A multi‐state model to improve the design of an automated system to monitor the activity patterns of patients with bipolar disorder

This paper describes the role of mathematical modelling in the design and evaluation of an automated system of wearable and environmental sensors called PAM (Personalised Ambient Monitoring) to monitor the activity patterns of patients with bipolar disorder (BD). The modelling work was part of an EPSRC‐funded project, also involving biomedical engineers and computer scientists, to develop a prototype PAM system. BD is a chronic, disabling mental illness associated with recurrent severe episodes of mania and depression, interspersed with periods of remission. Early detection of the onset of an acute episode is crucial for effective treatment and control. The aim of PAM is to enable patients with BD to self‐manage their condition, by identifying the person's normal ‘activity signature’ and thus automatically detecting tiny changes in behaviour patterns which could herald the possible onset of an acute episode. PAM then alerts the patient to take appropriate action in time to prevent further deterioration and possible hospitalisation. A disease state transition model for BD was developed, using data from the clinical literature, and then used stochastically in a Monte Carlo simulation to test a wide range of monitoring scenarios. The minimum best set of sensors suitable to detect the onset of acute episodes (of both mania and depression) is identified, and the performance of the PAM system evaluated for a range of personalised choices of sensors.