assumptions transparent and testable. He is particularly interested in settings where exposures and outcomes evolve over time (longitudinal and time-to-event data). Many of his works are inspired by applications in clinical medicine and epidemiology. Before he came to EPFL, Stensrud worked with Miguel Hernán and Jamie Robins at Harvard School of Public Health as a Kolokotrones Research Fellow and Fulbright Research Scholar. Before he became a full time academic, he had a short career as resident doctor in internal medicine. He received his MD, Dr.Philos in Neuroscience and BSc in Mathematics from the University of Oslo. He also holds a Msc in Statistics from the University of Oxford.

About the Causal Inference Interest Group: https://cls-data.github.io/CIIG/