Title: Are our predictions fair? Assessing and addressing algorithmic bias in a transdiagnostic risk calculator for psychosis

Abstract: Precision medicine aims to use data and clinical prediction models to inform patients' care. The data we use has inherent biases, which can therefore be perpetuated in precision medicine approaches. For ethical provision of precision medicine, we need to better understand the biases in our model outputs and, if possible, address them. We have developed, validated and implemented a clinical prediction model for identifying people at risk for psychosis in secondary health care but we do not know the biases associated with its predictions. In this talk, I will use this model to outline the limitations of current recommendations for assessing these biases, how to improve them and how we can address the biases we identify.

Biography: Dr Dominic Oliver is a postdoctoral researcher at Department of Psychiatry, University of Oxford and a visiting research associate at Department of Psychosis Studies, King's College London. His research involves using prediction modelling and digital health data, such as electronic health records, to improve prevention of severe mental disorders, including psychosis.

 Join the meeting with this Teams link.

Please contact maudsley.brc@kcl.ac.uk if you have any questions.