A Predictive Chronological Model of Multiple Clinical Observations

The expanding clinical information provided by the advent of electronic medical records offers an exciting opportunity to substantially improve the quality of health care. By examining the clinical observations (such as diagnoses, risk factors, and medications) mentioned in longitudinal EMRs, we can use patients´ medical chronologies to automatically predict the progression of their pathologies. In this paper, we present a novel probabilistic model which jointly learns how to (1)group patients based on the similarities between their clinical observations as well as how to (2) predict the way a new patient´s clinical observations might evolve in the future. We show that our model can be used to not only track how a patient´s clinical findings might change over time, but to also identify which patients are due for preventative visits. In addition, our model has the potential to improve the quality of over-all patient care in practice by predicting the most likely set of clinical observations at an arbitrary point in the future.