Inferring 'Therapeutic States' of Patients from Community Electronic Prescribing Data

Jim Warren, Jan Stanek, Svetla Gadzhanova, Ivan Iankov, Gary Misan


We set out to devise a method for analysis of chronic disease therapeutic decision making with specific emphasis on practice patterns across multiple consultations and providers within a single community-based practice. We examine treatment by abstracting each patient’s therapeutic state at any given time as a vector of n Boolean state variables, each representing a key decision in the domain under examination. We illustrate the method where the state variables are inferred from electronic prescribing data for treatment of hypertension at a rural practice. We find that graphs of therapeutic state transitions, at various levels of granularity, can provide an overview of pre-scribing practice or help to identify cohorts of patients that warrant further examination. The graphs, however, are sensitive to heuristic interpretation of the data. A direction for further research is to identify the principles for inference of therapeutic state that are adequately sophis-ticated for accurate classification of cases and yet interpretable for clinical audit.


Machine Learning; Family Practice; Chronic Disease; Therapy

Full Text:


::::::::::::::  eJHI - electronic Journal of Health Informatics - ISSN 1446-4381  ::::::::::::::

                                     Privacy Statement - Uptime