Project 463264

The digitization of health records mean that large volumes of data are now available for scientists to study disease processes, obtain real-world evidence regarding the effect of treatments, and to use data to gain insights into treatment choices that may be best suited for individuals or groups of patients. Electronic medical records from clinical cohorts of patients with autoimmune diseases at the Centre for Prognosis Studies in the Rheumatic Diseases motivate this research and will be used to illustrate the advances developed. This research program is first directed at the development of a rigorous methodological framework to study and address biases inherent in electronic health records (EHR) data including selection biases, and observation biases arising from disease-related clinic visits and loss to follow-up. Appropriate methods of dealing with confounding by indication in the real world setting require models for treatment selection by physicians. Modern machine learning algorithms will be developed to accommodate incomplete data on time-dependent confounders, while facilitating causal comparisons of different therapeutic classes based on "real-world" data. Predictive models for the occurrence of adverse effects and clinical response will be developed and validated, and based on these treatment guidelines will be developed to aid clinicians in decision making. This research team is comprised of internationally renowned biostatisticians and public health researchers and distinguished clinician scientists. The planned research is motivated by an established interdisciplinary research program founded on registries of patients with various autoimmune diseases. Our goal is to ensure personalized medical decision-making is based on powerful predictive models exploiting modern machine learning algorithms which have been rigorously assessed. In this way, judicious and transparent use of EHR data will improve patient care and outcomes.