The UCSF Clinical Ethics Journal Club is hosted by UCSF Bioethics. Join us this week as Dr. James Deardorff, a geriatrician and assistant professor in the Department of Geriatrics, will present on algorithmic fairness with a few brief readings. He will then workshop his research plan to address fairness in a prognostic model for people who are incarcerated in California.

Register for for the Zoom meeting [here]. You may access the readings at  [tiny.ucsf.edu/ClinEthicsJC].

A note from Dr. Deardorff:

"I'm a geriatrician and research faculty in the UCSF Division of Geriatrics. Our team is working on developing a 2-year mortality prediction model for individuals incarcerated in the California prison system. We have developed a model that includes predictors such as age, comorbidities, and hospitalizations in the past year which performs pretty well. The goal of this model is to flag individuals at high mortality risk to make referrals to palliative care, advance care planning discussions, and consideration of compassionate release.

We want to ensure that our model does not worsen healthcare disparities among different racial and ethnic groups (e.g., failing to identify truly high-risk individuals in minority groups leading to fewer palliative care referrals). There has been much discussion in the literature about how to ensure that models are "fair," and a variety of metrics have been proposed to assess this (e.g., equal opportunity, disparate impact, and demographic parity which are all really just different forms of measuring sensitivity, specificity, positive predictive value, and negative predictive value).

Our goal for this journal club is to get feedback on a proposed fairness evaluation that we are conducting. We will review a document that shows the details of the fairness assessment, including some text and 2 tables, which I will go over during the first part of our discussion. The footnotes of the tables include a further elaboration on how these measures are calculated.

If anyone is interested, there will also be notes on algorithmic fairness and a paper that does a great job of explaining how these analyses can be conducted. Looking forward to the discussion!"