About

Dr. Ruth Etzioni is a biostatistician who primarily focuses on cancer screening and early detection. Much of her work is in the area of prostate and breast cancer, where she develops methods for evaluating diagnostic tests; creates mathematical models to reflect the impact of screening tests on the incidence and mortality rates of these cancers; calculates costs and benefits of preventive screening; tracks population trends with regard to screening and related behaviors and works with investigators on trial design and analysis. Dr. Etzioni also researches overdiagnoses associated with certain screening tests — when screening finds cancers that would not cause symptoms or death within a patient’s natural lifetime. She also evaluates novel cancer biomarkers and tracks patterns and outcomes of cancer care. Dr. Etzioni leads the biostatistics core for the National Cancer Institute-funded multicenter Northwest Prostate Cancer Specialized Program of Research Excellence, or SPORE, and she has a longstanding interest in researching, tracking and working to eliminate health disparities.

Research topics

• Medicine
• Gynecology
• Internal medicine
• Pathology
• Oncology
• Medical physics
• Environmental health
• Demography
• Obstetrics
• Gerontology
• Family medicine
• Economic growth

Selected publications

• TrajOnco: a multi-agent framework for temporal reasoning over longitudinal EHR for multi-cancer early detection

  arXiv (Cornell University) · 2026-04-12

  preprintOpen access

  Accurate estimation of cancer risk from longitudinal electronic health records (EHRs) could support earlier detection and improved care, but modeling such complex patient trajectories remains challenging. We present TrajOnco, a training-free, multi-agent large language model (LLM) framework designed for scalable multi-cancer early detection. Using a chain-of-agents architecture with long-term memory, TrajOnco performs temporal reasoning over sequential clinical events to generate patient-level summaries, evidence-linked rationales, and predicted risk scores. We evaluated TrajOnco on de-identified Truveta EHR data across 15 cancer types using matched case-control cohorts, predicting risk of cancer diagnosis at 1 year. In zero-shot evaluation, TrajOnco achieved AUROCs of 0.64-0.80, performing comparably to supervised machine learning in a lung cancer benchmark while demonstrating better temporal reasoning than single-agent LLMs. The multi-agent design also enabled effective temporal reasoning with smaller-capacity models such as GPT-4.1-mini. The fidelity of TrajOnco's output was validated through human evaluation. Furthermore, TrajOnco's interpretable reasoning outputs can be aggregated to reveal population-level risk patterns that align with established clinical knowledge. These findings highlight the potential of multi-agent LLMs to execute interpretable temporal reasoning over longitudinal EHRs, advancing both scalable multi-cancer early detection and clinical insight generation.

  PublisherDOI

• Measuring and defining screening benefit in a new era of cancer early detection

  JNCI Journal of the National Cancer Institute · 2026-03-27

  article1st authorCorresponding

  We are at a watershed moment in the history of early cancer detection in which many novel tests are poised to become available for population screening. An ongoing debate concerns how to properly evaluate these tests and specifically whether a shorter-term, incidence-based outcome might substitute for cancer mortality as an endpoint in randomized trials of screening test efficacy. An incidence-based endpoint promises to reduce time and resources, but there is no framework for how studies using this endpoint should report results and how they should be interpreted in terms of clinical utility. We consider whether publication of incidence-based results ahead of any mortality results could result in adoption of new screening tests ahead of reliable mortality results becoming available. We argue that guardrails are needed for this scenario, including standards for conduct and reporting of trials with incidence-based endpoints to assure valid interpretation of clinical utility. For example, information regarding the type and timing of tests used for diagnostic workup in screen and control groups will be needed. Clinicians and policy makers will need to determine acceptable measurements and magnitudes of this modified measure of test efficacy. The roles of incidence-based and mortality-based endpoints in determining practice standards will need to be defined, along with specifications for permissible adjunct evidence, such as modeling studies and real-world data. As screening trials for new multi-cancer tests will soon begin to report incidence-based results, resolution of these questions is a matter of urgency.

  PublisherDOI

• TrajOnco: a multi-agent framework for temporal reasoning over longitudinal EHR for multi-cancer early detection

  ArXiv.org · 2026-04-12

  articleOpen access

  Accurate estimation of cancer risk from longitudinal electronic health records (EHRs) could support earlier detection and improved care, but modeling such complex patient trajectories remains challenging. We present TrajOnco, a training-free, multi-agent large language model (LLM) framework designed for scalable multi-cancer early detection. Using a chain-of-agents architecture with long-term memory, TrajOnco performs temporal reasoning over sequential clinical events to generate patient-level summaries, evidence-linked rationales, and predicted risk scores. We evaluated TrajOnco on de-identified Truveta EHR data across 15 cancer types using matched case-control cohorts, predicting risk of cancer diagnosis at 1 year. In zero-shot evaluation, TrajOnco achieved AUROCs of 0.64-0.80, performing comparably to supervised machine learning in a lung cancer benchmark while demonstrating better temporal reasoning than single-agent LLMs. The multi-agent design also enabled effective temporal reasoning with smaller-capacity models such as GPT-4.1-mini. The fidelity of TrajOnco's output was validated through human evaluation. Furthermore, TrajOnco's interpretable reasoning outputs can be aggregated to reveal population-level risk patterns that align with established clinical knowledge. These findings highlight the potential of multi-agent LLMs to execute interpretable temporal reasoning over longitudinal EHRs, advancing both scalable multi-cancer early detection and clinical insight generation.

  PublisherOA PDF

• The Vanguard Study: Assessing feasibility of conducting randomized trials of novel multi-cancer detection testsShort title: The Vanguard Study: Assessing feasibility of randomized multi-cancer detection clinical trials

  SSRN Electronic Journal · 2026-01-01

  preprintOpen access
  PublisherDOI

• Supplementary Figure S5 from Estimating the Number of Men Living with Metastatic Prostate Cancer in the United States

  2025-11-26

  articleOpen accessSenior author

  <p>Supplementary Figure S5: Population distributions by race and age group for years 2000, 2010, 2020, and 2030.Projections for 2020 and 2030 are taken from US Census population projections based on the 2010 census.</p>

  PublisherDOI

• Supplementary Figure S3 from Estimating the Number of Men Living with Metastatic Prostate Cancer in the United States

  2025-11-26

  articleOpen accessSenior author

  <p>Supplementary Figure S3: Population projections by race between 1990 and 2030 for all ages. Projections from2019 onward are taken from US Census population projections based on the 2010 census.</p>

  PublisherDOI

• Supplementary Table S9 from Estimating the Number of Men Living with Metastatic Prostate Cancer in the United States

  2025-11-26

  articleOpen accessSenior author

  <p>This supplemental table compares the prevalence by duration under the alternative scenarios considered in the sensitivity analysis.</p>

  PublisherDOI

• Supplementary Figure S1 from Estimating the Number of Men Living with Metastatic Prostate Cancer in the United States

  2025-11-26

  articleOpen accessSenior author

  <p>Supplementary Figure S1: Prevalence over time for all races, assuming worse recurrent survival. Solid black linerepresents estimated number of new metastatic prostate cancer (MPC) cases; dashed blue line representsobserved number of prostate cancer (PC) deaths; dotted red line indicates projections after 2018.</p>

  PublisherDOI

• Cancer Screening Trials

  2025-12-01

  book-chapter1st authorCorresponding

  Screening trials are performed to determine the impact of early detection tests offered or administered to persons without symptoms or signs of cancer. These trials involve randomizing participants to a screening or a control (usually non-screening) group. Screening trials for biomarker-based tests constitute the fifth phase of the Early Detection Research Network's Phases of Biomarker Development, 1 a framework similar to the three phases of treatment trials. By virtue of their randomized designs, screening trials are analogous to Phase 3 treatment trials.

  PublisherDOI

• Supplementary Figure S4 from Estimating the Number of Men Living with Metastatic Prostate Cancer in the United States

  2025-11-26

  articleOpen accessSenior author

  <p>Supplementary Figure S4: Population projections by race and age group (40+) between 1990 and 2030.Projections from 2019 onward are taken from US Census population projections based on the 2010 census.</p>

  PublisherDOI

Recent grants

• NIH Grant R29CA070227

  NIH · $606k · 2002

• Developing Data-Driven Cancer Researchers

  NIH · $13.6M · 1980–2029

• NIH Grant R01CA131874

  NIH · $1.1M · 2013

• NIH Grant U01CA088160

  NIH · $2.7M · 2012

• PROMISS - Prostate Modeling to Identify Surveillance Strategies

  NIH · $2.9M · 2013–2019

Frequent coauthors

• Roman Gulati

  Fred Hutch Cancer Center

  463 shared

• Scott D. Ramsey

  University of Washington

  246 shared

• Timothy R. Church

  University of Minnesota

  224 shared

• Louise C. Walter

  University of California, San Francisco

  214 shared

• Robert A. Smith

  Heart Hospital Baylor Plano

  209 shared

• Christopher R. Flowers

  The University of Texas MD Anderson Cancer Center

  208 shared

• Ya‐Chen Tina Shih

  University of California, Los Angeles

  208 shared

• Kevin C. Oeffinger

  208 shared

Labs

• Plein Center for AgingPI

Education

• PhD, Statistics

  Carnegie Mellon University

  1990