Machine Learning (ML) systems can make outcome and diagnosis predictions that have the same level of performance as expert clinicians in some domains. However, the success of these systems has been mainly limited to cases where large amount of clearly labelled data is available. In order to expand this success to other areas, the use of data sources with higher uncertainty and subjectivity should be investigated. In this regard, Patient Reported Outcome Measures (PROMs) are a potentially important data source for ML. A PROM is a questionnaire that aims to measure patient status and outcomes based on multiple questions filled in by the patient. Repetitive collection of multiple PROMs offers a useful resource to track the patient status over time, but this can also increase measurement uncertainty as it requires increased cognitive load and effort from the patient. This project focuses on using PROMs for ML and decision support tools. The project aims to develop a systematic approach to integrate PROMs into Bayesian Network models. These probabilistic PROM models will enable more efficient collection of PROMs as they will be able to deal with missing inputs, select questions specific to the patient’s condition and their previous answers, and stop asking questions when sufficient information is acquired.
