Grant: Project Grant

Chronic kidney disease is a frequent disorder associated with multiple complications. There are currently approximately 45,000 Canadians benefitting from kidney replacement therapy, with an estimated cost of $2.5 billion annually. Beyond economic considerations, end-stage kidney disease drastically reduces patients' quality of life and conveys significant risks for cardiac events, hospitalizations, and death. Intradialytic (i.e.: occurring during dialysis sessions) blood pressure variations are an important factor leading to poor outcomes. Although the responsibility of blood pressure variations on prognosis is not to be questioned, several of their aspects remain unstudied. In particular, no consensus definition exists for either intradialytic hypertension or hypotension. Different cutoffs have been used in previous studies, leading to great variability in results and difficult comparisons. Machine learning is a branch of artificial intelligence aiming at issuing predictions based on statistical relationships within a training dataset, without being explicitly programmed to do so. Latent class analysis is an unsupervised machine learning approach with recent applications in medical research, as it allows to identify "hidden" groups (latent classes) from "observed" data, by maximizing within-groups similarities and between-groups differences. We propose to use latent class analysis to identify phenotypes of patients with different blood pressure responses during dialysis. Once identified, we will determine if these phenotypes are associated with clinical outcomes, including mortality, myocardial infarction, stroke, heart failure, and hospitalizations. We expect results from this study to help identify pathological blood pressure variations during dialysis. This knowledge will in turn help designing future interventional trials to improve clinical outcomes for patients on dialysis.