Background
Generative artificial intelligence (AI) is increasingly integrated into healthcare and policy-making, including the development of clinical and operational practice guidelines. However, ethical concerns—such as algorithmic bias, accountability, and transparency—remain inadequately addressed, raising risks of inequitable recommendations and eroded trust in guidelines.

Objective
This study systematically evaluates the ethical frameworks and challenges associated with using generative AI in formulating practice guidelines, aiming to identify gaps and propose actionable recommendations for responsible integration.

Methods
A cross-sectional review was conducted by analyzing peer-reviewed articles, institutional reports, and AI ethics guidelines published between 2022 and 2024. Databases including PubMed, Embase, IEEE Xplore, and Google Scholar were searched using terms related to “generative AI,” “ethics,” and “practice guidelines.” Included studies were thematically analyzed to extract ethical themes, regulatory approaches, and stakeholder perspectives.

Results
Of 1,452 screened documents. Key ethical issues included: (1) lack of transparency in AI decision-making processes, (2) risks of perpetuating biases from training data, and (3) unclear accountability for guideline errors. Regulatory gaps were prominent in low-resource settings and non-clinical domains. Detailed results will be presented at meeting.

Discussion
Current ethical frameworks for GAI in guideline development emphasize technical validation but underprioritize human oversight, equity, and participatory design. Strengthening accountability mechanisms, ensuring diverse dataset curation, and mandating transparency in AI contributions are critical to safeguarding guideline integrity. Future research should focus on co-developing ethical standards with clinicians, policymakers, and marginalized communities to mitigate harms and enhance trust in AI-augmented guidelines.

Background
Decision Support Systems (DSS) help transform clinical guidelines into actionable recommendations within electronic health records (EHRs). Traditional DSS rely on structured data and rule-based logic, making it challenging to process unstructured information in guidelines and free-text EHR entries. This study explores the integration of generative artificial intelligence (AI) to enhance decision support for Youth Health Care (JGZ) professionals.

Methods
An AI model, based on large language models (LLM), was developed using JGZ guidelines, data protocols, and an ontology, i.e., the "basis dataset". It was validated by analysing 14 pediatric cases (ages 0-4 years), focusing on hip dysplasia, skin conditions, and congenital heart defects. The AI-generated recommendations included the identification of abnormalities, suggested follow-up actions, and assessment of intervention urgency. These AI recommendations were compared with 1) existing guideline-based recommendations and 2) real-world decisions made by JGZ professionals.

Results
* Full agreement between AI-generated recommendations and guidelines in distinguishing normal vs. abnormal findings.
* 60% agreement on follow-up actions, with discrepancies primarily due to missing contextual factors, such as patient history, regional guidelines, and professional expertise.
* AI-generated recommendations provided relevant and actionable educational guidance for parents, supporting informed decision-making.

Conclusions
AI-driven decision support can effectively assist YHC professionals, particularly in identifying abnormalities and providing tailored education and advice. However, human expertise remains essential for context-based decision-making. Future improvements should focus on integrating historical cliënt data, coping with regional variations, and an interactive user interface to enhance AI-human collaboration in clinical practice.

Background
Generative artificial intelligence (GAI) tools can enhance the quality and efficiency of medical research, but their improper use may result in plagiarism, academic fraud, and unreliable findings. Transparent reporting of GAI use is essential, yet existing guidelines from journals and institutions are inconsistent, with no standardized principles.

Objective
To address this, we developed the RAISE checklist (Reporting guideline for the use of chatbots and other generative Artificial Intelligence toolS in mEdical research) through an international, multidisciplinary expert group.

Methods
The development process included a scoping review, two Delphi rounds, and virtual meetings with 51 experts from 26 countries.

Results
The final checklist comprises nine reporting items: general declaration, GAI tool specifications, prompting techniques, tool’s role in the study, declaration of new GAI model(s) developed, AI-assisted sections in the manuscript, content verification, data privacy, and impact on conclusions.

Discussion for scientific abstracts
RAISE provides universal and standardised guideline for GAI use in medical research, ensuring transparency, integrity, and quality.

Background: In interpersonal and human-computer interaction (HCI), natural language processing significantly enhances communication efficiency, with large language models (LLMs) like GPT-4o demonstrating advanced capabilities in contextual analysis through deep learning architectures. While educational psychology substantiates the efficacy of positive feedback in human performance optimization, its impact on LLM functionality, particularly in critical healthcare applications such as clinical practice guideline evaluation using standardized tools like AGREE II, remains inadequately investigated.
Objectives: This study systematically examines whether encouraging linguistic prompts enhance GPT-4o's performance in assessing clinical guideline quality compared to neutral prompts.
Methods: Utilizing 28 clinical guidelines with human AGREE II evaluations from PubMed as the gold standard, we conducted a controlled experiment where GPT-4o's experimental group received encouraging prompts (e.g., "Please analyze meticulously") while the control group received neutral instructions. Methodological rigor was ensured through intraclass correlation coefficient (ICC) analysis, Bland-Altman agreement assessments, and paired sample t-tests comparing domain score variances between LLM and human evaluations.
Results: Quantitative analysis revealed superior alignment in the encouragement cohort, demonstrating a 0.632% mean score differential from human ratings (95% LoA: -35.354% to 47.996%) with 89.9% of deviations within ±33.3% clinical acceptability thresholds. Comparatively, neutral prompts yielded a 12.471% mean divergence (95% LoA: -30.566% to 55.508%) and 81.5% threshold compliance. Statistical confirmation via paired t-tests (p<0.05) established the significant mitigating effect of encouraging language on scoring discrepancies across AGREE II domains, substantiating LLM sensitivity to motivational linguistic cues in clinical evaluation contexts.

Background: Published systematic reviews (SRs) display an heterogeneous methodological quality. Poorly conducted SRs can have biased results and/or inadequate conclusions. Therefore, guideline developers should carefully assess the methodological quality of the SRs they use as sources of evidence. However, such an assessment can be time-consuming.
Objective: (i) To develop a tool to assess the reported methodological quality of SRs and (ii) to evaluate its implementation through large language models (LLMs).
Methods: We developed a new tool – ReMarQ – consisting of 26 dichotomous items to evaluate the reported methodological quality of SRs. We applied an Item Response Theory model to assess the difficulty and discrimination of ReMarQ items. We assessed the performance of five foundational and three fine-tuned LLMs in implementing ReMarQ in a sample of 100 medical SRs. In particular, we compared the answers provided by the LLMs to each of the 26 items of ReMarQ with those provided independently by human reviewers.
Results: The best performing LLM was a fine-tuned GPT-3.5 model (mean accuracy=96.5%; mean kappa coefficient=0.90; mean F1-score=0.91). When compared with a human reviewer, this model displayed an accuracy above 80% and a kappa coefficient higher than 0.60 for all individual items. The model produced consistent results when the analysis was repeated 60 times on the same sample.
Discussion for scientific abstracts: We have developed a new tool to assess the methodological quality of SRs. Its implementation through LLMs – which demonstrated high accuracy and reliability – may increase the efficiency of the methodological evaluation of SRs.