An Analysis of Frailty Progression in Elderly Patients using Process Mining and Machine Learning

Frailty is a geriatric medical condition which affect 26% of people with age over 85 in the UK. This distinctive health state happened as a result of cumulative deterioration in bodily systems and diminished clinical state functional reserve over lifetime. It is a transition state from healthy ageing to dependent elderly life. The transitioning happened as the elderly with frailty has low ability to cope with acute illnesses or daily stressors. Understanding frailty progression as part of early identification of frailty may offer great opportunity in lengthening the transition time and maximize healthy ageing period. Majority of user in the healthcare sector identified as the elderly. The abundance of data recorded in the electronic healthcare record (EHR) related to patient health status and conditions has potential to facilitate the understanding of frailty. This thesis used machine learning and process mining techniques which is an emerging data-driven analytic approaches to understand frailty progression, its association with three frailty deficits of concern namely fall, hypertension and polypharmacy and investigate the variation between two frailty scores cut-off points. The main objective in this work is to analyse frail elderly pathway with respect to frailty progression pathway using electronic frailty index (eFI) score and highlight the feasibility of employing process mining and machine learning in analysing frailty. This work used two real-life healthcare datasets to understand the variability of frailty progression pathway. The first dataset is a publicly open healthcare dataset from the tertiary hospital setting in the USA. The first dataset provides as a platform for preliminary work and develop methods in analysing frailty trajectories to support reproducibility of the work. The second dataset is a UK healthcare dataset from the primary care setting. The experiments in the second dataset used as an improvement to comprehensively study frailty progression from the previous dataset. Furthermore, the variation between the literature and proposed cut-off points in this study was investigated using the second dataset. Frailty progression pathway analysis used process mining technique while the identifying the cut-off points in the proposed approach used machine learning technique. The advantages of using process mining in healthcare domain especially in frail elderly has been proven. It is useful in presenting the visualization of frailty progression at different frailty stages and discover the variation of frailty progression with the associated deficits of concern. This thesis also explores the comparison in frailty progression between two cut-off points approaches used in the literature and the proposed approach in this study.