Evaluating large language models for selection of statistical test for research: A pilot study.

UNSTRUCTURED Misleading information has significant implications for society but can have disastrous impact for health matters. Transformative artificial intelligence (AI) tools such as large language models (LLMs) have the potential for limitless content generation (including fake), soon making internet information impossible to assess using traditional human approaches. We asked if the same LLMs (GPT4 and Gemini1-5-Pro) could be part of a more scalable solution. We tested 2 publicly available LLMs for their ability to identify misinformation in HealthReleases previously labeled by human experts. We found that simple prompts lead to overall low accuracy (F1 Macro 0,45 (GPT4) and 0,49 (Gemini1-5Pro)), but very different profiles for each LLM. Adding specific criteria used by experts to critically assess the Releases enhanced Gemini (0.66) but surprisingly reduced GPT4 (0,37) performances. We therefore developed a novel approach incorporating summaries of expert feedback into prompts and then observed major improvements in performance for both LLMs(GPT4;0.63 and Gemini1-5Pro; 0.96). Our study provides the first use case of LLMs as high throughput proofing of medical text, but more importantly provides insights into LLMs’ “truth biases”. We provide a novel paradigm integrating knowledge into the prompts which may reduce the need for LLM training, and the requirement for ever larger datasets and compute power. Importantly, we show how experts could and need to be involved in LLMs used to enhance their performance and potentially minimize the data wall issue.

To evaluate the impact of constraining proprietary and open large language models (LLMs) to structured outputs in processing radiology request forms (RRF). We evaluated five LLMs-two proprietary (GPT-5-Thinking, Gemini 2.5 Pro) and three open (Qwen3-235B-A22B-Thinking, gpt-oss-120b, medgemma-27b-it)-on 100 RRFs (50 computed tomography, 50 magnetic resonance imaging). Each model processed cases with and without constraints to structured outputs. Endpoints included accuracy for modality, anatomical region, contrast phase, urgency, "all correct" (all four categories correct), and "indication improved" (clarity of rewritten text). Outputs were evaluated against a reference standard defined by board-certified radiologists and compared with two radiology residents (first-year and third-year). Accuracies with 95% confidence intervals were calculated. Constraining to structured outputs had model-dependent effects: it improved Gemini 2.5 Pro (all correct: from 53.0% [43.3-62.5] to 66.0% [56.3-74.5]) but reduced GPT-5-Thinking accuracy (from 76.0% [66.8-83.3] to 53.0% [43.3-62.5]), with minimal influence on open models. Both proprietary LLMs outperformed the best open models (up to 41.0% [31.9-50.8]). All LLMs exceeded the unassisted first-year residents' performance (19.0% [12.5-27.8]). LLM assistance improved first-year residents' accuracy to 65.0% [55.3-73.6], approaching the third-year residents' performance (80.0% [71.1-86.7]), who performed comparably to the best LLMs. Across models, performance was highest for modality and anatomical region, and lowest for urgency. Indication reformulation was judged clearer in > 90% of cases across all models without hallucinations. Constraining to structured outputs exerted model-specific effects. Proprietary LLMs achieved the highest accuracy in RRF-based protocol selection and improved first-year resident performance to an experienced-resident level. LLMs may serve as valuable decision-support tools for radiology workflow. Constraining LLMs to structured outputs produced divergent, model-specific effects in radiology protocol selection-improving Gemini 2.5 Pro, reducing GPT-5-Thinking, and minimally affecting open models-highlighting the need for model-specific prompting strategies before adopting LLMs in radiology decision support. Structured output constraints affect LLM performance differently. Gemini 2.5 Pro benefits from structured prompting, while GPT-5-Thinking declines. Open-weight models show minimal impact from output constraining. Proprietary models outperform open models in radiology protocol selection.

To assess the performance of state-of-the-art large language models in classifying vertebral metastasis stability using the Spinal Instability Neoplastic Score (SINS) compared to human experts, and to evaluate the impact of task-specific refinement including in-context learning on their performance. This retrospective study analyzed 100 synthetic CT and MRI reports encompassing a broad range of SINS scores. Four human experts (two radiologists and two neurosurgeons) and four large language models (Mistral, Claude, GPT-4 turbo, and GPT-4o) evaluated the reports. Large language models were tested in both generic form and with task-specific refinement. Performance was assessed based on correct SINS category assignment and attributed SINS points. Human experts demonstrated high median performance in SINS classification (98.5% correct) and points calculation (92% correct), with a median point offset of 0 [0-0]. Generic large language models performed poorly with 26-63% correct category and 4-15% correct SINS points allocation. In-context learning significantly improved chatbot performance to near-human levels (96-98/100 correct for classification, 86-95/100 for scoring, no significant difference to human experts). Refined large language models performed 71-85% better in SINS points allocation. In-context learning enables state-of-the-art large language models to perform at near-human expert levels in SINS classification, offering potential for automating vertebral metastasis stability assessment. The poor performance of generic large language models highlights the importance of task-specific refinement in medical applications of artificial intelligence.

BackgroundThis study evaluated the feasibility of ChatGPT and Gemini, two off-the-shelf large language models (LLMs), to automate causality assessments, focusing on Adverse Events Following Immunizations (AEFIs) of myocarditis and pericarditis related to COVID-19 vaccines.MethodsWe assessed 150 COVID-19-related cases of myocarditis and pericarditis reported to the Vaccine Adverse Event Reporting System (VAERS) in the United States of America (USA). Both LLMs and human experts conducted the World Health Organization (WHO) algorithm for vaccine causality assessments, and inter-rater agreement was measured using percentage agreement. Adherence to the WHO algorithm was evaluated by comparing LLM responses to the expected sequence of the algorithm. Statistical analyses, including descriptive statistics and Random Forest modeling, explored case complexity (e.g., string length measurements) and factors affecting LLM performance and adherence.ResultsChatGPT showed higher adherence to the WHO algorithm (34%) compared to Gemini (7%) and had moderate agreement (71%) with human experts, whereas Gemini had fair agreement (53%). Both LLMs often failed to recognize listed AEFIs, with ChatGPT and Gemini incorrectly identifying 6.7% and 13.3% of AEFIs, respectively. ChatGPT showed inconsistencies in 8.0% of cases and Gemini in 46.7%. For ChatGPT, adherence to the algorithm was associated with lower string complexity in prompt sections. The random forest analysis achieved an accuracy of 55% (95% confidence interval: 35.7–73.5) for predicting adherence to the WHO algorithm for ChatGPT.ConclusionNotable limitations of ChatGPT and Gemini have been identified in their use for aiding causality assessments in vaccine safety. ChatGPT performed better, with higher adherence and agreement with human experts. In the investigated scenario, both models are better suited as complementary tools to human expertise.Supplementary InformationThe online version contains supplementary material available at 10.1007/s40264-025-01531-y.

SummarySustained engagement in HIV care and adherence to antiretroviral therapy (ART) are essential for achieving the UNAIDS “95-95-95” targets. Despite increased ART access, disengagement from care remains a significant issue, particularly in sub-Saharan Africa. Traditional machine learning (ML) models have shown moderate success in predicting care disengagement, which would enable early intervention. We develop an enhanced large language model (LLM) fine-tuned with electronic medical records (EMRs) to predict people at risk of disengaging from HIV care in Tanzania and to provide interpretative insights into modifiable risk factors.MethodsWe developed a novel AI model by enhancing a pre-trained LLM (LLaMA 3.1, an open-source pre-trained LLM released by Meta) using routinely collected EMRs from Tanzania’s National HIV Care and Treatment Program from January 1, 2018, to June 30, 2023 (4,809,765 records for 261,192 people) to identify people at risk of disengaging from HIV care or developing adverse outcomes. Outcomes included risk of ART non-adherence, non-suppressed viral load, and loss to follow-up. Models were evaluated internally (Kagera region) and externally (Geita region), with performance compared against state-of-art ML models and zero-shot LLMs. Additionally, a team of HIV physicians in Tanzania assessed the LLM’s predictions along with LLM provided justifications for a subset of patient records to evaluate their clinical relevance and reasoning.FindingsThe enhanced LLMs consistently outperformed the supervised ML model and zero-shot LLMs across all outcomes in both internal and external validation datasets. When focusing on the 25% of PLHIV predicted as most likely to lost-to-follow-up (LTFU), the model correctly identified 78% (2,515 of 3,224) of people living with HIV (PLHIV) genuinely at risk in internal validation and 73% (7,105 of 9,733) in external validation. Attention score analysis indicated that the enhanced LLM focused on keywords such as gaps in follow-up care and ART adherence. The human expert evaluation showed alignment between clinician assessments and the LLM’s predictions in 65% of cases, with experts finding the model’s justifications reasonable and clinically relevant in 92.3% of aligned cases.InterpretationIf implemented in HIV clinics, this LLM-based AI model could help allocate resources efficiently and deliver targeted interventions, improving retention in care and advancing the UNAIDS “95-95-95” targets. By functioning like a clinician–analyzing patient summaries, predicting risks, and offering reasoning–the enhanced LLM could be integrated into clinical workflows to complement human expertise, facilitating timely interventions and informed decision-making. If implemented widely, this human-AI collaboration has the potential to improve health outcomes for people living with HIV and reduce onward transmission.FundingThe study was supported by a grant from the US National Institutes of Health (NIH): NIH NIMH 1R01MH125746.

Large Language Models (LLMs) became wide-spread during only the last couple of years and are used in almost every scientific and non-scientific domain. Understanding opportunities, applications and limitations of LLMs is crucial for a risk-free, effective and useful implementation of LLMs into scientific workflows. We demonstrate that their effectiveness is maximized not through autonomous operation but through careful integration with specialized tools and contextual knowledge bases. Using local deployments of modern LLMs (QWen-2.5-coder, LLaMA, Mistral) comes with a number of benefits in a scientific context. Our approach employs vector embeddings for enhanced context retention and metadata databases for structured data access, enabling guided, context-aware interactions with the LLM. Local deployments allow for improved data handling and privacy, improved cost management and a higher degree of customization. Energy consumption can more easily be observed and managed, which can be a crucial property of such a system, especially compared to the newest generation of LLMs, which have extensive power  (and cost) requirements.Opportunities and limitations are explored through two case studies: (1) an LLM-driven system that queries metadata databases to retrieve data from common open data sources and harmonizes patio-temporal subsets into data-cubes, and (2) a VBA-to-Python code translation project to preserve a legacy selection-system forest management software, which was developed in ACCESS / VBA over more than two decades. The LLM's translation process and reasoning are preserved in a vector database for consistent context maintenance and the original as well as the ‚new‘ code is searchable using the LLM to aid rebuilding a modernized software. Results suggest that this tool-augmented approach leads to a more reliable and maintainable solution compared to purely LLM-driven implementations, suggesting a new paradigm for integrating AI in scientific workflows where LLMs rather facilitate than replace domain-specific tools and human expertise.

Although research evaluators and scientometricians have promoted the message of responsible bibliometrics through initiatives like the Leiden Manifesto, these do not mention Large Language Models (LLMs). LLMs can now make useful quality predictions for journal articles, giving values that correlate more strongly with expert judgements than do citation-based indicators in most fields. This has created the possibility that they could supplement or even replace citation-based indicators for some applications. As tested so far, LLMs predict the quality rating that a human expert would give a paper. They do this by reading the quality level descriptions and then processing the article title and abstract. This raises multiple new issues in comparison to the Leiden Manifesto. First, authors might try to trick LLMs into giving high scores by crafting LLM-friendly abstracts. Second, LLM models incorporate billions of parameters, so their scores are opaque. Third, it is not clear how LLMs work in terms of the main influences on their scores, so their biases are unknown. Fourth, whilst citations reflect tangible and permanent contributions to the scientific record, albeit of variable value, LLM-based predictions do not clearly link to progress. Fifth, LLM scores are ephemeral in the sense that newer LLMs may give substantially different scores and rankings.

In this article, we explore the application of Large Language Models (LLMs) in assessing the content validity of psychometric instruments, focusing on the Big Five Questionnaire (BFQ) and Big Five Inventory (BFI). Content validity, a cornerstone of test construction, ensures that psychological measures adequately cover their intended constructs. Using both human expert evaluations and advanced LLMs, we compared the accuracy of semantic item-construct alignment. Graduate psychology students employed the Content Validity Ratio to rate test items, forming the human baseline. In parallel, state-of-the-art LLMs, including multilingual and fine-tuned models, analyzed item embeddings to predict construct mappings. The results reveal distinct strengths and limitations of human and AI approaches. Human validators excelled in aligning the behaviorally rich BFQ items, while LLMs performed better with the linguistically concise BFI items. Training strategies significantly influenced LLM performance, with models tailored for lexical relationships outperforming general-purpose LLMs. Here we highlight the complementary potential of hybrid validation systems that integrate human expertise and AI precision. The findings underscore the transformative role of LLMs in psychological assessment, paving the way for scalable, objective, and robust test development methodologies.

BackgroundArtificial intelligence (AI) has rapidly permeated various sectors, including healthcare, highlighting its potential to facilitate mental health assessments. This study explores the underexplored domain of AI’s role in evaluating prognosis...

Multimodal large language models (MLLMs) capable of integrating visual and textual information represent a promising advancement for clinical applications requiring image interpretation. Wound care assessment, which demands simultaneous analysis of wound photographs and clinical data, provides an ideal domain to evaluate multimodal vs unimodal artificial intelligence capabilities against human expertise. This study aims to compare the performance of MLLMs, unimodal ChatGPT 5.0, and human clinical experts on a standardized wound care certification examination. This cross-sectional comparative study evaluated 3 participant groups on a 25-question wound care certification examination spanning 4 clinical domains (Diagnosis, Treatment, Complication Management, and Wound Subtype Knowledge). Participants included 3 MLLMs (Med-PaLM 2, LLaVA-Med, and BioGPT), 1 unimodal large language model (ChatGPT 5.0), and 4 human clinical experts (general surgeon, wound care nurse, and 2 internal medicine physicians). Statistical analyses included one-way ANOVA with Tukey post hoc tests and domain-specific Kruskal-Wallis comparisons. Human experts achieved the highest accuracy (mean 86%, SD 9.1%), followed by MLLMs (mean 78.7%, SD 12.2%), while ChatGPT 5.0 achieved 64% accuracy, failing the 70% certification threshold. Significant overall group differences were observed (F2,5=8.42, P=.02, η²=0.74). MLLMs significantly outperformed ChatGPT 5.0 (difference=14.7 percentage points, P=.03, Cohen d=1.38), with the multimodal advantage most pronounced in visually dependent domains: Diagnosis (81% vs 43%, P=.008) and Complication Management (72% vs 50%, P=.03). No multimodal advantage was observed for text-based Wound Subtype Knowledge (both 67%). Med-PaLM 2 achieved 92% accuracy, matching that of the wound care nurse, while the general surgeon achieved the highest overall performance (96%). MLLMs demonstrate significant performance advantages over unimodal artificial intelligence in wound care assessment, particularly for visually dependent clinical tasks. While human experts with specialized wound care experience maintain overall superiority, the point estimate of the top-performing MLLM (Med-PaLM 2, 92%) fell within the observed range of human scores; however, the underpowered comparison (power=0.52) and wide CIs preclude definitive conclusions regarding noninferiority or equivalence to human experts. These findings support the potential role of MLLMs as clinical decision-support tools, warranting further adequately powered validation studies.

The coding capabilities of large language models (LLMs) have opened up new opportunities for automatic statistical analysis in machine learning and data science. However, before their widespread adoption, it is crucial to assess the accuracy of code generated by LLMs. A major challenge in this evaluation lies in the absence of a benchmark dataset for statistical code. To fill in this gap, this paper introduces StatLLM, an open-source dataset for evaluating the performance of LLMs in statistical analysis. The StatLLM dataset comprises three key components: statistical analysis tasks, LLM-generated SAS code, and human evaluation scores. The first component includes statistical analysis tasks spanning a variety of analyses and datasets, providing problem descriptions, dataset details, and human-verified SAS code. The second component features SAS code generated by GPT-3.5, GPT-4, and Llama-3.1 70B for those tasks. The third component contains evaluation scores from human experts in assessing the correctness, effectiveness, readability, executability, and output accuracy of the LLM-generated code. We also illustrate the unique potential of the established benchmark dataset for (1) evaluating and enhancing natural language processing metrics, (2) assessing and improving LLM performance in statistical coding, and (3) developing and testing of next-generation statistical software – advancements that are crucial for data science and machine learning research. This paper has online supplementary materials.

Computation-intensive tensor operators constitute over 90% of the computations in Large Language Models (LLMs) and Deep Neural Networks. Automatically and efficiently generating high-performance tensor operators with hardware primitives is crucial for diverse and ever-evolving hardware architectures like RISC-V, ARM, and GPUs, as manually optimized implementation takes at least months and lacks portability. LLMs excel at generating high-level language codes, but they struggle to fully comprehend hardware characteristics and produce high-performance tensor operators. We introduce a tensor-operator auto-generation framework with a one-line user prompt (QiMeng-TensorOp), which enables LLMs to automatically exploit hardware characteristics to generate tensor operators with hardware primitives, and tune parameters for optimal performance across diverse hardware. Experimental results on various hardware platforms, SOTA LLMs, and typical tensor operators demonstrate that QiMeng-TensorOp effectively unleashes the computing capability of various hardware platforms, and automatically generates tensor operators of superior performance. Compared with vanilla LLMs, QiMeng-TensorOp achieves up to 1291× performance improvement. Even compared with human experts, QiMeng-TensorOp could reach 251% of OpenBLAS on RISC-V CPUs, and 124% of cuBLAS on NVIDIA GPUs. Additionally, QiMeng-TensorOp also significantly reduces development costs by 200× compared with human experts.

Skeletal dysplasias are a heterogeneous group of rare genetic disorders with diverse and overlapping clinical presentations, posing diagnostic challenges even for experienced clinicians. With the increasing availability of artificial intelligence (AI) in healthcare, large language models (LLMs) offer a novel opportunity to assist in rare disease diagnostics. This study aimed to compare the diagnostic accuracy of two advanced LLMs, ChatGPT (version GPT-4) and DeepSeek, with that of a clinical expert panel in a cohort of pediatric patients with genetically confirmed skeletal dysplasias. We designed a prospective vignette-based diagnostic benchmarking study including 45 children with confirmed skeletal dysplasias from two tertiary centers. Both LLMs were prompted to provide primary and differential diagnoses based on standardized clinical case vignettes. Their outputs were compared with those of two human experts (a pediatric endocrinologist and a pediatric orthopedic surgeon), using molecular diagnosis as the gold standard. ChatGPT and DeepSeek achieved a comparable diagnostic top-3 accuracy (62.2% and 64.4%, respectively), with a high intermodel agreement (Cohen's κ = 0.95). The expert panel outperformed both models (82.2%). While LLMs performed well on more common disorders, they struggled with ultra-rare and multisystemic conditions. In one complex case missed by experts, the DeepSeek model successfully proposed the correct diagnosis. LLMs offer a complementary diagnostic value in skeletal dysplasias, especially in under-resourced medical settings. Their integration as a supportive tool in multidisciplinary diagnostic workflows may enhance early recognition and reduce diagnostic delays in rare disease care.

The emergence of Large Language Models (LLMs), such as ChatGPT and Claude, has sparked discussions on their capacity to augment qualitative research data analysis. This study investigates using LLMs in futures research, where analysing large textual datasets is paramount. LLMs were employed experimentally to assess theme generation, patterns, and insights from infrastructure futures workshops. The study conducted a comparative analysis of Claude’s Sonnet 3.5 and ChatGPT 4o to examine workshop transcripts with and without contextual data. The data reveal that LLMs exhibit exceptional speed and efficiency but face challenges in precision, ethical implications, and the need for human supervision. This research adds to current discussions on integrating LLMs with qualitative research methods, enhancing futures analytical tools like Causal Layered Analysis, the Futures Triangle, Futures Wheel, Futures Cone, visioning, and Backcasting. The paper argues that LLMs offer a promising means of improving qualitative data analysis, maximised when used with conventional methods and human expertise.

Recent breakthroughs in Large Language Models (LLMs) have positioned them as a promising paradigm for agents, with long-term planning and decision-making emerging as core general-purpose capabilities for adapting to diverse scenarios and tasks. Real-time strategy (RTS) games serve as an ideal testbed for evaluating these two capabilities, as their inherent gameplay requires both macro-level strategic planning and micro-level tactical adaptation and action execution. Existing RTS game-based environments either suffer from relatively high computational demands or lack support for textual observations, which has constrained the use of RTS games for LLM evaluation. Motivated by this, we present TowerMind, a novel environment grounded in the tower defense (TD) subgenre of RTS games. TowerMind preserves the key evaluation strengths of RTS games for assessing LLMs, while featuring low computational demands and a multimodal observation space, including pixel-based, textual, and structured game-state representations. In addition, TowerMind supports the evaluation of model hallucination and provides a high degree of customizability. We design five benchmark levels to evaluate several widely used LLMs under different multimodal input settings. The results reveal a clear performance gap between LLMs and human experts across both capability and hallucination dimensions. The experiments further highlight key limitations in LLM behavior, such as inadequate planning validation, a lack of multifinality in decision-making, and inefficient action use. We also evaluate two classic reinforcement learning algorithms: Ape-X DQN and PPO. By offering a lightweight and multimodal design, TowerMind complements the existing RTS game-based environment landscape and introduces a new benchmark for the AI agent field.