Measuring Implicit Bias in ICU Notes Using Word-Embedding Neural Network Models

BACKGROUND

Language in nonmedical data sets is known to transmit human-like biases when used in natural language processing (NLP) algorithms that can reinforce disparities. It is unclear if NLP algorithms of medical notes could lead to similar transmissions of biases.

RESEARCH QUESTION

Can we identify implicit bias in clinical notes, and are biases stable across time and geography?

STUDY DESIGN AND METHODS

To determine whether different racial and ethnic descriptors are similar contextually to stigmatizing language in ICU notes and whether these relationships are stable across time and geography, we identified notes on critically ill adults admitted to the University of California, San Francisco (UCSF), from 2012 through 2022 and to Beth Israel Deaconess Hospital (BIDMC) from 2001 through 2012. Because word meaning is derived largely from context, we trained unsupervised word-embedding algorithms to measure the similarity (cosine similarity) quantitatively of the context between a racial or ethnic descriptor (eg, African-American) and a stigmatizing target word (eg, nonco-operative) or group of words (violence, passivity, noncompliance, nonadherence).

RESULTS

In UCSF notes, Black descriptors were less likely to be similar contextually to violent words compared with White descriptors. Contrastingly, in BIDMC notes, Black descriptors were more likely to be similar contextually to violent words compared with White descriptors. The UCSF data set also showed that Black descriptors were more similar contextually to passivity and noncompliance words compared with Latinx descriptors.

INTERPRETATION

Implicit bias is identifiable in ICU notes. Racial and ethnic group descriptors carry different contextual relationships to stigmatizing words, depending on when and where notes were written. Because NLP models seem able to transmit implicit bias from training data, use of NLP algorithms in clinical prediction could reinforce disparities. Active debiasing strategies may be necessary to achieve algorithmic fairness when using language models in clinical research.

Identification of dysregulation of atrial proteins in rats with chronic obstructive apnea using two-dimensional polyacrylamide gel electrophoresis and mass spectrometry

Obstructive sleep apnea (OSA) affects an estimated 20% of adults worldwide and has been associated with electrical and structural abnormalities of the atria, although the molecular mechanisms are not well understood. Here, we used two‐dimensional polyacrylamide gel electrophoresis (2D PAGE) coupled with nanoliquid chromatography‐tandem mass spectrometry (nanoLC‐MS/MS) to investigate the proteins that are dysregulated in the atria from severe and moderate apnea when compared to control. We found enzymes involved in the glycolysis, beta‐oxidation, electron transport chain and Krebs cycle to be down‐regulated. The data suggested that the dysregulated proteins may play a role in atrial pathology developing via chronic obstructive apnea and hypoxia. Our results are consistent with our previous 1D‐PAGE and nanoLC‐MS/MS study (Channaveerappa et al, J Cell Mol Med. 2017), where we found that some aerobic and anaerobic glycolytic and Krebs cycle enzymes were down‐regulated, suggesting that apnea may be a result of paucity of oxygen and production of ATP and reducing equivalents (NADH). The 2D‐PAGE study not only complements our current study, but also advances our understanding of the OSA. The complete mass spectrometry data are available via ProteomeXchange with identifier PXD011181.

Atrial electrophysiological and molecular remodelling induced by obstructive sleep apnoea

Obstructive sleep apnoea (OSA) affects 9-24% of the adult population. OSA is associated with atrial disease, including atrial enlargement, fibrosis and arrhythmias. Despite the link between OSA and cardiac disease, the molecular changes in the heart which occur with OSA remain elusive. To study OSA-induced cardiac changes, we utilized a recently developed rat model which closely recapitulates the characteristics of OSA. Male Sprague Dawley rats, aged 50-70 days, received surgically implanted tracheal balloons which were inflated to cause transient airway obstructions. Rats were given 60 apnoeas per hour of either 13 sec. (moderate apnoea) or 23 sec. (severe apnoea), 8 hrs per day for 2 weeks. Controls received implants, but no inflations were made. Pulse oximetry measurements were taken at regular intervals, and post-apnoea ECGs were recorded. Rats had longer P wave durations and increased T wave amplitudes following chronic OSA. Proteomic analysis of the atrial tissue homogenates revealed that three of the nine enzymes in glycolysis, and two proteins related to oxidative phosphorylation, were down regulated in the severe apnoea group. Several sarcomeric and pro-hypertrophic proteins were also up regulated with OSA. Chronic OSA causes proteins changes in the atria which suggest impairment of energy metabolism and enhancement of hypertrophy.