Citations

Accurate identification of surgical instruments is crucial for efficient workflows and patient safety within the operating room, particularly in preventing complications such as retained surgical instruments. Artificial Intelligence (AI) models have shown the potential to automate this process. This study evaluates the accuracy of publicly available Large Language Models (LLMs)—ChatGPT-4, ChatGPT-4o, and Gemini—and a specialized commercial mobile application, Surgical-Instrument Directory (SID 2.0), in identifying surgical instruments from images. The study utilized a dataset of 92 high-resolution images of 25 surgical instruments (retractors, forceps, scissors, and trocars) photographed from multiple angles. Model performance was evaluated using accuracy, weighted precision, recall, and F1 score. ChatGPT-4o exhibited the highest accuracy (89.1%) in categorizing instruments (e.g., scissors, forceps). SID 2.0 (77.2%) and ChatGPT-4 (76.1%) achieved comparable accuracy, while Gemini (44.6%) demonstrated lower accuracy in this task. For precise subtype identification of instrument names (like “Mayo scissors” or “Kelly forceps”), all models had low accuracy, with SID 2.0 having an accuracy of 39.1%, followed by ChatGPT-4o (33.69%). Subgroup analysis revealed ChatGPT-4 and 4o recognized trocars in all instances. Similarly, Gemini identified surgical scissors in all instances. In conclusion, publicly available LLMs can reliably identify surgical instruments at the category level, with ChatGPT-4o demonstrating an overall edge. However, precise subtype identification remains a challenge for all models. These findings highlight the potential of AI-driven solutions to enhance surgical-instrument management and underscore the need for further refinements to improve accuracy and support patient safety.

Background: Chordoma is a rare bone cancer with limited treatment options. Clinical trials are crucial for developing effective therapies, but their success depends on including diverse patient populations. The objective of this study was to systematically evaluate the reporting of racial, ethnic, and socioeconomic diversity in United States clinical trials exploring treatment for chordoma. Methods: A literature search was conducted through PubMed/Medline, Cochrane, Epistemonikos, and ClinicalTrials.gov databases for published US chordoma trials up until 19 August 2024. The data collected included trial characteristics and racial and ethnic data, as well as socioeconomic indicators when available. Methodological Index for Non-Randomized Studies (MINORS) and Revised Cochrane Risk-of-Bias Tool for Randomized Trials (RoB2) analyses were adopted to assess the methodological quality. The N-1 Chi-squared (χ2) test was implemented to compare the reported racial and ethnic data with the most recent US Census Bureau data. Results: Five trials involving 111 patients (median age: 63 years; 34% female) were included. Four studies (80%) were single-arm non-randomized studies with one study (25%) having a high methodological quality and three (75%) having a moderate quality based on the MINORS analysis. Most patients (91%, n = 82) were White/Caucasian, representing a proportion which was significantly higher than the reported 75% in the US population (p = 0.0005). Black/African American patients (2%, n = 2) were significantly underrepresented compared to the 14% in the US population (p = 0.0015). Regarding ethnicity, Hispanic/Latino patients (7%, n = 6) were significantly underrepresented compared to the 20% in the US population (p = 0.0021). No measures of socioeconomic status were reported. Conclusions: This systematic review highlighted the need for improved racial and ethnic diversity in chordoma trials and the better reporting of socioeconomic data. The underrepresentation of minority groups may obscure potential disparities in disease incidence, treatment access, and clinical outcomes.

Current clinical care relies heavily on complex, rule-based systems for tasks like diagnosis and treatment. However, these systems can be cumbersome and require constant updates. This study explores the potential of the large language model (LLM), LLaMA 2, to address these limitations. We tested LLaMA 2′s performance in interpreting complex clinical process models, such as Mayo Clinic Care Pathway Models (CPMs), and providing accurate clinical recommendations. LLM was trained on encoded pathways versions using DOT language, embedding them with SentenceTransformer, and then presented with hypothetical patient cases. We compared the token-level accuracy between LLM output and the ground truth by measuring both node and edge accuracy. LLaMA 2 accurately retrieved the diagnosis, suggested further evaluation, and delivered appropriate management steps, all based on the pathways. The average node accuracy across the different pathways was 0.91 (SD ± 0.045), while the average edge accuracy was 0.92 (SD ± 0.122). This study highlights the potential of LLMs for healthcare information retrieval, especially when relevant data are provided. Future research should focus on improving these models’ interpretability and their integration into existing clinical workflows.

Background: Effective pain management is crucial for both comfort and outcomes, yet predicting and managing this pain is difficult. This study aimed to analyze postoperative pain in patients undergoing hand surgery at the Mayo Clinic Florida, examining how patient characteristics and anxiety affect pain outcomes. Methods: We conducted a single-arm clinical trial at Mayo Clinic Florida, recruiting patients undergoing hand surgery. Preoperative pain and anxiety were assessed using the Pain Catastrophizing Scale (PCS) and State-Trait Anxiety Inventory (STAI). Postoperatively, patients used an iPhone app to record pain levels and medication use every four hours. Devices were collected three days after surgery. We analyzed the relationship between demographics, PCS, STAI scores, and pain levels using linear and logistic regression models. All statistical tests were two-sided with significance set at p < 0.05, analyzed with R4.2.2. Results: Data were collected from 62 patients (62.9% women, 37.1% men) undergoing hand surgery. Participants were mainly White (90.3%), with 50% being in the middle-aged adult group. Most had low anxiety levels (80.6% STAI-S, 82.3% STAI-T) and low catastrophizing (61.3% PCS). Postoperative pain was low, with median scores between 1.0 and 2.0 over three days. Demographics, anxiety, and catastrophizing were not significant predictors of pain levels. Logistic regression showed time as a significant factor, with pain levels peaking on Day 3. Conclusions: Postoperative pain after hand surgery was generally low, with time being a significant predictor of increased pain. Demographic factors, anxiety, and catastrophizing did not significantly affect pain levels. Pain management should emphasize time-sensitive interventions and ongoing monitoring.

OBJECTIVE To evaluate a regimen of inhaled Technosphere insulin (TI) plus insulin degludec in adults with type 1 diabetes, who prestudy were predominately using either an automated insulin delivery (AID) system or multiple daily insulin injections (MDI) with continuous glucose monitoring. RESEARCH DESIGN AND METHODS At 19 sites, adults with type 1 diabetes were randomly assigned to TI plus insulin degludec (N = 62) or usual care (UC) with continuation of prestudy insulin delivery method (N = 61) for 17 weeks. RESULTS Prestudy, AID was used by 48% and MDI by 45%. Mean ± SD HbA1c was 7.57% ± 0.97% at baseline and 7.62% ± 1.06% at 17 weeks in the TI group and 7.59% ± 0.80% and 7.54% ± 0.77%, respectively, in the UC group (adjusted difference 0.11%, 95% CI −0.10 to 0.33, P value for noninferiority = 0.01). HbA1c improved from baseline to 17 weeks by &gt;0.5% (5.5 mmol/mol) in 12 (21%) in the TI group and in 3 (5%) in the UC group and worsened by &gt;0.5% (5.5 mmol/mol) in 15 (26%) in the TI group and in 2 (3%) in the UC group. The most common TI side effect was a brief cough; eight participants discontinued TI due to side effects. CONCLUSIONS In adults with type 1 diabetes, HbA1c after 17 weeks with a regimen of TI and degludec was noninferior to UC, which consisted predominately of either AID or MDI. TI should be considered an option for people with type 1 diabetes, particularly those who are motivated to further reduce postprandial hyperglycemia.

Background: Recurrent pericarditis results from inflammasome activation and is associated with significant morbidity. Cannabidiol has been shown to block numerous inflammasome signaling pathways, including reducing transcription of pro-IL-1β and NLRP3, and inhibiting secretion of IL-1β and IL-6. Aim: To assess the impact of oral pharmaceutically manufactured (P-)cannabidiol in recurrent pericarditis patients (pts). Methods: Adult pts (≥18 years) with symptomatic, recurrent pericarditis (≥2 recurrences) were eligible for this open-label, multicenter study. Eligible pts had pericarditis chest pain with a numerical rating scale (NRS) pain score ≥4 within 7 days prior to enrollment, together with either elevated (≥1.0 mg/dL) C-reactive protein (CRP) within the prior 7 days or prior pericardial late gadolinium enhancement (LGE) on cardiac imaging. Pts had to be on chronic treatment with NSAIDs, colchicine or corticosteroids (any combination) with stable baseline (BL) doses. Oral P-cannabidiol was uptitrated to 10 mg/kg twice daily (BID), or maximum tolerated dose. The primary efficacy endpoint was improvement in NRS pain score at week 8 (WK8). A secondary efficacy endpoint among pts with elevated BL CRP was the percentage of pts with CRP normalization (≤0.5 mg/dL) at WK8. Results: 27 pts, mean age of 53 years (range 24-77), 18 of female sex, with a median of 3 prior pericarditis episodes were enrolled. CRP was ≥1.0 mg/dL in 10 pts (37%) at BL. Prior evidence of LGE was present in 21 pts (78%). At BL, 41% of pts were receiving corticosteroid therapy. The highest tolerated study medication dose was BID 10 mg/kg in 20 pts (74.1%), 7.5 mg/kg in 4 pts (14.8%), and 5 mg/kg in 3 pts (11.1%). All 27 pts (100%) completed the WK8 visit. Mean NRS score decreased by 3.7, from 5.8 at BL to 2.1 at WK8. In those with CRP elevation at BL, CRP normalized in 8 pts (80%) by WK8, with mean CRP decreasing from 5.71 to 0.31 mg/dL. Study medication was discontinued due to AEs in 2 pts (7.4%). 24 pts (89%) tolerated P-cannabidiol with a treatment response and proceeded into the 18-week extension period (EP) where background therapy is weaned. Conclusions: Among pts with symptomatic pericarditis recurrence, oral administration of P-cannabidiol led to a clinically relevant reduction in pericarditis chest pain (measured by NRS score) and in CRP. P-cannabidiol was shown to be safe and generally well tolerated, with the majority of pts reaching the target dose and continuing into the study EP.

Abstract Low-grade gliomas (LGG) are slowly growing brain cancers that ultimately undergo a malignant progression to lethal secondary glioblastomas with a dismal prognosis. Despite intensive research, very little progress has been made in our understanding of this transformation. Hence, outside of known alternations in well-characterized driver genes CDKN2A, PTEN and PDGFRA, genetic drivers that underlie LGG transformation are largely unknown. Contrary to other types of cancer, LGGs exhibit very few driver mutations (30-40 mutations) as they progress. Interestingly, LGGs also exhibit a highly prevalent and heterogenous pattern of copy-number-alterations (CNAs) suggesting that they might be driven by changes in gene dosages that result from widespread chromosomal instability observed within these tumors. Through analysis of the human Mayo Clinic and TCGA LGG datasets (~1500 patients), we have identified eleven highly recurrent CNAs affecting LGG patients, many of which correlate with patient survival. We hypothesized that these CNAs contain driver genes that confer a growth advantage resulting in the highly recurrent CNA pattern and prognostic associations. To investigate the tumor suppressive potential of the ~1800 genes located in recurrently deleted regions and the oncogenic potential of the ~1200 genes located in recurrently amplified regions, our lab developed a novel in vivo ‘CRISPR-Knock Out and Activation Linked Assay’ (CRISPR-KOALA). Stereotaxic delivery of arm-level CNA lentiviral gRNA libraries targeting the mouse orthologs of genes in CNAs into the brain of our animal model (rs557mut;IDHmut;P53KO;AtrxKO;Cas9GFP: Yanchus et. al Science 2022) resulted in rapid formation of tumors compared to animals injected with control libraries. Deconvoluting the screen revealed enrichment of immune modulators such as Them6 as well as a modulator of NMDA receptors (Grina) implicating immune system and glutamate dysregulation in LGG malignant transformation. Taken together, our study provides insights into mechanisms underlying LGG progression which may serve to inform intervention strategies for this lethal disease.

ABSTRACT Background Large language model (LLM) chatbots, a form of artificial intelligence (AI) that excels at prompt‐based interactions and mimics human conversation, have emerged as a tool for providing patients with information about urologic conditions. We aimed to examine the quality of information related to benign prostatic hyperplasia surgery from four chatbots and how they would respond to sample patient messages. Methods We identified the top three queries in Google Trends related to “treatment for enlarged prostate.” These were entered into ChatGPT (OpenAI), Bard (Google), Bing AI (Microsoft), and Doximity GPT (Doximity), both unprompted and prompted for specific criteria (optimized). The chatbot‐provided answers to each query were evaluated for overall quality by three urologists using the DISCERN instrument. Readability was measured with the built‐in Flesch–Kincaid reading level tool in Microsoft Word. To assess the ability of chatbots to answer patient questions, we prompted the chatbots with a clinical scenario related to holmium laser enucleation of the prostate, followed by 10 questions that the National Institutes of Health recommends patients ask before surgery. Accuracy and completeness of responses were graded with Likert scales. Results Without prompting, the quality of information was moderate across all chatbots but improved significantly with prompting (mean [SD], 3.3 [1.2] vs. 4.4 [0.7] out of 5; p < 0.001). When answering simulated patient messages, the chatbots were accurate (mean [SD], 5.6 [0.4] out of 6) and complete (mean [SD], 2.8 [0.3] out of 3). Additionally, 98% (39/40) had a median score of 5 or higher for accuracy, which corresponds to “nearly all correct.” The readability was poor, with a mean (SD) Flesch–Kincaid reading level grade of 12.1 (1.3) (unprompted). Conclusions LLM chatbots hold promise for patient education, but their effectiveness is limited by the need for careful prompting from the user and by responding at a reading level higher than that of most Americans (grade 8). Educating patients and physicians on optimal LLM interaction is crucial to unlock the full potential of chatbots.