Bipolar Disorder (BD) is a serious mental illness characterised by high and low moods. Individuals usually experience symptoms for years before being correctly diagnosed. This project aims to improve the recognition of BD through identifying earlier signs before diagnosis. We aim to examine the risk of development of adverse events and identify factors associated with delayed diagnosis.
Looking back at health records of people with BD, we will identify when the most frequently occurring health events were prior to BD diagnosis, such as depression, anxiety and antidepressant prescription. We will compare these health events with those occurring over the same time in individuals without BD. This will enable us to decide on events most likely to indicate the onset of BD and important earlier signs leading up to onset. A separate analysis using machine learning, which identifies patterns of events without prior knowledge of associated symptoms will also be adopted. To examine consequences of BD, we will look forward from diagnosis and calculate risk of developing adverse outcomes, including self-harm, cancer, diabetes, cardiovascular disease and premature death. In the BD individuals, calculating the time between onset and diagnosis will provide an estimate of delay so we may identify risk factors associated to short or long delay. We will examine the risks of developing subsequent disease and also premature death depending on the length of delay in diagnosis.
The study objectives are to improve early recognition of Bipolar Disorder (BD), examine major adverse outcomes as a consequence of BD and to examine risk factors for delayed diagnosis and the consequences of delay. Individuals with first diagnosis of BD, identified by Read codes between 1/1/2010 and 31/07/2017 in CPRD and eligible for IMD, HES and ONS-linkage will be extracted and matched on age, gender, practice and index date with unaffected control patients (ratio 1:20). Likely symptoms of onset and those earlier signs leading up to onset will be identified by investigating the incidence of each event of interest for each year prior to diagnosis from all contacts through primary and secondary care sources of individuals with and without BD. A separate analysis using machine learning, which identifies patterns of events without prior knowledge of associated symptoms will also be adopted. To handle the large number of event codes and with some codes very rarely used, semantic similarity and principal component analysis will be applied. In addition, density-based spatial clustering of applications with noise (DBSCAN) algorithm to identify clusters of patients exhibiting similar patterns will be implemented. Cox regression will be used to investigate whether individuals with BD have elevated morbidity and mortality risk. By establishing the likely onset date of BD and calculating the probable diagnosis delay, multivariable regression models (linear or logistic regression) will be used to identify factors independently associated with diagnosis delay such as subsequent mental health diagnoses or substance abuse. We will also examine elevated risk between delayed diagnosis and subsequent morbidity and mortality outcome by using Cox regression analysis to obtain hazard ratios.
Health Outcomes to be Measured:
Diabetes mellitus morbidity
Major cardiovascular event
HES A&E;HES Admitted;HES Outpatient;ONS;Patient IMD