Availability of information needed to evaluate algorithmic fairness — A systematic review of publicly accessible critical care databases
Authors' conclusion
This review demonstrates that critical care publicly available data used to train AI algorithms do not include enough information to properly look for intrinsic bias and fairness issues towards historically marginalized populations.
Study context
- Population
- Publicly accessible critical care electronic medical record (EMR) datasets used to train/validate machine learning (ML) algorithms
- Participants
- 7 publicly available databases
- Outcome
- Availability of 12 variables to identify historically marginalized populations (age, sex, gender identity, race and/or ethnicity, indigenous self-identification, payor, primary language, religion, place of residence, education, occupation, income)
Pooled result
As reported in the abstract
Availability of 12 equity-relevant variables in publicly available critical care databases
- Medical Information Mart for Intensive Care (MIMIC): variables of interest reported
- 7 of the 12
- SIVEP-Gripe: variables of interest reported
- 7
- COVID-19 Mexican Open Repository: variables of interest reported
- 4
- eICU: variables of interest reported
- 4
- Other datasets: variables of interest reported
- 2 or fewer
- Databases with sex and age
- All 7
- Databases with native or indigenous identification
- Four databases (57%)
- Databases with race and/or ethnicity
- 3 (43%)
- Databases with residence
- Two databases (29%)
- Databases with payor, language and religion
- one (14%)
- Databases with education and occupation
- One database (14%)
Original publication
Availability of information needed to evaluate algorithmic fairness — A systematic review of publicly accessible critical care databases. Anaesth Crit Care Pain Med. Published 20 May 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →