Advancing Precision Treatment Selection for Post-traumatic Stress Disorder

Post-traumatic stress disorder is a mental health condition that develops following exposure to one or more traumatic events. In line with clinical practice guidelines, National Health Service (NHS) Talking Therapies services deliver Trauma-focussed Cognitive Behavioural Therapy (Tf-CBT) and Eye Movement Desensitisation and Reprocessing (EMDR) for the treatment of PTSD. Despite being evidence-based, many patients do not respond to these treatments and rates of reliable improvement are lower for PTSD than other mental health problems. There is evidence to suggest that some patients are more likely to respond to one of these therapies than the other (i.e., Tf-CBT vs. EMDR), and it may be possible to identify patients' optimal treatment from their pre-treatment data using machine learning methods. This is known as precision treatment selection. This thesis investigated whether precision treatment selection could improve treatment outcomes for PTSD in NHS Talking Therapies. First, a systematic review of studies that applied machine learning methods to predict the outcome of psychological therapy for PTSD was conducted. This revealed significant limitations and omissions in the application and reporting of machine learning methods, and an almost complete lack of external validation of prediction models. Second, a previously published precision treatment selection model was externally validated in an independent sample of NHS trauma therapy cases. This found that the model did not generalise to new patients, potentially due to methodological limitations. Third, a range of machine learning methods were applied to predict Tf-CBT outcomes using a large sample of NHS Talking Therapies trauma cases. Models were optimised and out-of-sample performance was compared in a validation sample, and methodological recommendations were made. It may be possible to develop a precision treatment selection model for PTSD, but this will require reliable application of machine learning methods in adequate clinical datasets.