Building bayesian networks based on patient reported outcome questionnaires for musculo-skeletal conditions

Machine learning (ML) which is a branch of artificial intelligence (AI), has been an important approach used in the medical domain. ML approaches learn from historical data to evaluate and predict patient status. These approaches have been successful in medical domains, such as radiology and dermatology, where a large amount of data exists with clearly labelled patient outcomes. However, such clearly labelled outcome data do not exist in large amounts in most medical domains. Patient reported outcome measures (PROMS) are the primary way to assess patient outcomes in many medical areas. Filling in PROMs regularly and repetitively can be difficult due to time and cognitive-load requirements. Considering that some PROMs contain over 30 questions, collecting large amounts of patient outcome data can be difficult in these domains. This study proposes an approach for collecting patient outcome data with less time and cognitive-load requirements. In this context, an ML approach called Bayesian networks (BNs) is used to predict patient outcomes with missing PROM inputs, and to identify the most informative PROM questions for specific patients. Also, random questions were selected from the PROMs and these questions were used to determine the patient status. The obtained estimation results were compared with the estimation results obtained by using the most informative questions. The proposed approach has been applied to PROMS used in the musculo-skeletal domain. Results were evaluated by cross validation method. Cross-validation results show that the proposed approach can accurately predict patient outcomes with fewer PROM questions.