Ulrich Mansmann

Claude Lévi-Strauss’s structuralism offers a useful perspective on large language models (LLMs) through its emphasis on relationships within systems of meaning. This paper recalls relational meaning, oppositions and mediation, transformations, and bricolage, illustrating these ideas through Lévi-Strauss’s analysis of myth and an everyday example of gift-giving. It then examines their connections with contextual representations and generative processes in LLMs. These connections support a productive comparison, although similarities do not establish that language models implement or validate structuralist theory. Practical implications include assessing whether models preserve relevant relationships across reformulations, examining the cultural categories expressed in their responses, and distinguishing coherent explanations from empirically justified claims. Structuralism can inform the interpretation and evaluation of LLM behaviour, while LLMs offer opportunities to investigate how relationships are learned, represented and transformed. The comparison therefore provides a basis for exploring meaning and creativity, with attention to the differences between linguistic competence, cultural understanding and knowledge of the world.

Models for individualized treatment recommendations in relapsing–remitting multiple sclerosis often select therapies based on predicted risks of relapse and/or 3-month confirmed disability progression (3m-CDP). Recommendations should be interpretable, align with guidelines, and reflect patient perspectives. We validated recently developed prognostic algorithms in eight subgroups defined by reason for therapy switch: intolerance, lack of efficacy, pregnancy desire, serious adverse events, programmed stop, personal convenience, therapy initiation, drug-holiday end. From the OFSEP registry, we analysed 3768 therapy cycles from 3768 patients treated between 2017 and 2021 with interferon beta, glatiramer acetate, teriflunomide, dimethyl fumarate, fingolimod, natalizumab. One therapy cycle was randomly selected per patient. The algorithms produced ranked therapy lists per treatment cycle as switching option; certainty of the provided recommendation was quantified by entropy (0 = clear; 4.78 = maximal uncertainty). Recommendations were more consistent for relapse (median entropy 0.54, IQR 0.39–0.61) than for 3m-CDP (0.95, IQR 0.88–1.01). Across subgroups, natalizumab most often ranked highest for relapse, whereas teriflunomide or interferon beta frequently ranked highest for 3m-CDP. Calibration varied across switcher subgroups and outcomes, while discrimination was comparable to overall set: C-index 0.776 (95% CI 0.753–0.797) for 3m-CDP and 0.638 (95% CI 0.615–0.661) for relapse. Guideline and relapse-based recommendations did not always align, notably in the pregnancy desire subgroup. For 3m-CDP, guideline-based recommendations are scarce, precluding systematic comparison with model outputs. Agreement between the best-ranked therapy for relapse and 3m-CDP was zero (κ = 0), underscoring outcome-dependent divergence. Algorithm-generated recommendations depend on the chosen outcome and may diverge from guideline-based practice or patient perspectives. Transparent communication of uncertainty and outcome trade-offs is essential for shared decision-making.

Background: Clinical trial data sharing is increasingly endorsed by journals, funders, regulators, and research organisations. Nevertheless, access to individual participant data (IPD) remains inconsistent and may not be available when trial results are published. Access timelines can reflect several distinct components, including scientific and governance review, contractual and legal procedures, data preparation, de-identification, transfer, and subsequent harmonisation. When original data are unavailable, researchers may reconstruct approximate participant-level event-time data from published Kaplan-Meier curves. Trinquart and Stockler recently proposed that trialists instead release minimal analysable time-to-event data at publication. Their proposal raises a broader question: what kinds of data and documentation are required for different forms of reproducibility and scientific reuse? Main text: We argue that clinical trial data transparency should be considered in relation to the scientific claims and research activities that a particular data object can support. We propose a three-tier framework for data intentionally released or made accessible by trialists or data custodians. Tier 1 comprises original publication-time reduced datasets and supports endpoint-level numerical reproducibility and analyses that are functions of the released observations. Tier 2 comprises reproducibility packages containing analysis-ready data, code, derivation logic, and documentation sufficient to reproduce the reported analytic process. Tier 3 provides governed access to broader de-identified IPD for substantively new analyses and reassessment of operational and analytic decisions. Reconstructed pseudo-IPD are considered separately because they are generated externally from publications rather than shared from the trial database. Privacy protection, governance, and processing environment are cross-cutting dimensions and do not map uniquely to a particular tier. They also apply differently to each tier. Conclusions: Publication-time reduced datasets can provide meaningful and timely transparency when broader IPD are not yet accessible. Their value should be recognised without confusing endpoint-level numerical reproducibility with reproduction of the complete analytic process or broader scientific reuse. We propose that publication-time reduced datasets should form a practical floor for timely transparency, while reproducibility packages and governed access to broader IPD provide progressively richer opportunities for reproducing, interrogating, and extending clinical trial analyses.

Background: A tension exists between the open science imperative for data transparency through FAIR (Findable, Accessible, Interoperable, Reusable) data sharing and the stringent requirements of current data protection regulations for health data. Researchers lack efficient yet affordable tools to share sensitive data responsibly. Existing solutions are often either difficult to use, expensive, or lack features like variable-level access control and auditable and reproducible workflows. Our objective was to develop a solution that lowers the threshold for researchers to share sensitive data responsibly and that can be operated by existing institutional staff. Results: TRACE (Trusted Research Access & Collaboration Environment) was designed to support several data-sharing elements highlighted in the 2025 SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) and CONSORT (Consolidated Standards of Reporting Trials) statements by providing controlled and auditable workflows and supporting reproducibility at the level of data and analysis code. Its features for data governance, sharing, reproducibility and security allow institutions to manage study data access and to establish consistent, transparent standard operating procedures through granular variable selection, integrated study visualisation, and provisioning into a sealed analysis environment. TRACE provides an automated workflow for selected data-access steps that had previously relied on manual curation and ad-hoc sharing. A core feature versions datasets together with the corresponding analysis scripts, supporting reproducibility at the level of data and code; the computational environment is not captured. TRACE has supported data governance across seven clinical studies and processed 56 data access applications. Across three studies, an uncontrolled before-and-after comparison found 42.4 manual interface exports per 30 days before TRACE and 4.7 after implementation; causality cannot be inferred. Conclusions: Platforms like TRACE may help shift resource-intensive, expert-dependent data governance towards automated processes. Operated by existing institutional staff, TRACE may make controlled data sharing feasible for institutions with limited resources. By automating previously manual steps, TRACE is intended to reduce the administrative burden of governed data sharing; whether it shortens the interval from data request to analysis has not been measured. TRACE is used as a training environment for preparing and using data in sharing processes.

The selection of the optimal antithrombotic regimen in patients with atrial fibrillation and acute coronary syndrome remains challenging. Previous trials have demonstrated that dual antithrombotic therapy (DAT), consisting of direct oral anticoagulants (DOACs) plus a P2Y12 inhibitor, reduces bleeding compared to a triple-therapy regimen using vitamin K antagonists. However, subsequent meta-analyses have suggested an increased risk of ischemic events with DAT, particularly within the first month of treatment. Clopidogrel has been the predominant P2Y12 inhibitor used across these studies, despite the risk of high on-treatment platelet reactivity when using this drug. In this study, we conducted an open-label, randomized controlled trial (EPIDAURUS) in patients with atrial fibrillation and acute coronary syndrome, designed to assess the efficacy and safety of a 1-month regimen of DOAC plus potent P2Y12 inhibitor (prasugrel or ticagrelor) compared to DOAC plus clopidogrel and in-hospital aspirin. The primary outcomes of the trial were an efficacy endpoint, recurrent ischemic events and safety endpoints, including death and major bleeding, which were evaluated 6 weeks after randomization using separate win−loss ratio analyses. The study was prematurely terminated after enrollment of 602 patients (154 female) of an expected 1,474 patients, owing to safety concerns raised by the Data and Safety Monitoring Board. Exploratory analyses of secondary safety endpoints, including bleeding type ≥2 and ≥3 according to the Bleeding Academic Research Consortium scale, revealed that, compared to clopidogrel and in-hospital aspirin, treatment with a potent P2Y12 inhibitor was associated with higher bleeding rates without a clear reduction in the risk of ischemic complications. These findings do not support the routine use of potent P2Y12 inhibitors in combination with DOACs in this patient population. ClinicalTrials.gov identifier: NCT04981041 .

Clinical trials occupy a central position in modern medicine because they generate evidence used to decide whether interventions are effective, safe, and worth implementing. Their social value depends not only on scientific ambition but also on integrity: the assurance that trials are ethically justified, methodologically sound, operationally controlled, statistically credible, transparently reported, and open to scrutiny. Clinical trial integrity is broader than fraud prevention and broader than regulatory compliance. It includes the relevance of the research question, the protection of participants, the precision with which the treatment effect is defined, the faithful execution of the protocol, reliable data governance, prespecified statistical analysis, balanced interpretation, and responsible public communication. This article argues that trial integrity is best understood as a life-cycle concept. It begins with a clinically meaningful question and a design capable of answering it, continues through trial conduct, data management, statistical analysis, interpretation, and transparent reporting, and extends after publication through data sharing, public audit, post-publication review, and correction of the scientific record. Particular attention is given to prespecification, trial registration, statistical analysis plans, the query process, negative trials, early termination, and the interaction between personal, institutional, and community-driven integrity. The central claim is that trustworthy clinical trials require more than procedural compliance. They require a professional culture in which patient welfare, methodological rigor, reproducibility, and public accountability remain aligned throughout the entire trial life cycle.

27 July, 2026. This preprint has been retracted at the request of the authors due to a conflict with the European Society of Cardiology conference embargo policy. The retraction is unrelated to the scientific content of the manuscript.