United States

Background: Although safety net coverage programs like Medicaid exist to fund health care for low-income patients, the administrative burden of the lengthy application limits uptake by eligible individuals. Therefore, many patients do not attempt to enroll until they become ill, which is common among uninsured adults diagnosed with cancer. Further, key groups, including many non-citizen immigrants (~8% of the US population), are not eligible for Medicaid, and must be covered by local charity care programs. The literature reports worse cancer outcomes for those with Medicaid, but reasons for this are unclear. The delays in enrollment for patients with uncertain coverage at the time of diagnosis – those in limbo – are understudied and may explain some of this persisting disparity. 

Objective: We estimated the patient and health system burdens for those with uncertain coverage status at the time of cancer diagnosis at Parkland Health, the main provider of cancer treatment for the uninsured in Dallas, Texas. 

Methods: We assembled the electronic health record financial encounter data of patients with an incident cancer diagnosis in the Parkland tumor registry, diagnosed from 2010 – 2022. We defined patients as being “in limbo” if they had self-pay, pending, or missing insurance plans at the date of diagnosis. We also included patients whose coverage verification date occurred after diagnosis date, since coverage programs can apply retroactively for 90 days. We used plan verification dates to identify when limbo status resolved. We used coarsened exact matching to balance across age, race, Spanish-language preference, Charlson, and screenable cancer type, then multivariate regression to compare patients that were in limbo to those that had established coverage before cancer diagnosis. 

Results: We identified 17,468 adults newly diagnosed with cancer, 80% of whom were &lt; age 65, 55% female, 73% Hispanic and/or Black, and 30% were Spanish-language preferring. The most common cancer types were GI/colorectal (19%), breast (17%), and lung (9%); 23% were advanced stage at diagnosis and 13% died within 180 days of diagnosis. 52.1% were already enrolled in Medicaid/charity care, but one-third (35%) were in limbo at diagnosis, which lasted for a median 31 days (IQR 13 – 126). Over half (55%) of patients in limbo resolved to Medicaid or charity care. Patients who experienced limbo had a median of 596 financial account notes in the first year after diagnosis, 121 more than the comparison group (95% CI 86.7 - 156.1) and began treatment 10 days later (95% CI 7.7 - 12.7). They also had a median of 1 ED visit in the year after diagnosis, 0.2 more than patients never in limbo (95% CI 0.11 - 0.26). 

Conclusions: The administrative burden of enrolling into coverage programs for the uninsured may impose direct and indirect costs on patients and safety-net health systems and may contribute to worse cancer care outcomes. A more comprehensive accounting of the effects of administrative burdens might suggest ways to redesign coverage enrollment for the uninsured to improve care and reduce costs.

AMA Research Challenge 2024 

Impacts of the administrative burden of establishing healthcare coverage for the uninsured​

Background: Although safety net coverage programs like Medicaid exist to fund health care for low-income patients, the administrative burden of the lengthy application limits uptake by eligible individuals. Therefore, many patients do not attempt to enroll until they become ill, which is common among uninsured adults diagnosed with cancer. Further, key groups, including many non-citizen immigrants (~8% of the US population), are not eligible for Medicaid, and must be covered by local charity care programs. The literature reports worse cancer outcomes for those with Medicaid, but reasons for this are unclear. The delays in enrollment for patients with uncertain coverage at the time of diagnosis – those in limbo – are understudied and may explain some of this persisting disparity. 

Objective: We estimated the patient and health system burdens for those with uncertain coverage status at the time of cancer diagnosis at Parkland Health, the main provider of cancer treatment for the uninsured in Dallas, Texas. 

Methods: We assembled the electronic health record financial encounter data of patients with an incident cancer diagnosis in the Parkland tumor registry, diagnosed from 2010 – 2022. We defined patients as being “in limbo” if they had self-pay, pending, or missing insurance plans at the date of diagnosis. We also included patients whose coverage verification date occurred after diagnosis date, since coverage programs can apply retroactively for 90 days. We used plan verification dates to identify when limbo status resolved. We used coarsened exact matching to balance across age, race, Spanish-language preference, Charlson, and screenable cancer type, then multivariate regression to compare patients that were in limbo to those that had established coverage before cancer diagnosis. 

Results: We identified 17,468 adults newly diagnosed with cancer, 80% of whom were < age 65, 55% female, 73% Hispanic and/or Black, and 30% were Spanish-language preferring. The most common cancer types were GI/colorectal (19%), breast (17%), and lung (9%); 23% were advanced stage at diagnosis and 13% died within 180 days of diagnosis. 52.1% were already enrolled in Medicaid/charity care, but one-third (35%) were in limbo at diagnosis, which lasted for a median 31 days (IQR 13 – 126). Over half (55%) of patients in limbo resolved to Medicaid or charity care. Patients who experienced limbo had a median of 596 financial account notes in the first year after diagnosis, 121 more than the comparison group (95% CI 86.7 - 156.1) and began treatment 10 days later (95% CI 7.7 - 12.7). They also had a median of 1 ED visit in the year after diagnosis, 0.2 more than patients never in limbo (95% CI 0.11 - 0.26). 

Conclusions: The administrative burden of enrolling into coverage programs for the uninsured may impose direct and indirect costs on patients and safety-net health systems and may contribute to worse cancer care outcomes. A more comprehensive accounting of the effects of administrative burdens might suggest ways to redesign coverage enrollment for the uninsured to improve care and reduce costs.

poster

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.
 

[![](https://assets.underline.io/markdown_image/1/image/20e1e77aa954efeac7a15d84c85eb383.jpg)](https://underline.io/events/468/sessions?searchGroup=all)
 
[![](https://assets.underline.io/markdown_image/1/image/068522d81aca0b68a369787c7dfccd1b.jpg)](https://underline.io/events/468/posters?searchGroup=all)
 
[![](https://assets.underline.io/markdown_image/1/image/d3971ece79d5e5575b178bdc2c8bbfae.jpg)](https://underline.io/events/468/help-desk)
 

#### 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. 

### 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/)

**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 or Log-in 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).

Background: The catastrophic Crowdstrike patch failure of July 19, 2024, exposed criticalvulnerabilities in global healthcare systems, stemming from a memory safety issue in C++ code. This null pointer error, a common pitfall in languages without automatic memorymanagement, led to system-wide failures in Microsoft-based environments while GNU/Linux and Apple systems remained unaffected. This event underscores the urgent need for robust,quantum-resistant cryptographic solutions in healthcare IT infrastructure.

Methods: We developed a protocol for building and benchmarking National Institute of Standards and Technology (NIST)-endorsed classical and post-quantum encryption algorithmson-premesis, using consumer grade Linux computers to prioritize viability for underservedregions & underfunded institutions. We compiled OpenSSL with Open Quantum Safe (OQS) Clibrary to enable post-quantum encryption development that allowed the same level of accessas Crowdstrike's faulty driver code while allowing for bindings with numerous memory safeprogramming languages. Our focus on post-quantum Key Encapsulation Mechanism (KEM) encryption reflects the ubiqutious protection that these protocols provide to securecommunication and knowledge-work as well as the relative ease of hybridization with classicalencryption protocols like Elliptical Curve Diffie-Hellman (ECDH). Following on-devicecompilation and installation of the encryption binaries, we built and executed an evaluationscript with OpenSSL's native toolkit for twenty-four NIST-endorsed KEM protocols consisting ofclassical, quantum, and hybrid KEM implementations. We evaluated the KEMs on the numberand rate of key generations (keygen), key encapsulation (encap) rate, and key decapsulations(decap) and rated their NIST post-quantum security level according to NIST advancedencryption standard (AES) exaustic key search levels.

Results: We successfully benchmarked all 24 KEM protocols, producing an examplepublic/private key pair following the evaluation. The 24 KEM protocols are evenly split acrossNIST security levels 1, 3, and 5, with 8 protocols at each.We made all relevant code, regulatoryinformation, and the example cryptographic key pairs available on the Qompass AI Github page.We released them under the GNU Affero General Public License (AGPL) to maintain the freeavailability of these encryption tools to benefit communities.Conclusion: Out of the evaluated KEMs, we propose hybrid combinations of ECDH and Kyber for most acute adoption of enhanced encryption protocols due to the layered security of nascentpost-quantum encryption with established efficient classical protocols. Currently, Google Chrome implements Kyber768/X25519 hybrid encryption as part of its Transport Layer Security (TLS), offering a familiar and accessible platform to perform institutional assessements.

Kyber Odyssey: Charting a course for secure innovation in a post-Crowdstrike world

Unhoused trauma patients present with unique injury patterns and comorbidities. While this population has a higher prevalence of substance use, the existing literature does not characterize rates of polysubstance use. This study aims to evaluate differences in substance use in housed and unhoused trauma patients at a national scale, particularly for rates of polysubstance use. 

Methods
All patients included from 1/1/2021 to 12/31/2021 in the National Trauma Database (NTDB) were retrospectively sorted into housed and unhoused groups, and their drug and alcohol screening results were collected. The NTDB reports the screening of alcohol and 13 types of drugs, of which tricyclic antidepressants were excluded. A BAC greater than or equal to 0.08% was considered alcohol positive. Polysubstance use was defined as positive screening results of 2 or more substances (drugs and/or alcohol). All statistical analyses were performed in R4.4.0.

The NTDB 2021 dataset contains a total of 1,209,097 patients, of which 9537 (0.79%) identified as unhoused. 82.64% of unhoused patients were male compared to 57.98% of housed patients. 

Unhoused patients underwent alcohol testing more frequently than housed patients (RR=1.80, 95% CI [1.78, 1.82]). Of those screened for alcohol, the unhoused had a higher likelihood of positive test results (RR=1.36, 95% CI [1.31,1.41]). Unhoused patients were also more likely to be drug tested than housed patients (RR=2.12, 95% CI [2.08, 2.15]). Of those screened for drugs, unhoused patients were more than twice as likely to test positive (RR=1.73, 95% CI [1.70,1.76]).

Compared to housed patients, unhoused patients were more likely to be screened for both drugs and alcohol (RR=2.21, 95% CI [2.17, 2.26]). In patients screened for both drugs and alcohol, a positive drug and alcohol screening was use more likely in the unhoused (RR=2.83, 95% CI [2.65, 3.01]), and negative testing for all substances was 30% less likely in the unhoused (RR=0.70, 95% CI [0.70, 0.71]). When accounting for all polysubstance combinations, the unhoused had a significantly higher risk of preinjury polysubstance use. (RR=2.59, 95% CI [2.51, 2.66]). 


Conclusion
Unhoused trauma patients have higher rates of positive tests for alcohol, drug, and polysubstance use than housed patients; however, there is a discrepancy in substance screening based on housing status.

Polysubstance Use in Unhoused Trauma Patients: Results from Drug
and Alcohol Screening in the National Trauma Data Bank

Background:
The incidence of endometrial cancer in the United States has increased, disproportionately affecting medically underserved women of lower socioeconomic backgrounds and minority ethnicities. Previous studies have shown that women with endometrial cancer residing in rural communities have a worse prognosis and lower survival rate than those of urban settings, but few have evaluated the relationship between location of residence and stage at diagnosis. We aim to determine whether rural versus urban residence influences stage of endometrial cancer at diagnosis.

Methods:
Data from the National Cancer Institute’s Surveillance Epidemiology and End Results (SEER) database from 2000-2019 was included. Eligible participants were women ages 18 and older with a diagnosis of endometrial cancer (ICD-0-3 C54.1). Rural versus urban population was the exposure variable, with stage of the endometrial cancer at diagnosis (localized vs regional / distant) as the outcome. Statistical analysis methods included exploratory and descriptive to characterize the population, Chi-Square test to determine the unadjusted association of the exposure and outcome, and logistic regression to calculate the unadjusted and adjusted odds ratio (OR) and 95% confidence intervals (CI).

Results:
192,973 women were included (59.1% aged 60+, 70.0% of white race, 88.5% residing in urban areas). Advanced stage of diagnosis of endometrial cancer was seen in 28.1% of women and was more frequent among those residing in urban areas, aged 80+, of Non-Hispanic Black ethnicity, or with Type 2 histology (p<0.001). The unadjusted odds of being diagnosed at regional/distant stage decreased by 6% in individuals of rural areas compared to those in urban areas (OR 0.94, (95% CI 0.91-0.97, p<0.001)). When adjusting for confounders, this risk became nonsignificant (OR 0.99, 95% CI, 0.95-1.02,) p=0.41)). Women aged 80+, of Non-Hispanic Black ethnicity, or with Type 2 histology had an increased odds of advanced stage at diagnosis of 41%, 45%, and nearly 400% respectively.

Conclusion:
Residential status (urban vs rural) was not associated with the stage of endometrial cancer at diagnosis. However, our study found significant associations between advanced stage and age, race, and histology type, therefore suggesting the presence of additional health disparities affecting the diagnosis of endometrial cancer.

The Effect of Rural versus Urban Residence on Stage of Endometrial Cancer at Diagnosis

With limited healthcare resources, clinics and physicians face productivity challenges. Currently, minimal workflow standardization exists. As electronic health record systems (EHR) offer more capabilities to track times for which clinic stakeholders – physicians, patient flow coordinators, nurses, technicians, trainees, and other staff – interact with the system, such secondary data can reflect clinic activities. Despite the large data repository, little is known about its accuracy and applications. Therefore, this study aimed to understand the accuracy of time-motion metrics within EHR data and demonstrate preliminary trends in the clinic workflow.

Methods
This study was conducted at the Ambulatory Cardiac Clinics at Peter Munk Cardiac Center, University Health Network in Toronto, Canada. Two trained observers collected a convenient sample of over 60 hours of observational data using a time-motion approach during 7 clinic days. Automated data captured by this clinic’s EHR, EPIC, of all in-person outpatient visits during regular hours between January to December 2023 were used. Time-related and descriptive variables with >10% missingness were excluded. The acceptable error for accurate timestamp and interval variables was ±5 minutes and ±5% of the total interval time, respectively, for ≥ 90% of EPIC data compared to observational data. Using acceptable variables as metrics, EPIC data was analyzed using summary statistics and multivariable regressions to capture preliminary clinic workflow trends.

Results
Across 9 cardiologists, 158 observational data records were compared with EPIC data. Of 49 time- and date-related variables, only 6 had less than 10% missingness (appointment date, appointment weekday, appointment time, appointment length, patient check-in time, and department-recorded check-in time), all of which were accurate. 15 descriptive variables also had less than 10% missingness. During the study period, 51,887 records from EPIC during weekdays 7 am - 5 pm were analyzed. The median patient appointment lasted 20 minutes [interquartile range 15, 60]. Furthermore, morning and afternoon clinic transitioned from 12 – 1pm and the 30-minute appointment load peaked during the middle of each morning and afternoon clinic. 

Conclusion
Secondary data automatically collected as part of the EPIC system is accurate and can be used to measure, manage, and improve clinic efficiency. However, further standardization of when stakeholders engage with the EHR is necessary to reduce missingness and increase the number of variables as clinic efficiency metrics. Additionally, more variables allow for regionalization, physician workload, and patient satisfaction to be determined for clinic function and management in addition to appointment-related metrics.

Standardizing Secondary Data from Electronic Health Record System to Assess Clinic Workflow – A Time-Motion Approach

Background
Developments in artificial intelligence (AI) have enabled programs to mimic human behaviors, including thought and speech. Capabilities of AI (e.g. ChatGPT) include summarizing findings of previous research, generating and analyzing data, and assisting in writing and editing. However, these capabilities also raise concerns surrounding copyright, attribution, plagiarism, authorship, accuracy. This cross-sectional study aims to assess how dermatology journals address the use of AI for authors and reviewers and determine if there is a difference in impact factor (IF) between journals with and without AI statements.

Methods
Dermatology journals, ranks, and IFs were identified from The Observatory of International Research in July 2024. Journals’ statements on AI were collected from their guidelines pages. When the journal deferred to the publisher for AI guidance, statements were also collected from the publishers’ guidelines pages. The Shapiro-Wilk test assessed the distribution of the data. The Mann-Whitney-Wilcoxon test examined differences in IFs between journals with and without AI statements. Significance was set at p<0.05.

Results
A total of 59 journals were included in the study, 29 (49.2%) and 41 (69.5%) of which did not have AI statements for authors and reviewers, respectively. The most common author guidelines included: requiring disclosure of AI use (n=29, 49.2%), prohibiting listing AI as author (n=27, 45.8%), and allowing AI to enhance readability and language (n=12, 20.3%). For reviewers, the most common guidelines included: cautioning against AI use due to confidentiality concerns (n=16, 27.1%), requiring declaration of AI use in peer review reports (n=5, 8.5%), and prohibiting AI use in peer review (n=3, 5.1%). No significant difference in IF was found between journals with and without AI statements for authors (p=0.26); however, journals with higher IF were more likely to have AI statements for reviewers (p=0.03).

Conclusion
We found that nearly half of the dermatology journals included in our study lacked AI guidelines for authors, and even more did not have guidelines for reviewers. Journals of all impacts have begun to release AI statements for authors, but higher impact journals are more likely to provide guidance to reviewers. As AI technology continues to evolve, it is crucial for journals to establish comprehensive and clear guidelines to ensure ethical and accurate use of AI in research. In the meantime, dermatology researchers should adhere to these guidelines to ensure transparency in their work, and readers should be aware of these standards when interpreting research findings.

A cross-sectional study assessing AI guidelines across impact factors in dermatology journals: implications for authors and reviewers

Background: The average cost for tuition, fees, and living expenses at 50% of United States medical schools exceeds $275,000. In addition, a career in medicine requires 3-7 years of residency training and sometimes 1-3 years of fellowship training at a fraction of their projected income. Virtually no data exists on the credit profiles of internal medicine borrowers and the reasons for their borrowing. In the first analysis of its kind, we assessed the credit profiles of physician borrowers by FICO score, training or career stage, and reasons for borrowing.

Methods: This cross-sectional study was based on pooled FICO scores from 974 internal medicine trainee and attending borrowers from 2019-2023 sourced from a privately held physician and dentist financial services company in compliance with the Equal Credit Opportunity Act (ECOA). Borrowers applied for loans from October 1, 2019 to May 31, 2023. Descriptive analyses were performed characterizing the median borrower FICO score based on status (trainee vs. attending), years from training, and reasons for loan requests. FICO score excludes age, employment history, gender, location, income, marital status, national origin, and race to protect customers from potentially discriminatory credit decisioning. The FICO score pulls data from the three credit bureaus (Equifax, Experian, and Transunion) weighing the following categories when being calculated: payment history (35%), amounts owed (30%), length of credit history (15%), recent credit applications (10%), and credit mix (10%).

Results: The national median FICO scores were 679 (n=500; IQR=648-721) among trainees and 731 (n=474; IQR=683-778) among attending physicians (p<0.001). There was no statistically significant difference among physician FICO scores by geography. Credit card consolidation was the most common reason for borrowing across both cohorts followed by moving/relocation among trainees and home improvement among attendings. 

Conclusion: The median FICO scores of trainees is 11 points lower than the average American of their generation; for attendings, the median FICO score is 4 points higher. This is surprising given the personal character and comparative salary of physicians. Financial stresses reflected in FICO score can contribute to broader physician burnout and further the cycle of stress and shortage.

Financial Wellness in Medicine: A Fair Isaac Corporation (FICO) Score Cross-Sectional Analysis of United States Internal Medicine Trainees and Attendings

Studies also show that the effect of social risk factors on health is cumulative such that patients who experience more social risk factors have poorer health-related quality of life. Although low educational attainment, specifically, not completing high school, is known to lead to poor health and possible premature death; little is known about the association between low educational attainment and other social risk factors that can impact health. We identified and compared the prevalence of social risk factors in patients with less than high school education and patients with high school education or greater to determine what factors may be further contributing to poor health outcomes in patients with low educational attainment.

We reviewed results from social determinants of health screenings performed from October 2021 to October 2023 at a primary care clinic in Hartford, Connecticut. We compared the prevalence of social risk factors among patients who did and did not complete high school using Chi-square test of independence (p <0.05).

Volunteers screened 2,153 patients, 386 (17.9%) of whom did not complete high school. Compared to patients with high school education or greater, patients with less than high school education had a higher prevalence of unemployment (17.4% v. 13.4%, p=0.041), food insecurity (15.5% v. 9.8%, p=.0067), difficulty affording utilities (16.9% v. 9.8%, p=.0001), lack of transportation (30.8% v. 14.8%, p<0.001), and difficulty accessing healthcare/medication (5.1% v. 2.4%, p=0.007). The two groups had a similar prevalence of homelessness (9.0% v. 6.7%, p =0.17).

Our results show that failing to complete high school is associated with a higher prevalence of numerous social risk factors. Independently low educational attainment and the individual risk factors we investigated negatively impact health, but together their effects can compound and contribute to further worsening health outcomes. These findings argue that patients with low educational attainment should be routinely screened for social risk factors so, all factors that negatively impact their health can be identified and addressed.

Low Educational Attainment Linked to Social Risk Factors that Worsen Health Outcomes

Background: Health literacy (HL) is a cornerstone for favorable and equitable health outcomes, yet nearly 9 in 10 U.S. adults struggle with limited HL associated with higher rates of hospital readmissions, medication nonadherence, and all-cause mortality. This disproportionately affects marginalized populations, especially those from racial and ethnic minorities. While the 21st Century Cures Act aims to mitigate differential access to health information by providing patients with access to their clinical notes in electronic health records (EHRs), most are unable to utilize them for health-based decisions and adequate management of care. As health systems shift towards a culture of shared decision-making, limited readability of EHRs remains a key barrier in our goal of empowering patients as agents of their own care. Thus, there is a critical need to improve comprehensibility of clinical notes and facilitate equitable outcomes at a systems level. Here, we evaluate the utility of GPT4-based plain language translation of discharge summary notes (DSNs) on subjective and objective comprehension, confidence, and time spent reading as well as its differential effect on various patient demographics. 

Methods: 553 patients were enrolled from December 2023 to February 2024 and read four DSNs (two untranslated, two GPT4-translated) on common admission diagnoses on an inpatient general medicine service: congestive heart failure (CHF), community acquired pneumonia (CAP), diabetic ketoacidosis (DKA), acute ischemic stroke (AIS). After reading each DSN, patients answered questionnaire items assessing subjective perceptions and objective accuracy of comprehension. Linear mixed models were used to analyze effects of translation on comprehension outcomes. 

Results: Across all four DSNs, GPT4-based translation improved objective comprehension by 61% (β=1.18, p<0.001), reduced time spent reading by 51% (β=-0.57, p<0.001), improved self-reported confidence by 45% (β=1.94, p<0.001), and improved subjective comprehension by 18% (β=2.32, p<0.001). It also strengthened the association between patient confidence and objective comprehension by 67% (p<0.01). Improvements were greater in Black and Hispanic patients, older patients, male patients, and those who reported limited health knowledge (p<0.01). In particular, objective comprehension improved an additional 22% and 39% among Black and Hispanic patients respectively (p<0.01). 

Conclusion: GPT4-based translation of DSNs substantially improved patient comprehension of disease course and management while optimizing time spent reading them, and to a greater degree in marginalized populations with historically low rates of HL. Implementation of GPT4-translation in EHR interfaces for a patient-friendly supplement to provider notes offers great promise in closing the health literacy gap across a spectrum of patient demographics in medicine.

Using OpenAI to Improve Patient Comprehension of Clinical Notes and Bridge Health Literacy Gaps in Internal Medicine

Summary of Background Data 
Concordance with evidence-based medicine (EBM) guidelines is associated with improved clinical outcomes in spine surgery. The North American Spine Society (NASS) has published coverage guidelines on indications for lumbar fusion surgery, with a recent survey demonstrating a 60% concordance rate across its members. GPT-4 is the latest popular deep learning model that receives knowledge training across public databases including those containing EBM guidelines. There is a paucity of information regarding the utility of artificial intelligence in the clinical decision-making process for patients with lumbar pathology. 

Purpose 
To assess GPT-4’s responses to validated clinical vignettes and evaluate them for concordance with NASS clinical guidelines and recommendations for lumbar fusion surgery. 

Study Design 
Comparative Analysis and Narrative Review 

Patient Sample 
Adult patients in the United States evaluated for lumbar spine fusion surgery. 

Methods 
Seventeen well validated clinical vignettes with specific indications for or against lumbar fusion based on NASS criteria were obtained from a prior published research study. Each case included history of present illness, physical exam findings, previous treatments attempted, and definitive imaging interpretations. The cases were transcribed into a standardized prompt and entered into GPT-4 to obtain a decision whether fusion is indicated. The query for each case was repeated three times under identical conditions to evaluate the model’s inter-query reliability. If there was internal disagreement between the queries for a given case, GPT-4’s decision for that case was determined by majority rule. Queries were all entered in separate strings to ensure that no contextual memory in the software influenced subsequent responses. The investigator entering the prompts was blinded to the NASS-concordant decisions for the cases prior to complete data collection. GPT-4 inter-query reliability was assessed using the Fleiss’ Kappa statistic. Differences in decision-making between GPT-4 and NASS guidelines were analyzed with Chi-square analysis. 

Results 
GPT-4 responses for 14/17 (82.4%) of the clinical vignettes were in concordance with NASS EBM lumbar fusion guidelines. There was a significant association in clinical decision-making when determining indication for spine fusion surgery between GPT-4 and NASS guidelines (χ² = 7.14; p < 0.01). There was substantial agreement among the sets of responses generated by GPT-4 for each clinical case (K = 0.71; p < .001). 

Conclusions 
There is significant concordance between GPT-4 responses and NASS EBM indications for lumbar fusion surgery. Artificial intelligence and deep learning models may prove to be an effective adjunct tool for clinical decision-making and medical education within modern spine surgery practices.

Evaluation of GPT-4 Concordance with North American Spine Society Guidelines for Lumbar Fusion Surgery

Academic outpatient clinics report appointment non-adherence rates between 8-20%.As unfilled appointments represent a financial loss to the system, patient-directed financial penalties emerged as a strategy to offset departmental losses. 
We aimed to examine the impact of patient financial penalties on appointment non-adherence within a single academic institution's obstetrics and gynecology (OBGYN) outpatient clinics.
This retrospective policy effectiveness-implementation hybrid study included administrative data from all appointments at Eastern Virginia Medical School's OBGYN outpatient clinics from May 1, 2018 to April 30, 2022.
The institution-wide patient financial penalty policy for appointment non-adherence was implemented on May 1, 2020. Institutional Review Board approval was waived for this study. 
We conducted an interrupted time series analysis (ITS) stratified by clinic and insurance after adjusting the model for the effect of COVID-19.
There were 414,006 OBGYN outpatient clinic appointments; 58,473 (14.1%) were not attended. The mean (standard deviation) per month was 8391 (727) appointments and 14.3% (1.48%) non-adherence rate. Medicare/Medicaid patients had a higher non-adherence rate (36,872/188,343 [19.6%]).
COVID-19 was a significant confounder across all clinics and insurances. Adjusted ITS models observed a significant sustained effect on percent of non-adherent appointments per month (+0.18%; p<0.01). Stratified by clinic, maternal-fetal medicine (-1.47%; p=0.01), minimally invasive gynecology (+4.62%; p<0.01) and urogynecology (+4.42%; p=0.01) observed a significant immediate effect. Maternal-fetal medicine (+0.09%; p=0.02) and generalists (-0.33%; p<0.01) observed a significant sustained effect. 
Stratified by insurance, publicly insured patients (-3.99%; p=0.048) observed a significant immediate effect. Publicly (+0.21%; p<0.01) and non-publicly (+0.09%; p=0.03) insured patients observed a significant sustained effect.
A clinically significant reduction in appointment non-adherence was not observed 
in a single academic center’s OBGYN outpatient clinics after institutional implementation of patient-directed financial penalties for appointment non-adherence. The impact may differ based on clinic or patient insurance type.

Premium content

Impacts of the administrative burden of establishing healthcare coverage for the uninsured

Next from AMA Research Challenge 2024

Kyber Odyssey: Charting a course for secure innovation in a post-Crowdstrike world

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Premium content

Impacts of the administrative burden of establishing healthcare coverage for the uninsured​

Next from AMA Research Challenge 2024

Kyber Odyssey: Charting a course for secure innovation in a post-Crowdstrike world

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Impacts of the administrative burden of establishing healthcare coverage for the uninsured