There is an increasing demand for authors of research publications to provide sufficient detail to enable readers to reproduce the reported results. When studies are reproducible they become building blocks for future research, for example by acting as tutorials for carrying out analyses and by providing reusable analytical code. In this presentation, we highlight ongoing efforts at the MIT Laboratory for Computational Physiology to work towards reproducibility in critical care research. We present several freely available critical care datasets shared by the laboratory, including the Medical Information Mart for Intensive Care (MIMIC), and we discuss our experiences of hosting international ‘datathons’. We also report on a recent study in which we attempt to reproduce the cohorts of 28 published mortality prediction studies that use MIMIC. We discuss the challenges in reproducing the cohorts, highlighting the importance of clearly reported methods (e.g. data cleansing, variable selection, cohort selection) and the need for open code and publicly available benchmarks.

Skip to content