Recent Publications by CFE Educators

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.

PURPOSE

The discovery of X-rays was followed by a variety of attempts to treat infectious diseases and various other non-cancer diseases with ionizing radiation, in addition to cancer. There has been a recent resurgence of interest in the use of such radiotherapy for non-cancer diseases. Non-cancer diseases for which use of radiotherapy has currently been proposed include refractory ventricular tachycardia, neurodegenerative diseases (e.g. Alzheimer's disease and dementia), and Coronavirus Disease 2019 (COVID-19) pneumonia, all with ongoing clinical studies that deliver radiation doses of 0.5-25 Gy in a single fraction or in multiple daily fractions. In addition to such non-cancer effects, historical indications predominantly used in some countries (e.g. Germany) include osteoarthritis and degenerative diseases of the bones and joints. This narrative review gives an overview of the biological rationale and ongoing preclinical and clinical studies for radiotherapy proposed for various non-cancer diseases, discusses the plausibility of the proposed biological rationale, and considers the long-term radiation risks of cancer and non-cancer diseases.

CONCLUSIONS

A growing body of evidence has suggested that radiation represents a double-edged sword, not only for cancer, but also for non-cancer diseases. At present, clinical evidence has shown some beneficial effects of radiotherapy for ventricular tachycardia, but there is little or no such evidence of radiotherapy for other newly proposed non-cancer diseases (e.g. Alzheimer's disease, COVID-19 pneumonia). Patients with ventricular tachycardia and COVID-19 pneumonia have thus far been treated with radiotherapy when they are an urgent life threat with no efficient alternative treatment, but some survivors may encounter a paradoxical situation where patients were rescued by radiotherapy but then get harmed by radiotherapy. Further studies are needed to justify the clinical use of radiotherapy for non-cancer diseases, and optimize dose to diseased tissue while minimizing dose to healthy tissue.

Tissue stem-progenitor cell frequency has been implicated in tumor risk and progression, but tissue-specific factors linking these associations remain ill-defined. We observed that stiff breast tissue from women with high mammographic density, who exhibit increased lifetime risk for breast cancer, associates with abundant stem-progenitor epithelial cells. Using genetically engineered mouse models of elevated integrin mechanosignaling and collagen density, syngeneic manipulations, and spheroid models, we determined that a stiff matrix and high mechanosignaling increase mammary epithelial stem-progenitor cell frequency and enhance tumor initiation in vivo. Augmented tissue mechanics expand stemness by potentiating extracellular signal-related kinase (ERK) activity to foster progesterone receptor-dependent RANK signaling. Consistently, we detected elevated phosphorylated ERK and progesterone receptors and increased levels of RANK signaling in stiff breast tissue from women with high mammographic density. The findings link fibrosis and mechanosignaling to stem-progenitor cell frequency and breast cancer risk and causally implicate epidermal growth factor receptor-ERK-dependent hormone signaling in this phenotype.

The Pooled Uranium Miners Analysis (PUMA) study is the largest uranium miners cohort with 119,709 miners, 4.3 million person-years at risk and 7754 lung cancer deaths. Excess relative rate (ERR) estimates for lung cancer mortality per unit of cumulative exposure to radon progeny in working level months (WLM) based on the PUMA study have been reported. The ERR/WLM was modified by attained age, time since exposure or age at exposure, and exposure rate. This pattern was found for the full PUMA cohort and the 1960 + sub-cohort, i.e., miners hired in 1960 or later with chronic low radon exposures and exposure rates. The aim of the present paper is to calculate the lifetime excess absolute risk (LEAR) of lung cancer mortality per WLM using the PUMA risk models, as well as risk models derived in previously published smaller uranium miner studies, some of which are included in PUMA. The same methods were applied for all risk models, i.e., relative risk projection up to 95 years of age, an exposure scenario of 2 WLM per year from age 18-64 years, and baseline mortality rates representing a mixed Euro-American-Asian population. Depending upon the choice of model, the estimated LEAR per WLM are 5.38 × 10 or 5.57 × 10 in the full PUMA cohort and 7.50 × 10 or 7.66 × 10 in the PUMA 1960 + sub-cohort, respectively. The LEAR per WLM estimates derived from risk models reported for previously published uranium miners studies range from 2.5 × 10 to 9.2 × 10. PUMA strengthens knowledge on the radon-related lung cancer LEAR, a useful way to translate models for policy purposes.