Machine Learning in Mental Health Research: Why the Early Adopter Status is So Important
Machine learning is here to stay. Considered part of artificial intelligence, it broadly involves computers discovering patterns in data and improving the predictions they make without being explicitly programmed to do so. Machine learning is integrated in our everyday life, even though we may not realise it. The keyboard in our smartphones uses machine learning to try and predict the next word we are going to type, and the insurance quotes we get are generated from machine learning algorithms that learn from historical data what our likelihood of making a claim are. Even the online retailers from which we shop recommend products to us using machine learning algorithms which identify what products our specific user profiles are most likely going to like.
So, what does all this have to do with mental health?
Mental health provision has long been secondary to physical health. In England, mental health trusts received just 5.5% income increases between 2012 and 2017 compared to a 16.8% budget increase in acute hospitals during the same period. This leads to people with mental health problems facing increasing difficulties in accessing mental health provision. When they do access mental health care, their journey towards treatment is laid with obstacles. People rarely fit into neat boxes and due to the fact that diagnoses are not based on clear biological markers it is difficult to successfully predict the best course of action. Eventually, when patients receive treatment, this is often a game of trial and error. Clinicians know that certain medications work well for certain disorders, but it is currently difficult to predict which individual will respond successfully to which course of treatment. The difficulties in treatment get even more compounded by the fact that what constitutes mental illness still remains elusive.
A powerful tool to predict treatment outcomes
Machine learning is a powerful tool which may be able to help such diagnostic and therapeutic conundrums. By collecting a wealth of different forms of data (such as brain scans, blood samples, and clinical questionnaires) and feeding them into machine learning algorithms, scientists can attempt to uncover what truly constitutes mental illness and furthermore create individualized treatment profiles. An increasing number of initiatives in psychiatry are using machine learning for these reasons and the results so far have been promising. The PRONIA consortium developed a machine learning model that combined neuroimaging and clinical data, derived from clinical interviews, to predict how patients who are at risk for psychosis would fare in terms of their social life one year after the start of their treatment. The model could successfully predict which patients would have an increase in their social outcomes such as their ability to work, or create and maintain relationships and which ones would not, with an accuracy of 82.7%, outperforming the predictions made by human groups of experts. More recently, in a machine learning study led by the lab I am part of, we found that depression is inherent in early phases of schizophrenia and linked to specific brain structures. This has the potential to inform novel treatments in schizophrenia with the use of anti-depressant medication alongside anti-psychotic medication in patients who fit the profile.
However, these recent advances in the use of machine learning in mental health research have not yet been translated into clinical practice, and a similar pattern can be observed across the medical field. There are various explanations regarding why clinical application of machine learning has not yet been materialized with the most obvious being that huge advances have only occurred recently in the field which is especially true of mental health research. Nevertheless, machine learning seems to be incorporated into our everyday lives in almost every other aspect, so it is only a matter of time before attention turns to how it can be used in the clinic.
The early adopter status
As with every technological advancement, people are classed into either; innovators, early adopters, early majority, late majority or laggards. These distinctions can apply both to individuals/companies and fields/companies. I would argue that the field of mental health is in the early adopter category with regards to machine learning. Taking into account the fact that it has always been secondary to physical health, this is one of the few if not the only occasion where the field of mental health finds itself in the same category as physical health. Taking advantage of this early adopter status will be crucial in closing the gap between the two. Technological advances in machine learning are growing rapidly and a field that is an early adopter can find itself in the late majority or laggard category quite quickly.
Within the field of mental health, clinicians and scientists can also be categorized in the aforementioned way. There will be individuals who are sceptical of the use of machine learning in mental health and will be averse to implementing the new technological opportunities that will arise. I believe that it is the responsibility of innovators and the early adopters in the field, to communicate the opportunities that machine learning presents, in order to encourage people to support the inevitable transformation of clinical care it will lead to.
Scientists need to be transparent, reproducible, open, and robust in the way they do science in this field in order to keep the early adopter status and take advantage of the opportunities that machine learning presents. By educating ourselves and others in its promises and pitfalls, scientists will hopefully be able to see a transformation of clinical care in mental health that will lead to faster access to mental health provision, lower rates of misdiagnosis, targeted treatments, and truly personalized care.
