Bipolar disorder is a life-long, recurrent, episodic illness with high rates of hospitalisation, suicide and physical health problems, affecting around 2% of the population and causing major disability worldwide. Medication is the mainstay of treatment, and it is often required long-term. Lithium remains the best maintenance mood stabiliser for bipolar disorder. However, only 30% of individual's prescribed lithium will have a good therapeutic response. Concerns about adverse effects of lithium, especially kidney failure, have led to a decline in its use. Presently, there is no reliable way to predict lithium responders, or those who will require an alternative drug, e.g., a second-generation antipsychotic. Furthermore, it is not possible to predict who is at risk of kidney failure with lithium or will experience adverse effects whilst taking second generation antipsychotics (most of which are related to weight gain). Other reasons for patients stopping medication are also poorly understood.
To address these issues, We will use CPRD to identify bipolar disorder characteristics that predict response and adverse events during drug treatment. Results will ultimately aid personalised treatment choices, leading to improved medication effectiveness and reduced side effect burden.
AIMS
To personalise prescribing for people with bipolar disorder via prediction models that quantify the potential benefits and risks of existing treatments based on phenotypic characteristics of the individual.
OBJECTIVES
1. Identify early, individualised clinical predictors of lithium and second-generation antipsychotic response
2. Determine clinical predictors of renal failure in individuals taking lithium
3. Determine clinical predictors of pathological weight gain in individuals taking second-generation antipsychotic medication
4. Identify predictors of early tolerability issues related to lithium and second-generation antipsychotic medication
METHODOLOGY
We will complete analyses using standard epidemiological methods (penalised Cox proportional hazards regression with exploration of suitable variable selection approaches such as Lasso) and machine learning methods (such as multi-task Gaussian process prediction), drawing on the strengths of each approach.
Treatment response defined as:
1. Time to hospitalisation
2. Time to change in treatment
Renal Failure defined as:
1. Categories of chronic kidney disease
2. Trajectories of estimated glomerular filtration rate
Weight change defined as:
1. Percentage weight change
2. Trajectory of body mass index
Joseph Hayes - Chief Investigator - University College London ( UCL )
Joseph Hayes - Corresponding Applicant - University College London ( UCL )
Christina Dalman - Collaborator - Karolinska Institute Sweden
David Osborn - Collaborator - University College London ( UCL )
Emma Francis - Collaborator - University College London ( UCL )
Fehmi Ben Abdesslem - Collaborator - University College London ( UCL )
Gareth Ambler - Collaborator - University College London ( UCL )
Glyn Lewis - Collaborator - University College London ( UCL )
Ian Wong - Collaborator - University College London ( UCL )
Irene Petersen - Collaborator - University College London ( UCL )
John Geddes - Collaborator - University of Oxford
Kate Walters - Collaborator - University College London ( UCL )
Kenneth Man - Collaborator - University College London ( UCL )
Laurie Tomlinson - Collaborator - London School of Hygiene & Tropical Medicine ( LSHTM )
Magnus Boman - Collaborator - KTH Royal Institute of Technology
Mihaela van der Schaar - Collaborator - University of Oxford
Rumana Omar - Collaborator - University College London ( UCL )
Sarah Hardoon - Collaborator - University College London ( UCL )
HES Accident and Emergency;HES Admitted Patient Care;HES Outpatient;Mental Health Services Data Set (MHSDS);ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation