Write Like You See It
![a data distribution graph](https://ssrc.mit.edu/wp-content/uploads/2023/04/WriteLikeYouSeeIt.png)
Clinical notes are becoming an increasingly important data source for machine learning (ML) applications in healthcare. Prior research has shown that deploying ML models can perpetuate existing biases against racial minorities, as bias can be implicitly embedded in data.
In this study, we investigate the level of implicit race information available to ML models and human experts and the implications of model-detectable differences in clinical notes. Our work makes three key contributions.
First, we find that models can identify patient self-reported race from clinical notes even when the notes are stripped of explicit indicators of race.
Second, we determine that human experts are not able to accurately predict patient race from the same redacted clinical notes.
Finally, we demonstrate the potential harm of this implicit information in a simulation study, and show that models trained on these race-redacted clinical notes can still perpetuate existing biases in clinical treatment decisions.
RESEARCH AREAS
![line drawing of the cross section of a brain when looking from the top with a square with the letters "AI" in the middle of it](https://ssrc.mit.edu/wp-content/uploads/2023/11/Artificial-Intelligence.png)
Artificial Intelligence
![line drawing of a computer microchip - a box with four lines coming out of each side and a smaller box in the center with the bottom right corner angled](https://ssrc.mit.edu/wp-content/uploads/2023/11/Computation-1.png)
Computation
![line drawing of a magnifying glass with ones and zeroes inside of it](https://ssrc.mit.edu/wp-content/uploads/2023/11/Data-Science.png)
Data Science
![line drawing of two arrows pointing in different directions with a magnifying glass looking at the tail end of one of them](https://ssrc.mit.edu/wp-content/uploads/2023/11/Info-and-Decision.png)
Info & Decision
![a line drawing graphic of four people figures (one at the top, one at the right, one at the bottom, and one at the left) connected by circular lines between them](https://ssrc.mit.edu/wp-content/uploads/2023/11/InstitutionalBehavior.png)
Institutional Behavior
![line drawing of a DNA double helix](https://ssrc.mit.edu/wp-content/uploads/2023/11/Healthcare-Medicine.png)
Life Sciences
![line drawing of a head in profile with a microchip overlaying the brain area of the head](https://ssrc.mit.edu/wp-content/uploads/2023/11/Machine-Learning.png)
Machine Learning
![line drawing of a bar graph with a line graph over it](https://ssrc.mit.edu/wp-content/uploads/2023/11/Statistics.png)
Statistics
![line drawing of a light bulb with the outline of a gear around the glass part of the bulb](https://ssrc.mit.edu/wp-content/uploads/2023/11/Systems-Innovation.png)
Systems Innovation
![line drawing of a cloud in the top of the image with three lines coming down out of it going to line drawings of a phone, laptop, and tablet](https://ssrc.mit.edu/wp-content/uploads/2023/11/Systems-and-Networking.png)
Systems & Networking
IMPACT AREAS
![line drawing of a stack of computer servers connected to each other](https://ssrc.mit.edu/wp-content/uploads/2023/11/Big-Data.png)
Big Data
![line drawing of three intersecting circles, one with a dollar symbol, one with a euro symbol, one with a bitcoin symbol](https://ssrc.mit.edu/wp-content/uploads/2023/11/Finance.png)
Finance
![line drawing of a heart with a one pulse line in the middle of it](https://ssrc.mit.edu/wp-content/uploads/2023/12/Healthcare-1.png)
Healthcare / Medicine
![line drawing of a train trestle bridge with three columns underneath and an arched span across the top](https://ssrc.mit.edu/wp-content/uploads/2023/11/Infrastructure-2.png)
Infrastructure
![line drawing of three people figures (one top center, one bottom left, and one bottom right) with a circle in the center between them and lines connecting each figure to the center circle](https://ssrc.mit.edu/wp-content/uploads/2023/11/SocialNetworks.png)
Social Networks
![](https://ssrc.mit.edu/wp-content/uploads/2023/11/Urban-Systems.png)