Cliff Wong

Cliff Wong

Auto-generated profile, based on publicly available information

Cliff Wong

Microsoft Research, Redmond, WA

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2025 in NEJM AI
Verified
Journal Article
Closed access
SZSheng Zhang
Sheng Zhang
YXYanbo Xu
Yanbo Xu
NUNaoto Usuyama
Naoto Usuyama
HXHanwen Xu
Hanwen Xu
JBJaspreet Bagga
Jaspreet Bagga
HPHoifung Poon
Hoifung Poon

25 September 2025 in NEJM AI
Verified
Journal Article
Closed access
JGJavier González
Javier González
RURisa Ueno
Risa Ueno
CWCliff Wong
Cliff Wong
ZGZelalem Gero
Zelalem Gero
JBJaspreet Bagga
Jaspreet Bagga
HPHoifung Poon
Hoifung Poon

1 November 2024 in JCO Oncology Practice
Verified
Journal Article
Open access
ADAlexa K. Dowdell
Alexa K. Dowdell
RMRyan C. Meng
Ryan C. Meng
AVAnn Vita
Ann Vita
BBBela Bapat
Bela Bapat
DHDouglas Hanes
Douglas Hanes
Brian PieningBrian Piening
Brian Piening
PURPOSE: Precision therapies and immunotherapies have revolutionized cancer care, with novel genomic biomarker-associated therapies being introduced into clinical practice rapidly, resulting in notable gains in patient survival. Despite this, there is significant variability in the utilization of tumor molecular profiling that spans the timing of test ordering, comprehensiveness of gene panels, and clinical decision support through therapy and trial recommendations. METHODS: To standardize testing, we designed a pathologist-directed test ordering system at the time of diagnosis using a 523-gene DNA/RNA hybrid comprehensive genomic profiling (CGP) panel and extensive clinical decision support tools. To comprehensively characterize the clinical impact of this protocol, we developed a novel natural language processing (NLP)–based approach to extract clinical features from physician chart notes. We assessed test actionability rates, therapy choice, and outcomes across a set of 3,216 patients with advanced cancer. RESULTS: We observed 49% of patients had at least one actionable genomic biomarker-driven–approved and/or guideline-recommended targeted or immunotherapy (IO) and 53% of patients would have been eligible for a precision therapy clinical trial from three large basket trials. When assessing CGP versus an in silico 50-gene panel, 67% of tumors compared with 33% harbored actionable alterations including clinical trials. Among patients with 6 months or more of follow-up, over 52% received a targeted therapy (TT) or IO, versus 32% who received conventional chemotherapy alone. Furthermore, patients receiving TT had significantly improved overall survival compared with patients receiving chemotherapy alone ( P < .001). CONCLUSION: Overall, these data represent a major shift in standard clinical practice toward molecularly guided treatments (targeted and immunotherapies) over conventional systemic chemotherapy. As guidelines continue to evolve and more precision therapeutics gain approval, we expect this gap to continue to widen.

25 July 2024 in ESMO Real World Data and Digital Oncology
Verified
Journal Article
Open access
RMR. Meng
R. Meng
ADA.K. Dowdell
A.K. Dowdell
AVA. Vita
A. Vita
DHDouglas Hanes
Douglas Hanes
BBB. Bapat
B. Bapat
Brian PieningBrian Piening
Brian Piening
New breakthroughs in precision therapies are transforming cancer care for patients with advanced non-small-cell lung cancer (NSCLC). To this effect, comprehensive genomic profiling (CGP) has emerged as a streamlined workflow to test for all relevant tumor biomarkers within a single assay. Despite this, there are still significant gaps in access to CGP testing, with many patients only tested for a subset of biomarkers or not tested at all. We assessed the clinical impact of wide deployment of an in-house CGP assay at a large health care system in the United States by analyzing a cohort of advanced-stage NSCLC patients who received CGP testing and a retrospective cohort that was tested with a prior-generation 50-gene assay (small panel). Seventy-seven percent of CGP-tested NSCLC patients had one or more tumor biomarkers that were actionable for a precision therapy compared to 63% of small panel-tested patients. CGP-tested patients with an actionable biomarker received appropriate precision therapies at a higher rate than small panel-tested patients (64% versus 50%, P < 0.001), and there were marked improvements in survival outcomes for patients tested with CGP [median overall survival (OS) 15.7 months versus 7 months, P < 0.0001]. These data demonstrate clear precision therapy selection and patient OS benefits from universal access to CGP testing. Despite this, not all patients with an actionable biomarker received a precision therapy, suggesting that there are still gaps between access to CGP testing and access to precision therapies.

13 April 2026
Verified
Conference Paper
Open access
JLJiachen Li
Jiachen Li
AHAmanda K Hall
Amanda K Hall
RZRuican Zhong
Ruican Zhong
SESelin S. Everett
Selin S. Everett
AUAlyssa Unell
Alyssa Unell
EHEric Horvitz
Eric Horvitz
Multidisciplinary tumor boards (MTBs) bring specialists together to identify therapies for complex cancer cases, but preparing for them is time-intensive. Clinicians must extract key details from extensive records and evaluate treatment options. While large language models (LLMs) show promise in medicine for basic tasks like summarizing notes, little is known about their role in high-stakes tasks like MTB preparation. We conducted a mixed-methods study with 16 oncologists using two AI systems to prepare patient cases for MTB: an off-the-shelf assistant (Copilot) and a task-specific multi-agent system (Healthcare Agent Orchestrator, HAO). We analyzed oncologist prompts, AI responses, and oncologists’ perception of AI. Participants showed greater willingness to adopt HAO but were often overconfident in AI summaries and skeptical of AI-recommended therapies. Trust calibration strategies, such as source links and agent-trajectories, failed to align trust with system capabilities. We conclude with how AI systems should be built to support clinicians in high-stakes tasks.