United States

Background
Peripheral intravenous (IV) catheters are fundamental to inpatient care but are commonly associated with complications such as thrombophlebitis, leading to discomfort, prolonged hospitalization, and higher healthcare costs. Preventive measures like routine 8-hourly saline flushes and transparent Tegaderm dressings remain underused despite their effectiveness and affordability. This quality improvement (QI) project assessed the impact of these two interventions in reducing peripheral IV catheter-related thrombophlebitis among hospitalized adults by enhancing catheter patency and allowing for early visual detection of complications.

Methods
In a prospective pre–post intervention study at a tertiary care hospital, 112 adult inpatients with peripheral IV catheters were observed. During the baseline phase (n=62), usual care included opaque Dynaplast dressings and inconsistent saline flushing. The intervention phase (n=50) implemented standardized 8-hour saline flushes and exclusive use of Tegaderm dressings. Multivariable logistic regression adjusted for confounders including age, sex, and comorbidities.

Results
Of the 112 patients (mean age 55.1 ± 15.6 years; 58% male), common comorbidities included hypertension (52%), diabetes (41%), CKD (18%), and CHF (15%). Thrombophlebitis occurred in 32.3% of the pre-intervention group and 8.0% of the post-intervention group, reflecting a 75% relative reduction (p = .006). In adjusted analysis, irregular saline flushing was associated with a fourfold increase in thrombophlebitis risk (AOR: 3.9, 95% CI: 1.6–9.1, p = .003), and Dynaplast dressings carried an even higher risk (AOR: 4.6, 95% CI: 1.8–10.8, p = .001). Conversely, routine saline flushes reduced risk by 76% (AOR: 0.24, 95% CI: 0.08–0.79, p = .019), and Tegaderm use reduced risk by 79% (AOR: 0.21, 95% CI: 0.07–0.71, p = .013). Every 10 years of increase in age beyond 60 years showed a mild association with thrombophlebitis (AOR: 1.2, 95% CI: 1.0–1.6, p = .048).


Conclusion
This quality improvement initiative demonstrates that standardizing peripheral IV catheter care with routine 8-hourly saline flushes and transparent Tegaderm dressings significantly reduces catheter-related thrombophlebitis. The intervention’s effectiveness stems from improved site visibility, consistent catheter patency, and early detection of inflammation—critical factors in preventing progression to complications. Requiring no advanced technology and minimal added cost, it is both scalable and sustainable across various healthcare settings. These findings support replacing opaque dressings and irregular maintenance with evidence-based, preventive care. Broader adoption of such simple, cost-effective practices can improve patient safety, reduce complications, and enhance IV therapy quality, particularly in resource-limited environments. This project highlights the value of embedding routine, preventive strategies into standard clinical protocols to achieve durable improvements in outcomes and system efficiency.



2025 AMA Research Challenge – Member Premier Access

Reducing Peripheral IV Catheter-Related Thrombophlebitis: Impact of Simple Interventions


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)
[![](https://assets.underline.io/markdown_image/1/image/0ca1b933deefd7731529477b45345166.png)](https://underline.io/events/506/schedule)

#### 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 will announce the top five finalists who will advance to the AMA Research Challenge final round, present their research to an elite panel of judges, and compete to win a $10,000 grand prize 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).

Abstract Title

Low CSF-Pressure Diagnostic Criterion in Spontaneous Intracranial Hypotension is Potentially Misleading in Clinical Practice

Background

Low CSF pressure (Pcsf) has classically been considered a defining feature in the pathogenesis of spontaneous intracranial hypotension (SIH) and is part of the International Headache Society (IHS) diagnostic criteria for SIH. However, normal or elevated Pcsf is commonly measured in SIH during myelography. This suggests that the low Pcsf criterion may be potentially misleading in clinical practice. We report on Pcsf and its correlations in patients with SIH due to CSF-venous fistula (CVF). 

Methods
Following IRB approval, consecutive SIH patients undergoing myelography between 2021 and 2025 at a single quaternary care center were retrospectively analyzed. Pcsf was defined as opening pressure (OP) during myelography, with patient in decubitus positioning with mild hip elevation. In line with IHS, Low OP was defined as < 6 cm H2O. Elevated OP was defined as > 20 cm H2O.

Results
Of the 86 SIH patients analyzed, 35 (41%) had a CVF [average age 59±13; 12/35 (34%) male; 23/35 (66%) female]. Of the 35, OP was reported for 32 patients. The mean ± SD OP was 13.97 ± 3.91 cm H2O (range: 6-25 cm H2O). None (0%) of the patients had a low OP. 29/32 (91%) had OP in the normal range, and 3/32 (9%) had elevated OP. Pcsf was positively correlated with BMI (p = 0.045); but not correlated with age, sex, prior epidural blood patching, symptom duration, or mean arterial pressure. High variability in Pcsf values was not fully accounted for by predictors included in our model (pseudo-R^2 = 0.188).

Conclusions

First, an absence of low Pcsf should not be used to rule out SIH, as our analysis of patients with CVF shows normal/elevated Pcsf in all included patients. As such, the current IHS criteria for SIH need to be reappraised. Second, such Pcsf readings in patients with active CSF leaks suggest that the development of CVF and incidence of post-treatment rebound headaches might be due to a pre-existing intracranial hypertension. Further research is needed to establish this link, as it has implications both for close post-treatment management of rebound hypertension to prevent potential recurrence of CVF, and also for management of patients with idiopathic intracranial hypertension, to monitor for potential incidence of CVF. Third, our findings support the more appropriate recharacterization of the syndrome of ‘spontaneous intracranial hypotension’ as ‘CSF hypovolemia’. Finally, while BMI is positively correlated with Pcsf, low explanatory power of our model suggests influence of other factors not considered in our analysis.


Low CSF-Pressure Diagnostic Criterion in Spontaneous Intracranial Hypotension is Potentially Misleading in Clinical Practice

Background:
Hormone replacement therapy (HRT) is a cornerstone treatment for menopausal symptoms, yet its impact on gastrointestinal health is not well defined—particularly in women who undergo premature menopause. Given estrogen’s role in modulating gastrointestinal physiology, this study evaluates whether HRT increases the risk of gastroesophageal reflux disease (GERD) and related esophageal pathology differently in women with premature menopause versus postmenopausal women.
Methods:
This retrospective observational study utilized TriNetX, a global federated health research network. Two cohorts of HRT users were created: (1) women ≥45 years with natural menopause (ICD-10 N95.1) and (2) women ≤45 years with premature menopause (ICD-10 E28.31, E28.310, E28.319). All patients had a diagnosis of HRT use (Z79.890). Exclusion criteria included prior diagnosis of GERD or esophageal malignancy. Propensity score matching adjusted for confounders including pregnancy-related diagnoses. Outcomes included new diagnoses of GERD, Barrett’s esophagus, esophageal ulcers, and esophageal cancer following HRT initiation.
Results:
Postmenopausal women on HRT had a significantly higher risk of GERD than premature menopausal women (17.13% vs. 9.44%; Risk Ratio 1.81, 95% CI [1.38–2.39], p < 0.0001). Interestingly, esophageal cancer was more prevalent in the premature menopause group (<0.93% vs. 0%, p = 0.0015). No significant differences were observed in Barrett’s esophagus or esophageal ulcer incidence between groups.
Conclusions:
In this large-scale EHR-based analysis, HRT was associated with a higher incidence of GERD in postmenopausal women, while women with premature menopause experienced a modestly increased rate of esophageal cancer. These findings suggest the need for personalized risk stratification when prescribing HRT. Further research should examine how HRT administration routes influence GI risk profiles and explore these trends in larger, more diverse populations.


Hormone Replacement Therapy and Gastrointestinal Risk in Premature vs. Postmenopausal Women: A TriNetX Cohort Study

Abstract:
Background: Brucella species are zoonotic pathogens responsible for
brucellosis, a systemic bacterial infection primarily transmitted through
direct contact with infected animals or consumption of unpasteurized
dairy products. While Brucella infections following total knee arthroplasty
(TKA) are rare, they pose significant diagnostic and therapeutic challenges
with limited reported cases.

Methods: A systematic review, following the Preferred Reporting in
Systematic Reviews andMeta-Analyses guidelines,was performed on February
22, 2025, using the databases PubMed and Google Scholar for Brucella TKA
periprosthetic joint infections (PJIs) in patients older than 18 years. The search
was further narrowed by excluding articles before 2015 to reflect the most
current trends and practices. Our eligibility criteria were guided by the
Population, Intervention, Comparison, and Outcome framework. We
considered outcomes including, but not limited to, successful eradication of
infection, complications, and functional outcomes following intervention.

Results: Fifteen studies met inclusion criteria. Brucella PJIs in a TKA typically
present late with nonspecific symptoms, often mimicking aseptic loosening or
culture-negative PJIs.Most patients had identifiable riskfactors, including travel
to endemic regions, animal exposure, or consumption of unpasteurized dairy.
Two-stage revision was used in 11 of 15 cases. Conservativemanagement with
implant retentionwas successfulwhen no looseningwas present in 3 out of the
15 cases. The most common antibiotic treatment was doxycycline plus
rifampicin with duration ranging from 3 to 12 months, most commonly for a
total of 6 months. Successful outcomes are possible with prolonged
combinationantibioticsand often require 2-stage revisionarthroplasty, though
diagnostic and treatment approaches vary widely.

Conclusion: This is an updated systematic review of Brucella infections
following TKA within the last 10 years. Given the insidious onset and
potential for chronic infection, orthopaedic surgeons and infectious
disease specialists must be aware of the possibility of Brucella PJIs in
patients with the appropriate history and clinical examination.

A 10-Year Systematic Review of Brucella Periprosthetic Joint Infections Following Total Knee Arthroplasty

Mood Prediction in Perioperative VA Patients Using Machine Learning Detection of Facial Action Units  

Background 
The Facial Action Coding System (FACS) provides an objective framework for coding facial muscle movements, identifying combinations of Action Units (AUs) linked to emotions considered universal across cultures. Detecting these expressions could provide real-time insight into patients’ moods, helping clinicians reinforce positive experiences, improve patient–monitor interactions, and guide behavioral interventions. This study presents early results on identifying which AUs are most strongly associated with self-reported happiness in perioperative patients. 

Methods 
Prospective observational data from perioperative patients at the SF VAMC from 2021-2023 were collected, including self-reported moods. Three independent annotators classified responses into seven categories (anger, disgust, fear, happiness, sadness, surprise, neutral) using a modified Nominal Group Technique to reach consensus. Responses deemed irrelevant or ambiguous (e.g., “drowsy”) were excluded. For each happiness episode, a 30-second window (15 seconds before and after the mood report) was labeled “happy,” with adjacent 15-second segments as internal “not happy” controls. Videos were analyzed at 10 FPS using two Py-Feat pipelines: (1) continuous AU probabilities from Feat-XGB; and (2) binary AU activation from Feat-SVM. Logistic regression identified AUs associated with happiness, reporting odds ratios and 95% confidence intervals (Wald method). P-values <0.01 were considered statistically significant. 

Results 
Across 130 patients, 28 unique happiness episodes were isolated, analyzing 49,508 frames and 20 AUs. The Feat-XGB model identified AU09, AU12, AU24, AU25, and AU28 as positively associated with happiness (p < 0.01), while the Feat-SVM model identified AU02, AU09, AU11, AU15, AU23, AU24, and AU25 (p < 0.01). Notably, AU09, AU24, and AU25 were significant in both models. Although AU25 is canonically linked to happiness, AU09 is typically associated with disgust, underscoring the complexity of interpreting AU co-activation in real-world contexts. AU14 and AU20, linked to contempt and fear respectively, were negatively associated with happiness in both models (p < 0.01). 

Conclusion 
Interpreting the relationship between facial expressions and subjective mood remains inherently complex, especially when using models trained on prototypical expressions and short video segments without broader context. The finding that different machine learning AU detection models identified different sets of happiness-related AUs underscores how methodological choices can shape mood prediction outcomes. Future work should validate automated predictions against human-coded facial annotations and explore deep learning approaches (e.g., ResMaskNet) that better capture contextual and relational interactions across anatomical landmarks. Incorporating additional patient data (e.g., pain, mental health, vital signs) could further improve predictive accuracy and help monitor patients’ emotional progression and comfort during recovery.

Mood Prediction in Perioperative VA Patients Using Machine Learning Detection of Facial Action Units

Lung cancer is the leading cause of cancer-related death globally. Early detection and staging of cancers such as non-small cell lung cancer (NSCLC) is crucial for improved patient prognosis. Endobronchial ultrasound (EBUS) is a minimally invasive technique that allows real-time lymph node visualization, access, staging, and diagnosis of disease in the chest. Even though EBUS-transbronchial needle aspirations (TBNA) have become more available to clinicians, procedural and technical variability has contributed to the lack of EBUS-TBNA standardization. This qualitative study employed semi-structured interviews with three oncologists, three pulmonologists, and three thoracic surgeons to investigate variations in their practices, perspectives, and comprehension of EBUS-TBNA. This study revealed consensus on the central role of EBUS-TBNA and PET/CT for mediastinal staging with divergences in nodal sampling, procedural prioritization, technical challenges, and institutional practices. The EBUS-TBNA procedure can be improved by focusing on two primary areas: integrating multidisciplinary expertise and implementing standardized protocols. These implementations in the standard of care could help us streamline workflow, optimize procedure execution, and ultimately improve patient outcomes.

Physician Perspectives on Endobronchial Ultrasound – Transbronchial Needle Aspiration (EBUS-TBNA): A Qualitative Study

Background:
Wilson’s disease (WD) is an autosomal recessive disorder characterized by copper accumulation in the liver, brain, and other tissues. Neurological features usually include tremors, dystonia, and parkinsonism. Central pontine myelinolysis (CPM) is rarely reported in WD and is typically associated with rapid correction of hyponatremia or fulminant liver failure. Early recognition of CPM as a presenting sign of WD is crucial for timely intervention.
Case Presentation:
A 19-year-old male presented with a history of chronic liver disease of unclear etiology, later developing melena and acute flaccid quadriparesis. Neurological examination revealed proximal weakness, diminished reflexes, and hypophonia with preserved sensation. MRI brain demonstrated symmetric T2 hyperintensities in the central pons consistent with CPM. Slit-lamp examination showed Kayser–Fleischer rings, serum ceruloplasmin was low (8 mg/dL), and 24-hour urinary copper excretion was elevated. Liver function tests revealed mildly deranged transaminases. A diagnosis of WD with hepatic and neurological involvement was established, and the patient was started on zinc acetate, D-penicillamine, and supportive care.
Discussion:
This case highlights CPM as a rare initial neurological manifestation of WD. While commonly associated with osmotic shifts, CPM may also arise from copper-induced neurotoxicity and hepatic dysfunction. The acute motor presentation mimicked peripheral neuropathy but was centrally mediated. Recognition of this atypical presentation can prevent diagnostic delays and facilitate early initiation of chelation therapy.

Wilson’s Disease Presenting with Acute Flaccid Paraparesis: A Rare Neurological Manifestation

Title: Pennsylvania’s 30-Day Rule and Involuntary Commitment: A Case Review of Commonwealth v. Gibson

Background: Involuntary psychiatric commitment laws balance protecting individual liberties while ensuring safety. Pennsylvania’s Mental Health Procedures Act (MHPA) of 1976 requires “clear and convincing evidence” that an individual poses a “clear and present danger” to themselves or others within the past 30 days for long-term involuntary commitment under Section 304. This 30-day dangerousness standard can pose a challenge for psychiatrists who base risk assessments on longer-term clinical history. The 1981 Pennsylvania Supreme Court case Commonwealth ex rel. Gibson v. DiGiacinto highlights the differences between legal criteria and psychiatric judgment.

Case Presentation: Russell Gibson had a documented history of paranoid schizophrenia, medication non-adherence, psychosis, prior violent behavior, and failed community treatment. Multiple clinicians concluded he posed a danger to himself and others. However, legal review of his case focused on the absence of a specific overt act or threat within the 30 days preceding the commitment hearing. In 1978, the Pennsylvania Superior Court agreed with expert opinion. However, in 1981, The Pennsylvania Supreme Court ruled that despite psychiatric testimony, no legal criteria for involuntary commitment were met without recent behavior fitting the 30-day dangerousness standard. The Supreme Court decision emphasized reliance on recent behavioral evidence over clinical risk assessments extending beyond 30 days.

Discussion: This case illustrates the disconnect between clinical psychiatric evaluations, which incorporate long-term behavioral patterns and risk factors, and the MHPA’s 30-day dangerousness standard required by law. Psychiatric testimony assesses risk through multiple factors including past violence, psychiatric history, and medication adherence. Research supports that individuals with psychosis, paranoia, and history of violence are significantly more likely to commit future violent acts. Moreover, studies suggest that in cases involving severe mental illness, repeat involuntary commitment can be predicted up to 64 to 70 days ahead, exceeding the statutory timeframe. The MHPA’s narrow window may limit the effectiveness of psychiatric testimony in legal decisions, potentially undermining patient safety and clinical judgment. This clinical case vignette emphasizes the need for clinicians to understand legal constraints on risk assessment and invites consideration of legal reform to update Pennsylvania’s 30-day dangerousness standard with evidence-based psychiatric risk evaluations.

Pennsylvania’s 30-Day Rule and Involuntary Commitment: A Case Review of Commonwealth v. Gibson

Background:
Community-engaged learning allows an immersive, context-specific experience for medical students to appreciate the impact of social drivers of health (SDOH) (1-4). Community organizations (COs) and community partners are important stakeholders in these learning experiences as they are directly and indirectly impacted by the work of students. Despite the critical role COs play, their perspectives are not traditionally incorporated in curriculum development. In contrast, co-designed curriculum incorporates the knowledge and expertise of COs to produce a practical, context-specific curriculum with shared ownership. We partnered with COs to co-design a revised community-engaged learning curriculum. 


Methods:
Representatives from COs, medical students, residents, and faculty members were recruited to be part of a SDOH curriculum co-design team. Five co-design sessions were held between January to March 2025. Aligned with co-design methodology, participants shared their experiences with the current community-engaged learning curriculum and ideas for improvement, practicality, and sustainability within the community. Summative transcriptions of our co-design sessions underwent rapid qualitative analysis to identify group priorities, and specific revisions were proposed back to the team for further discussion. A finalized curriculum was made with the team and is planned for implementation in the upcoming academic school year. 


Results:

Our co-design team included 14 CO representatives, 4 medical students, 1 resident, and 5 medical school faculty. 

Group priorities included:
1) creating sessions to develop skills in cultural humility and community trust-building prior to working with the community 
2) developing longitudinal partnerships between COs and students to develop a meaningful and tailored educational experience 
3) assessing students’ growth in understanding and appreciation of SDOH 

Our revised curriculum incorporates these priorities in three areas: 
1) Pre-clinical foundational years: Students first learn SDOH concepts, undergo empathy skills training, and simulate response to community needs through case-based discussion and exposure to lived experience narratives. Students are also paired with a CO to begin formulating a longitudinal community project. 
2) Clinical years: Students participate in a required community clerkship where they work with COs on longitudinal community projects. 
3) Fourth year: Students reflect on experiences with community and patients to state the impact of this curriculum on personal and professional growth. 


Conclusion:
Partnering with the community in designing an SDOH community-engaged learning curriculum provides stronger alignment between medical institutions and COs to broaden social impact, enhance transformative learning around SDOH, and more effectively address community needs.


References:
1. Harding D, Uppal S, Meiring S, Golding B, Dutta N, Parekh R. Community-engaged primary care medical education. Clin Teach. 2023;20(3):e13573. doi:10.1111/tct.13573 

2. Bamdas JAM, Averkiou P, Jacomino M. Service-Learning Programs and Projects for Medical Students Engaged With the Community. Cureus. 2022;14(6):e26279. Published 2022 Jun 24. doi:10.7759/cureus.26279 

3. Teherani A, Nikjoo A, den Boer A, Tong MS, Desai A. Community-engaged sustainable health care education. Clin Teach. 2021;18(1):62-68. doi:10.1111/tct.13234 

4. Gimpel N, Kindratt T, Dawson A, Pagels P. Community action research track: Community-based participatory research and service-learning experiences for medical students. Perspect Med Educ. 2018;7(2):139-143. doi:10.1007/s40037-017-0397-2

Co-Designing a Social Determinants of Health Medical School Curriculum in Partnership with Community Organizations

Background: High-quality multiple-choice questions (MCQs) are essential for medical education and licensing exams such as the USMLE Step 1. However, faculty face increasing time constraints and cognitive burden in creating psychometrically sound items aligned with both curriculum and national standards. Generative AI, such as GPT-4o, offers the potential to automate this process, yet no published studies have rigorously compared AI-generated items with faculty-authored questions using blinded expert evaluation and psychometric analysis.
Methods: We conducted a randomized controlled study involving preclinical medical students at a U.S. medical school (2024–2025). Using a structured prompt framework and NBME-aligned rubric, GPT-4o generated 20 USMLE Step 1-style MCQs based on six faculty lectures. Two faculty content experts independently authored 20 matched questions from the same lectures. All questions were evaluated using a standardized 12-point rubric by a blinded national assessment expert. Ten students per course (N=20) completed both AI and faculty questions in randomized order. Student accuracy, response time, and source attribution were recorded. Item-level point-biserial correlations and distractor analysis were conducted.
Results: AI-generated items scored similarly to faculty-authored questions on expert review (mean rubric scores: 11.1 vs. 11.4 out of 12, p=0.3). Student performance did not differ significantly between AI and faculty items (mean accuracy difference: 1.5 points, p=0.8). Point-biserial correlations for correct answers and distractors were equivalent. Students were unable to reliably distinguish between AI and faculty-authored questions (AI: 70%, Faculty: 60%, p=0.5). Notably, performance on AI-generated questions correlated more strongly with final course grades (r=0.665, p=0.001) than faculty-written items (r=0.152, p=0.522). AI-generated explanations were shorter and rated lower in reasoning accuracy. GPT-4o required under 2 minutes to generate 10 questions; faculty required ~3 hours.
Conclusion: GPT-4o, when guided by structured prompts and institutional lecture content, can generate USMLE Step 1-style questions that match faculty-authored items in quality, psychometric performance, and student usability. These findings support scalable integration of generative AI into medical assessment workflows, with implications for faculty efficiency, curricular alignment, and learner preparation. While AI-generated questions matched faculty in structure and performance, their explanations consistently lacked depth and nuance, highlighting a critical area where human oversight remains essential. The custom GPT-4o profile used in this study is publicly available, and we encourage others to adapt it and generate NBME-style questions directly from lecture content.

Generative AI as a Medical Educator: GPT-4o Matches Faculty in Writing Step 1-Style Questions

Background
Texas is experiencing a severe health crisis as Galveston County has the second-highest rate of drug overdose fatalities. Limited access to education about life-saving intervention worsens circumstances for unhoused individuals. Despite interventions such as naloxone – an opioid antagonist that reverses overdoses – becoming more accessible, low health literacy and drug use stigma continue to impede effective use. This project aims to evaluate naloxone literacy among the unhoused in Galveston and assess the impact of targeted educational and street distribution interventions on awareness and utilization.

Methods
From July 2025 to May 2026, participants will be recruited via convenience sampling during biweekly outreach in partnership with the Street Medicine Interest Group and receive care kit compensation. Student volunteers will be trained on proper naloxone use and sensitive surveying. The intervention contains three phases: (1) baseline health literacy assessment asking the questions (“Do you know what naloxone is?” & “Do you know how to use naloxone?”) and post-round interviews of student volunteers, (2) distribution of naloxone kits with culturally sensitive educational materials and referrals to local substance abuse treatment resources, and (3) follow-up surveys and focus group interviews to evaluate changes in knowledge and attitudes. Each participant will be assigned a unique identifier (UID) to classify responses anonymously to assess longitudinal tracking of changes in knowledge and engagement across study phases.

Results
We anticipate enrolling a maximum of 100 unhoused participants over the study period. Outcome variables include quantitative data: the number of participants who can correctly identify naloxone, describe how to administer it, and the accepted distribution rate. Qualitative data includes: volunteer focus groups feedback on stigma, system-level barriers, and the cultural relevance of harm reduction education materials. Survey data was analyzed with descriptive statistics to summarize baseline health literacy, naloxone knowledge, and distribution rates. Pre- and post-intervention surveys are compared using paired t-tests with a significance threshold of p≤ 0.05 applied.
Conclusion
Phase 1 of this project shows that while 72% of the unhoused could identify naloxone, a majority (60%) of those surveyed did not know how to use the drug. Thus, this study offers a scalable, community based model for increasing naloxone access among vulnerable populations through street medicine outreach and peer-driven education. Findings will inform future harm reduction strategies and guide sustainable partnerships between medical institutions and underserved communities to address the ongoing opioid crisis in Texas and beyond.

Downloads

Next from 2025 AMA Research Challenge – Member Premier Access

Low CSF-Pressure Diagnostic Criterion in Spontaneous Intracranial Hypotension is Potentially Misleading in Clinical Practice