United States

Background
Acute hip fractures are a public health problem that affect primarily older adults. Given the subtleties in different patient indications and classifications such as fracture type, preoperative mobility/functional status, and presence or absence of preexisting/symptomatic arthritis, Chat Generative Pre-trained Transformer (ChatGPT) may be a helpful clinical decision-making tool in the treatment plan to optimize patient outcomes. The purpose of this study was to evaluate the accuracy of ChatGPT-4.0 by comparing its appropriateness scores for acute hip fractures with the American Academy of Orthopedic Surgeons (AAOS) Appropriate Use Criteria (AUC) given various patient scenarios. 

Methods
Using the AUC as the gold standard, a numerical scale was utilized to quantify the appropriateness of 6 different treatments. Quantitative scores ranging from 1-9 were assigned as follows: 7-9 signifies “Appropriate,” 4-6 signifies “May Be Appropriate,” and 1-3 signifies “Rarely Appropriate.” ChatGPT-4.0 was prompted separately and asked to assign an appropriate score for each treatment. To measure any notable errors, AAOS AUC scores were subtracted from the ChatGPT-4.0 scores, and the mean error, mean absolute error, and mean squared error were calculated. Spearman’s rank correlation coefficients and paired t-tests were used to determine statistical significance.

Results
30 patient scenarios were evaluated among 6 different treatment options for a total of 180 paired scores. Spearman’s rank correlation testing found that there was a very strong positive correlation between AAOS and ChatGPT-4.0 scores for Multiple Cannulated Screw Fixation, a strong positive correlation between scores for THA and for Hemiarthroplasty, and a moderate positive correlation for Long Cephalomedullary Nails. There was a non-significant weak positive correlation between scores for Short Cephalomedullary Nails and for Sliding Hip Screw ± Anti-Rotation Screw. Using a paired t-test, there were significant differences between scores for Long Cephalomedullary Nails.

Conclusion
The AAOS AUC is designed to optimize treatment outcomes for acute hip fractures in adults by improving the quality of care, and minimizing short- and long-term mortality. The appropriateness scores for THA, Hemiarthroplasty, Long Cephalomedullary Nails, and Multiple Cannulated Screw Fixation determined by ChatGPT-4.0 were positively correlated with the AUC scores. ChatGPT-4.0 overestimated the appropriateness of THA, Hemiarthroplasty, and Long Cephalomedullary Nails. The scores to treat with Long Cephalomedullary Nails were non-equivalent. Although it demonstrated some utility in recommending the appropriateness of treatment options for acute hip fractures in older adults, clinicians should refrain from fully relying on ChatGPT-4.0. The AI model needs improvement to assist physicians and must be further evaluated.

AMA Research Challenge 2023

Can Large Language Models Recommend Treatment for Hip Fractures?

poster

Welcome to the **largest national, multi-specialty research** event featuring research from nearly 700 medical students, residents, fellows, and international medical graduates.

Use the buttons below or navigation menu to access the event.
 


[![](https://assets.underline.io/uploads/markdown_image/1/image/3e33a8c754cc49610b30950617d813e7.png)](https://underline.io/events/434/sessions?searchGroup=all)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/551404d429c4e274f9a54cbdd7081918.png)](https://underline.io/events/434/posters?searchGroup=all)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/677d92f566b880cebd36905fc738fc10.png)](https://underline.io/events/434/lounge)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/c7392833c7261c3874c5b43fa2f4df19.png)](https://underline.io/events/434/help-desk)
 

#### Event Details

The 2023 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
* **Attend education sessions** with physician experts on key research topics

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 sponsored 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 the judging or scoring process to select the winner. The winner will be announced on February 4, 2024 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)

**There is no registration fee for this event! **

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/about/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, score posters in the semifinals (an AMA member exclusive opportunity) and attend education sessions with experienced physicians to learn tips to advance your research and career. 

INTRODUCTION:
Leveraging electronic health records (EHR) and linked biorepositories has the potential to elevate quality, efficiency and
sample size of studies evaluating etiology of rotator cuff tears (RCT). To date, no functionally validated algorithm reliably
identifies rotator cuff tear cases and controls form EHR systems. To facilitate this major need, we developed universal
phenotypic algorithms to classify RCT cases and controls from EHRs.
METHODS:
Using ICD9/10 diagnostic, CPT procedure codes and natural language processing (NLP), two versions of the algorithm
were developed to assess case and control status: a platinum, with image verification, and a silver, without image
verification. Both versions are ICD/CPT based with the option to add NLP. The algorithms were developed and validated
through manual blinded chart review of 525 medical records in the Vanderbilt University Medical Center (VUMC) Synthetic
Derivative.
RESULTS:
Only algorithms performing at 80% predictive value (PV) or higher were considered. Algorithms that only relied on ICD
diagnostic codes for rotator cuff tear had PV < 80%, and were substantially improved by adding evidence of imaging (+PV
=87%). Code based Algorithms relying on surgical CPT codes had the greatest PV of code based algorithms
(+PV=88.7%). For both cases and controls, NLP improved PV (+PV cases: ICD/CPT only= 82%, w/NLP = 98%; -PV
Controls: ICD/CPT only= 96.7%, w/NLP= 94.7%). For cases, addition of NLP increased sample size (ICD/CPT
only=12,777, w/NLP=66,907). Generally, requiring imaging marginally increased PV but substantially decreased sample
size for both cases and controls (Platinum: Cases n= 10,048, Controls n= 14,236; Silver Cases n=69,615, Controls
n=1,825,797).
DISCUSSION AND CONCLUSION:
We successfully developed algorithms that can be applied to any EHR system to identify RCT case/control status. These
phenotypic algorithms will allow the field to efficiently scale up high-quality data from EHRs, paving the way for previously
unimaginable big-data studies in understanding the etiology of RCT.

Universal Algorithms for the Identification of Rotator Cuff Cuff Tear Status From 
Electronic Health Record Systems

Background
Anatomic Total Shoulder Arthroplasty (ATSA) and Reverse Total Shoulder Arthroplasty (RTSA) reliably alleviate pain and restore shoulder function for a variety of indications. However, these procedures are not well-studied in patients with neurocognitive impairment. Therefore, the purpose of this study was to investigate whether patients with dementia or mild cognitive impairment (MCI) have increased odds of surgical or medical complications following arthroplasty.

Methods
The PearlDiver database was queried from April 2010 through October 2021 to identify a cohort of patients who underwent either ATSA or RTSA and had a minimum 2-year follow-up. Current Procedural Terminology (CPT) and International Classification of Diseases (ICD) codes were used to stratify this cohort into three groups: (1) patients with dementia, (2) patients with MCI, and (3) patients with neither condition. Surgical and medical complication rates were compared among these three groups.

Results
The overall prevalence of neurocognitive impairment among patients undergoing TSA was 3.0% in a cohort of 92,022 patients. Patients with dementia had increased odds of sustaining a periprosthetic humerus fracture (OR = 1.46, p <0.001), developing prosthesis instability (OR = 1.72, p < 0.001), and undergoing revision arthroplasty (OR = 1.55, p = 0.003) after RTSA compared to patients with normal cognition. ATSA patients with dementia did not have an elevated risk of surgical complications or revision. Conversely, RTSA patients with MCI did not have an elevated risk of complications or revision, although ATSA patients with MCI had greater odds of prosthesis instability (OR = 2.51, p = 0.008). Additionally, patients with neurocognitive impairment had elevated odds of medical complications compared to patients with normal cognition, including acute myocardial infarction and cerebrovascular accident.

Conclusion
Compared to patients with normal cognition, RTSA patients with pre-operative dementia and ATSA patients with pre-operative MCI are at increased risk for surgical complications. Moreover, both ATSA and RTSA patients with either pre-operative MCI or dementia are at increased risk for medical complications. As the mean age in the U.S. continues to rise, special attention should be directed towards patients with neurocognitive impairment to minimize post-operative complications after TSA.

Total Shoulder Arthroplasty in Patients with Dementia or Mild Cognitive Impairment

Background 

Increased preoperative shoulder dislocations are associated with recurrent instability and greater glenoid bone loss. Initial instability events and subsequent episodes can result in 6.8% and 20% increased glenoid bone loss respectively. Glenoid bone loss increases the risk of instability recurrence and poor outcomes following shoulder stabilization surgery. Low socioeconomic status (SES) indicators, such as high Area Deprivation Index (ADI) and non-commercial insurance status, correlate with emergency room visits and delays to orthopaedic care. This study assesses the correlation between ADI, insurance status, and preoperative glenoid bone loss. We hypothesized higher national ADI and non-commercial insurance would correlate with greater levels of radiographic glenoid bone loss. 



Methods 

339 Bankart repair patients were identified. Data obtained included demographics, national ADI, insurance status, and glenoid bone loss. National ADI was obtained utilizing the Neighborhood Atlas Website and patients’ home addresses. Patients were excluded for prior ipsilateral shoulder surgery (N=16), lack of viewable imaging (N=97), and low-quality magnetic resonance imaging (MRI)(N=34). Glenoid bone loss was measured using the Best-fit circle method on high-quality MRIs (i.e., Fast imaging employing steady-state acquisition (FIESTA)). Three-dimensional alignment of images provided the best glenoid en face view. Researchers were blinded to ADI and insurance status during radiographic analysis and inter- and intra-observer reliability testing. Glenoid bone loss was compared between insurance types using one-way ANOVAs. Before evaluating the ADI and glenoid bone loss association, a scatterplot was generated to qualitatively assess the nature of the relationship. A curve fitting tool determined a quadratic non-linear regression best characterized the association between ADI and glenoid bone loss. 



Results 

192 patients (23% female; 23.9±7.5 years of age; national ADI=16.7%±16.0%) were included. Patients experienced 9.50±6.59% glenoid bone loss. A significant quadratic association between national ADI and glenoid bone loss (R2=0.483, p<0.001) was observed. Individuals with non-commercial insurance experienced greater glenoid bone loss compared to commercial insurance (commercial=8.80±6.46%, other=12.52±6.36%, p=0.003). Interobserver reliability was .910; intraobserver reliability was .948. 



Conclusion 

Our findings indicate a positive association between area deprivation and MRI-measured glenoid bone loss in patients with anterior shoulder instability. Patients with low SES indicators, such as high ADI and non-commercial insurance, are more likely to have greater glenoid bone loss and could be at risk for increased rates of pre- and postoperative instability. While a correlation between low SES indicators and glenoid bone loss has been demonstrated; the association of SES with surgical complexity and functional outcomes in this context should be explored.

Low Socioeconomic Indicators Correlate with Greater Pre-Operative Glenoid Bone Loss

**Scoring is only possible in the semifinals hall**

**Abstract:**



Hospital Characteristics and Perioperative Complications of Hispanic Patients Following Reverse Shoulder Arthroplasty – A Large Database Study

Incidence of Venous Thromboembolism in Fracture Below the Knee with and without Chemical Thromboprophylaxis: A Meta-analysis

INTRODUCTION 
Preoperative anemia is associated with increased postoperative transfusion and complication rates after total knee arthroplasty (TKA). We aimed to create TKA-specific data-driven preoperative hemoglobin strata that quantify the likelihood of 90-day postoperative blood transfusion and evaluate whether these strata are associated with increased risk of 90-day major complications and 2-year prosthetic joint infection (PJI).  

METHODS 
Primary TKA patients from 2013-2022 were identified using a national database. Stratum-specific likelihood ratio (SSLR) analysis defined hemoglobin strata associated with the risk of 90-day blood transfusion. Each stratum was propensity-score matched to the highest identified hemoglobin strata. Unmatched incidence rates and matched risk of 90-day major complications and 2-year PJI between strata were compared.   

RESULTS 
SSLR identified four 90-day blood transfusion hemoglobin strata for males [Strata (g/dL), Likelihood Ratio (5.0-11.4, 8.06; 11.5-11.9, 4.34; 12.0-12.9, 1.70; 13.0-17.0, 0.54)] and females (5.0-10.4, 8.22; 10.5-11.4, 2.84; 11.5-12.4, 1.38; 12.5-17.0, 0.50). Increased 2-year PJI risk was associated with three male strata (12.0-12.9, 11.5-11.9, 5.0-11.4; all p<0.001) and three female strata (11.5-12.4, 10.5-11.4, 5.0-11.4; all p<0.001). Increased 90-day major complication risk was associated with three male strata (12.0-12.9, 11.5-11.9, 5.0-11.4; all p<0.001) and three female strata (11.5-12.4, 10.5-11.4, 5.0-10.4; all p<0.001). 

CONCLUSION 
We identified TKA-specific hemoglobin strata that maximize the likelihood of 90-day blood transfusions and predict the risk of 90-day major complications and 2-year PJI. When using preoperative hemoglobin as a target for anemia optimization, we recommend using these thresholds, noting the increase in complication risk with hemoglobin (g/dL) below 13.0 in males and 12.5 in females.

Stratum-Specific Preoperative Hemoglobin Intervals Quantify the Likelihood of Blood Transfusion, Major Complications, and Periprosthetic Joint Infection Following Total Knee Arthroplasty

Introduction: 
National joint replacement registries allow for evaluation and improvement of joint arthroplasty interventions through extensive data reporting. Many registries now include patient-reported outcome measure (PROM) data. PROM surveys and reporting methods vary widely between registries, and data comparison between them is challenging. The purpose of this study is to: 1) summarize how national joint registries publish patient-reported outcome measure (PROM) survey results differently in their annual reports; 2) explain the utility of being able to compare PROMs data between registries; and 3) identify possible solutions to facilitate harmonization in reporting.

Methods: 
The PROMs reported in the open access 2022 annual reports of five registries were summarized and compared. The five registries were: the American Joint Replacement Registry, Australian Orthopaedic Association National Joint Replacement Registry, Canadian Joint Replacement Registry, National Joint Replacement Registry of the United Kingdom, and the Swedish Arthroplasty Register. 

Results: 
Twelve unique PROMs surveys were identified among the five registries: EQ-5D, EQ-VAS, PROMIS-10, VR-12, Oxford Knee Score, Oxford Hip Score, KOOS-12, HOOS-12, KOOS-JR, HOOS-JR, a patient satisfaction question, and a patient perceived change question. All registries reported different scoring methods. For risk-adjustment, registries used a wide range of different variables to stratify patients.

Conclusions:
Harmonization of the PROMs reporting process will facilitate direct comparison between registries, allowing for identification of which regions have the greatest improvement in PROMs following intervention. Registry data from participating regions can inform future approach, implant, and technique selections worldwide if consistently reported and compared to other registries. Harmonization of reporting could be achieved by standardizing how a minimally important clinical difference (MCID) is calculated or selecting PROMs that fulfill the Outcome Measures in Arthritis Clinical Trials (OMERACT) criteria.

Harmonizing the Presentation of Patient-Reported Outcome Measure Data in Joint Registries

Background:
Oral cavity cancer (OCC) and laryngeal cancer are among the most common cancers worldwide. This study investigated survival in non-Hispanic (NH) Black, NH White, Asian, and Hispanic patients with OCC and laryngeal cancer of low, intermediate, and high neighborhood socioeconomic status (nSES).

Methods:
We used data from the SEER 18 Census Tract-level SES and Rurality Database of the National Cancer Institute to create cohorts of OCC and laryngeal cancer patients from 2013 to 2018. Univariate survival analysis was performed with Kaplan–Meier curves and log-rank P values by nSES and then the cross-classification of race, ethnicity, and nSES. We used Cox proportional hazards regression model for multivariable analysis.

Results:
Higher nSES was associated with better OCC survival for NH White, NH Black, and Asian patients, and better laryngeal cancer survival for NH White, NH Black, Hispanic, and Asian patients. In the multivariable analyses of both OCC and laryngeal cancer survival, NH Black patients had worse survival than NH White patients in the high nSES tertile. NH Black patients with OCC were at higher risk of death than NH White patients at all nSES levels. Conversely, Asian patients with laryngeal cancer demonstrated better survival than other races within the high nSES.

Conclusions:
Overall survival differs between racial and ethnic groups of similar nSESs. These health disparities in patients with OCC and laryngeal cancer reflect broader inequities in the cancer control continuum.

Neighborhood Socioeconomic Status and Racial and Ethnic Survival Disparities in Oral Cavity and Laryngeal Cancer

Background: In the United States, the incidence of head and neck cancer (HNC) is over 54,000 cases per year, resulting in over 11,000 deaths annually. The average hospitalization cost for patients undergoing HNC treatment is $18,371 and the average hospital length of stay (LOS) is 6.6 days. Although much literature focuses on the effect of hemoglobin A1C, weight loss, and smoking status contributing to LOS, less attention has been focused on socioecological metrics. This study aims to identify socioecological factors that may predict increased LOS for patients
undergoing surgical treatment for HNC in Western New York (WNY).
Methods: A retrospective review of patients in WNY with a diagnosis of HNC treated between October 2017 and November 2022. The CDC’s Social Vulnerability Index (SVI) was used at the census tract level. Statistical analysis was conducted using Microsoft Excel and R Studio.
Results: LOS data for 419 patients was analyzed, 31% (n=131) were female, 65% (n=280) identified as white, and 76% (n=319) lived in urban counties. Median LOS was 8 days (range:1-96) for patients from rural counties and 8 days for patients from urban counties (range: 1-189). For areas that had a low or low-medium SVI, average LOS was 13.3 days (SD=18.6). For medium-high and high areas, average LOS was 13.4 days (SD=18.7), demonstrating no significant difference. Poisson regression modeling revealed that distance to hospital, ethnicity, gender, and insurance carrier all contribute significantly to LOS (p<0.0001).
Conclusions: SVI and geographic region independently do not influence LOS for HNC patients, but when socioecological factors are combined into a regression model, there is an influence. This suggests that multiple social determinants of health impact LOS for patients who undergo complex procedures for HNC. Further investigation with multiple institutions in WNY may provide insight into the impact of socioecological factors on LOS in this region.

Socioecological Determinants of Length of Stay for Head and Neck Cancer Patients

Skull base osteomyelitis (SBO) is a rare, but serious condition which can be challenging to diagnose due to broad initial symptomatology, such as headache, fever, or tinnitus, a relatively silent clinical progression which untreated, can cause notable long-term deficits and death. Long courses of antibiotics to target infection and surgical management are potential treatment options, but mortality is noteworthy given the hidden disease nature and ease of progression under delayed treatment. Our paper seeks further insight into predisposing factors of SBO such as comorbidities, gender, age and other clinical factors of significance. We identified 110 patients, 70 of whom had diabetes, and 29 of which had preexisting hypertension, with a mean age of 61.66 ± 10.89 years. The majority (51/110) of patients had solely bacterial pathogens isolated in cultures, 16 were solely fungal, and 11 were a mix of both, while 30 patients had cultures with no isolated organism. Pseudomonas accounted for over half of the bacterial cases, followed by Streptococcus. Most common fungal pathogens include Aspergillus and Rhizopus which were isolated in 9 patients each of 27 total patients with fungal SBO (33.3%). Our population was 73% male and 64% diabetic, supporting the population that is typically regarded being affected by this condition. To best differentiate SBO, initial workup should include CBC with differential, C-reactive protein (CRP) and Erythrocyte Sedimentation Rate (ESR), followed by MRI imaging and nuclear imaging such as Technetium scan, if possible, for stronger evaluation. Treatment has been variable, from broad spectrum antibiotics like ciprofloxacin or ceftriaxone with metronidazole for minimum of 6 weeks IV with up to a year orally, to surgical debridement, if severe and in need of source control, although a strong correlation in surgical treatment to mortality has not been associated.

Premium content

Can Large Language Models Recommend Treatment for Hip Fractures?

Next from AMA Research Challenge 2023

Universal Algorithms for the Identification of Rotator Cuff Cuff Tear Status From Electronic Health Record Systems