Grant: Travel Awards - Institute Community Support

The need for mental illness care continues to grow, adding onto to the already strained mental health care system. Algorithms have the potential to help people get the care they need but can also be biased and unfair. Currently, tests of new algorithms in mental illness care focus on accuracy (how often the algorithm can predict the "right" decision) but are not being regularly checked for bias & fairness problems. Our plan is to put together a way of testing as many of the ways that bias and fairness that can affect an algorithm's prediction that we can and then showing those answers in a way that everyone can understand. We will gather ways in which bias and fairness can impact decision making from health sciences research as well as from other fields such as psychology, economics, and law. We will then pick the ways which can be measured from health record data (which is what the algorithmic predictions we are testing will be based on). We will also keep track of any ways that we cannot measure so that we can include those limitations with the results. We will then test the collection ways that bias and fairness can affect algorithm predictions in a real-world case. The algorithm we will test is designed to predict who is more likely to need more intense psychotherapy. After ensuring that our bias and fairness testing works on the algorithm's predictions, we will engage in focus group discussion with people who have mental illness, carers, healthcare providers, and hospital administrators. These will provide us feedback so that we can update the way the end results of the bias and fairness analysis is presented so that the people who will be affected by an algorithm can understand its impacts and make informed decisions about its use.