18 November 2024 in Nature Communications
Patient recruitment is challenging for clinical trials. We introduce TrialGPT, an end-to-end framework for zero-shot patient-to-trial matching with large language models. TrialGPT comprises three modules: it first performs large-scale filtering to retrieve candidate trials (TrialGPT-Retrieval); then predicts criterion-level patient eligibility (TrialGPT-Matching); and finally generates trial-level scores (TrialGPT-Ranking). We evaluate TrialGPT on three cohorts of 183 synthetic patients with over 75,000 trial annotations. TrialGPT-Retrieval can recall over 90% of relevant trials using less than 6% of the initial collection. Manual evaluations on 1015 patient-criterion pairs show that TrialGPT-Matching achieves an accuracy of 87.3% with faithful explanations, close to the expert performance. The TrialGPT-Ranking scores are highly correlated with human judgments and outperform the best-competing models by 43.8% in ranking and excluding trials. Furthermore, our user study reveals that TrialGPT can reduce the screening time by 42.6% in patient recruitment. Overall, these results have demonstrated promising opportunities for patient-to-trial matching with TrialGPT.
1 April 2022 in International Journal of Infectious Diseases
Vaccination is an essential measure to stop the coronavirus disease 2019 (COVID-19) pandemic. We report a case of viral activation and CD4+ T cell loss in a treatment-naïve HIV-positive patient after receiving inactivated COVID-19 vaccine (Sinopharm). The vaccine should probably be given only to people living with HIV receiving antiretroviral therapy.
27 September 2023 in Clinical and Translational Gastroenterology
INTRODUCTION: Persistent villous atrophy (VA) is not uncommon in celiac disease (CeD) while patients take a gluten-free diet (GFD). METHODS: We conducted a retrospective study with 122 serum samples collected from controls and patients with CeD either at the initial diagnosis or at the follow-up during endoscopy. These samples were assigned to 3 groups: nonceliac control, non-VA CeD (Marsh score 0–2), and VA CeD (Marsh score 3a-3c). We established an in-house multiplex assay to identify potential serological biomarkers for VA. We assessed autoantibodies reported to affect the small intestine, including IgA and IgG antibodies against tissue transglutaminase (tTG), interferons, villin, actin, autoimmune enteropathy-related 75 kDa antigen (AIE-75), and tryptophan hydroxylase (TPH)-1, as well as 27 cytokines. The apolipoproteins quantified included apo A1, apo B-100, and apo A4, which were produced predominantly by the intestinal epithelium or expressed specifically in villi. RESULTS: Autoantibody levels were high only for tTG antibodies, which performed well in initial CeD diagnosis, but suboptimally for VA prediction during follow-up, because 14.6% of the follow-up patients with VA had low tTG-IgA. Increasing dilution improved tTG-IgA quantification, particularly when the antibody levels were extremely high but did not significantly improve VA detection. Among those with low tTG-IgA and persistent VA, high proinflammatory cytokines were observed in 2 patients. Median low-density lipoprotein cholesterol levels were significantly lower in the VA CeD group (P = 0.03). Apolipoprotein levels were similar in patients with and without VA but diverged between those on a GFD or not. DISCUSSION: tTG-IgA as a biomarker is suboptimal for VA prediction while on a GFD. Persistent VA is associated with low low-density lipoprotein cholesterol levels and partially related to persistent high proinflammatory cytokines.
19 September 2026 in Communications Medicine
Background: The remarkable capabilities of large language models make them increasingly compelling for use in real-world healthcare applications. However, the risks associated with using these artificial intelligence systems in medicine are not systematically understood. The aim of this study is to characterize these risks by applying five key principles for safe and trustworthy medical artificial intelligence: truthfulness, resilience, fairness, robustness, and privacy. Methods: We introduce MedGuard-Bench: a safety benchmark featuring one thousand expert-verified questions covering ten specific aspects of our five core principles. We use this comprehensive corpus to systematically evaluate sixteen commonly used large language models, assessing their safety and reliability in medical contexts. Results: We show that current large language models generally perform poorly on most of our safety tests, regardless of their safety alignment mechanisms. Our evaluation demonstrates that these models fall significantly short when compared to the high performance and reliability of human physicians. Conclusions: Despite reports indicating that advanced large language models can match or exceed human performance in various medical tasks, this study reveals a significant safety gap in current technology. This underscores the crucial need for ongoing human oversight and the implementation of strict safety guardrails before deploying these tools in clinical practice.