United States

Stick to the Script: 
Large Language Model (LLM) Performance on the Hinting Task Highlights the Role of Language in Theory of Mind

 Alexander Schemmer, A.B., Ryan Partlan, M.S., Jiefei Li, M.P.S., Amir H Nikzad, M.D., Sunny X. Tang, M.D.

Background / Objective:

Theory of Mind (ToM)—the ability to reason about others’ mental states—is impaired in schizophrenia spectrum disorders (SSD), contributing to poorer social functioning and quality of life. While disorganized speech and pragmatic comprehension deficits characterize schizophrenia, their relationship to ToM deficits remains unclear, limiting targeted interventions.

No current language-based interventions specifically address ToM deficits in schizophrenia. However, two non-clinical domains offer promising models: Konstantin Stanislavski’s &quot;script analysis&quot; in modern acting, which enhances inductive reasoning and subtextual understanding through textual analysis; and Large Language Models (LLMs), which produce contextually relevant responses by modeling deep semantic relationships among words.

This study investigates whether ToM deficits in SSD reflect impaired semantic modeling—difficulty aligning language with underlying intention. To test this, we administered the Hinting Task (Corcoran et al., 1995)—a validated ToM measure using narrative vignettes with indirect speech—to individuals with SSD, healthy controls (HC), and three high-performing LLMs (GPT-4o, Claude, LLaMA). We hypothesized that semantic alignment—measured via cosine similarity between prompt and response embeddings—would predict successful ToM inference in humans and LLMs.

Methods:

Participants included 36 individuals with SSD, 38 HC, and three groups of 30 simulated responses from GPT-4o, Claude, and LLaMA. Each completed the 10-item Hinting Task, scored from 0–2 (maximum total = 20). Semantic alignment was calculated using cosine similarity between sentence embeddings (sentence transformer model embeddings) of each participant’s responses and their corresponding prompts. Group comparisons and accuracy-based differences were analyzed via t-tests.

Results:

LLMs achieved exceptional performance (GPT: 19.6 ± 0.6; Claude: 19.5 ± 0.7), significantly surpassing HC (17.5 ± 2.1) and SSD (13.6 ± 4.3). Cosine similarity mirrored this performance, with pooled LLM responses demonstrating significantly higher semantic alignment than pooled human responses (p &lt; .001, d = 1.56). Within HC and SSD, accurate responses exhibited significantly greater semantic alignment than inaccurate ones (HC: p &lt; .001, d = 0.79; SSD: p &lt; .001, d = 0.58). No significant semantic alignment differences emerged between correct human responses and those generated by GPT or Claude.

Discussion / Conclusion:

Findings support the hypothesis that impaired semantic modeling is closely linked to Theory of Mind deficits in schizophrenia. The association between semantic similarity and ToM accuracy highlights the potential of language-based therapeutic approaches inspired by Stanislavski’s methods and LLM semantic modeling strategies. Future research should explore whether specific training modules targeting semantic comprehension, drawing on these approaches, can effectively enhance ToM abilities and improve social functioning outcomes in individuals with schizophrenia spectrum disorders.

AMA Research Challenge 2025 

Stick to the Script: Large Language Model (LLM) Performance on the Hinting Task Highlights the Role of Language in Theory of Mind

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair:<br>
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for  Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025.  ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am  deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.*<br>
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)<br><br><br>



To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

ACL 2025

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Stick to the Script: 
Large Language Model (LLM) Performance on the Hinting Task Highlights the Role of Language in Theory of Mind

 Alexander Schemmer, A.B., Ryan Partlan, M.S., Jiefei Li, M.P.S., Amir H Nikzad, M.D., Sunny X. Tang, M.D.

Background / Objective:

Theory of Mind (ToM)—the ability to reason about others’ mental states—is impaired in schizophrenia spectrum disorders (SSD), contributing to poorer social functioning and quality of life. While disorganized speech and pragmatic comprehension deficits characterize schizophrenia, their relationship to ToM deficits remains unclear, limiting targeted interventions.

No current language-based interventions specifically address ToM deficits in schizophrenia. However, two non-clinical domains offer promising models: Konstantin Stanislavski’s "script analysis" in modern acting, which enhances inductive reasoning and subtextual understanding through textual analysis; and Large Language Models (LLMs), which produce contextually relevant responses by modeling deep semantic relationships among words.

This study investigates whether ToM deficits in SSD reflect impaired semantic modeling—difficulty aligning language with underlying intention. To test this, we administered the Hinting Task (Corcoran et al., 1995)—a validated ToM measure using narrative vignettes with indirect speech—to individuals with SSD, healthy controls (HC), and three high-performing LLMs (GPT-4o, Claude, LLaMA). We hypothesized that semantic alignment—measured via cosine similarity between prompt and response embeddings—would predict successful ToM inference in humans and LLMs.

Methods:

Participants included 36 individuals with SSD, 38 HC, and three groups of 30 simulated responses from GPT-4o, Claude, and LLaMA. Each completed the 10-item Hinting Task, scored from 0–2 (maximum total = 20). Semantic alignment was calculated using cosine similarity between sentence embeddings (sentence transformer model embeddings) of each participant’s responses and their corresponding prompts. Group comparisons and accuracy-based differences were analyzed via t-tests.

Results:

LLMs achieved exceptional performance (GPT: 19.6 ± 0.6; Claude: 19.5 ± 0.7), significantly surpassing HC (17.5 ± 2.1) and SSD (13.6 ± 4.3). Cosine similarity mirrored this performance, with pooled LLM responses demonstrating significantly higher semantic alignment than pooled human responses (p < .001, d = 1.56). Within HC and SSD, accurate responses exhibited significantly greater semantic alignment than inaccurate ones (HC: p < .001, d = 0.79; SSD: p < .001, d = 0.58). No significant semantic alignment differences emerged between correct human responses and those generated by GPT or Claude.

Discussion / Conclusion:

Findings support the hypothesis that impaired semantic modeling is closely linked to Theory of Mind deficits in schizophrenia. The association between semantic similarity and ToM accuracy highlights the potential of language-based therapeutic approaches inspired by Stanislavski’s methods and LLM semantic modeling strategies. Future research should explore whether specific training modules targeting semantic comprehension, drawing on these approaches, can effectively enhance ToM abilities and improve social functioning outcomes in individuals with schizophrenia spectrum disorders.

poster

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>.

Please log in to explore 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

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).

Activating mutations in the receptor tyrosine kinase c-Kit, particularly the D816V substitution, are implicated in several malignancies including systemic mastocytosis and gastrointestinal stromal tumors, and are known to confer resistance to type-II tyrosine kinase inhibitors (TKIs) such as imatinib and dasatinib. We investigated the structural and thermodynamic mechanisms underlying drug resistance and constitutive activation in c-Kit mutants, focusing on the interplay between local rigidification and intrinsic disorder.
Through in silico saturation mutagenesis and vibrational entropy analysis, we demonstrate that hydrophobic and aromatic substitutions at position D816 (e.g., V, F, W, Y, L) induce enthalpy-driven rigidification of critical regulatory elements such as the juxtamembrane latch, N-lobe wedges, and activation loop. This rigidified core pre-aligns the catalytic spine, locks the kinase in an active conformation, and prevents transition to the inactive DFG-out state required for type-II TKI binding. In contrast, substitutions at Asn819 lead to increased conformational flexibility, partial destabilization, and retention of inhibitor sensitivity.
Notably, despite increased rigidity at D816, a segment of intrinsic disorder persists within the activation loop and may serve as a dynamic interface for protein interactions. Structural modeling also reveals a shift in dimerization preference, with D816 mutants favoring heterodimer formation with wild-type receptors, further amplifying oncogenic signaling.
These findings redefine rigidification—not flexibility—as a key driver of drug resistance and constitutive activity in c-Kit, and provide a mechanistic framework for the development of next-generation inhibitors targeting rigidified active conformations or disordered regulatory interfaces.


Constitutive activation of the c-Kit oncoprotein induced by a rigidifying hydrophobic interaction between D816V and N819 in the presence of imatinib

An Unusual Case of Post-Coital Takotsubo Syndrome in a Female Patient on Venlafaxine 
Takotsubo cardiomyopathy is characterized by new EKG changes of ST elevation, T wave inversions or cardiac troponin elevation; left-ventricular dysfunction and apical-ballooning in the absence of obstructive coronary artery disease; and reversible LV dysfunction. (Mayo Clinic criteria). The syndrome was first recognized in Japan and named Takotsubo, meaning octopus- catching pot. Although precise mechanisms are unclear, takotsubo syndrome is often triggered by sudden or overwhelming physical or emotional stress, with attendant catecholamine excess. Generally, negative stressors have induced stress cardiomyopathy, but there are rare instances of positive experiences also eliciting it. We present a rare case of Takotsubo syndrome in a female, with no prior cardiac history, thirty minutes after sexual intercourse. Temporal association does not prove causality, but our report suggests that Takotsubo may follow either a negative or positive event. Sexual activity is associated with a catecholamine surge of dopamine and noradrenaline. Our patient was on venlafaxine, which may synergistically trigger catecholamine excess alongside sexual intercourse. 
A 47-year-old female with depression on venlafaxine presented with sudden onset of chest tightness after sexual intercourse. On admission, blood pressure was elevated at 170/100, and heart rate elevated at 90- 100’s. An electrocardiogram demonstrated T-wave inversions in V2-5 and a CT angiogram excluded pulmonary embolism. Cardiac troponin levels rose from .099 to .646 to .74 ng/ml. She was suspected of having NSTEMI. She received aspirin, nitroglycerin drip, heparin drip, metoprolol, and atorvastatin. Subsequent cardiac catheterization showed normal coronary arteries. The left ventriculogram demonstrated normal left-ventricular function, apical ballooning, and left-ventricular akinesis. The rest of her hospitalization was uneventful, and she was discharged in good condition on nifedipine and atorvastatin. 
We compiled six total patients with post-coital takotsubo syndrome. All were female, average age of 53.8 years (range 47-61). All presented with chest pain or dyspnea, and one presented in cardiac arrest. One took atomoxetine (selective-norepinephrine-reuptake-inhibitor) for ADHD. Ours is the youngest and had two possible triggers – venlafaxine (serotonin-noradrenaline- reuptake- inhibitor) and sexual intercourse. Her elevated blood pressure and heart rate may reflect sympathetic or catecholamine excess. Venlafaxine has had reported associations with Takotsubo syndrome through a postulated catecholamine excess. Rarer complications of sexual intercourse include cardiocerebral infarction and coronary artery dissection. 
This case reminds the provider to include takotsubo syndrome in the differential of acute coronary syndromes in patients presenting post-coital or who are on venlafaxine. Takotsubo syndrome may follow positive or negative stressors, with an underlying catecholamine basis. 
 
References 
Athanassopoulos P, Y-Hassan S. A case report: 'happy heart' syndrome in a patient treated with atomoxetine for attention deficit hyperactivity disorder. Eur Heart J Case Rep. 2019 Dec;3(4):1- 5. doi: 10.1093/ehjcr/ytz151. Epub 2019 Sep 30.PMID: 31911993; PMCID: PMC6939811. 
Tomcsányi J, Marosi A, Arabadzisz K, Zsoldos A, Bózsik B. Tako-tsubo syndrome associated with sexual intercourse. Int J Cardiol. 2007 Oct 18;121(3):e28-9. doi: 10.1016/j.ijcard.2007.06.131. Epub 2007 Sep 19. PMID: 17884207. 
Recurrent Sexual Intercourse-Associated Takotsubo Cardiomyopathy, Ishak, M. et al. Heart, Lung and Circulation, Volume 32, S174. 
Madias JE. Follow-up of a patient who had suffered a "heart breaking pleasure"-triggered Takotsubo syndrome. Int J Cardiol. 2016 Nov 1;222:1057. doi: 10.1016/j.ijcard.2016.01.067. Epub 2016 Mar 31. PMID: 27038689. 
Mathew B, Villarreal D. Two unusual cases of Takotsubo cardiomyopathy presenting with sudden cardiac death. Am J Med Sci. 2009 Jun;337(6):473-5. doi: 10.1097/MAJ.0b013e3181956f49. PMID: 19390432. 
Liu G, Khalid U, Birnbaum Y, Hamzeh IR. A heartbreaking pleasure. Int J Cardiol. 2016 Feb 1;204:177-8. doi: 10.1016/j.ijcard.2015.11.115. Epub 2015 Nov 19. PMID: 26666345. 
SG, Faulx M, Kapadia SR, Griffin BP, Desai MY. Characteristics and Outcomes of Patients With Takotsubo Syndrome: Incremental Prognostic Value of Baseline Left Ventricular Systolic Function. J Am Heart Assoc. 2020 Aug 18;9(16):e016537. doi: 10.1161/JAHA.120.016537. Epub 2020 Aug 5. PMID: 32755253; PMCID: PMC7660826. 
Pastore C, White M, Henry M, Filler L. Simultaneous Cardiocerebral Infarction Associated with Postcoital Activity. J Emerg Med. 2023 Dec;65(6):e554-e558. doi: 10.1016/j.jemermed.2023.08.008. Epub 2023 Aug 25. PMID: 37852811. 
Möller J, Ahlbom A, Hulting J, Diderichsen F, de Faire U, Reuterwall C, Hallqvist J. Sexual activity as a trigger of myocardial infarction. A case-crossover analysis in the Stockholm Heart Epidemiology Programme (SHEEP). Heart. 2001 Oct;86(4):387-90. doi: 10.1136/heart.86.4.387. PMID: 11559674; PMCID: PMC1729949. 
Finocchiaro G, Westaby J, Behr ER, Papadakis M, Sharma S, Sheppard MN. Association of Sexual Intercourse With Sudden Cardiac Death in Young Individuals in the United Kingdom. JAMA Cardiol. 2022 Mar 1;7(3):358- 359. doi: 10.1001/jamacardio.2021.5532. PMID: 35019956; PMCID: PMC8756354. 
Madias JE. Venlafaxine and takotsubo syndrome: Can we learn more from published patient cases? Int J Cardiol. 2016 Dec 15;225:73-74. doi: 10.1016/j.ijcard.2016.09.133. Epub 2016 Oct 1. PMID: 27716552. 
Woronow D, Suggs C, Levin RL, Diak IL, Kortepeter C. Takotsubo Common Pathways and SNRI Medications. JACC Heart Fail. 2018 Apr;6(4):347-348. doi: 10.1016/j.jchf.2017.12.011. PMID: 29598939. 



An Unusual Case of Post-Coital Takotsubo Syndrome in a Female Patient on Venlafaxine

Background:
Nonalcoholic steatohepatitis (NASH), sarcoidosis, and hypothyroidism are often associated with immune dysregulation and may coexist in complex clinical scenarios. Immunoglobulin G4-related disease (IgG4-RD) is a systemic fibroinflammatory condition affecting multiple organs—including the pancreas, thyroid, and hepatobiliary system—and can mimic malignancy or other autoimmune disorders. Recognizing multisystem IgG4-RD is critical for timely diagnosis and treatment.
Case Presentation:
A 56-year-old female with known NASH, hypothyroidism, and sarcoidosis presented with gastrointestinal discomfort, right shoulder pain, and intolerance to L-ornithine L-aspartate (Microliv Forte). Lab studies revealed elevated liver enzymes, vitamin D deficiency, low folic acid, and prediabetes (HbA1c 6.0%). TSH was elevated, but anti-TPO antibodies were negative. Imaging showed hepatomegaly with fatty infiltration, cholelithiasis, and a 1 cm ill-defined lesion in the pancreatic head/uncinate process with upstream ductal dilatation on MRCP—raising concern for malignancy. EUS revealed two small pancreatic cysts without a discrete mass or lymphadenopathy.
Further workup showed elevated IgG4 (136 mg/dL) and complement C4, with normal CA 19-9 and negative autoantibodies (dsDNA, RF, anti-TPO, celiac serologies). Bone density testing showed osteopenia. In the context of elevated IgG4 and thyroid dysfunction without TPO antibodies, IgG4-related thyroid disease (IgG4-RTD) was considered. The constellation of findings suggests possible systemic IgG4-RD with pancreatic and hepatobiliary involvement.
Discussion:
This case underscores the need to consider IgG4-RD in patients with overlapping metabolic and autoimmune disorders who develop atypical pancreatic or hepatobiliary findings. IgG4-RD can resemble malignancy or other inflammatory conditions, making diagnosis challenging. A combination of serology, imaging, and histopathology is essential. Early recognition is vital, as corticosteroids are highly effective in preventing irreversible organ damage. The MITIGATE trial has shown promising results using B-cell depletion therapies like inebilizumab. Additionally, hypothyroidism has been shown to increase the risk of metabolic-associated steatotic liver disease (MASLD), with a 1.7-fold higher risk independent of other metabolic factors.



Beyond NASH, Sarcoidosis, Hypothyroidism, and Pancreatic Lesions: A Case Highlighting the Overlap Between Autoimmune, Metabolic, and IgG4-Related Pathology


Background
Microscopic polyangiitis (MPA) is a pauci-immune small-vessel vasculitis associated with MPO-ANCA positivity, characterized by necrotizing small-vessel inflammation. Though lung and kidney involvement is most common, systemic symptoms may precede or overshadow the classic features, thereby complicating the diagnostic process. As the incidence is low and symptoms often vague, MPA can mimic infection, malignancy, or other autoimmune conditions. Early identification is key to preventing irreversible organ damage.

Case Presentation
A 48-year-old woman with hypertension, type 2 diabetes mellitus, and a history of post-surgical pulmonary embolism presented to the ED with progressive dyspnea, dry cough, fatigue, subjective fever, abdominal pain and dark urine over three weeks. On presentation, she was febrile (38.5°C) and hypoxic with an oxygen saturation of 93% on room air and bibasilar rhonchi. She was empirically treated with ceftriaxone and azithromycin for community-acquired pneumonia. 
On hospital day 3, she developed submandibular swelling, a facial rash, and exacerbation of a chronic symmetric pruritic, hyperpigmented rash on her dorsal forearms. These findings prompted consideration of autoimmune and systemic inflammatory etiologies. Chest CT revealed multifocal bilateral nodular opacities and a small left pleural effusion. Thoracentesis performed on hospital day 5 showed exudative fluid with lymphocytic predominance and gram-positive cocci on gram stain with no growth on culture. Due to concern for a complicated parapneumonic effusion, ceftriaxone was continued for 14 days. 
Despite fluid resuscitation, her serum creatinine continued increasing (peak 3.2); her abdominal pain and fatigue persisted. Urinalysis revealed nephrotic-range proteinuria and macroscopic hematuria, prompting nephrology consultation for glomerulonephritis workup. Her autoimmune panel displayed a positive direct Coombs test indicating IgG-mediated hemolytic anemia, MPO-ANCA positivity, and weakly positive rheumatoid factor, anti-dsDNA, anti-cardiolipin IgA, and β2-glycoprotein IgA. IV methylprednisolone was initiated, and renal biopsy revealed pauci-immune crescentic glomerulonephritis consistent with microscopic polyangiitis. Rituximab was initiated for induction therapy. The patient demonstrated marked clinical improvement and was weaned off oxygen prior to discharge on hospital day 14. After one month on rituximab and prednisone, her creatinine levels normalized and respiratory symptoms resolved.

Discussion
This case highlights the diagnostic challenges of MPA, particularly when systemic features such as hypoxia, abdominal pain, cutaneous findings, and glandular swelling obscure classic presentations of pulmonary-renal vasculitis. It underscores the importance of maintaining a high index of suspicion for vasculitis in patients with pulmonary infiltrates, rash, and unexplained kidney injury. Prompt and accurate distinction between infectious and autoimmune etiologies is critical to avoid delays in initiating appropriate treatment.


From Cough to Crescents: Microscopic Polyangiitis Uncovered Through Systemic Cues

Background
People with disabilities (PWD) face numerous barriers to equitable healthcare, despite utilizing the healthcare system at three times the rate of people without disabilities. In 2005, the Surgeon General identified the lack of provider training as one major barrier to high-value care for PWD, citing disability awareness education for healthcare providers as a key solution. Nevertheless, a 2017 survey found that only 52% of medical schools provided a disability awareness program, with significantly fewer reporting an experiential component or the incorporation of disability community input. As the disabled population continues to grow, it is increasingly important to implement and evaluate disability curricula for medical trainees to close this gap. 

Methods
A four-week, full-time (118.5 contact hours) elective course for third- and fourth-year medical students was developed by faculty with disabilities at Washington University, in partnership with disability-centered community groups. The course simultaneously aims to: (1) increase knowledge about conceptual frameworks of disability and community support services, (2) enhance skills in accommodating patient needs and adapting clinical assessments, and (3) change attitudes through engagement with PWD sharing their lived experiences. Attitudes and perceived self-competence were assessed via completion of identical pre- and post-intervention surveys consisting of Likert-type items adapted from the validated Medical Student Attitudes toward People with Disabilities (MSA-PWD) instrument, as well as open-ended feedback questions administered at course-end. Knowledge was assessed dichotomously using students’ scores on course assessments.

Results
Across seven cohorts of students enrolled in the course from January 2023 – September 2025, 41 students completed the pre- and post-intervention surveys. Wilcoxon signed-rank tests, conducted to compare mean pre-intervention and post-intervention scores, indicated significant increases in positive attitudes towards PWD, comfort interacting with PWD, and perceived self-competence in caring for PWD (all: W=0, p<0.001), regardless of whether or not students had prior personal or professional experience working with PWD. Exam scores demonstrated that all students achieved competency on course assessments. Qualitative feedback revealed themes focused on attitudinal change and appreciation of experiential coursework and non-clinical interactions with disability communities.

Conclusion
This disability elective represents a novel experiential curriculum that has demonstrated the ability to effectively educate medical students about disability-inclusive health and positively change attitudes toward PWD. Future work will develop and evaluate expanded disability-focused content for undergraduate medical education, with the goal of implementing a longitudinal curriculum to ensure that medical trainees become healthcare professionals who are equipped to provide high-value care to PWD and chronic illnesses.

Efficacy of a 4-week Immersive Disability Elective in Increasing Knowledge and Changing Attitudes

Background: The integration of artificial intelligence (AI) into healthcare has accelerated, with an increase in its use for clinical decision support and medical education. In dermatology, large language models are often used to answer patients’ inquiries and support learning for medical professionals, which raises concerns about their reliability and clinical safety. Given the nuanced interpretation required in dermatologic assessment and the potential consequences of incorrect guidance—whether for students or patients—category-specific evaluation of these models is essential to ensure responsible and effective use in dermatologic practice and education. The objective of this study is to assess and compare the performance of 8 publicly available AI models in answering dermatology-related multiple-choice questions across 6 clinical categories.

Methods: A board-certified internal medicine physician developed 11 multiple-choice American Board of Internal Medicine Certification dermatology questions covering 6 dermatologic categories: hair disorders, skin/nail infections, acneiform eruptions, nail disorders, leg ulcers, and geriatrics. All 8 AI models were prompted with the same 11 questions. Each question had 4 – 5 answer choices with a single correct answer. A response was marked correct if the AI model selected the correct choice.

Results: Gemini and DeepSeek scored highest, each with 10/11 questions answered correctly (90.91%). Le Chat scored lowest with 8/11 questions answered correctly (72.73%). All others (ChatGPT, Perplexity, Copilot, Claude, Meta) scored 9/11 (81.82%). No statistically significant differences were found between any models. Fisher’s Exact Test found that the acneiform eruptions category had a statistically significant lower correct rate (62.5%) compared to the hair disorders category (95%) across all models (p ≈ 0.0025). No statistically significant differences between other category pairs.

Conclusion: AI models show promise in dermatology-related question answering, but their performance remains inconsistent and dependent on the question category. No single model demonstrated statistically superior performance overall, and all models struggled significantly more with questions related to acneiform eruptions compared to hair disorders. These findings highlight the need for continued refinement and rigorous evaluation of AI models in dermatology. Further research using a larger, more comprehensive question bank across a broader range of dermatologic topics is warranted to better assess model capabilities and guide their integration into clinical and educational settings.

Evaluating the Performance of Large Language Models in Answering American Board of Internal Medicine Certification Dermatology Questions Across Different Categories

Background:
With the transition of STEP 1 and pre-clinical coursework to pass/fail grading recently, research has been emphasized in residency applications. At New York Medical College (NYMC), the pre-clinical curriculum became pass/fail beginning in Fall 2020; STEP 1 became pass/fail in January 2022. Since these changes are recent, limited data exists on how they have influenced medical student behavior. This study aims to evaluate how recent curricular changes have impacted student engagement in research. 

Methods:
Since 2018, NYMC has offered student-led research seminars as part of its Biomedical Area of Concentration (AoC) to graduate with distinction in research, first implemented for the Class of 2020. Each year, approximately 16 seminars are organized by medical students. Applications to present at seminars are open to the student body, with two students selected per seminar based on themed topics. Seminars are held via Zoom, and attendance is recorded using a Turning Point code distributed mid-session. To receive AoC credit, students must attend 16 seminars during their pre-clinical years. 

To assess trends in AoC participation and seminar attendance from 2020 to 2025, a linear regression was performed using year as the independent variable and number of participants as the dependent variable. An ordinary least squares model was used, with statistical significance set at p < 0.05. 

Results:
There was a statistically significant increase in the number of students completing the AoC from 2020 to 2025 (p<0.05): 18 in 2020, 22 in 2021, 34 in 2022, 35 in 2023, 42 in 2024, and 41 in 2025. 

Average seminar attendance by academic year was as follows: 56 (2018–2019), 58 (2019–2020), 94 (2020–2021), 78 (2021–2022), 57 (2022–2023), 49 (2023–2024), and 71 (2024–2025). Attendance peaked in 2020–2021, which was significantly higher than other years (p < 0.05).

Conclusion:
The increase in number of students completing the Biomedical AoC suggests growing interest in research as a means of enhancing residency applications since pre-clinical curriculum and STEP 1 became pass/fail. However, an increase in research seminar attendance was only observed during the 2020-2021 academic year. The consistent attendance suggests students may attend for other reasons besides completing the AoC such as general research interest or supporting their peers. The increase during 2020-2021 may reflect the newly established pass/fail curriculum or increased flexibility in students’ schedule and need for a sense of connection due to remote learning during the COVID-19 pandemic. 


Student Lead Research Seminars at an LCME-accredited Medical School

Background

Chronic kidney disease (CKD) affects 15% of US adults, with a disproportionate burden among racial minorities and socioeconomically disadvantaged populations. Despite established national evidence linking social determinants of health (SDOH) to CKD outcomes, granular local level analyses remain unexplored. This knowledge gap hinders the development of targeted interventions and equitable resource allocation. We hypothesized that zip-code level SDOH metrics would demonstrate strong associations with CKD prevalence, enabling identification of modifiable upstream drivers for precision public health interventions.

Methods

We conducted a comprehensive cross-sectional analysis of 70 zip codes across the Kansas City Metropolitan area (Missouri and Kansas) using 2022-2024 data. CKD prevalence and 37 validated SDOH indicators were systematically compiled from the American Community Survey, CDC PLACES, EPA, National Center for Health Statistics, Neighborhood Atlas, and Open Data KC databases. Our analytic approach employed Pearson correlations for bivariate associations, mutual information regression for predictor selection, and variance inflation factor filtering (VIF<5) to address multicollinearity. The final ordinary least squares regression model incorporated nine predictors, optimized through ten-fold cross-validation using minimum mean squared error. Benjamini-Hochberg correction controlled the false discovery rate.

Results

Striking correlations emerged between SDOH metrics and CKD prevalence. The strongest associations included mobility impairment (r=0.95), disability (r=0.84), physical inactivity (r=0.77), obesity (r=0.70), and smoking (r=0.72). Protective factors included life expectancy (r=−0.70), median household income (r=−0.54), and college graduation rates (r=−0.47), all statistically significant (FDR-adjusted p<0.001).

Our multivariable model demonstrated exceptional explanatory power (R²=0.882, F=49.89, p<0.0001), with life expectancy (β=−0.28) and median household income (β=−0.12) as primary predictors. Notably, systemic environmental factors outweighed traditional behavioral risk factors, suggesting CKD risk is fundamentally shaped by structural determinants rather than individual choices alone.



Conclusion

This preliminary study provides the first granular evidence that CKD prevalence across Kansas City is strongly determined by community-level social and structural vulnerability indicators. Our findings challenge conventional individual-focused prevention models by demonstrating that neighborhood-level factors—particularly life expectancy, economic stability, and educational attainment—are primary drivers of kidney disease risk. Health systems can leverage these risk profiles to implement targeted screening programs, allocate resources to high-vulnerability zip codes, and develop community-level interventions addressing upstream social determinants. While this cross-sectional analysis establishes foundational associations, future longitudinal studies will determine causal pathways and validate interventions. This precision public health approach represents a paradigm shift toward addressing the root causes of health disparities, potentially transforming kidney disease prevention and management strategies nationwide.

Mapping Social Vulnerability of Chronic Kidney Disease in Urban Communities

Purpose: 
Prevalence of diabetes among women in South Carolina is growing. Uncontrolled maternal diabetes is associated with significant maternal and neonatal morbidity and mortality. The Wellness and EmpOwerment for Women (WOW) program provides primary care and diabetes management to women with or at risk for diabetes. This study examines the association of continuous glucose monitors (CGM) and FoodShare as supplements to medical diabetes management, with change in HbA1c over 12 months. 

Methods: 
Women of childbearing age (18-45 years old) with a diagnosis of diabetes mellitus (T1DM or T2DM) were referred to the WOW program. Participants were included if they had a baseline and 12-month HbA1c measurement. Additionally, clinical assessments and questionnaires also were taken at baseline and 12-month appointments to measure comorbidities, demographics, and socioeconomic factors. Patients also were offered continuous glucose monitors (CGM) and access to a food share program. The collected data was summarized using means, standard deviations, and tabulations. Multivariate regression analysis was performed to examine the association between participation in continuous glucose monitoring and Foodshare programs on changes in HbA1c, controlling for demographics, socioeconomic factors, comorbid diagnoses of hypertension and depression. 

Results: 
This study found that CGM utilization was statistically significantly associated (0.04) with a reduction in HbA1c (-1.46). Participation in a food share program was not significantly associated with a change in HbA1c. 

Conclusions: 
The study demonstrates the clinical promise of a program focusing on CGMs as a supplement to medical management for glycemic control. Future efforts should focus on expanding use of such tools with the establishment of primary care and regular follow up for women with diabetes in South Carolina. 


Analyzing the Impact of Supplemental Continuous Glucose Monitoring and Foodshare Programs on Glycemic Control Among Women With Diabetes Mellitus

Introduction:
Only 23% of musculoskeletal referrals result in surgery, highlighting the need for decision support tools that assist providers in interpreting imaging when orthopaedic input is not readily available. Primary care providers often navigate complex radiology narratives without subspecialty guidance, leading to variability in referral quality. Some patients are referred prematurely, while others face delays despite surgical need. We developed OrthoScreen, a transformer-based NLP model built to support referral prioritization using solely imaging text, for potential application in arthroplasty triage. We hypothesized that OrthoScreen would demonstrate high sensitivity in identifying patients likely to be considered for surgery within 1 year, based solely on imaging reports.

Methods:
We retrospectively analyzed patients presenting with knee pathology to an urban safety-net hospital between 2017 and 2020. Each had MRI or X-ray imaging and documented one-year surgical follow-up. Patients with missing imaging or outcome data were excluded. Text from radiology “Findings” and “Impression” sections was extracted from each initial report. OrthoScreen was developed by fine-tuning Bio_ClinicalBERT to predict whether surgery was considered within one year and evaluated using stratified 10-fold cross-validation. We compared performance to a Bag-of-Embeddings MLP model (BGE M3) using identical inputs. Metrics included AUC, sensitivity, specificity, PPV, NPV, and F1 score. No formal statistical comparisons were performed.

Results:
A total of 453 patients were included, 60% of whom underwent or were recommended surgery within one year. OrthoScreen achieved AUC 0.745, sensitivity 0.933, specificity 0.352, PPV 0.689, NPV 0.837, and F1 score 0.787. BGE M3 yielded AUC 0.713, sensitivity 0.922, specificity 0.375, PPV 0.699, NPV 0.754, and F1 score 0.787.

Conclusion:
OrthoScreen demonstrated strong performance using only imaging text and may help streamline surgical triage. Broader validation could support scalable integration into orthopaedic surgical referral workflows.

Premium content

Next from AMA Research Challenge 2025

Constitutive activation of the c-Kit oncoprotein induced by a rigidifying hydrophobic interaction between D816V and N819 in the presence of imatinib

Similar lecture

Beyond Words: Integrating Theory of Mind into Conversational Agents for Human-Like Belief, Desire, and Intention Alignment