BACKGROUND: Large Language Models (LLMs) like ChatGPT have become increasingly prevalent. In medicine, many potential areas arise where LLMs may offer added value. Our research focuses on the use of open-source LLM alternatives like Llama 3, Gemma, Mistral, and Mixtral to extract medical parameters from German clinical texts. We concentrate on German due to an observed gap in research for non-English tasks.
OBJECTIVE: To evaluate the effectiveness of open-source LLMs in extracting medical parameters from German clinical texts, specially focusing on cardiovascular function indicators from cardiac MRI reports.
METHODS: We extracted 14 cardiovascular function indicators, including left and right ventricular ejection fraction (LV-EF and RV-EF), from 497 variously formulated cardiac magnetic resonance imaging (MRI) reports. Our systematic analysis involved assessing the performance of Llama 3, Gemma, Mistral, and Mixtral models in terms of right annotation and named entity recognition (NER) accuracy.
RESULTS: The analysis confirms strong performance with up to 95.4% right annotation and 99.8% NER accuracy across different architectures, despite the fact that these models were not explicitly fine-tuned for data extraction and the German language.
CONCLUSIONS: The results strongly recommend using open-source LLMs for extracting medical parameters from clinical texts, including those in German, due to their high accuracy and effectiveness even without specific fine-tuning.

Objective: To evaluate and compare the quality and comprehensibility of answers produced by five distinct artificial intelligence (AI) chatbots-GPT-4, Claude, Mistral, Google PaLM, and Grok-in response to the most frequently searched questions about kidney stones (KS). Materials and Methods: Google Trends facilitated the identification of pertinent terms related to KS. Each AI chatbot was provided with a unique sequence of 25 commonly searched phrases as input. The responses were assessed using DISCERN, the Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P), the Flesch-Kincaid Grade Level (FKGL), and the Flesch-Kincaid Reading Ease (FKRE) criteria. Results: The three most frequently searched terms were \"stone in kidney,\" \"kidney stone pain,\" and \"kidney pain.\" Nepal, India, and Trinidad and Tobago were the countries that performed the most searches in KS. None of the AI chatbots attained the requisite level of comprehensibility. Grok demonstrated the highest FKRE (55.6 ± 7.1) and lowest FKGL (10.0 ± 1.1) ratings (p = 0.001), whereas Claude outperformed the other chatbots in its DISCERN scores (47.6 ± 1.2) (p = 0.001). PEMAT-P understandability was the lowest in GPT-4 (53.2 ± 2.0), and actionability was the highest in Claude (61.8 ± 3.5) (p = 0.001). Conclusion: GPT-4 had the most complex language structure of the five chatbots, making it the most difficult to read and comprehend, whereas Grok was the simplest. Claude had the best KS text quality. Chatbot technology can improve healthcare material and make it easier to grasp.

Large language models (LLMs) are transformer-based neural networks that can provide human-like responses to questions and instructions. LLMs can generate educational material, summarize text, extract structured data from free text, create reports, write programs, and potentially assist in case sign-out. LLMs combined with vision models can assist in interpreting histopathology images. LLMs have immense potential in transforming pathology practice and education, but these models are not infallible, so any artificial intelligence generated content must be verified with reputable sources. Caution must be exercised on how these models are integrated into clinical practice, as these models can produce hallucinations and incorrect results, and an over-reliance on artificial intelligence may lead to de-skilling and automation bias. This review paper provides a brief history of LLMs and highlights several use cases for LLMs in the field of pathology.

Health care-associated infection (HAI) surveillance is vital for safety in health care settings. It helps identify infection risk factors, enhancing patient safety and quality improvement. However, HAI surveillance is complex, demanding specialized knowledge and resources. This study investigates the use of artificial intelligence (AI), particularly generative large language models, to improve HAI surveillance.
We assessed 2 AI agents, OpenAI\'s chatGPT plus (GPT-4) and a Mixtral 8×7b-based local model, for their ability to identify Central Line-Associated Bloodstream Infection (CLABSI) and Catheter-Associated Urinary Tract Infection (CAUTI) from 6 National Health Care Safety Network training scenarios. The complexity of these scenarios was analyzed, and responses were matched against expert opinions.
Both AI models accurately identified CLABSI and CAUTI in all scenarios when given clear prompts. Challenges appeared with ambiguous prompts including Arabic numeral dates, abbreviations, and special characters, causing occasional inaccuracies in repeated tests.
The study demonstrates AI\'s potential in accurately identifying HAIs like CLABSI and CAUTI. Clear, specific prompts are crucial for reliable AI responses, highlighting the need for human oversight in AI-assisted HAI surveillance.
AI shows promise in enhancing HAI surveillance, potentially streamlining tasks, and freeing health care staff for patient-focused activities. Effective AI use requires user education and ongoing AI model refinement.

Despite promising results obtained in the early diagnosis of several pathologies, breath analysis still remains an unused technique in clinical practice due to the lack of breath sampling standardized procedures able to guarantee a good repeatability and comparability of results. The most diffuse on an international scale breath sampling method uses polymeric bags, but, recently, devices named Mistral and ReCIVA, able to directly concentrate volatile organic compounds (VOCs) onto sorbent tubes, have been developed and launched on the market. In order to explore performances of these new automatic devices with respect to sampling in the polymeric bag and to study the differences in VOCs profile when whole or alveolar breath is collected and when pulmonary wash out with clean air is done, a tailored experimental design was developed. Three different breath sampling approaches were compared: (a) whole breath sampling by means of Tedlar bags, (b) the end-tidal breath collection using the Mistral sampler, and (c) the simultaneous collection of the whole and alveolar breath by using the ReCIVA. The obtained results showed that alveolar fraction of breath was relatively less affected by ambient air (AA) contaminants (p-values equal to 0.04 for Mistral and 0.002 for ReCIVA Low) with respect to whole breath (p-values equal to 0.97 for ReCIVA Whole). Compared to Tedlar bags, coherent results were obtained by using Mistral while lower VOCs levels were detected for samples (both breath and AA) collected by ReCIVA, likely due to uncorrected and fluctuating flow rates applied by this device. Finally, the analysis of all data also including data obtained by explorative analysis of the unique lung cancer (LC) breath sample showed that a clean air supply might determine a further confounding factor in breath analysis considering that lung wash-out is species-dependent.

Several documents and guidelines provide recommendations for effective management of COPD patients. However, there is often a significant imbalance between recommended treatment of COPD patients and the actual care provided both in primary care and specialty setting. This imbalance could result in a significant negative impact on patients\' health status and quality of life, leading to increased hospitalisations and health resource utilisation in COPD patients METHODS: MISTRAL was an observational, longitudinal, prospective cohort study, designed to assess the overall pharmacological approach of COPD in routine clinical practice in Italy. Eligible patients were divided into two cohorts based on their exacerbation history in the year prior to the enrolment, frequent exacerbators (FEs; ≥2 exacerbations), and non-frequent exacerbators (NFEs; ≤1 exacerbation). The primary objective was to assess adherence to Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2011 treatment recommendations in FEs and NFEs at baseline and follow-up visits RESULTS: Of the 1489 enrolled patients, 1468 (98.6%; FEs, 526; NFEs, 942) were considered evaluable for analyses. At baseline, 57.8% of patients were treated according to GOLD 2011 recommendations; a greater proportion of FEs were treated according to GOLD recommendations, compared with NFEs patients at baseline (77.1% versus 46.7%; P < 0.0001), and all study visits. At baseline, GOLD group D patients were the most adherent (81.2%) to treatment recommendations, while group A patients were the least adherent (30.3%) at baseline, attributed mainly to overuse of inhaled corticosteroids in less severe GOLD groups. Triple therapy with long-acting muscarinic antagonist (LAMA) + long-acting β2-agonist/inhaled corticosteroid (LABA/ICS) was the most frequent prescribed treatment at all study visits, irrespective of patient\'s exacerbation history. Changes in treatment were more frequent in FEs versus NFEs CONCLUSIONS: The Mistral study reports a scarce adherence to the GOLD 2011 treatment recommendations in routine clinical practice in Italy. The adherence was particularly low in less severe, non-frequent exacerbating patients mostly for ICS overuse, and was higher in high-risk, frequent exacerbating COPD patients.