Back

Half the picture: Word frequencies reveal racial differences in clinical documentation, but not their causes

Penn, J. A.; Newman-Griffis, D. R.

2022-02-13 health informatics
10.1101/2022.02.10.22270631 medRxiv
Show abstract

Clinical notes are the best record of a providers perceptions of their patients, but their use in studying racial bias in clinical documentation has typically been limited to manual evaluation of small datasets. We investigated the use of computational methods to scale these insights to large, heterogeneous clinical text data. We found significant differences in negative emotional tone and language implying social dominance in clinical notes between Black and White patients, but identified multiple contributing factors in addition to potential provider bias, including mis-categorization of some healthcare vocabulary as emotion-related. We further found that notes for Black patients were significantly less likely to mention opioids than for White patients, potentially reflecting both inequitable access to medication and provider bias. Our analysis showed that computational tools have significant potential for studying racial bias in large clinical corpora, and identified key challenges to providing a nuanced analysis of bias in clinical documentation.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.