Using Case- based Reasoning for Creating Intelligent Systems in Healthcare

Healthcare research is an emerging field of application of machine learning techniques to investigate complex health datasets. Patients are at the center of any healthcare system. There is a growing realisation of the patientcentred healthcare system and the notion is slowly changing the healthcare scenario from a one glove fits all approach to a more personalised approach. Data collected in population-based and intervention-based studies has immense potential in supporting primary caregivers in providing patient-centred care by facilitating clinical decision-making. From identifying patients at a high risk of post-surgical complications to forecasting their quality of life, recent developments in healthcare informatics suggest that leveraging the capabilities of machine learning techniques on complex health data can have a significant impact on the decision-making process in clinical settings. To this end, we explored supervised and unsupervised machine learning methods, predominantly, Case-Based Reasoning (CBR) methodology. The overall theme of this research is exploring the potential of healthcare datasets using CBR methodology. We used two unique and innovative datasets—a population-based dataset consisting of objectively measured physical behaviour data collected using body-worn sensors in HUNT4 cohort study, and an intervention-based dataset comprising patient-reported outcome measurements collected during clinical trials to test the efficacy of tailored interventions in SELFBACK mobile app—and applied both supervised and unsupervised learning to glean valuable information.The goal is to develop intelligent modules that can be incorporated into clinical decision support systems to support clinicians in the informed decision-making process or as standalone systems. Focus is placed on applying the casebased methodology to learn from the data without making assumptions. The HUNT4 physical behaviour dataset was investigated to get insights into the physical behaviour characteristics of the population and identify clusters of similar behaviour profiles using a new clustering approach. The clustering approach was valuable in identifying groups of similar physical behaviour, which can be used further by primary caregivers to underpin the amount of physical activity tailored to the individual’s needs. The SELFBACK intervention datasets were explored to determine the predictors of various patient-reported outcomes and investigate the predictive potential of the patient-reported outcome measurements using case-based and conventional machine learning methods. The methods used show the potential to predict pain-related patient-reported outcomes. Overall, our results indicate that a close liaison between healthcare data, clinicians, and machine learning methods can promote a better understanding of achieving patient-centred care through the addition of intelligent systems in clinical decision support. The results also provide grounds for further research and development of evidence-based clinical decision support systems