| Summary: | <p>It is estimated that half of the health issues in the United Kingdom are linked to mental health for individuals under the age of 65, with the majority of these health issues being attributed to depression or anxiety. One of the most prevalent mental disorders is bipolar disorder, with patients experiencing mood instability with extended periods of depression, associated with feelings of unhappiness and dejection. They also experience periods of mania, associated with euphoria, delusions and overactivity. Borderline personality disorder is the most common personality disorder and is linked with disrupted mood and altered personal interactions. Current clinical practice in the management of bipolar disorder and borderline personality disorder involves self-reported monitoring and interviews with clinicians. This practice is based on research focused on clinical laboratory settings and single modality sensors to record data. Smartphones and mobile devices allow for at home monitoring in a natural setting and can acquire multiple signal modalities over very long periods of time, thus providing more information about the links between behaviour and physiology in mental health disorders. These objective measures could provide passive measures of sleep and circadian rhythms which could be used to determine the health of patients at home. This thesis explores the relationship between circadian rhythms, mood symptoms and depression within individuals and patient groups. </p> <p>A longitudinal study of 141 individuals, including bipolar disorder, borderline personality disorder and healthy control participants, monitored behavioural, physiological and mood measures for a mean of 83±31 weeks. Circadian features of activity and heart rate, in addition to measures of sleep were extracted from the signals through fitting of sinusoids and application of time varying hidden Markov models.</p> <p>Internal desynchronisation of circadian rhythms were investigated through differences in the phase of the rhythms of a four day period. Significant desynchronisation between the rhythms of heart rate and both activity and sleep were seen in the patients groups. Borderline personality disorder participants had the greatest desynchrony with a three hour lag between heart rate and activity and one hour between heart rate and sleep.</p> <p>The variability of circadian rhythms in healthy mood states, known as euthymia, were greatest in the borderline personality disorder patients, with bipolar disorderparticipants also having more variable sleep. Significant correlations were seen between circadian variability and mood variability in borderline personality disorder participants. The variability in activity rhythms correlated strongly with negative mood (ρ = 0.819, p = 0.04) in addition to the variability in heart rate amplitudes and negative mood (ρ = 0.852, p = 0.02).</p> <p>New features of circadian activity and sleep were extracted from longitudinal smartphone accelerometry using time varying hidden Markov models. Personalised regression models were developed which are capable of predicting levels of depression, with average mean absolute errors of depression scores (ranging from 0 to 27) of 0.80 ± 0.79 in bipolar disorder participants, 1.33 ± 1.03 in borderline personality disorder participants, and 0.39 ± 0.40 in healthy control participants. These errors are smaller than the thresholds for diagnosing episodes of depression (0 − 5 <em>no depression</em>, 6 − 10 <em>mild depression</em>, 11 − 15 <em>moderate depression</em>, and > 15 <em>severe depression</em>), with bipolar disorder and borderline personality disorder participants entering all levels of depression, but healthy controls remaining mainly in the no depression category. The personalised regression models developed were able to classify depression with an accuracy of 88% in bipolar disorder, 79% in borderline personality disorder, and 98% in healthy participants.</p> <p>The results presented in this thesis show both bipolar disorder and borderline personality disorder to be associated with disrupted circadian rhythms. A major finding of this thesis is that the quantification of circadian rhythm disruption may be performed over a short-term high intensity period, reducing the need for multiple obtrusive devices to be worn over long periods of time, with shorter term monitoring have less barriers for adoption in to clinical practice. It was also found that the use of multiple modalities (acceleration and heart rate) allows for novel desynchronisation metrics to be performed, which are not possible from single sensor recordings, providing further insight into the levels of illness.</p> <p>The findings also highlight the relationship between mood and circadian rhythms and show longitudinal monitoring of acceleration may act as an objective marker to aid monitoring of symptoms of depression. Due to the activity being recorded from smartphones, which are already in widespread use in current society, there is a reduction in the barriers to uptake for patients to begin using these monitoring tools. These markers of circadian activity may have a considerable impact in the future of clinical care through providing automated objective measures to compliment the current clinical practice and reduce some of the subjectivity in diagnosis and monitoring of depression.</p>
|
|---|