Virginia Hospital Center

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

This study evaluated the effectiveness of large language models (LLMs) and vision-language models (VLMs) in gastroenterology. We used board-style multiple-choice questions to assess the performance of both proprietary and open-source LLMs and VLMs—including GPT, Claude, Gemini, Mistral, Llama, Mixtral, Phi, and Qwen, across different interfaces, computing environments, and levels of compression (quantization). Among the proprietary models, o1-preview (82.0%) and Claude3.5-Sonnet (74.0%) had the highest accuracy, outperforming the top open-source models: Llama3.3-70b (65.7%) and Qwen-2.5-72b (61.0%). Among the small quantized open-source models, the 8-bit Llama 3.2-11b (51.7%) and 6-bit Phi3-14b (48.7%) performed the best, with scores comparable to their full-precision counterparts. Notably, VLM accuracy on image-containing questions improved (~10%) when given human-generated captions, remained unchanged with original images, and declined with LLM-generated captions. Further research is warranted to evaluate model capabilities in real-world clinical decision-making scenarios.

This paper presents a comprehensive framework that leverages a hierarchical federated learning (HFL) architecture to enhance the performance and privacy of facial recognition and behavior recognition systems within Human-Computer Interaction (HCI) applications. The core of this framework is the Actor-Critic-based client selection mechanism, which uses reinforcement learning to dynamically optimize client participation in federated learning. This approach intelligently balances the exploration of new clients with the exploitation of high-performing ones, improving both model efficiency and accuracy. Additionally, the framework introduces a multi-dimensional weighted aggregation method that incorporates factors such as data volume, loss values, and uncertainty, further enhancing the robustness and accuracy of the global model. Experimental results on the CIFAR-10, CelebA, and WISDM datasets demonstrate that the proposed method significantly outperforms traditional random client selection strategies, especially in heterogeneous data environments, achieving approximately a 4% improvement in final accuracy. The results highlight that the framework not only enhances training efficiency and model performance but also ensures robust privacy protection, making it particularly suitable for sensitive data applications such as facial recognition in HCI systems.

ALIF (anterior lumbar interbody fusion) and other spinal fusion surgeries are among the most common orthopaedic procedures requiring blood transfusions. However, blood transfusions have been associated with various complications, including adverse reactions and infections. The present study aims to identify independent risk factors for blood transfusions in patients undergoing single-level ALIF specifically to better identify high risk patients and optimize perioperative management. All patients who had undergone single-level ALIF patients for the treatment of degenerative spinal conditions, excluding traumatic, pathologic, and infectious etiologies, were identified by querying a multi-institutional surgical registry from 2005 to 2018. Multi-level fusions, PLIF/TLIF, and posterior procedures were also excluded. Mann-Whitney-U-Tests were used to analyze continuous variables, while Fisher's-Exact-Tests/Bonferroni-Corrected-Tests were used for categorical variables. Multivariate logistic regression analysis with alternating backward stepwise elimination and forward entry was implemented to identify significant predictors for blood transfusions within 72 hours after incision. The predicted probabilities were used in post-regression diagnostics to generate a Receiver Operating Characteristic (ROC) curve to assess model performance. 4,792 single-level ALIF patients met inclusion criteria - 183 (3.82%) had received blood transfusions within 72 hours after incision and 4,609 (96.18%) had not. Age ≥60 years (OR 1.954, p<0.001), preoperative transfusions (OR 33.758, p=0.023), extended operative times (≥197.0 minutes; 75th percentile) (OR 4.645, p<0.001), ASA≥3 (OR 1.395, p<0.001) and preoperative hematocrit levels (Hct) 30.00-37.99 (OR 1.562, p=0.016) and Hct <30.00 (OR 6.334, p<0.001) were shown to be significant independent risk factors for perioperative blood transfusions. The area under the ROC curve (AUROC; C-statistic) was 0.759 (p<0.001), indicating relatively strong discriminatory ability/predictability of the final model. Several independent risk factors including age ≥60 years, preoperative blood transfusions and extended operative times increased risk for blood transfusion following single-level ALIF. The present study aims to help surgeons identify high-risk patients to better communicate postoperative expectations and optimize patients to reduce the risk of transfusions and secondary complications. Level of Evidence: III.

Introduction: The overall survival of cardiac arrest patients is very poor. There is a higher incidence of cardiac arrest on weekends especially on Saturdays as per national statistics. Also, hospitals experience reduced staffing on the weekends. We examined the outcomes of cardiac arrests on weekends compared to weekdays. Hypothesis: There is no difference in mortality in cardiac arrests which happen over the weekday compared to over the weekend. Methods: This retrospective cohort study used a national inpatient sample database to identify patients from 2016 to 2019 presenting with cardiac arrest on weekdays vs on weekends. SAS 9.3 was used for data analysis. The data were weighted to generate national estimates of the entire US hospitalized population. Results: Out of 57,300 patients who presented with cardiac arrest during the study period, 41,410 were on weekdays (72.2%; mean age 65.9; 56.2% males) and 15,890 were on weekends (27.8%, mean age 65.2; 56.7% males). Inhospital mortality was less on weekdays than on weekends (72.4% vs 76.2%). The younger age group (18-40 years) had a higher share of of cardiac arrest on weekends when compared to weekdays (8.2% vs 7.1%). Weekday admissions had a higher rate of prior MI (8.9% vs 7.6%) and a higher rate of comorbities including DM (39.4% vs 36%), CHF (32.2% vs 28.9%), Renal Failure (29.5% vs 26.6%), CAD (34.8% vs 31.7%). Weekend admissions had a higher rate of smoking history (16.9% vs 15.1%). All p values &lt;0.001. Conclusions: Cardiac arrest on weekends has a higher mortality rate compared to weekdays. We found a higher burden of comorbidities on patients with weekday cardiac arrest.