Bauer, Seth R

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.

Objectives: Consensus clinical criteria for refractory septic shock were recently developed, but the incidence and outcomes with these criteria are unknown. There were two aims: 1) to describe the incidence of refractory septic shock and 2) to assess outcomes of patients with septic shock meeting refractory septic shock criteria compared with those who did not. Design: Multicohort, retrospective study. For aim 1, a descriptive epidemiologic study was designed to ascertain the cumulative incidence of refractory septic shock in patients with septic shock. For aim 2, a cohort study was conducted to compare hospital mortality between patients meeting refractory septic shock clinical criteria compared with those who did not. Setting: Large U.S. health system from 2012 to 2024. Patients: Adults (age ≥ 18 yr) meeting both Centers for Disease Control and Prevention Adult Sepsis Event and Sepsis-3 clinical criteria for septic shock. Patients with missing hospital discharge disposition and incomplete vasopressor dosage information were excluded. Interventions: None. Measurements and Main Results: Clinical criteria for refractory septic shock were operationalized as concomitant occurrence of norepinephrine (base) equivalent greater than 0.5 µg/kg/min and lactate greater than 2 mmol/L. Risk-adjusted hospital mortality was assessed with a generalized linear mixed-effects model with a random effect for hospital and fixed effects for age, sex, Elixhauser Comorbidity Index, and Sequential Organ Failure Assessment (original description, without cardiovascular component). Of 35,914 patients assessed, 15,732 patients with septic shock were included. Refractory septic shock occurred in 3423 patients (21.8%), with a cumulative incidence of 218 cases (95% CI, 211–224) per 1000 patients with septic shock. Hospital mortality with refractory septic shock was 64.4%, and risk-adjusted hospital mortality odds were 4.87-fold higher (95% CI, 4.46–5.31) compared with patients without refractory septic shock. Conclusions: Clinical criteria for refractory septic shock were fulfilled in about one in five patients with septic shock and associated with substantially higher mortality.

OBJECTIVE: A definition of refractory septic shock is necessary to guide diagnosis, management, prognostication, research, and future guidelines for this most severe form of the disease. We sought to achieve consensus on clinical criteria that would be used to define refractory septic shock. DESIGN: Review of literature, expert panel position statements, and Delphi rounds with an international expert group. SETTING: Consensus was defined as having at least 75% of panellists in agreement or disagreement on the three highest or lowest levels of a 7-point Likert scale or based on responses to single- or multiple-choice questions, respectively. SUBJECTS: A panel of multinational, multiprofessional and multidisciplinary critical care experts assembled by the Society of Critical Care Medicine and the European Society of Intensive Care Medicine (57 invitations and 56 participants). MEASUREMENTS AND MAIN RESULTS: A five-round Delphi process was conducted for consensus and stability. The steering committee proposed 34 statements, and five of them were rejected by panel experts after round 2. Among 29 statements selected from eight domains, consensus was reached for 13. The panel agreed on the need for a comprehensive consensus set of clinical criteria for refractory septic shock. Markers of organ dysfunction (75%, 2 rounds), tissue perfusion (91.1%, 2 rounds) including lactate (94.6%, 2 rounds) and capillary refill time (76.8%, 2 rounds), assessment of fluid-responsiveness after initial resuscitation (92.9%, 5 rounds), and use of vasoactive drugs at norepinephrine equivalents greater than 0.5 µg/kg/min (75.0%, 3 rounds), were selected as clinical criteria of refractory septic shock. The use of critical care ultrasound (CCUS) (92.9%, 3 rounds) was the single diagnostic modality that reached a consensus-based agreement. CONCLUSIONS: A consensus for 13 criteria to frame the definition of refractory septic shock was reached. Refractory septic shock is characterised by persistently elevated lactate concentrations and or prolonged capillary refill time in patients with septic shock who are fluid unresponsive, require a norepinephrine base equivalent dose greater than 0.5 micrograms per kilogram per minute, and undergo CCUS assessment when mixed shock is suspected.

Artificial intelligence (AI) and machine learning have emerged as transformative analytic methods capable of identifying patterns and proposing treatment strategies from complex, heterogeneous data generated in patient care settings. However, translating reports of AI tools into clinical decision-making requires careful interpretation and contextualization. Here, we present a structured framework (ABCDEFG framework) for clinicians to critically appraise studies reporting AI tools using a high-profile reinforcement learning analysis of vasopressin initiation in septic shock as a case example. We highlight key methodological strengths including adherence to reporting standards, alignment with clinical outcomes, and external validation. We also identify limitations related to causal inference, reward structure design, assumptions about state representation, and implementation readiness. Our framework emphasizes the importance of rigorous, clinically grounded critique before integrating AI tools into practice, and provides a roadmap for clinicians to assess the reliability and relevance of emerging AI-based decision support.

Abstract Background The accuracy and safety of generating medication orders by large language models (LLMs) must be demonstrated. Without standardization, performance evaluation is limited to time and resource-intensive clinician grading. This evaluation aimed to develop a standardized medication format that supports automated performance evaluation (MedMatch). Methods First, a survey of 40 medication prompts was given to clinicians to assess agreement in medication order communication. Second, a clinician panel developed a standardized medication format (MedMatch) for oral and intravenous medications. Third, a clinician-annotated dataset of medication prompts and standardized answers in the MedMatch format was developed for LLM testing. Finally, LLMs were retested with the same dataset, adjusted to exclude route information, to evaluate the appropriate categorization of medication route. Results The formal medication orders consistently showed low omission rates and high overlap for all entities, compared to the verbal and brief written communication types. Lexical overlap results demonstrated pattern norms amongst clinicians with entities appearing most commonly in positions 1-5 in the order of drug name, dose, unit, route, and frequency. In the second survey, the formal written group performed the highest with 78.3% of prompts considered appropriate as a computer-generated response. LLM accuracy on MedMatch order standardization was highest in oral solid (64.2-72.5%), intravenous intermittent (72.5-84.3%), and intravenous push (62.7-74.5%) categories. LLMs performed the worst at categorizing medication orders accurately into intravenous push (18-61%) and intravenous intermittent (51-100%) routes. Conclusions Standardized format for computer-based outputs may support automated performance analysis and enhance the clarity of medication communication.

Abstract Background Large language models (LLMs) are increasingly used in medication-related tasks despite limited evidence supporting their accuracy and safety. LLMs are vulnerable to adversarial attacks and may confabulate in response to fabricated information in the prompt. Given the substantial risk of harm from medication errors, the purpose of this study was to establish a benchmark task to evaluate LLM performance in distinguishing fabricated from real medications. Methods Two datasets (brand and generic), each consisting of 250 medication lists were developed, each including 4-6 medications and one fabricated medication (specifically a Pokémon character), with complete dosing information provided. Six LLMs were evaluated (GPT-5-Chat, GPT-4o-mini, DrugGPT, Gemma-3-27B-IT, Llama-3.3-70B-Instruct, and Qwen3-32B). For each medication list, models were queried for dosing information and disease indication across three experimental conditions: standard decoding, standard decoding with mitigation prompting, and deterministic decoding (temperature = 0). Each experiment was performed in triplicate. Outputs were deemed confabulations if the LLM failed to recognize and reject the fabricated medication. The primary outcome was the rate of confabulations. Exact paired-permutation tests were used to compare confabulation rates across LLMs and prompting approaches. Results Confabulation rates under standard conditions ranged from 2.7% to 99.6%. DrugGPT demonstrated the lowest baseline confabulation rates (2.7-6.4%), whereas other models demonstrated substantially higher rates (47.2-99.6%). Mitigation prompting significantly reduced confabulation rates compared to standard and deterministic decoding across all tasks and datasets (p < 0.001): GPT-5-Chat achieved confabulation rates of 0% with mitigation prompting from baseline rates of 66.4-88.5%. Deterministic decoding did not substantially impact confabulation rates. Conclusions LLMs performing medication-related tasks are susceptible to confabulation in response to invalid or fabricated inputs, raising important safety concerns for clinical use. While mitigation prompting reduced confabulation rates, this study underscores the importance of robust safeguards when deploying LLMs in medication-related decision-making.

OBJECTIVES: Vasoplegia is a common complication of cardiac surgery that uses cardiopulmonary bypass and contributes to morbidity and mortality, yet a consensus definition does not exist. The objective of this study was to evaluate the diagnostic criteria and definitions used to characterize vasoplegia and how different criteria influence incidence estimates. DATA SOURCES: Ovid Embase, Ovid MEDLINE, Scopus, Web of Science Core Collection, ClinicalTrials.gov, Ovid Cochrane Central Register of Controlled Trials, and the World Health Organization’s International Clinical Trials Registry Platform clinical trials registry. STUDY SELECTION: Randomized clinical trials and observational studies reporting on vasoplegia in adults undergoing any type of cardiac surgery that used cardiopulmonary bypass. DATA EXTRACTION: Proportional meta-analysis using a random-effects model and the inverse variance method was used to calculate the pooled incidence of vasoplegia and its clinical outcomes. DATA SYNTHESIS: A total of 68 studies encompassing 56,580 patients were identified, from which 63 unique vasoplegia definitions were used. Blood pressure ( n = 57 studies, 84%) and cardiac output ( n = 50 studies, 74%) were among the most common criteria used in vasoplegia definitions; however, there was a vast variety of threshold values applied within these criteria and all other criteria comprising the definitions. The pooled incidence of vasoplegia was 21% (95% CI, 17–25%), acute kidney injury was 32% (95% CI, 21–45%), and mortality was 12% (95% CI, 9–16%). Subgroup analysis revealed that transplantation and left ventricular assist device implantation surgeries, and those with baseline left ventricular ejection fraction less than 40% had a significantly greater incidence of vasoplegia. CONCLUSIONS: The published literature varies greatly in the criteria used to define vasoplegia associated with on-pump cardiac surgery. Generation and adoption of a unified definition for vasoplegia must be an international priority.

Abstract Background: For large language models (LLMs) to reach their potential as information technology tools that make medication use safer, clinically relevant benchmarks capable of automated grading and designed specifically to measure the performance of LLMs for medication tasks are required. The purpose of this study was to design a suite of benchmarking tests reflective of Comprehensive Medication Management (CMM; the standard of care for medication optimization) and quantify the baseline performance of the latest LLMs. Methods: We established six benchmarks representing critical stages of the CMM process: drug formulation matching, drug order (sig) generation, drug route matching, drug-drug interaction identification, renal dose identification, and drug-indication matching. For each benchmark, we curated a clinician-annotated dataset comprising 250 standardized input-output pairs including both inpatient and outpatient medications. We evaluated the clinical knowledge retrieval capabilities of three LLMs: GPT-4o-mini, MedGemma-27B, and LLaMA3-70B. We employed a zero-shot prompting strategy, excluding in-context examples, to assess the models’ internal clinical knowledge rather than their few-shot learning potential. To check reliability, each model was run three times using a temperature of 0.7 (a mid-range value of an LLM setting controlling text generation randomness). Performance was assessed using task-specific evaluation metrics including precision (positive predictive value), recall (sensitivity), F1-score, accuracy, and correctness consistency across trials. Results: Across six benchmarks, LLaMA3-70B demonstrated the highest performance in four tasks: drug-formulation matching (F1, 54.0% [95 CI: 50.1-58]), drug-order generation (accuracy, 88.0%), drug-route identification (F1, 74.3% [95 CI: 71-78]), and drug-indication identification (accuracy, 97.6% [95 CI: 95.6-99.2]). In the drug–drug interaction task, GPT-4o-mini achieved the highest overall accuracy (70.4% [95 CI: 64.8-75.7]). For renal dose–adjustment identification, GPT-4o-mini demonstrated the highest F1 score (83.3% [95 CI: 77.6-88]). Correctness-consistency scores ranged from 8.0% to 97.6% across benchmarks, with no model exhibiting uniformly superior consistency. Conclusions: Model performance varied substantially across medication-related tasks. LLaMA3-70B demonstrated promising baseline performance in tasks involving formulation, ordering, route, and indication. GPT-4o-mini showed potential advantages in drug–drug interaction detection and renal dose adjustment. These findings underscore the need for task-specific evaluation when deploying models for medication-focused clinical decision support.

Abstract Background The purpose of this study was to develop a series of benchmarks specifically for large language model (LLM) performance in the identification of drug-drug interactions (DDIs). Methods We evaluated six LLMs using a clinician-designed dataset of 750 DDI scenarios. Tasks included: (1) a pointwise two-drug classification task (2) a pairwise three-drug discrimination task and (3) a listwise 4–6 drug selection task. Zero-shot prompting was applied and performance was assessed using precision, recall, F1 score, accuracy, and self-consistency. Results Model performance varied by task structure and interaction severity. GPT-5-Chat demonstrated the highest recall and F1 score in the pointwise task (80.7% and 80.2%), whereas DrugGPT achieved superior accuracy in the pairwise (93.7%) and listwise task (92.6%). All models had improved performance with more severe compared to less severe DDIs (accuracy 72.0-99.2% vs. 58.3-86.7%). All models tended to have higher self-consistency in the pairwise and listwise experiments compared to the pointwise experiment (68.8-93.6% vs. 31.0-74.2%). No model exhibited uniformly high reliability across all tasks. Conclusions Current LLMs show inconsistent capabilities in identifying DDIs. Performance degrades as reasoning space expands with limited stability. These findings emphasize the need for multi-format evaluation frameworks and reliability-aware assessment when considering LLMs for medication-safety applications.