United States

### Background
As large language models (LLMs) gain multimodal capabilities to interpret both images and text, their potential to assist in clinical decision-making continues to grow. While prior research has shown LLMs can match and even outperform individual physicians on standardized exams, little is known about their performance relative to collective human reasoning where varied perspectives contribute to more accurate conclusions. This study evaluates the performance of state-of-the-art multimodal LLMs in solving complex medical challenges compared to both individual and aggregate responses from a large pool of medically literate readers.

### Methods
We compiled 100 ophthalmology-related cases from the New England Journal of Medicine (NEJM) Image Challenge archive (2006–2024), each containing a clinical vignette, image, and multiple-choice question. Six multimodal LLMs (GPT-4 Turbo, GPT-4o, GPT-4o mini, Gemini 1.5 Flash, Gemini 1.5 Pro, and Claude 3.5 Haiku) were prompted to answer each case and self-report confidence (1–4 scale). Human performance was assessed using (1) the average accuracy of NEJM respondents and (2) the accuracy of the most commonly selected (modal) answer per case, representing collective judgment. LLM accuracy and confidence were compared using Kruskal-Wallis and post hoc Dunn’s tests (p &lt; 0.05 considered significant).

### Results
All LLMs except Claude 3.5 Haiku outperformed the average individual respondent (mean human accuracy: 50.4%, 95% CI: 40.8–60.0%), with GPT-4o achieving the highest model accuracy at 72.0% (95% CI: 66.7–76.8%). However, the collective human response—the most frequently selected answer—was correct in 93.0% of cases (95% CI: 86.3–96.6%), significantly outperforming every LLM. The most popular response across LLMs (i.e., the answer most frequently selected by the models) was correct in 65.3% of cases (95% CI: 56.0–74.7), significantly lower than the collective human accuracy. Correct answers were associated with higher confidence scores for all LLMs except Gemini Flash, suggesting self-rated confidence may be a useful proxy for model reliability.

### Conclusion
Multimodal LLMs can outperform individual human respondents on image-based diagnostic tasks but fall short of the accuracy achieved by collective human reasoning. This underscores a key strength of human cognition: the diversity of thought across individuals produces more accurate outcomes than any single decision-maker, human or machine. In contrast, LLMs often rely on a uniform reasoning process that can propagate the same errors consistently. While LLMs hold promise as diagnostic aids, further refinement may be required for them to match the collective intelligence of human experts in interpreting clinical vignettes.

AMA Research Challenge 2025 

Accuracy of Multimodal Large Language Models vs. Human Consensus in Interpreting Clinical Vignettes

Welcome to the **largest national, multi-specialty research** event featuring research from more than 750 medical students, residents, fellows, and international medical graduates.

Use the buttons below or navigation menu to access the event.
<br><br><br>

[![](https://assets.underline.io/markdown_image/1/image/20e1e77aa954efeac7a15d84c85eb383.jpg)](https://underline.io/events/468/sessions?searchGroup=all)
<br>
[![](https://assets.underline.io/markdown_image/1/image/068522d81aca0b68a369787c7dfccd1b.jpg)](https://underline.io/events/468/posters?searchGroup=all)
<br>
[![](https://assets.underline.io/markdown_image/1/image/d3971ece79d5e5575b178bdc2c8bbfae.jpg)](https://underline.io/events/468/help-desk)
<br><br>

#### Event Details

The 2024 American Medical Association Research Challenge includes a poster symposium hall and a semifinals hall with the top scored research. **Log in at any time throughout the event to:**

* **View posters and presentations** on a variety of topics and specialties
* **Score posters in the semifinals** to help decide the top five finalists (AMA member exclusive opportunity)
* **Connect with participants** and provide feedback to help advance their research

At the conclusion of this event, the AMA will announce the top five finalists who will advance to the AMA Research Challenge finals, present their research to an elite panel of judges, and compete to win a $10,000 grand prize presented by Laurel Road.<br><br>

### Grand Prize Sponsor

AMA would like to thank Laurel Road for sponsoring a $10,000 grand prize for the winner of the AMA Research Challenge. Laurel Road is not involved in winner selection process. The winner will be announced on February 20, 2025 during the AMA Research Challenge finals. Payment of the grand prize will be directly from Laurel Road. See the AMA Research Challenge Official Rules for terms and conditions.

[![](https://assets.underline.io/uploads/markdown_image/1/image/611e331d4a0eba0920f240cfc04b4d77.png)](https://www.laurelroad.com/)

<br><br>

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

**There is no registration fee for this event! **

If you have used the AMA Research Challenge platform previously, please click the ‘Log-in’ button below. If you do not remember your password, please click “forgot password” and follow prompts to reset your password. After resetting your password, please refresh your webpage to access the platform.

Register to explore research in a variety of topics. AMA members get exclusive access to score posters in the semifinals. 

**Please use this link [here](https://www.ama-assn.org/member-benefits/events/ama-research-challenge) to register for free.**

Registration Is Required

AMA Research Challenge 2024 

Join the AMA Research Challenge poster symposium and semifinals to explore top research posters and score posters in the semifinals (an AMA member exclusive opportunity).

### Background
As large language models (LLMs) gain multimodal capabilities to interpret both images and text, their potential to assist in clinical decision-making continues to grow. While prior research has shown LLMs can match and even outperform individual physicians on standardized exams, little is known about their performance relative to collective human reasoning where varied perspectives contribute to more accurate conclusions. This study evaluates the performance of state-of-the-art multimodal LLMs in solving complex medical challenges compared to both individual and aggregate responses from a large pool of medically literate readers.

### Methods
We compiled 100 ophthalmology-related cases from the New England Journal of Medicine (NEJM) Image Challenge archive (2006–2024), each containing a clinical vignette, image, and multiple-choice question. Six multimodal LLMs (GPT-4 Turbo, GPT-4o, GPT-4o mini, Gemini 1.5 Flash, Gemini 1.5 Pro, and Claude 3.5 Haiku) were prompted to answer each case and self-report confidence (1–4 scale). Human performance was assessed using (1) the average accuracy of NEJM respondents and (2) the accuracy of the most commonly selected (modal) answer per case, representing collective judgment. LLM accuracy and confidence were compared using Kruskal-Wallis and post hoc Dunn’s tests (p < 0.05 considered significant).

### Results
All LLMs except Claude 3.5 Haiku outperformed the average individual respondent (mean human accuracy: 50.4%, 95% CI: 40.8–60.0%), with GPT-4o achieving the highest model accuracy at 72.0% (95% CI: 66.7–76.8%). However, the collective human response—the most frequently selected answer—was correct in 93.0% of cases (95% CI: 86.3–96.6%), significantly outperforming every LLM. The most popular response across LLMs (i.e., the answer most frequently selected by the models) was correct in 65.3% of cases (95% CI: 56.0–74.7), significantly lower than the collective human accuracy. Correct answers were associated with higher confidence scores for all LLMs except Gemini Flash, suggesting self-rated confidence may be a useful proxy for model reliability.

### Conclusion
Multimodal LLMs can outperform individual human respondents on image-based diagnostic tasks but fall short of the accuracy achieved by collective human reasoning. This underscores a key strength of human cognition: the diversity of thought across individuals produces more accurate outcomes than any single decision-maker, human or machine. In contrast, LLMs often rely on a uniform reasoning process that can propagate the same errors consistently. While LLMs hold promise as diagnostic aids, further refinement may be required for them to match the collective intelligence of human experts in interpreting clinical vignettes.

technical paper

Congratulations to all the researchers who have qualified for the AMA Research Challenge virtual poster symposium and semifinals!

Welcome to the **largest national, multi-specialty research event** featuring research from nearly 1,000 medical students, residents, fellows, and international medical graduates across six topic areas - basic science, clinical and translational research, clinical vignettes, health systems science, medical education, and public health and health policy.

#### Access the Event
Use the buttons below or navigation menu to access the event. 

[![](https://assets.underline.io/markdown_image/1/image/d4076f2146939317a29eb9f69c53ad41.png)](https://underline.io/events/506/sessions?searchGroup=all)
[![](https://assets.underline.io/markdown_image/1/image/6bcee0b523a4874c4292ca0415d88dde.png)](https://underline.io/events/506/posters?searchGroup=all)

#### Event Details
The 2025 AMA Research Challenge includes the following symposium halls:
* **Semifinals hall:** Features research from 54 medical students, residents and international medical graduates that received top scores during first round scoring. 
* **Poster symposium hall:** Features research from nearly 1,000 medical students, residents and international medical graduates categorized by topics and subtopics.

**Here’s how to participate:**
* **View posters and presentations** on a variety of topics and specialties
* **Score posters in the semifinals** to help decide the top five finalists (AMA member exclusive opportunity)
* **Connect with presenters** and provide feedback to help advance their research
* **New this year: Watch live poster presentations** by topic area and engage directly with researchers

At the conclusion of this event, the AMA announced the top five finalists who advanced to the AMA Research Challenge final round, presented their research to an elite panel of judges, and competed to win a $10,000 grand prize presented by Laurel Road.

**Addison Shenk, a fourth-year medical student from Edward Via Virginia College of Osteopathic Medicine, was the winner of the 2025 AMA Research Challenge and was awarded the grand prize of $10,000, presented by Laurel Road.**

#### A message from Dr. Bobby Mukkamala, AMA President
<div style="position: relative; width: 100%; padding-bottom: 56.25%; height: 0; overflow: hidden;">
  <iframe 
    src="https://underline.io/embed/136894-ama-video?t=0"
    title="AMA Welcome Video"
    allowfullscreen
    style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border: 0;">
  </iframe>
</div>


#### Recognized Institutions
In 2025, 1,400 abstracts were submitted across six topic areas representing 174 medical schools and more than 150 residency and health care institutions. [See if your institution is represented this year!](https://drive.google.com/file/d/1eFrTqCmCeobx9pr5drJK2hllvsIpFNOY/view?usp=drive_link)

#### Grand Prize Sponsor
AMA would like to thank Laurel Road, a brand of KeyBank, for sponsoring a $10,000 grand prize for the winner of the AMA Research Challenge. Laurel Road and KeyBank are not involved in winner selection process. The winner will be announced during the AMA Research Challenge final round. Payment of the grand prize will be directly from Laurel Road/KeyBank. See the [AMA Research Challenge Official Rules for terms and conditions](https://www.ama-assn.org/system/files/research-challenge-official-rules.pdf).

[![](https://assets.underline.io/uploads/markdown_image/1/image/611e331d4a0eba0920f240cfc04b4d77.png)](https://www.laurelroad.com/)

#### Code of Conduct
<html>
AMA has a code of conduct for its meetings – we ask all attendees to adhere to this code to participate in this year’s meeting. Please read and <a href="https://www.ama-assn.org/about/code-conduct" target="_blank">learn more here</a> about your conduct requirements.
<br><br>
	</html>
	

#### Semifinals Judges

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Need help? Contact us at <support@underline.io>.

If you have used the AMA Research Challenge platform previously, please click the ‘Log-in’ button below. If you do not remember your password, please click “forgot password” and follow prompts to reset your password. After resetting your password, please refresh your webpage to access the platform. 

Register to explore research in a variety of topics. AMA members get exclusive access to score posters in the semifinals. 

**Please use [this link here](https://www.ama-assn.org/member-benefits/events/ama-research-challenge) to register for free. **

### Background
Infertility is a rising health concern that affects approximately 13% of reproductive age women, but advancements in treatment are limited by our understanding of endometrial function. Decidualization is the process of endometrial stromal cell (eSC) differentiation from a fibroblastic state to a secretory state that occurs during the mid-secretory phase of the menstrual cycle and is associated with a dramatic influx of uterine natural killer (uNK) cells to the endometrium. Defects in decidualization and uNK cell function are associated with infertility, implantation failure, and pregnancy loss, but mechanisms of eSC-uNK cell interaction are not well defined. In this study, we developed and validated uNK cell isolation and expansion techniques using menstrual effluent (ME) as a source of endometrial tissue, then used these ME-derived uNK cells to investigate interactions between uNK cells and decidualizing eSCs to further understand cellular processes occurring prior to embryo implantation.

### Methods
ME samples were collected from participants of the Research OutSmarts Endometriosis (ROSE) study at the Feinstein Institutes for Medical Research. uNK cells were purified by sequential magnetic isolations and expanded in vitro using IL-15 and IL-2. These ME-derived uNK cells were transcriptionally verified by scRNAseq and compared to reference datasets. For co-culture experiments, eSC decidualization was induced by cAMP + MPA. Interactions between uNK cells and decidualizing eSCs were characterized by live-imaging microscopy, eSC staining, and scRNAseq. Functional mechanisms of uNK-eSC interaction were described by protein and gene expression using ELISA and qPCR analyses.

### Results
This novel uNK cell isolation method achieved a purity of up to 92% CD56bright CD16- uNK cells. Cultured ME-derived uNK cells exhibited up to 30-fold expansion, maintained viability for approximately 30 days, and retained phenotypes of reference decidua-derived uNK cells as assessed by flow cytometry and single cell RNA sequencing. In an ME-derived model of uNK-eSC interaction, uNK cells mediated contact-dependent eSC remodeling during decidualization as determined by live-imaging microscopy and cell density analysis. uNK cells concurrently stimulated a pro-inflammatory eSC response in part via IFN release as determined by transcriptomic and proteomic analysis.

### Conclusion
These results indicate that ME is a viable, non-invasive source of uNK cells that can be used for downstream analyses and suggest that uNK cells mediate both eSC remodeling and inflammation during decidualization. Interactions between uNK cells and eSCs are critical for endometrial receptivity, and this novel ME-derived system greatly expands the available tools to investigate mechanisms contributing to endometrial homeostasis and female fertility.

Waste to Wonder: Understanding endometrial interactions using a menstrual-effluent derived model

### Background
Veterans of the armed forces are the embodiment of our nation's security and freedom. The health challenges they face are often exacerbated by service-related issues including psychological trauma, excess repetitive noise, radiation exposure, and subjection to dangerous chemicals from Agent Orange, burn pit smoke, water contamination, asbestos, and more. [ref PMID: 34070145] Veterans experience a higher prevalence of complex health conditions, including increased risk of heart and lung diseases, diabetes mellitus, and various cancers. [ref PMID: 39696828] Providing high-quality equitable care to veterans requires targeted healthcare strategies that address their specific needs.

### Methods
To enhance the quality of care provided to our veteran population at Geisinger, we integrated inquiries about military service and health requirements into an 8-question “About Me” questionnaire. This initiative aimed to establish a clearer connection between veterans and the medical services they require. Utilizing the SlicerDicer® tool in the Epic electronic health record system, we analyzed our veteran population to identify prevalent health risks with the goal of using these data to develop targeted care programs.

### Results
Veterans (n=35,648) had higher overall risk scores for various health conditions compared to the non-veteran population (n=507,773). Specifically, the risk ratios for chronic obstructive pulmonary disease (COPD) were 2.09, for hypertension 1.60, for heart failure 2.47, for diabetes 1.74, and for other cancers 2.19. Among Vietnam veterans (n=5,441), the risk ratios for certain conditions indicated increased vulnerabilities: hypertension at 1.79, type 2 diabetes at 1.99, prostate cancer at 2.89, skin cancer at 2.44, and atherosclerosis at 2.18. Gulf War veterans (1990-present, n=7,586) showed significantly higher relative risks for post-traumatic stress disorder (PTSD) at 7.00, COPD at 3.55, hypertension at 2.84, asthma at 1.58, and lung cancer at 2.14.

### Conclusion
Our findings highlight the critical need for tailored healthcare initiatives for Vietnam and Gulf War veterans, who are the most at-risk populations among veterans. These findings support the development of specialized screening, prevention, and treatment protocols that recognize both the physical and psychological effects of military service as seen in previous studies. This information will be used to create a value-based care guide aimed at optimizing care for our veteran population, as many utilize community healthcare facilities outside of the VA, and this fills an urgent gap in educating healthcare teams about veteran health needs.

Leveraging Patient-Provided Data in the Electronic Health Record to Improve Care for Veterans

### Background
The Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) provides essential nutritional, health, and educational support to low-income pregnant women; however, little is known about trends in program participation and associated maternal characteristics this past decade. We aim to investigate temporal trends and maternal predictors of WIC participation during pregnancy in California from 2010-2021, focusing on sociodemographic, geographic, and behaviors. Understanding these predictors is critical for informing equitable access and targeted policy interventions.

### Methods
We conducted a retrospective cohort analysis of 5,529,722 singleton births among California residents using population-based birth cohort data from 2010 to 2021. Descriptive statistics, linear trend analyses, and multivariate logistic regression models were used to assess prevalence and predictors of WIC utilization during pregnancy. Odds ratios (ORs), 95% confidence intervals (CIs), and p-values were reported in quantified associations. Statistical significance was defined as p < 0.05.

### Results
WIC participation declined significantly over the study period, from 54.1% in 2010 to 35.0% in 2021 (β=–1.83, p<.0001, R²=0.98). Multivariate analysis identified several independent predictors of WIC utilization. Compared to mothers aged 40–54 years, those under 20 had the highest odds of WIC use (OR = 3.12; 95% CI: 3.06–3.19). African American (OR = 3.126), Hispanic (OR = 2.96), and American Indian mothers (OR = 1.53) were significantly more likely to participate than White mothers. A strong inverse gradient was observed; mothers with less than high school education had nearly sixfold higher odds (OR = 5.84) than those with a bachelor’s degree or higher. Utilization was higher among foreign born mothers (OR = 1.53) and in regions such as San Joaquin Valley (OR = 2.72) and Los Angeles County (OR = 2.55) compared to Greater Bay Area. Mothers with Medi-Cal (OR=8.09), first-trimester prenatal care (OR = 1.07), higher prepregnancy BMI (e.g., OR = 1.91 for Obese III), vaginal deliveries (OR = 1.02), and maternal smoking during both trimesters (OR = 1.17) were significantly more likely to participate.

### Conclusion
WIC utilization during pregnancy has declined markedly in California, despite persistent disparities by age, race/ethnicity, education, nativity, geographic region, and healthcare access. Addressing structural barriers and improving outreach is essential to enhance equitable WIC engagement and improve maternal-child health.

Trends and Determinants of WIC Participation Among Pregnant Women in California: A Decade-Long Population Study

Levodopa, co-administered with the dopa decarboxylase inhibitor carbidopa, remains the gold standard for treating Parkinson's Disease; however, due to rapid degradation, only about 1% of orally administered levodopa crosses the blood-brain barrier.
This study developed and validated a robust high-performance liquid chromatography method to quantify levodopa, achieving reliable linearity across multiple replicates. Various buffer systems and pH conditions were tested, with phosphate buffered saline (PBS) at pH 6 and supplemented with glutamic acid hydrochloride offering the most stable environment. With stability confirmed, levodopa was successfully incorporated into a hydrogel matrix using poly(vinyl alcohol) (PVA), and hydrogel-forming microneedles using poly(methyl vinyl
ether/maleic acid), polyethylene glycol, and sodium bicarbonate. These microneedles
demonstrated sufficient mechanical strength to penetrate Parafilm “M” skin simulants, suggesting potential for effective skin permeation.
This work represents a significant step toward creating an alternative levodopa delivery system that could reduce gastrointestinal side effects, improve patient adherence, and offer more consistent plasma drug levels, thereby alleviating the motor fluctuations common with oral levodopa therapy. Although not curative, the finding supports further investigation into in vitro permeation studies and eventual clinical translation of microneedle-based patch system for Parkinson’s disease management.

Development and Characterization of Levodopa Microneedle Patches for Parkinson's Disease

### Background
Despite current treatment strategies for atherosclerotic vascular disease focusing on lifestyle modification and lowering cholesterol, a significant residual risk of major atherosclerotic complication remains, prompting investigation into lipoprotein(a) [Lp(a)] as a potential predictive biomarker. The objective of this study was to determine the utility of Lp(a) in identifying patients at high risk of incident extracoronary atherosclerotic vascular disease and complications. 

### Methods
Data from 460,544 participants in the UK Biobank with prospectively measured Lp(a) concentrations were included in this analysis. Cox proportional hazards regressions modeled the associations of Lp(a) concentrations with incident peripheral arterial disease (PAD) and incident carotid artery stenosis, and progression to the first major adverse limb event (MALE) and the first stroke, respectively.

### Results
Of the study participants, the median [IQR] age at study enrollment was 58 [51 – 64] years, 54.2% were male, 94.4% European, 5.5% had diabetes, and 10.5% were smokers. Over a median follow-up time of 13.6 [12.9 – 14.4] years, 6,347 (1.4%) and 1,972 (0.43%) developed the first incidence of PAD and carotid stenosis, respectively. Among participants with prevalent PAD and carotid stenosis at enrollment, 196 (2.7%) and 67 (1.9%) progressed to the first incidence of MALE and stroke, respectively. Median Lp(a) concentrations were significantly different in those without atherosclerotic vascular disease at 19.5 nmol/L [7.6-73.5] compared to incident PAD at 25.3 nmol/L [8.3-107.3], progression to MALE at 33.3 nmol/L [8.7-158.2], incident carotid stenosis at 29.5 nmol/L [8.5-116.3], and carotid stenosis progression to stroke at 37.8 nmol/L [11.1-158.3]. The risk estimate per 75 nmol/L Lp(a) for incident PAD (HR 1.18, 95% CI 1.15-1.20, p-value<0.0001) was similar to incident carotid stenosis (HR 1.17, 95% CI 1.13-1.20, p-value<0.0001). Among participants with PAD, those with high Lp(a) concentrations were at 1.57 times the risk of developing MALE than participants with normal Lp(a) concentrations (95% CI 1.14-2.16, p-value=0.006). Among participants with carotid stenosis, participants had 1.40 times the risk of developing an ischemic stroke with high Lp(a) concentrations, although this was not significant (95% CI 0.81-2.40, p-value=0.228).

### Conclusion
High concentrations of Lp(a) are associated with both incident extracoronary atherosclerotic vascular disease and progression to major complications.


Lipoprotein(a) is a Prognostic Marker of Extracoronary Atherosclerotic Vascular Disease Progression

Alzheimer’s disease (AD) is a debilitating dementia that affects about seven million Americans. Oligodendrocytes and myelin degeneration have been noted as features of AD, with demyelination beginning in preclinical disease. However, the mechanisms underlying oligodendrocyte and myelin decay remain poorly understood. Here, we show how transcriptional stress coupled-DNA damage accumulates in preclinical disease, and outline a novel pathway downstream of this DNA damage resulting in activation of the Type I interferon response. In vitro, we find that upon DNA damage, oligodendrocytes activate Stimulator of Interferon Genes (STING), a mediator of the DNA damage-Type I interferon signaling axis, and this drives myelin deficits in vitro. Oligodendrocyte lineage-specific deletion of STING in the 5XFAD murine AD model reduces demyelination, microgliosis, neurodegeneration, and plaque formation. Finally, loss of STING in oligodendroglial cells rescues deficits in spatial memory and learning in the 5XFAD model, as shown by Barnes Maze latencies. Our findings uncover a previously overlooked and key role for oligodendrocytes as active drivers of neurodegeneration in Alzheimer’s Disease.

Oligodendrocyte genomic stress-induced interferon responses drive progression of Alzheimer’s Disease

### Background 
White matter hyperintensities (WMH) are MRI-detected markers of small vessel disease associated with cognitive decline and Alzheimer’s disease (AD). African Americans (AAs) are disproportionately burdened by cerebrovascular risk factors, yet remain significantly underrepresented in WMH research. This disparity limits the generalizability of findings and may obscure key determinants of disease progression in diverse populations. Identifying how race and related factors influence WMH burden is critical to improving equity in Alzheimer’s care and prevention. This study aimed to examine whether racial differences exist in WMH burden across the AD spectrum, while accounting for other demographic and clinical variables. 

### Methods 
We analyzed cross-sectional MRI and clinical data from 1,649 participants in the Alzheimer’s Disease Neuroimaging Initiative (ADNI): 682 cognitively normal (CN), 584 with mild cognitive impairment (MCI), and 383 with AD. Of these, 124 participants (7.5%) identified as African American: CN (n=70), MCI (n=39), and AD (n=15). Linear mixed effects models were used to evaluate the effect of race, cognitive status, time since first MRI, age, sex, education, intracranial volume (ICV), MRI acquisition type, and APOE ε4 status on total WMH volume. 

### Results 
Older age, greater time since MRI, and larger ICV were significantly associated with increased WMH burden across cognitive groups. Use of FLAIR imaging was negatively associated with WMH volume, reflecting differences in image acquisition sensitivity. Higher educational attainment was linked to lower WMH burden, suggesting a possible neuroprotective effect. The most pronounced associations were observed within the AD subgroup. After adjusting for all covariates, race was not a statistically significant independent predictor of WMH burden. 

### Conclusion 
Although race did not independently predict WMH burden, other structural and demographic factors, including age, education level, and MRI technique, significantly influenced WMH volume. These findings highlight the complex interplay between biological, social, and technical variables in shaping cerebrovascular brain aging. Greater inclusion of African American participants in future studies is essential to better understand race-related mechanisms and to design interventions that equitably address WMH accumulation and its cognitive consequences.

Understanding the Impact of Race on White Matter Hyperintensities in Alzheimer’s Disease: Findings from the ADNI Study

### Background
Pheochromocytomas and paragangliomas (PPGLs) are rare neural-crest derived tumors with one of the highest germline heritability rates of all neoplasias (40%) and lead to lifelong surveillance. Succinate dehydrogenase (SDHx)-deficient PPGLs are associated with an aggressive disease course, with an overall survival of 50% at 5 years for individuals with metastatic disease. Predicting primary tumor development or metastatic disease progression throughout the life course with imaging/plasma metanephrines and histopathological scoring systems have limitations. The discovery of metabolic rewiring of SDHx deficiency at the tumor level holds promise as a non-invasive disease biomarker.

### Methods
Harnessing the connection of SDHx to metabolic dysfunction, we performed prospective plasma metabolomics (liquid chromatography/mass spectrometry) on 122 well-phenotyped patients, 49.1% of whom were SDHx carriers, at one or multiple time points during a patient’s clinical journey. Concurrently, tissue (adrenal, muscle, heart, brain) and plasma metabolomics of a novel SDHx deficient murine model were utilized for validation of patient-derived biomarkers.

### Results
Metabolomics of patient plasma samples revealed that succinate is a sensitive and specific biomarker of SDHx deficiency (ROC 0.825 – 75% sensitivity and 79.5% specificity), and succinate:glycine ratios improved the ROC to 0.856. With deep phenotyping, we found that age, BMI, and sex trended with succinate. Specifically, succinate levels were significantly correlated with female sex and increasing age. Of those with tumors, succinate had a ROC of 0.933, distinguishing SDHx deficient (including WT gastrointestinal stromal tumors (GISTs), which lack clinical biomarkers of disease) and proficient tumors. Five individuals with SDHx deficient tumors with multiple blood samples developed progressive disease, all of which were correspondingly reflected in succinate elevations. Succinate was not sensitive to treatment response (of both radiotherapy and surgery). Murine SDHx deficient adrenal glands had significantly elevated succinate concentrations however there were no significant changes in succinate plasma concentrations compared to WT mice.

### Conclusion
These findings suggest that plasma metabolomics holds promise as a screening test, especially in resource-poor settings or when rapid clinical decision making is necessary, as genetic testing can often take weeks, if not months, in some clinical settings. Prior to clinical application, further definition of the normal distribution of succinate levels in the general population is required. The mechanism underlying the unexpected finding of elevated succinate in SDHx carriers is a focus of future research. Overall, plasma metabolomics is a novel companion diagnostic for asymptomatic SDHx carriers and individuals with SDHx deficient tumors, of not only PPGLs but also GISTs.

Plasma Biomarkers for SDHx Deficient Pheochromocytomas and Paragangliomas

### Introduction
Estrogen-related receptors (ERRs) are nuclear receptors essential for postnatal cardiac maturation. Our recent studies have shown that ERRs are necessary for coordinated activation of cardiomyocyte metabolic and structural gene programs through interactions with PGC-1 coactivators and cardiogenic factors such as GATA4. In heart failure (HF), ERR transcriptional programs may revert to a fetal-like state, leading to metabolic inefficiency and energy starvation.

### Methods
To better understand ERRs’ role in HF, 176 SNPs from a recent HF GWAS were mapped onto the human cardiomyocyte cistrome to identify potential HF-associated variants proximal to ERRγ and/or ERRα binding regions. Overlaps were analyzed for ERR binding motifs with an 80% position-weight matrix threshold. Permutation testing was performed to determine if HF-associated variants were found more frequently than expected near ERR binding regions. Hypergeometric testing assessed whether HF-associated SNPs were enriched in ERR binding regions compared to genome-wide SNPs, with per-chromosome p-values combined using Fisher’s method. Stratified linkage disequilibrium score regression (S-LDSC) was performed to quantify whether SNPs within ERR binding regions disproportionately contributed to HF heritability compared to genome-wide averages.

### Results
Both permutation and hypergeometric testing showed significant enrichment of HF-associated variants proximal to ERR binding regions (p < 0.05). The number of overlaps increased with the size of the flanking region, ranging from 48 at 5 kb to 135 at 50 kb. Consensus ERR motifs were identified in 6.4% of ERRγ overlaps and 4.0% of ERRα overlaps. S-LDSC demonstrated notable HF heritability enrichment within ERRγ binding regions (enrichment score: 8.8 ± 3.0, p = 0.010), whereas ERRα showed non-significant enrichment (3.8 ± 3.3, p = 0.426), and GATA4 showed negligible enrichment (−0.4 ± 5.27, p = 0.786).

### Conclusion
This computational analysis reveals significant enrichment of HF-associated SNPs in or near genomic regions containing ERR transcription factors, highlighting ERR-driven transcriptional networks as potentially important mechanisms underlying heart failure development. These results establish a platform for future functional experimental studies to determine the impact of HF-associated variants on ERR-mediated transcriptional regulatory function and additional computational analyses to clarify the precise biological roles of ERRs in HF progression.

Heart Failure-Associated SNPs Are Enriched in Estrogen-Related Receptor Binding Regions

### Background 
Quality assessments relying solely on structured EHR data do not capture note-documented explanations for apparent non-adherence to lipid management guidelines. Large language models (LLMs) present a promising approach for extracting this information at scale.

### Purpose 
To evaluate the accuracy and impact of LLMs when applied to assess adherence to lipid management guidelines in CAD. 

### Methods 
We used structured EHR data to analyze LDL-C levels and statin use of patients with ASCVD seen in cardiology clinics at a large academic medical center from 2022-2024. We then applied GPT-4o with zero-shot, chain-of-thought prompting to free-text cardiology clinic notes to (1) identify external LDL-C values documented within the notes of patients without a recent in-system lipid panel and (2) infer reasons for statin non-use among those with LDL-C ≥70 mg/dL. We validated GPT-4o's accuracy through manual review of 500 randomly selected charts. 

### Results 
Among 8,595 patients (median age 71.0 years, 61.5% male, 70.8% White) initially identified, 53.6% were on high-intensity statins and 38.1% had LDL-C <70 mg/dL. GPT-4o confirmed a CAD diagnosis in 6,534 (76.0%) patients (Figure); within this group, GPT-4o identified external lipid panels in n=403 (6.2%), high-intensity statin use (n=59, 0.7%), and statin intolerance (n=804, 12.3%) documented in notes but not captured in structured fields (PPV 0.91-1.00). Among patients with GPT-4o-confirmed CAD and updated medication/laboratory data had higher measured rates of high-intensity statin use (61.4% vs. 53.6%) and LDL-C <70 mg/dL (44.1% vs. 38.1%). Excluding statin-intolerant patients further increased high-intensity statin use to 68.4%. Among patients with LDL-C ≥70 mg/dL not on high-intensity statins (N=1,164), GPT-4o-extracted reasons for nonuse were clinician decision (13.0%), intolerance (38.5%), patient preference (8.2%), and undocumented (40.4%). GPT-4o classification performance was strong (F1 > 0.71 across all categories), with the highest performance observed for statin intolerance (F1 = 0.96).

### Conclusions
GPT-4o can accurately identify clinical factors not captured in structured EHR data. This approach enables more accurate guideline-based quality assessments and yields actionable insights into reasons for apparent gaps in care.

Premium content

Downloads

Next from AMA Research Challenge 2025

Waste to Wonder: Understanding endometrial interactions using a menstrual-effluent derived model

Similar lecture

Accuracy Benchmark Highlights AI's Potential to Transform Medical Education and Curricula Development