Contributions to
Scientific Knowledge.
A collection of journal publications, conference abstracts, and scientific posters documenting collaborative contributions across clinical research, evidence generation, oncology, data, and technology.
AI-Assisted Automated Abstraction for Enhanced Patient Insights in Gastrointestinal Cancers
Comparison of AI-assisted, human-validated abstraction with manual clinical data abstraction for gastrointestinal cancers.
Journal of Clinical Oncology · 2026 · Abstract 843
Publication analysis and contribution context
Research Purpose
The conference abstract examines whether an AI-assisted extraction workflow, reviewed by clinical experts, can generate structured clinical information from gastrointestinal oncology records with performance comparable to traditional manual abstraction.
Methodology and Approach
The reported framework included a gastrointestinal cancer data dictionary covering 13 event types in six clinical categories. A reference cohort of 50 patients was assessed using manually abstracted datasets and LLM-supported extraction with human validation; discrepancies were subsequently adjudicated.
Findings and Significance
The authors reported high agreement across selected clinical fields, with an average concordance of approximately 90% before additional review. The findings support further evaluation of hybrid automation and clinical oversight, but a 50-patient comparison should not be interpreted as evidence of universal performance across diseases, populations, or production settings.
My Contribution
Credited as a co-author of this 2026 ASCO GI conference abstract.
A Comparative Benchmarking of Large Language Models for Clinical Data Abstraction From Real-World Oncology Records
Evaluation of different large language models for extracting oncology information from unstructured patient records.
ESMO Real World Data & Digital Oncology · 2025 · Abstract 308P
Publication analysis and contribution context
Research Purpose
The study addresses a practical problem in real-world oncology research: converting narrative clinical records into reliable, structured research variables at scale without assuming that any one language model performs best for all clinical concepts.
Methodology and Approach
The abstract describes a retrieval-augmented generation approach with custom prompts and structured validation across 13 clinical event types. It used 50 manually curated oncology records and compared model performance using accuracy and agreement measures alongside cost and processing time.
Findings and Significance
The authors found that relative performance varied across abstraction categories. This cautions against selecting an AI model based solely on an overall score and points to the value of clinically specific evaluation, fit-for-purpose acceptance criteria, and continued human accountability.
My Contribution
Credited as a co-author of this 2025 ESMO conference abstract.
Real-World Testing Patterns of Circulating Tumor DNA in Early-Stage Triple-Negative Breast Cancer: A US Nationwide Database Study
A nationwide real-world analysis examining ctDNA testing patterns among patients with early-stage triple-negative breast cancer.
ESMO Open · 2025 · Abstract 115P
Publication analysis and contribution context
Research Purpose
The abstract considers how circulating tumor DNA testing is being used in routine oncology care for patients diagnosed with early-stage triple-negative breast cancer.
Methodology and Approach
Using a nationwide real-world data source, the investigators described testing patterns in a defined cohort of patients with stage I–III disease and multiple ctDNA assessments.
Findings and Significance
Real-world testing-pattern studies can help explain observed clinical use and inform study feasibility, data requirements, and interpretation of longitudinal biomarker information. Descriptive practice patterns do not by themselves establish that a testing strategy improves patient outcomes.
My Contribution
Credited as a co-author of this 2025 ESMO Breast Cancer Congress abstract.
Validation of a SNP-Based Prenatal Cell-Free DNA Screening for Fetal RhD Status in a Large U.S. Cohort
Scientific poster examining clinical validation of noninvasive prenatal cell-free DNA testing for fetal RhD status.
ACOG Annual Clinical & Scientific Meeting · May 2024 · Poster IP05-F
Publication analysis and contribution context
Research Purpose
The poster explores the performance of a prenatal single-nucleotide polymorphism–based cell-free DNA screening approach to determine fetal RhD status in a U.S. cohort.
Methodology and Approach
The meeting abstract describes a clinical validation using samples associated with prior SNP-based cell-free DNA testing and available RhD reference information. The specific poster methods and performance estimates should be interpreted from the original meeting record.
Findings and Significance
Accurate noninvasive identification of fetal RhD status may help inform decisions about RhD-related prenatal management. This conference poster is a separate scholarly work; it is not presented as authorship of a later journal article with a similar subject.
My Contribution
Credited as a co-author of the 2024 ACOG scientific poster.
Development of CancerLinQ, a Health Information Learning Platform From Multiple Electronic Health Record Systems to Support Improved Quality of Care
Development of a multi-EHR oncology learning-health platform that standardizes data for quality improvement and research.
JCO Clinical Cancer Informatics · 2020 · 4:929–937
Publication analysis and contribution context
Research Purpose
This full journal article describes the development of CancerLinQ, an oncology learning-health information platform intended to use routinely generated electronic health record information to support care quality and secondary research.
Methodology and Approach
The authors outline how data from different electronic record systems are standardized into a common representation and how selected unstructured clinical content is abstracted. The paper reported participation from 63 organizations using nine EHR systems as of March 2020, comprising over 1.4 million patients with a primary cancer diagnosis.
Findings and Significance
The work illustrates both the opportunity and difficulty of transforming heterogeneous clinical records into usable data. Its implications extend to semantic standardization, source-data interpretation, privacy controls, and the need to assess data fitness for each research question.
My Contribution
Credited as a co-author. The journal’s published author-contributions statement also lists Raven Brothers under conception and design.
Vital Status Ascertainment in CancerLinQ Discovery: Improvement in Mortality Capture With a Supplemental Data Source
Assessment of whether supplemental external mortality information improves vital-status completeness in oncology real-world data.
Journal of Clinical Oncology · 2020 · Abstract 7064
Publication analysis and contribution context
Research Purpose
Overall survival and mortality are central endpoints in oncology research. This meeting abstract focuses on limitations in vital-status ascertainment from clinical data sources and considers whether linking supplemental mortality information improves outcome capture.
Methodology and Approach
The authors examined agreement between vital-status or death-date information in CancerLinQ Discovery and an external data source, including the implications for estimating overall survival.
Findings and Significance
The study illustrates why incomplete or inconsistently recorded outcomes can affect evidence interpretation. Combining sources may improve coverage, but linkage reliability, temporal completeness, provenance, and the validity of the endpoint definition must be considered.
My Contribution
Credited as a co-author of this 2020 ASCO Annual Meeting abstract.
No publications match your search. Try a different keyword or category.
Entries are classified by publication type. Conference abstracts and posters are distinct from full-length research articles; descriptions summarize the published work, and authorship is attributed without overstating individual study responsibilities.