A machine learning approach to understanding comorbidity between mental and physical health conditions

In the UK, mental ill health is the biggest cause of disability and the leading reason why people take time off work when they are unwell. People with mental disorders (such as schizophrenia, bipolar disorder and major depressive disorder) die on average 15-20 years earlier than those without, often from preventable causes such as suicide and cardiovascular disease. We now have increasing amounts of 'big data' that can help us understand why people develop mental disorders and why those with poor mental health also experience more physical illnesses such as diabetes, heart disease, cancer and stroke.

Big data are defined by volume, variety and velocity. Volume refers to the quantity of data, variety refers to the numerous types of data and velocity refers to the speed of data processing. We now have the ability to combine large amounts of data that are routinely collected by the NHS and other novel studies such as the UK Biobank and Generation Scotland cohorts. By integrating a range of big data relating to mental health, such as prescriptions for medications (e.g. antidepressants), history of hospitalisation for psychiatric reasons and diagnostic information for physical illnesses and looking at the patterns within these, we may be able to better predict who will become mentally and physically unwell and what course their illnesses will take. It is likely that a complex interaction of environmental (e.g. deprivation), lifestyle (e.g. smoking) and biological (e.g. genetics) factors across people's life courses are involved. This is why projects like UK Biobank are so important to help us uncover the factors that matter the most. UK Biobank is a study which collects a range of biological, social and lifestyle data for around 500,000 people living in the UK. We can now apply novel methods to explore the vast amount of data available. For example, techniques such as machine learning, a branch of computer science that provides computers with the ability to learn from patterns in data and adapt on their own, allow us to take into account the complexity of the available data.

The analysis of big data in mental health has great potential to enhance our understanding of mental health problems, so that patients can be better supported and their conditions managed more effectively. In particular, to provide better treatment that is tailored to an individual's needs we need to be able to better identify early on who will develop comorbidities between physical and mental health conditions and we need to look for warning signs of who is most in need of help and when. By using data in this way we have the potential to greatly improve outcomes for both patients (e.g. increased survival) and wider society (e.g. reduced healthcare costs).

The research will be carried out by Dr Claire Niedzwiedz based at the Institute for Health and Wellbeing at the University of Glasgow. She is an inter-disciplinary quantitative researcher in public health with interests in understanding and preventing inequalities in mental health. She will analyse a range of data sources including UK Biobank, Generation Scotland and anonymised data that are routinely collected within the NHS (e.g. prescriptions, hospitalisation records and mortality records) and explore the potential of machine learning to better predict the comorbidity between mental and physical health conditions.

The key aim of the research is to enhance our understanding of the comorbidity between mental and physical health conditions by combining the disciplines of public health, computer science and psychiatry. The objectives of the research are to improve the prediction of the onset of co-morbidities between mental and physical health conditions (such as major depressive disorder and diabetes) and related adverse outcomes (e.g. hospitalisation and mortality). A range of data sources will be used to test the utility of implementing a machine learning approach to prediction. For example, using the UK Biobank and Generation Scotland cohorts will allow the exploration of a range of biological, environmental and lifestyle factors. Linked administrative health data (including the Scottish Morbidity Records, disease registers, Prescribing Information System, mortality records and the census) will facilitate the exploration of patients' complex medical histories and social characteristics (e.g. occupations). Machine learning algorithms (e.g. deep neural networks and random forest) will be used to learn from patterns in a range of big data by splitting the data into training and test datasets, assessing the algorithm performance and comparing the results with other methodological approaches. By identifying factors that are highly predictive of physical and mental health comorbidities and adverse outcomes, there is great potential to develop new approaches to patient stratification and novel precision medicine interventions. Collecting these data within medical settings may facilitate the development of improved diagnostic, treatment and preventative measures in clinical and public health practice.