Jun 01, 2018 | 14:00—14:45
The availability of large scale healthcare data promises to revolutionise medical care towards a precision medicine where treatments are tailored individually to each patient. In the last decade, the analysis of medical records, often recorded for billing purposes, was centred around binary relations between diseases, such as the comorbidity between them. The concept of comorbidity quantifies the probability that a patient with disease A also has disease B, but does not capture the long term history of a patient. For example, the subsequent development of diseases A,B and C could indicate the presence of a particular genetic defect, increasing the likelihood for disease D. The aim of our work is to bridge this gap by developing a method to forecast future diagnoses of a patient which takes into account the entire history, i.e., the specific sequence, of his or her diagnoses. To this end, we analyse a data set containing for each inpatient hospital stay of patients in Austria over two years the main and side diagnoses and the admission and release dates. The data thus contains a trajectory on the set of diseases for each of the approximately 8M people insured via an Austrian social security provider. We represent the health history of each patient by a binary vector, and divide the set of patients into categories by using a hierarchical clustering algorithm. When patients acquire new diseases, they can change their cluster. We analyse the pattern of transitions between the different clusters, in particular, we see that there exists a stream of patients into clusters acting as sinks, characterised by highly multimorbid patients.