Evaluating LLMs and AI agents for health

2026

1. npj Vaccines

   Evaluating large language models on multilingual vaccine knowledge: a benchmark study

   Siyuan Chen, Lily Wass, Zhengdong Wu, Lucas Garay, José Vizoso, Kathy Leung, Joseph Wu, and Leesa Lin

   npj Vaccines, Jun 2026

   Abs DOI Bib HTML

   Large language models (LLMs) are increasingly used by clinicians and the public for vaccine information, yet their factual accuracy across languages and vaccine domains remains insufficiently characterized. We evaluated 13 LLMs using VaxEval, a multilingual vaccine-knowledge benchmark of 1886 vaccine-related multiple-choice questions spanning 14 vaccines in English (71%), Spanish (13%), and Chinese (16%). All items underwent quality control, with reference answers verified against authoritative guidance and peer-reviewed sources. Model performance was evaluated under zero-shot, few-shot, and chain-of-thought (CoT) prompting, with exact-match accuracy defined as selecting the pre-specified reference option. We used mixed-effect logistic regression to estimate associations between model group (newer flagship models vs earlier models), prompting strategy, language, and vaccine type, and answer correctness. Mean accuracy across models was 86.0% in English, 83.7% in Spanish, and 80.0% in Chinese. Flagship models had higher odds of correctness than earlier versions (OR 1.57; 95% CI 1.50-1.65; P < .001). Few-shot prompting was associated with higher correctness (OR 1.17; P < .001), whereas CoT prompting was associated with lower correctness (OR 0.79; P < .001). Performance varied by vaccine type and question category, underscoring the need for rigorous evaluation, structured guardrails, and targeted refinement before using LLMs for vaccine communication.

   @article{chen2026llmvaccine,
     title = {Evaluating large language models on multilingual vaccine knowledge: a benchmark study},
     author = {Chen, Siyuan and Wass, Lily and Wu, Zhengdong and Garay, Lucas and Vizoso, José and Leung, Kathy and Wu, Joseph and Lin, Leesa},
     journal = {npj Vaccines},
     volume = {11},
     pages = {187},
     year = {2026},
     month = jun,
     doi = {10.1038/s41541-026-01500-1},
   }

2. medRxiv

   Artificial Intelligence Agents in Mental Health: A Systematic Review and Meta Analysis

   Lexuan Zhu, Wenkong Wang, Zhiying Liang, Wenjia Tan, Bingyi Chen, Xinxin Lin, Zhengdong Wu, Huizi Yu, Xiang Li, Jiyuan Jiao, Sijia He, and 14 more authors

   medRxiv preprint, Apr 2026

   Abs DOI Bib HTML

   The rapid rise of large language models (LLMs) and foundation models has accelerated efforts to build artificial intelligence (AI) agents for mental health assessment, triage, psychotherapy support and clinical decision assistance. Yet a gap persists between healthcare and AI-focused work: while both communities use the language of “agents,” clinical research largely describes monolithic chatbots, whereas AI studies emphasize agentic properties such as autonomous planning, multiagent coordination, tool and database use and integration with multimodal mental health data streams. In this Review, we conduct a systematic analysis of mental health AI agent systems from 2023 to 2025 using a six-dimensional audit framework: (i) system type (base model lineage, interface modality and workflow composition, from rule-based tools to role-aware multi-agent foundation-model systems), (ii) data scope (modalities and provenance, from elicited self-report and chatbot dialogues to electronic health records, biosensing and synthetic corpora), (iii) mental health focus (mapped to ICD-11 diagnostic groupings), (iv) demographics (age strata, geography and sex representation), (v) downstream tasks (screening/triage, clinical decision support, therapeutic interventions, documentation, ethical–legal support and education/simulation) and (vi) evaluation types (automated metrics, language quality benchmarks, safety stress tests, expert review and clinician or patient involvement). Across this corpus, we find that most systems (1) concentrate on depression, anxiety and suicidality, with sparse coverage of severe mental illness, neurocognitive disorders, substance use and complex comorbidity; (2) rely heavily on text-based self-report rather than clinically verified longitudinal data or genuinely multimodal inputs; (3) are implemented as single-agent chatbots powered by general-purpose LLMs rather than role-structured, workflow-integrated pipelines; and (4) are evaluated primarily via offline metrics or vignette-based scenarios, with few prospective, clinician- or patient-in-the-loop studies. At the same time, an emerging class of agentic systems assigns foundation models explicit roles as planners, retrieval agents, safety auditors or supervisors coordinating other models and tools. These multiagent, tool-augmented workflows promise personalization, safety monitoring and greater transparency, but they also introduce new risks around reliability, bias amplification, privacy, regulatory accountability and the blurring of clinical versus non-clinical roles. We conclude by outlining priorities for the next generation of mental health AI agents: clinically grounded, role-aware multi-agent architectures; transparent and privacy-preserving use of clinical and elicited data; demographic and cultural broadening beyond predominantly Western adult samples; and evaluation pipelines that progress from offline benchmarks to longitudinal, real-world studies with routine safety auditing and clear governance of responsibilities between agents and human clinicians.

   @article{zhu2026agentsreview,
     title = {Artificial Intelligence Agents in Mental Health: A Systematic Review and Meta Analysis},
     author = {Zhu, Lexuan and Wang, Wenkong and Liang, Zhiying and Tan, Wenjia and Chen, Bingyi and Lin, Xinxin and Wu, Zhengdong and Yu, Huizi and Li, Xiang and Jiao, Jiyuan and He, Sijia and Dai, Guangxin and Niu, Jiahui and Zhong, Yi and Zheng, Yongbo and Sun, Jie and Han, Andi and Li, Lingyao and Zhou, Jiayan and Hua, Wenyue and Chan, Ngan Yin and Lu, Lin and Wing, Yun Kwok and Ma, Xin and Fan, Lizhou},
     journal = {medRxiv preprint},
     year = {2026},
     month = apr,
     doi = {10.64898/2026.04.21.26351365},
   }

3. arXiv

   An evidence-guided reinforcement learning method to improve psychiatric reasoning in small language models

   Xinxin Lin, Guangxin Dai, Yi Zhong, Xiang Li, Xue Xiao, Jian Liu, Yixin Zhang, Lingming Hu, Zhengdong Wu, Yongbo Zheng, Runchuan Zhu, and 17 more authors

   arXiv preprint arXiv:2602.06449, Feb 2026

   Abs DOI arXiv Bib

   Privacy and computational constraints limit the use of large language models in psychiatry, while adapting small language models (SLMs) often requires substantial data and expert annotation. We developed ClinMPO, an evidence-guided reinforcement-learning framework guided by the psychiatrist-defined Clinical Psychiatry Thinking Strategy (CPTS). ClinMPO uses ClinRM, a reward model trained on 18,569 question–answer pairs from 4,474 psychiatry articles. We evaluated four Qwen3 sizes on 1,737 model-screened questions. ClinMPO outperformed Base, supervised fine-tuning and standard group relative policy optimization across scales. From responses by 300 senior pre-licensure medical students, we established the human baseline, a medical-student reference. The 4B model approached this baseline, whereas the 8B model surpassed it and ranked first among 31 models and post-training variants. ClinMPO improved performance across two complementary schemes covering ICD-11 diagnostic categories and psychiatric practice competencies. Blinded assessment by three clinicians showed improved rationale quality across CPTS criteria. These findings highlight how existing clinical evidence and specialist knowledge can be incorporated into the development of medical AI systems through evidence-guided learning.

   @article{lin2026psychrl,
     title = {An evidence-guided reinforcement learning method to improve psychiatric reasoning in small language models},
     author = {Lin, Xinxin and Dai, Guangxin and Zhong, Yi and Li, Xiang and Xiao, Xue and Liu, Jian and Zhang, Yixin and Hu, Lingming and Wu, Zhengdong and Zheng, Yongbo and Zhu, Runchuan and Zhao, Ming and Yu, Huizi and Zhang, Yi and Lu, Fangting and Wu, Shuo and Zhao, Jun and Yin, Ping and Chan, Joey W. Y. and Chan, Ngan Yin and Wang, Yumei and Yang, Lejin and Xing, Yanqiu and Chen, Sijing and Wing, Yun Kwok and Lu, Lin and Ma, Xin and Fan, Lizhou},
     journal = {arXiv preprint arXiv:2602.06449},
     year = {2026},
     month = feb,
     doi = {10.48550/arXiv.2602.06449}
   }