Reimagining Our Understanding of Self-Harm in Young People
Trigger warning: this article discusses sensitive topics around self-harm which might be distressing for some readers.
I am currently a visiting researcher at Imperial College London as part of the Mood Instability Research group. I am particularly interested in applying computational approaches, such as machine learning to complex data, to better understand mental health. Machine learning is a computational approach that can examine multiple factors simultaneously and identify patterns in data sets, that may not be otherwise apparent when using traditional statistical approaches.
My research focuses on the different experiences of young people who self-harm, with the ultimate aim of informing more effective support. During my Master's at Imperial College, I explored how we can use the concept of mental imagery in relation to self-harm behaviour, and used machine learning methods on a data from young people with experiences of self-harm.
The Impact of Self-Harm in Young People
Self-harm is a complex behaviour and it can be a way of coping with overwhelming emotional distress. The NICE 2022 guidelines define self-harm as deliberate self-injury or poisoning, regardless of intent.
Self-harm is a major public health concern and disproportionately affects young people, with its prevalence highest during adolescence. Adolescence and early adulthood are periods of enormous change as we navigate many new challenges and experiences. It is a formative time where we develop our identities, navigate relationships, and face new responsibilities. These experiences can become very overwhelming and, in contexts of severe distress, self-harm can sometimes emerge as a response to these emotions.
Around one in five people under 25 report experiences of self-harm, with rates continuing to rise. Importantly, self-harm is also associated with an increased risk of suicide. Together, this highlights the urgent need to better understand self-harm and improve the current support available.
There is no single explanation for the onset and persistence of self-harm. Despite the alarming statistics and findings surrounding self-harm, there is still a lot to learn about it. For example, we have limited evidence for effective treatments, and people’s responses to the same treatment or support can vary considerably. We do not yet understand who benefits from which types of support or why. A more personalised approach to mental health research may help address these issues.
Precision Psychiatry Approaches for Self-Harm Research
Across psychiatry, researchers are adopting a concept known as ‘precision psychiatry’. Precision psychiatry asks whether we can use information about an individual, such as lifestyle, biomarkers, or clinical information, to identify what treatment or support will be most effective for them. This aims to move away from trial-and-error treatment approaches which might take several attempts before finding an effective solution, prolonging a person’s mental health difficulties and increasing demands on mental health services. Precision psychiatry takes a more personalised approach to care.
Previously, research has often examined individual factors associated with self-harm in isolation. However, people's experiences are shaped by multiple factors that can interact with one another, and these factors may look very different from one person to another. Therefore, it is important we recognise the complexity of self-harm, rather than reducing it to a single cause or characteristic.
Mental Imagery in Self-Harm
An under-explored area in self-harm research is the role of mental imagery. These are images that come to mind when thinking about or experiencing an urge to self-harm. Up to 90% of people who self-harm report experiencing self-harm related mental imagery. There is growing evidence that self-harm related mental images can be a target for reducing a young person’s urge to self-harm. Targeting imagery in other psychiatric conditions such as PTSD is well established, but evidence for self-harm is only recently growing. Modifying self-harm imagery to reduce its emotional intensity may potentially reduce self-harm urges.
Where Does My Research Fit In?
I completed a six-month research project as part of the IMAGINE study, exploring which factors may predict response to a novel cognitive approach targeting self-harm related mental imagery. I worked with young people aged 16-25 years old with a history of self-harm and recent experience of self-harm related mental imagery.
Eligible young people were invited to attend an in-person session consisting of structured interviews, questionnaires, computer tasks assessing different cognitive traits, and a cognitive exercise. The cognitive exercise targeted self-harm imagery by modifying the image’s properties, aiming to reduce the image’s emotional intensity and the person’s urge to engage in self-harm.
Cognitive Manipulation of Self-Harm Imagery
Imagery-based interventions are recently emerging in self-harm research. For my research, cognitive manipulation targets self-harm imagery by changing the features of the image, such as its content or meaning, or through the sensory and structural features (i.e., changing colours, shape, or size of the images) to provide tools to manage self-harm urges.
During the cognitive manipulation exercise, the young person is guided through several rounds to alter their self-harm imagery. During the first round, the young person is ‘exposed’ to their self-harm imagery. An image they previously described when they think about self-harm is repeated back to them. They are asked to imagine the image as ‘real as it usually is’.
In the following round, the young person is asked to bring this image back to mind. However, several techniques are now introduced to manipulate or change the properties of their images to reduce emotional intensities associated with the image. Participants try a range of techniques, such as changing the colour or the size of the image, or they can use their own techniques.
Following this first imagery manipulation round, the young person chooses the technique they found most useful for changing their image. The subsequent rounds focus on using their chosen technique to manipulate and reduce the image’s emotional intensity.
During each round, the young person gives an emotional intensity rating of their images on a 0-100 rating scale, with 100 being very intense and 0 not at all intense.
Using Machine Learning to Predict Response
My results were based on a sample of 63 young people. Information from the young people’s in-person sessions, such as their cognitive, clinical, and lifestyle information, were used to identify which factors were most predictive of response to the cognitive manipulation exercise. I applied machine learning techniques to this dataset.
For the analysis, I categorised young people as responders to the cognitive manipulation if they demonstrated a decrease in emotional intensity ratings following the cognitive exercise. Those who showed no change or an increase in emotional intensity were categorised as non-responders. In total, there were 46 responders and 17 non-responders. Machine learning models were created to predict whether young people were responders or non-responders.
Using the best performing models, I identified the most influential factors for predicting response, which were found to be a measure for cognitive flexibility and greater emotion regulation difficulties.
The Implications of Findings and Future Steps
The research findings take an important step in expanding self-harm research to tailor support for young people and indicate possible factors to focus on for future studies. For example, it highlights the role cognitive flexibility has in a person’s ability to adapt or change the meaning of their imagery. Additionally, it demonstrated a role for emotion regulation in the outcomes of cognitive manipulation.
However, many more steps are required before these results can inform clinical practices. For example, further studies are required to replicate and generalise these findings, and it is important to explore the long-term effectiveness of cognitive manipulation. Importantly, we also need to further understand why certain individuals did not respond to the manipulation, and understand how we can better support them.
Overall, this research highlights the importance of integrating computational approaches with clinical methods to improve our understanding of mental health and inform more personalised care.
