Phenotype extraction
Merijn Beeksma, Iris Hendrickx and Antal van den Bosch

‘Phenotypes’ can be used to characterize clinical phenomena on the levels of individual patients, specific diseases, and illness trajectories. Phenotypes can be extracted from medical records automatically, but common approaches to phenotype extraction do not take the temporal order of events into account, and tend to yield ‘flat’ representations of phenotypes. The temporal order of events is necessary to fully understand patterns of co-morbidity and disease progression, and to enable comparisons between illness trajectories of different patients. We will explain what phenotyping is in the medical context, discuss the techniques borrowed from NLP to extract phenotypes automatically from both structured data (e.g. diagnostic codes) and unstructured data (e.g. clinical free-text notes), and how we can apply the same principles to model time-series data in other domains.