OHDSI Health Equity Workgroup
We aim to engage critically and intentionally in all of our work, considering not only the results but the potential interpretation and impact of results, steering clear of work that reinforces health disparities and misinterpretations that generate stigma, and lifting up work which is likely to contribute to health equity.
Objectives and Key Results
Generate and disseminate real-world evidence about the substantial public health issue of health inequities
- 3 fully-reproducible study packages executed across at least 20 OHDSI data partners
- 5 publications accepted in peer-reviewed journals
- 7 instances of presentations of our work
- 1 uses of OHDSI results by internal or external stakeholders that demonstrate influence in policy or clinical decision-making
Operationalize individual-level Social Determinants of health, Risk factors, and Needs (SDRN), and other data elements relevant to health equit work in OHDSI network studies [1]
- Identify OHDSI sites that are collecting SDRN and gather a report of what exists in those source systems, along with the maturity of collection and standardization. Publish a catalog.
- Identify 3 priority research questions with actionable results where individual-level SDRN is needed
- Provide recommendations for mapping and storing relevant SDRN data elements
- Release tools to assess and record data quality, gaps, and biases for SDRN data collection
- Engage with NLP team to release tools/methods for extracting SDRN from notes
- 1 validation / methods study, evaluating the use of individual-SDRN in the context of a network study.
Operationalize place-based public data sources in OHDSI network studies
- Identify 3 external datasets useful for incorporation in health equity studies
- Identify a priority research question with actionable results that requires linking group-level data sources to OMOP data
- Identify a priority use case for rolling up individual-level OMOP data to describe spatial-population-level properties
- Release a study package using OHDSI GIS tools
Extend OHDSI tools to make a health equity perspective the default and/or an option
- Augment Patient Level Prediction (PLP) to expose the differences of predictions, predictive power, and other fairness metrics of the predictive models it creates.
- Implement fairness metrics as part of phenotype evaluation
- Develop a framework for best practices in health equity across OHDSI study design / a guide on how to use the developed extensions.
Engage the broader community on issues related to health equity
- Update directory of accessible educational resources and research relevant to health equity
- Continue health equity reading group / journal club
- Invite 8 presentations from external groups in our meeting (ideas Fairness and Bias group in N3C (N3C SDoH Domain Team, HL7 Gravity, other OHDSI workgroups, RADx-UP, SIREN, Multi-stakeholder engaged groups, community organizations, NACHC / AAPCHO)
- Create a directory of membership and external group affiliations
-
Support the work of the group
- Create a boilerplate description of assets of this group to be useful for creating quick responses to grants
- 3 collaborative grant applications
- 1 funded collaborative grant
- 1 Health Equity hackathon, with at least 10 participants
Accessible educational resources on health equity
This document contains a few readings, lectures, papers, and content that might otherwise be helpful for people to understand more about health equity research. This list will grow and change - feel free to suggest edits!
Good introductions to health equity and fairness research:
A lecture by Pilar Ossorio at MLHC Professor of Law and Bioethics at the University of Wisconsin Law School that summarizes an overview of what is equity, fairness, and how these concepts apply to healthcare
For those that are text-based learners, this paper by Dunkelau and Leuschel in 2019 summarizes Fairness-aware Machine Learning, available online.
A smattering of research papers, recent news, and readings:
- Dissecting racial bias in an algorithm used to manage the health of populations - Science, 2019. Available here. Fair ML Keynote talk + slides available here. This is Obermeyer’s paper that was much talked-about in the press about how there were differences in outcomes across Black/white patients from the Optum health-risk algorithm that could potentially lead to differences in healthcare treatment. The major takeaway was to adjust outcome variables away from using cost as a metric.
- How scientists are subtracting race from medical risk calculators - Science, 2021. Available here. As the title suggests, the article summarizes the effort to now remove race from medical risk calculations. The authors provide a good summary of how race has previously been used in medicine [for good, or for scientifically-inappropriate measurements like the eGFR calculation].
- Fairness through awareness - 2011. Available here. This is an old-ish paper in computer science that summarizes how being aware of a protected class can lead to differences in fairness outcomes; the seminal paper that summarizes
Various books that discuss the implications of algorithms trained on big data:
- Race After Technology, by Ruha Benjamin [Professor of African American studies at Princeton University], summarizes how technology [from data collection, data imputation, government policy, etc] can play a role in different outcomes in society.
- Weapons of Math Destruction, by Cathy O’Neil [mathematician, data scientist, author] discusses the implications of algorithms and how big data can reinforce pre-existing inequities.
Questions? Comments? Additional materials?
Reach out to the OHDSI Health Equity Working Group. You can join the listserv and Teams environment by filling out this form.