Building Fair Machine Learning Models: Using Big Data to Explore Inequities in Risk Assessment at the Centre for Addiction and Mental Health (CAMH)

Big Data and advances in machine learning (ML) provide opportunities to improve risk assessments in mental health. Risk assessments can help manage uncertainty and account for past misfortunes. While few ML-based risk assessment tools are currently in use, it is easy to anticipate a future where such tools are used to help predict risk of inpatient violence, suicide, or psychiatric readmission. However, there is growing concern that ML tools amplify existing inequities, such as racial bias, often because they are trained on biased datasets. Few studies examine how risk assessments in mental health intersect with racism and other determinants of health, such as class, housing, and immigration status. Also, we know little about the experience of risk assessments from the patient’s perspective. Serious questions remain about how to account for biases and build fair models.