Grounding AI-Driven Clinical Decision Support in Evidence-Based Medicine: Visual Analytics for Evidence Construction and Application

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Grounding AI-Driven Clinical Decision Support in Evidence-Based 
Medicine: Visual Analytics for Evidence Construction and Application"

By

Mr. Rui SHENG


Abstract:

Artificial intelligence (AI) for clinical decision support has become 
increasingly prevalent in healthcare. However, those AI models always remain 
opaque, limiting clinicians' ability to understand and appropriately 
integrate AI recommendations into practice. Prior work has introduced 
explainable AI techniques, such as feature attribution and counterfactual 
explanations, but these methods are largely model-centric and poorly aligned 
with clinicians' realistic reasoning processes. Specifically, clinical 
decision-making is typically grounded in evidence-based medicine (EBM), where 
patient-specific findings are interpreted in the context of prior clinical 
trials, guidelines, and medical literature. This thesis investigates visual 
analytics for clinical decision support through the lens of evidence-based 
medicine, aiming to enable human-AI collaboration that better aligns with 
clinicians' common decision-making practice.

The thesis makes three complementary contributions to evidence-based clinical 
decision support through visual analytics. First, it investigates how visual 
analytics can support the construction of high-quality clinical evidence in 
clinical trials. It introduces an interactive system that helps clinicians 
design more appropriate eligibility criteria for clinical trials by exposing 
trade-offs among population coverage, cohort representativeness, and outcome 
validity. Second, before designing AI-assisted decision-making systems that 
are meaningfully integrated with clinical evidence, the thesis systematically 
explores the current designs of clinical human-AI interfaces. Through a 
systematic review of 43 papers, it synthesizes 15 core information entities 
and 12 reusable design patterns for AI-driven clinical systems. In addition, 
the thesis conducts interviews with healthcare professionals to supplement 
the justifications and considerations of the 12 design patterns, and 
organizes workshops to investigate how designers apply these design patterns 
in practice. These findings provide a design foundation for building more 
effective clinical decision support systems and reveal that existing systems 
rarely incorporate literature-based evidence as a first-class, richly 
structured design element in AI-driven workflows. Third, the thesis explores 
how clinicians can more effectively use clinical evidence in AI-assisted 
decision-making. It presents a visual analytics system that integrates 
patient-specific AI explanations with evidence from the clinical literature, 
thereby supporting informed and balanced clinical decisions.

Taken together, these contributions advance a unified research agenda that 
spans evidence construction, design knowledge for human-AI systems in 
clinical decision-making, and evidence-centered clinical decision support 
with AI. The findings suggest that effective human-AI collaboration requires 
not only more transparent AI models, but also careful integration with 
clinicians' common practice.


Date:                   Tuesday, 25 August 2026

Time:                   10:00am - 12:00noon

Venue:                  Room 3494
                        Lifts 25/26

Chairman:               Prof. Jiguang WANG (LIFS)

Committee Members:      Prof. Huamin QU
                        Dr. Xiaojuan MA
                        Prof. Ke YI
                        Dr. Wenhan LUO (AMC)
                        Prof. Jinwook SEO (Seoul National University)