The development and evaluation of a connected health exercise biofeedback platform to support orthopaedic rehabilitation

Home exercise programmes play a key role in patient rehabilitation following knee replacement surgery. The demand for these operations is increasing significantly due to the aging population, and in an attempt to reduce costs to healthcare providers, growing numbers of patients are being discharged directly home from hospital. These patients are typically provided with a home exercise programme which they are expected to complete on a routine basis, placing greater responsibility on the self-management skills of the patient. However, adherence to home exercise programmes is poor, with many reasons for lack of engagement, leading to sub-optimal rehabilitation outcomes and negative implications for the healthcare provider. Connected health technologies utilising ubiquitous mobile devices, alongside sensing platforms such as inertial measurement units, can be used to provide exercise biofeedback to patients in an automated manner. By harnessing machine learning to interpret sensor data captured during the exercise programme, the patient and clinician can receive personalised, accurate and timely information to support the rehabilitation process. Despite a number of systems being developed, acceptance and uptake remains poor. This is partly due to the lack of technical and user evaluation being undertaken, limiting the understanding of key components such as usability, feasibility and functionality. Hence, there is a need to thoroughly evaluate any newly developed systems from an early stage with key stakeholders. The focus of this programme of research was to develop and evaluate a prototype connected health exercise biofeedback system comprising of an Android tablet application and single inertial sensor for use in home exercise rehabilitation following knee replacement surgery. The research presented throughout this thesis suggests that whilst the developed prototype was easy to use and may aid engagement, there are numerous technical challenges in providing technique-based biofeedback using supervised machine learning. Furthermore, this thesis provides preliminary evidence that a more suitable and feasible method of providing feedback may be based on measuring joint angle, as opposed to the current classification approach.