Quantitative Evaluation of Large Language Models to Streamline Radiology Report Impressions: A Multimodal Retrospective Analysis.

Background The complex medical terminology of radiology reports may cause confusion or anxiety for patients, especially given increased access to electronic health records. Large language models (LLMs) can potentially simplify radiology report readability. Purpose To compare the performance of four publicly available LLMs (ChatGPT-3.5 and ChatGPT-4, Bard [now known as Gemini], and Bing) in producing simplified radiology report impressions. Materials and Methods In this retrospective comparative analysis of the four LLMs (accessed July 23 to July 26, 2023), the Medical Information Mart for Intensive Care (MIMIC)-IV database was used to gather 750 anonymized radiology report impressions covering a range of imaging modalities (MRI, CT, US, radiography, mammography) and anatomic regions. Three distinct prompts were employed to assess the LLMs' ability to simplify report impressions. The first prompt (prompt 1) was "Simplify this radiology report." The second prompt (prompt 2) was "I am a patient. Simplify this radiology report." The last prompt (prompt 3) was "Simplify this radiology report at the 7th grade level." Each prompt was followed by the radiology report impression and was queried once. The primary outcome was simplification as assessed by readability score. Readability was assessed using the average of four established readability indexes. The nonparametric Wilcoxon signed-rank test was applied to compare reading grade levels across LLM output. Results All four LLMs simplified radiology report impressions across all prompts tested (P < .001). Within prompts, differences were found between LLMs. Providing the context of being a patient or requesting simplification at the seventh-grade level reduced the reading grade level of output for all models and prompts (except prompt 1 to prompt 2 for ChatGPT-4) (P < .001). Conclusion Although the success of each LLM varied depending on the specific prompt wording, all four models simplified radiology report impressions across all modalities and prompts tested. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Rahsepar in this issue.

Trends and Determinants of Location of Death Due to Colorectal Cancer in the United States : A Nationwide Study.

BACKGROUND: Colorectal cancer (CRC) is the second leading cause of cancer-related mortality in the United States (US); however, there are limited data on location of death in patients who die from CRC. We examined the trends in location of death and determinants in patients dying from CRC in the US. METHODS: We utilized the Centers for Disease Control and Prevention Wide-Ranging Online Data for Epidemiologic Research database to extract nationwide data on underlying cause of death as CRC. A multinomial logistic regression was performed to assess associations between clinico-sociodemographic characteristics and location of death. RESULTS: There were 850,750 deaths due to CRC from 2003 to 2019. There was a gradual decrease in deaths in hospital, nursing home, or outpatient facility/emergency department over time and an increase in deaths at home and in hospice. Relative to White decedents, Black, Asian, and American Indian/Alaska Native decedents were less likely to die at home and in hospice compared with hospitals. Individuals with lower educational status also had a lower risk of dying at home or in hospice compared with in hospitals. CONCLUSIONS: The gradual shift in location of death of patients who die of CRC from institutionalized settings to home and hospice is a promising trend and reflects the prioritization of patient goals for end-of-life care by healthcare providers. However, there are existing sociodemographic disparities in access to deaths at home and in hospice, which emphasizes the need for policy interventions to reduce health inequity in end-of-life care for CRC.

Contemporary Patterns of Financial Toxicity Among Patients With Rheumatologic Disease in the United States.

BACKGROUND/OBJECTIVE: Rheumatologic diseases encompass a group of disabling conditions that often require expensive clinical treatments and limit an individual's ability to work and maintain a steady income. The purpose of this study was to evaluate contemporary patterns of financial toxicity among patients with rheumatologic disease and assess for any associated demographic factors. METHODS: The cross-sectional National Health Interview Survey was queried from 2013 to 2018 for patients with rheumatologic disease. Patient demographics and self-reported financial metrics were collected or calculated including financial hardship from medical bills, financial distress, food insecurity, and cost-related medication (CRM) nonadherence. Multivariable logistic regressions were used to assess for factors associated with increased financial hardship. RESULTS: During the study period, 20.2% of 41,502 patients with rheumatologic disease faced some degree of financial hardship due to medical bills, 55.0% of whom could not pay those bills. Rheumatologic disease was associated with higher odds of financial hardship from medical bills (adjusted odds ratio, 1.29; 95% confidence interval, 1.22-1.36; p < 0.001) with similar trends for patients suffering from financial distress, food insecurity, and CRM nonadherence (p < 0.001 for all). Financial hardship among patients with rheumatologic disease was associated with being younger, male, Black, and uninsured (p < 0.001 for all). CONCLUSION: In this nationally representative study, we found that a substantial proportion of adults with rheumatologic disease in the United States struggled with paying their medical bills and suffered from food insecurity and CRM nonadherence. National health care efforts and guided public policy should be pursued to help ease the burden of financial hardship for these patients.

Colon Cancer Disparities in Stage at Presentation and Time to Surgery for Asian Americans, Native Hawaiians, and Pacific Islanders: A Study with Disaggregated Ethnic Groups.

BACKGROUND: Vast differences in barriers to care exist among Asian American, Native Hawaiian, and Pacific Islander (AANHPI) groups and may manifest as disparities in stage at presentation and access to treatment. Thus, we characterized AANHPI patients with stage 0-IV colon cancer and examined differences in (1) stage at presentation and (2) time to surgery relative to white patients. PATIENTS AND METHODS: We assessed all patients in the National Cancer Database (NCDB) with stage 0-IV colon cancer from 2004 to 2016 who identified as white, Chinese, Japanese, Filipino, Native Hawaiian, Korean, Vietnamese, Laotian, Hmong, Kampuchean, Thai, Asian Indian or Pakistani, and Pacific Islander. Multivariable ordinal logistic regression defined adjusted odds ratios (AORs), with 95% confidence intervals (CI), of (1) patients presenting with advanced stage colon cancer and (2) patients with stage 0-III colon cancer receiving surgery at ≥ 60 days versus 30-59 days versus < 30 days postdiagnosis, adjusting for sociodemographic/clinical factors. RESULTS: Among 694,876 patients, Japanese [AOR 1.08 (95% CI 1.01-1.15), p < 0.05], Filipino [AOR 1.17 (95% CI 1.09-1.25), p < 0.001], Korean [AOR 1.09 (95% CI 1.01-1.18), p < 0.05], Laotian [AOR 1.51 (95% CI 1.17-1.95), p < 0.01], Kampuchean [AOR 1.33 (95% CI 1.04-1.70), p < 0.01], Thai [AOR 1.60 (95% CI 1.22-2.10), p = 0.001], and Pacific Islander [AOR 1.41 (95% CI 1.20-1.67), p < 0.001] patients were more likely to present with more advanced colon cancer compared with white patients. Chinese [AOR 1.27 (95% CI 1.17-1.38), p < 0.001], Japanese [AOR 1.23 (95% CI 1.10-1.37], p < 0.001], Filipino [AOR 1.36 (95% CI 1.22-1.52), p < 0.001], Korean [AOR 1.16 (95% CI 1.02-1.32), p < 0.05], and Vietnamese [AOR 1.55 (95% CI 1.36-1.77), p < 0.001] patients were more likely to experience greater time to surgery than white patients. Disparities persisted when comparing among AANHPI subgroups. CONCLUSIONS: Our findings reveal key disparities in stage at presentation and time to surgery by race/ethnicity among AANHPI subgroups. Heterogeneity upon disaggregation underscores the importance of examining and addressing access barriers and clinical disparities.

Diminishing Basic Science Research Experience Among United States Cardiothoracic Surgery Trainees.

INTRODUCTION: There is growing concern regarding the attrition of surgeon-scientists. To understand the decline of basic science research (BSR), it is essential to examine trends in research conducted by trainees. We hypothesized that, over recent decades, cardiothoracic (CT) surgery trainees have published fewer BSR articles. MATERIALS AND METHODS: CT surgeons at United States training institutions in 2020 who completed training in the past three decades, excluding international trainees, were analyzed (1991-2000: n = 148; 2001-2010: n = 228; 2011-2020: n = 247). Publication records were obtained from Scopus. Articles with medical subject heading terms involving molecular/cellular or animal research were classified as BSR using the National Institutes of Health iCite Translation module. Data were analyzed using Fisher's exact test or the Wilcoxon rank-sum test. RESULTS: While the proportion of surgeons who published a first-author paper during training remained stable over the past two decades (178/228 [78.1%] versus 189/247 [76.5%], P = 0.7427), the proportion who published a first-author BSR paper decreased significantly (135/228 [59.2%] versus 96/247 [38.9%], P < 0.0001). Among surgeons who published a first-author paper in training, the total papers published by each trainee did not change over the past two decades (3.5 versus 3.3 first-author papers per 10 y of training, P = 0.8819). However, the number of BSR papers published during training decreased significantly (1.7 versus 0.8 first-author papers per 10 y of training, P < 0.0001). CONCLUSIONS: CT surgery trainees are publishing fewer BSR papers. Additional efforts are needed to increase exposure of trainees to BSR and reaffirm that BSR is a valuable and worthwhile pursuit for academic surgeons.

Analyzing the Scholarly Impact of Cardiothoracic Surgery Research Using the Relative Citation Ratio.

INTRODUCTION: The National Institutes of Health (NIH) recently developed the relative citation ratio (RCR), calculated as article citations benchmarked to NIH-funded publications in the same field. Here, we characterized the scholarly impact of academic cardiothoracic (CT) surgeons and their research using the RCR. MATERIALS AND METHODS: Using a database of 992 CT surgeons, we calculated the RCR for all articles published by each surgeon since 1980 using the NIH iCite database. All data were collected from publicly available online sources. Data are presented as median (interquartile range) or as odds ratios (ORs) for multivariable logistic regression analysis. RESULTS: Where RCR 1.00 indicates equal impact as an NIH-funded publication, the RCR among all 37,402 CT surgery articles was 0.84 (0.33-1.83) and the RCR among NIH-funded CT surgery articles was 1.07 (0.53-2.17). CT surgeons exhibited a career median RCR of 0.82 (0.54-1.13) and maximum RCR of 6.20 (3.04-13.57). Predictors of career median RCR >1.00 included female gender (OR 2.23, P = 0.001), thoracic subspecialization (OR 2.50, P < 0.001), full professor rank (OR 1.89, P = 0.001), and NIH funding (OR 1.75, P = 0.001). Predictors of career maximum RCR >50th percentile among CT surgeons included male gender (OR 1.87, P = 0.030), thoracic subspecialization (OR 2.05, P < 0.001), full professor rank (OR 4.89, P < 0.001), NIH funding (OR 3.17, P < 0.001), and career duration (OR 1.03, P = 0.002). CONCLUSIONS: We present the first assessment of the NIH-validated RCR for academic CT surgery. CT surgery research is highly impactful, although gender disparities persist with respect to the highest-impact research of our specialty.

Impact of PhD Degree Versus Non-PhD Research Fellowship on Future Research Productivity Among Academic Cardiothoracic Surgeons.

BACKGROUND: A PhD degree can offer significant research experience, but previous studies yielded conflicting conclusions on the relationship between a PhD degree and future research output. We compared the impact of a PhD degree versus research fellowship (RF) training on research productivity in cardiothoracic surgeons, hypothesizing that training pathways may influence potential associations. METHODS: CT surgeons practicing at all accredited United States CT surgery training programs in 2018 who pursued dedicated time for research (n = 597), including earning a PhD degree (n = 92) or completing a non-PhD RF (n = 505), were included. To control for training pathways, we performed subanalyses of U.S. medical school graduates (n = 466) and international medical school graduates (IMGs) (n = 131). Surgeon-specific data were obtained from publicly available sources (e.g., institutional webpages, Scopus). RESULTS: PhD surgeons published greater total papers (68.5 vs. 52.0, p = 0.0179) and total papers per year as an attending (4.6 vs. 3.0, p = 0.0150). For U.S. medical school graduates, there were 40 PhD surgeons and 426 non-PhD RF surgeons; both groups published a similar number of total papers (64.5 vs. 54.0, p = 0.3738) and total papers per year (3.2 vs. 3.0, p = 0.7909). For IMGs, there were 52 PhD surgeons and 79 non-PhD RF surgeons; the PhD surgeons published greater total papers (80.5 vs. 45.0, p = 0.0101) and total papers per year (5.7 vs. 2.7, p = 0.0037). CONCLUSION: CT surgeons with dedicated research training are highly academically productive. Although a PhD degree may be associated with enhanced career-long research productivity for IMGs, this association was not observed for U.S. medical school graduates.

Weight gain after in vitro fertilization: a potential consequence of controlled ovarian stimulation.

In the USA, 42% of adult women were estimated to have obesity, and 13% of women of childbearing age similarly have impaired fecundity. Obesity is associated with infertility such that patients with obesity often seek out in vitro fertilization (IVF) services. Here, we report on the case of a woman with childhood-onset class II obesity who had been undergoing treatment with phentermine and topiramate prior to undergoing 3 cycles of IVF. With each cycle, the patient temporarily gained 13-15 lbs. during controlled ovarian stimulation (COS). Weight gain from COS may be clinically relevant and merits further study to optimize weight status across women's reproductive life and to better assist women who gain weight secondary to IVF. Incorporating weight monitoring into IVF protocols may also help better characterize the scope of weight gain from COS.

Characterization of Cardiothoracic Surgeons Actively Leading Basic Science Research.

BACKGROUND: There is increasing concern regarding the attrition of surgeon-scientists in cardiothoracic (CT) surgery. However, the characteristics of CT surgeons who are actively leading basic science research (BSR) have not been examined. We hypothesized that early exposure to BSR during training and active grant funding are important factors that facilitate the pursuit of BSR among practicing CT surgeons. MATERIALS AND METHODS: We created a database of 992 CT surgeons listed as faculty at accredited United States CT surgery teaching hospitals in 2018. Data regarding each surgeon's training/professional history, publication record, and National Institutes of Health funding were acquired from publicly available online sources. Surgeons who published at least one first- or last-author paper in 2017-2018 were considered to be active, lead researchers. RESULTS: Of the 992 CT surgeons, 73 (7.4%) were actively leading BSR, and 599 (60.4%) were actively leading only non-BSR. Only 2 women were actively leading BSR. Surgeons actively leading BSR were more likely to have earned a PhD degree (20.5% versus 9.7%, P = 0.0049), and more likely to have published a first-author BSR paper during training (76.7% versus 40.9%, P< 0.0001). Surgeons actively leading BSR were also more likely to have an active National Institutes of Health grant (34.2% versus 5.8%, P< 0.0001), especially an R01 grant (21.9% versus 2.5%, P< 0.0001). CONCLUSIONS: A small minority of CT surgeons at academic training hospitals are actively leading BSR. In order to facilitate the development of surgeon-scientists, additional support must be given to trainees and junior faculty, especially women, to enable early engagement in BSR.

Career Research Productivity Correlates With Medical School Ranking Among Cardiothoracic Surgeons.

BACKGROUND: The foundation for a successful academic surgical career begins in medical school. We examined whether attending a top-ranked medical school is correlated with enhanced research productivity and faster career advancement among academic cardiothoracic (CT) surgeons. MATERIALS AND METHODS: Research profiles and professional histories were obtained from publicly available sources for all CT surgery faculty at accredited US CT surgery teaching hospitals in 2018 (n = 992). We focused on surgeons who completed medical school in the United States during or after 1990, the first-year US News & World Report released its annual medical school research rankings (n = 451). Subanalyses focused on surgeons who completed a research fellowship (n = 299) and those who did not (n = 152). RESULTS: A total of 124 surgeons (27.5%) attended a US News & World Report top 10 medical school, whereas 327 (72.5%) did not. Surgeons who studied at a top 10 medical school published more articles per year as an attending surgeon (3.2 versus 1.9; P < 0.0001), leading to more total publications (51.5 versus 27.0; P < 0.0001) and a higher H-index (16.0 versus 11.0; P < 0.0001) over a similar career duration (11.0 versus 10.0 y; P = 0.1294). These differences in career-long research productivity were statistically significant regardless of whether the surgeons completed a research fellowship or not. The surgeons in both groups, however, required a similar number of years to reach associate professor rank (P = 0.6993) and full professor rank (P = 0.7811) after starting their first attending job. CONCLUSIONS: Attending a top-ranked medical school is associated with enhanced future research productivity but not with faster career advancement in academic CT surgery.

Increasing Voter Participation Through Health Care-Based Voter Registration.

Importance: Young people and historically marginalized racial and ethnic groups are poorly represented in the democratic process. Addressing voting inequities can make policy more responsive to the needs of these communities. Objective: To assess whether leveraging health care settings as venues for voter registration and mobilization is useful, particularly for historically underrepresented populations in elections. Design, Setting, and Participants: In 2020, nonpartisan nonprofit Vot-ER partnered with health care professionals and institutions to register people to vote. This cross-sectional study analyzed the demographics and voting behavior of people mobilized to register to vote in health care settings, including hospitals, community health centers, and medical schools across the US. The age and racial and ethnic identity data of individuals engaged through Vot-ER were compared to 2 national surveys of US adults, including the 2020 Cooperative Election Study (CES) and the 2020 American National Election Study (ANES). Exposure: Health care-based voter registration. Main Outcomes and Measures: The main outcomes were age composition, racial and ethnic composition, and voting history. Results: Of the 12 441 voters contacted in health care settings, 41.9% were aged 18 to 29 years, 15.9% were identified as African American, 9.6% as Asian, 12.7% as Hispanic, and 60.4% as White. This distribution was significantly more diverse than the racial and ethnic distribution of the ANES (N = 5447) and CES (N = 39 014) samples, of which 72.5% and 71.19% self-identified as White, respectively. Voter turnout among health care-based contacts increased from 61.0% in 2016 to 79.8% in 2020, a turnout gain (18.8-percentage point gain) that was 7.7 percentage points higher than that of the ANES sample (11.1-percentage point gain). Demographically, the age distribution of voters contacted in health care settings was significantly different from the ANES and CES samples, with approximately double the proportion of young voters aged 18 to 29 years. Conclusion and Relevance: This cross-sectional study suggests that health care-based voter mobilization reaches a distinctly younger and more racially and ethnically diverse population relative to those who reported contact from political campaigns. This analysis of the largest health care-based voter mobilization effort points to the unique impact that medical professionals may have on voter registration and turnout in the 2024 US elections. In the long term, health equity initiatives should prioritize expanding voting access to address the upstream determinants of health in historically marginalized communities.