AI-driven evidence integration
Investigating how large language models can support retrieval, synthesis and updating while preserving traceability.
Chuan Yin
Research focused on AI (large language model)-driven integration of evidence for evidence-based medicine, exploring global clinical Live Evidence and Patient-Centered Personalized Evidence production.
Founder and Chief Scientist, Evidence OSMedical doctorate · Clinical research and evidence methods
Research pathway: AI-driven evidence integration → global living evidence → patient-centered personalized evidence.
EOS research focus
Global clinical Live Evidence
Exploring systems and methods that connect global clinical research, continuously update evidence and track changes.
Patient-Centered Personalized Evidence
Investigating how patient questions, goals, preferences and comorbidity can inform relevant, transparent and reviewable evidence.
Research contributions to EOS
A patient-centred research framework
Organises evidence generation and use around patient goals, comorbidity and specific decisions, and identifies the steps requiring evaluation between medical answers and living evidence.
Methods applied to clinical research
Works with co-authors on the review of preoperative social connection and postoperative outcomes, organising 20,399 deduplicated records and 445 studies; now advancing musculoskeletal–metabolic comorbidity synthesis.
Traceable evidence governance
Works with co-authors to connect sources, effect selection, statistical contributions, corrections and released versions. Internal evaluation in one review case is complete, with independent comparisons planned.
RESEARCH JOURNEY
From a framework to reviewable research
Clinical synthesis becomes public research
The systematic review and meta-analysis of preoperative social connection and postoperative outcomes was released as a manuscript. Chuan Yin is first author, working with the research team; 72 studies enter at least one quantitative synthesis.
Read the manuscriptA patient-centred Evidence OS framework
An author manuscript with Zhicheng Zhang proposes a framework for evidence generation and use, starting with patient goals, comorbidity and clinical questions.
Read the manuscriptMapping medical LLM research
Published the single-author review manuscript, Large Language Models in Medical Decision Making from Clinical Answers to Living Evidence, connecting research questions across retrieval, synthesis, updating and individual decisions.
Read the manuscriptInternal evaluation of an auditable system
The second version of the methods manuscript with Zehao Jing and Zhicheng Zhang evaluates source, correction and version links in one review case: 50 root-cause events, 46 resolved and 4 retained limitations.
Read the manuscriptComorbidity research and the next evaluations
The musculoskeletal–metabolic comorbidity systematic review and meta-analysis is in the review stage. Conventional synthesis will provide the baseline for evaluating mixed-information synthesis and individual applicability.
CURRENT WORK
Read the daily evidence briefCurrent work & next steps
Musculoskeletal–metabolic comorbidity
A traditional systematic review and meta-analysis assisted by LLMs is in the review stage.
Next stepComplete conventional synthesis, then use the case to evaluate mixed-information synthesis and individual applicability.
Mixed-information synthesis
Method boundaries and an evaluation design compare conventional and mixed-information synthesis for coverage and credibility.
Next stepTest additional information, statistical coherence and clinical value in the comorbidity review.
Evidence to individual decisions
A decision-material evaluation design is established; comorbidity evidence will support further evaluation of individual applicability.
Next stepEvaluate applicability, traceability, justified abstention and decision quality; confidence alone is insufficient.
Complete the comorbidity comparison baseline
Complete LLM-assisted conventional systematic review and meta-analysis, retaining eligibility, extraction and reasons for non-pooling to support subsequent method comparisons.
Test the contribution of mixed information
Compare how much credible information can be retained when measures, time windows or statistical reporting differ, and assess statistical coherence and clinical value.
Evaluate evidence for specific decisions
Assess individual applicability, traceability and justified abstention around comorbidity and patient goals, then develop independent comparisons and evaluations of evidence updating.