A team of epidemiologists, clinicians, statisticians and data scientists focusing on applying artificial intelligence to routinely collected clinical data.
Randomized controlled trials are the gold standard for establishing causal relationships between observations. However, observational studies on routinely collected data (RCD) offer potential advantages in representativeness, generalizability, cost, and convenience over RCTs. RCD studies may fill an evidence gap, particularly for realistically structured populations and follow-up schedules (Hemkens, 2016). To gain from hidden structure and complexity of data, we need to go beyond current routine epidemiological approaches.
Routinely collected data containing time series, such as body mass index and blood serum analytes, are characterized by sparse, multivariate, irregular data. Techniques are used from machine learning, statistics and artificial intelligence, to further better assessment of patient risk and recommendations for policy improvement. The aim is to develop individually personalized models which capture important features of medical history that traditional population-based models do not deal adequately with.
Our hope is that improved analysis methods will uncover hidden patterns in data, improve understanding and, ultimately, inform policymaking.