How Machine Learning Could Help Mental Health Diagnoses
Authors: Dr Paris Lalousis, Dr Sian Lowri Griffiths, Professor Rachel Upthegrove et al., 2021
Background: It is a big challenge for clinicians to make an accurate diagnosis based on symptomatology, seeing as people with depression or psychosis rarely experience symptoms of just one illness. Capturing the complexity of an individual’s experience or the neurobiology behind their symptoms is especially difficult. This presents a challenge when delivering treatments as the majority of patients do experience co-morbidities (i.e. symptoms of another illness on top of their primary illness), such as depression, which may go untreated if a clinician chooses to focus on managing the symptoms of the ‘primary’ illness of psychosis (e.g. hallucinations).
Current Research: A team at the University of Birmingham’s Institute for Mental Health (#IMH) and the Centre for Human Brain Health (#CHBH) worked with researchers on the PRONIA Consortium (www.pronia.eu), to explore the use of machine learning techniques to create accurate models of ‘pure’ forms of psychosis or depression. They then tested these models on a cohort of patients with mixed symptoms, to investigate their diagnostic accuracy.
Data Collected: Researchers collected data from a total of 300 patients taking part in the PRONIA study. Data obtained included questionnaire responses of patients’ lived experience of their symptomatology, detailed clinical interviews, as well as structural magnetic resonance imaging data (MRI).
Results: It was found that machine learning models could indeed identify models of ‘pure’ psychosis or ‘pure’ depression. They then applied these models to patients with symptoms of both illnesses, to see if the model was accurate against the diagnosis the patient had received. What they found was that those with depression as their ‘primary’ illness were more accurately identified. Interestingly, patients with psychosis as their ‘primary’ illness with the addition of comorbid depression symptoms were more likely to be grouped into the ‘pure’ depression diagnosis model.
Future Directions: This suggests that depression may play a greater part in the experience of psychosis than previously thought. With this in mind, the PRIMED+ Lab is also running another study called #ADEPP which is a controlled clinical trial investigating the use and feasibility of anti-depressants for the prevention of depression following a first episode of psychosis.
If you’re interested in participating in our ADEPP trial, you can find further information here. We are recruiting people aged from 18-65 who have started treatment for a first episode of psychosis in the last 12 months. We are recruiting from sites in the West Midlands, Northwest, and Wales.
Twitter Socials: Primed Lab, Dr Paris Lalousis, Dr Sian Lowri Griffiths, Professor Rachel Upthegrove
Reference: Lalousis et al (2021) Heterogeneity and Classification of Recent Onset Psychosis and Depression: A Multimodal Machine Learning Approach. Schizophrenia Bulletin, Vol 47, 1130–1140. DOI: https://doi.org/10.1093/schbul/sbaa185
Blog written by Melanie Lafanechere (Research Associate for the PRIMED+ Lab)