Zach Cohen: Predicting Outcomes and Treatment Course in Child and Young People’s Mental Health
Despite an increase in access to psychological interventions for children and young people in the UK, a significant subset of those who access treatment do not experience positive outcomes. The Child Outcomes Research Consortium (CORC), led by Miranda Wolpert, has spearheaded recent initiatives to increase access, introduce outcome monitoring into services, and improve treatment quality in mental health services across the UK. In recent years, CORC have supported the Child and Young People’s Improving Access to Psychological Therapies (CYP IAPT) initiative, which has collected outcome data on over 96,000 cases seen across 75 services over a 5-year period.
How can we use this enormous dataset to improve outcomes?
Drawing on the success of precision medicine in areas like cancer treatment, mental health has begun to seek answers to the question, “What works for whom?” Precision medicine aims to help an individual select the specific intervention that is most likely to have the best outcome, based on his/her/their unique constellation of personal factors. For example, someone with depression might be unsure of whether to pursue antidepressants or psychotherapy. Precision medicine approaches can help people and clinicians make more informed treatment decisions.
First steps towards precision medicine in CYP-IAPT
One way in which we can try to improve outcomes in child and youth mental health is to help identify factors that predict response to specific interventions and use these factors to guide treatment selection. The CYP IAPT study contains patient-level data that could help predict treatment response. One complicating factor in this dataset is that individuals could have received one (or more) of over a dozen psychological interventions (as well as medications), and the number of therapy sessions children received ranged from one to over one hundred.
To account for this, we have built statistical models that associated individual characteristics (e.g., diagnosis) with the amount (number of sessions) and type of treatment that people accessed. These analyses revealed that certain characteristics (e.g., specific diagnoses) were associated with accessing specific types of treatment, and that specific interventions were associated with larger numbers of session than others.
The next steps will be to see whether individual differences predict the likelihood of having a positive outcome in treatment, and to determine whether this information can be used to help target specific treatments to individuals to improve response. Identifying these relationships will allow us to build predictive statistical models that can help determine, for each individual, which treatment is likely to lead to the best outcome. If successful, these efforts will help remove the guesswork from treatment recommendations.
Zach Cohen, from University of Pennsylvania joined us at CORC in June 2017 and presented on his research using the CORC dataset on predictive modelling of treatment.