Citations

Abstract Purpose Large language models (LLMs) have proven performance for certain diagnostic tasks; however, limited studies have evaluated their consistency in recommending appropriate medication regimens for a given diagnosis. Medication management is a complex task that requires synthesis of drug formulation and complete order instructions for safe use. Here, the performance of GPT-4o, an LLM available with OpenAI’s ChatGPT, was tested on 3 medication management tasks. Methods GPT-4o performance was tested on 3 medication tasks: identifying available formulations for a given generic drug name, identifying drug-drug interactions (DDIs) for a given medication regimen, and preparing a medication order for a given generic drug name. For each experiment, the model’s raw text response was captured exactly as returned and evaluated using clinician evaluation in addition to standard LLM metrics, including Term Frequency-Inverse Document Frequency (TF-IDF) vectors, normalized Levenshtein similarity, and Recall-Oriented Understudy for Gisting Evaluation (ROUGE-1/ROUGE-L) F1 score between each response and its reference string. Results For the first task of drug-formulation matching, GPT-4o had 49% accuracy for generic medications being matched to all available formulations, with an average of 1.23 omissions per medication and 1.14 hallucinations per medication. For the second task of drug-drug interaction identification, the accuracy was 54.7% for identifying the DDI pair. For the third task, GPT-4o generated order sentences containing no medication or abbreviation errors in 65.8% of the cases. Conclusion Model performance for basic medication tasks was consistently poor. This evaluation highlights the need for domain-specific training through clinician-annotated datasets and a comprehensive evaluation framework for benchmarking performance.

Abstract Purpose We compared perioperative morbidity after repeat transurethral resection (re-TURP) versus primary TURP, focusing on transfusion and broader bleeding-related complications relevant to counseling and risk stratification. Methods We analyzed the American College of Surgeons National Surgical Quality Improvement Program (2013–2023) to identify men undergoing primary TURP (Current Procedural Terminology [CPT] 52601) or re-TURP (CPT 52630). The primary outcome was 30-day red blood cell transfusion. Secondary outcomes included a prespecified postoperative hematuria endpoint based on ICD-10 codes. Multivariable logistic regression evaluated associations with bleeding endpoints and downstream complications. Results Among 83,539 TURP procedures, 76,983 (92.2%) were primary and 6,556 (7.8%) were repeat. Re-TURP patients were older, had higher ASA class, and more often underwent urgent surgery. Transfusion occurred in 1.9% after re-TURP versus 1.5% after primary TURP; hematuria-coded events occurred in 3.5% versus 1.3%, respectively. After adjustment, re-TURP was associated with postoperative hematuria (adjusted odds ratio [aOR] 2.42, 95% CI 2.03–2.88) but not with transfusion (aOR 1.09, 95% CI 0.87–1.36). Older age, urgent/emergent status, bleeding disorder, and lower hematocrit predicted both bleeding endpoints. Re-TURP was associated with slightly lower odds of readmission (aOR 0.87; 95% CI 0.76–0.99) and was not associated with return to the operating room. Transfusion strongly predicted return to the operating room (aOR 7.21) and readmission (aOR 2.05), whereas hematuria predicted return to the operating room only (aOR 1.50). Conclusions Compared with primary TURP, re-TURP was associated with more hematuria-coded events but not higher adjusted transfusion risk, suggesting a relative shift toward lower-severity bleeding morbidity.

The conventional workup for suspected early-stage lung cancer requires multiple visits, anesthetics, and procedures. Single anesthetic event lung cancer surgery integrates shape-sensing robotic bronchoscopy, cone-beam CT, rapid on-site cytologic evaluation (ROSE), endobronchial ultrasound staging, lesion localization, and minimally invasive resection into one operative encounter. In this narrative review, we describe the technical components of this pathway, summarize the supporting literature, and offer expert, institution-based troubleshooting guidance for common intraoperative dilemmas. Reported diagnostic yields for shape-sensing robotic bronchoscopy range from 80 to 96%, with approximately 90% concordance between ROSE and final pathology. Published single-institution series report reductions in time from detection to resection of 15–51 days, cost savings of approximately $3000–$10,000, and perioperative outcomes (length of stay 1.8–3.6 days; complication rates comparable to traditional pathways) similar to staged care. The supporting evidence is retrospective and derived from small, single-institution, high-volume referral cohorts; no prospective comparative trials or long-term survival data yet exist. Single anesthetic event lung cancer surgery is therefore best regarded as an emerging, resource-intensive care pathway for carefully selected patients at experienced centers rather than an established standard of care, pending prospective, multicenter validation.

BACKGROUND: It is unknown whether cardiac resynchronization therapy (CRT) improves systemic metabolomic profile by enhancing left ventricular function. The observational cohort study aimed to investigate the effect of CRT on left ventricular ejection fraction (LVEF) and circulating plasma metabolites in patients who had heart failure with reduced LVEF. METHODS: We prospectively screened patients with ischemic cardiomyopathy and nonischemic cardiomyopathy who received CRT with a defibrillator for LVEF ≤35% according to current guidelines. Clinical assessment included echocardiography and device interrogation. Blood samples for metabolomic analysis were collected before CRT and at 6-month follow-up. Plasma was subjected to gas chromatography-mass spectrometry and 1 H nuclear magnetic resonance-based metabolomic analysis. RESULTS: Totally 92 patients were enrolled, with a mean age of 67.3±11.3 years (37.0% female). LVEF was significantly improved from 28.9±7.6% at baseline to 36.0±11.3% at 6 months ( P <0.001) and to 40.1±12.6% at 12 months ( P <0.001). After CRT, 42 metabolite features were significantly decreased compared with the baseline. The ketone bodies, including 3-hydroxybutyrate ( P <0.001) and acetone ( P =0.01), were reduced. A branched-chain amino acid, isoleucine, was also decreased after CRT ( P =0.01). These metabolomic changes were mainly observed in the nonischemic cardiomyopathy group, while the metabolomic profile in the ischemic cardiomyopathy group was characterized by enhanced amino acid oxidation and elevated levels of lactate and pyruvate. The improvement in LVEF positively correlated with the ratio of changes in ketone bodies and isoleucine. CONCLUSIONS: CRT may modulate the systemic plasma metabolomic profile, which correlated with improvement in left ventricular function in patients with severe heart failure with reduced ejection fraction.

ABSTRACT Background Immune checkpoint inhibitors (ICIs) enhance anti‐tumour immunity by modulating immune pathways but can also trigger immune‐related adverse events. We aimed to evaluate the risk of developing coeliac disease (CeD) among patients with malignancy treated with ICIs. Methods Using the TriNetX US Collaborative Network, we identified adults (≥ 18 years) diagnosed with malignancies for which ICIs are routinely used between January 2010 and December 2024 and stratified them by exposure to immune checkpoint inhibitors (nivolumab, pembrolizumab, ipilimumab, atezolizumab, durvalumab, tremelimumab, or avelumab). Patients with pre‐existing celiac disease (CeD) were excluded. Cohorts were propensity score matched for demographics and autoimmune comorbidities. The primary outcome was incident CeD occurring ≥ 90 days after the index event. Adjusted hazard ratios (aHRs) were calculated, with subgroup analyses by age and sex. Statistical significance was defined as p < 0.05. Results After propensity score matching, 158,208 patients were included in each cohort. CeD was diagnosed more frequently among patients exposed to ICIs than among matched controls (0.152% vs. 0.071%; aHR 3.95, 95% CI 3.11–5.03; p < 0.001), corresponding to a number needed to harm (NNH) of 1235. The increased risk persisted across sex and age subgroups, including females (aHR 3.43, 95% CI 2.55–4.62), males (aHR 5.36, 95% CI 3.53–8.12), with the risk being highest among patients older than 60 years (0.14% vs. 0.06%; aHR 4.24, 95% CI 3.17–5.67; p < 0.001). Conclusion In this large real‐world cohort, ICI exposure was associated with nearly a fourfold increase in the risk of being diagnosed with CeD compared with matched controls.

ABSTRACT Background Optimizing access to cost‐effective care in capacity‐constrained health systems is a contemporary imperative. We evaluated whether machine learning (ML) models using patient‐reported data could optimize access for patients requiring surgical care in rhinology clinics, without need for CT imaging. The outcome was defined as any rhinologic surgery within 90 days of initial evaluation. Methods A de‐identified electronic dataset from patients seen at five distinct sites within an integrated healthcare system between 2018 and 2025 was used to train models. Demographic data and 22‐item Sinonasal Outcome Test (SNOT‐22) responses were studied. Models were trained using stratified 5‐fold cross‐validation with an 80% development cohort and validated in a 20% held‐out validation cohort. Hyperparameters were optimized with Optuna. The primary outcome was performance of any rhinologic surgical intervention within 90 days of initial SNOT‐22. Results Data from 35,170 patients were evaluated. Item‐level responses outperformed use of total SNOT‐22 score in the models. Among models, XGBoost demonstrated best discrimination with AUC of 0.70 (95% CI, 0.69–0.71), outperforming logistic regression (0.66), random forest (0.63), and TabNet (0.66) (all p < 0.001). At optimal threshold, XGBoost achieved 66% sensitivity, 64% specificity, 28% PPV, and 90% NPV. Top predictors for surgery were age, nasal blockage, facial pain or pressure, and decreased sense of smell or taste. Validation on the held‐out cohort remained stable (AUC 0.70), with strong discrimination across sites despite surgical rates ranging from 11.5% to 27.4%. Conclusions Demographics and item‐level SNOT‐22 responses were successful in developing an ML model that demonstrated moderate discrimination and high NPV (90%) for performance of surgery within next 90 days. ML models balanced with human oversight may accelerate triage and optimize surgical yield for rhinology clinics.