About

Constantine A. Gatsonis is the Henry Ledyard Goddard University Professor of Statistical Sciences at Brown University and serves as the Director of the Center for Biostatistics and Health Data Science. Educated at Princeton and Cornell, he joined the Brown faculty in 1995. Dr. Gatsonis is a leading authority on the evaluation of diagnostic and screening tests, with significant contributions to methods for medical technology assessment, health services, and outcomes research. He is a world leader in applying and synthesizing evidence on diagnostic tests in medicine and is currently developing methods for Comparative Effectiveness Research in diagnosis, prediction, and radiomics. His work includes extensive involvement in statistical methods for the evaluation of diagnostic tests and biomarkers, with a focus on ROC analysis, Bayesian methods, and study design in diagnostic test validation. Dr. Gatsonis has contributed to hierarchical regression models analyzing variations in healthcare utilization and outcomes, meta-analyses of diagnostic test accuracy, and variability in diagnostic performance among radiologists and institutions. He has held leadership roles in national committees, including chairing the Committee on Applied and Theoretical Statistics of the National Academies, and is a fellow of the American Statistical Association. Since 2016, he has served as a statistical consultant for the New England Journal of Medicine and has held editorial positions in prominent journals.

Research topics

• Medicine
• Internal medicine
• Psychology
• Nuclear medicine
• Neuroscience
• Radiology

Selected publications

• DuCTSNet: MA‑SAM based fully automated DCIS segmentation and radiomic phenotyping for patient risk stratification in the ECOG‑ACRIN E4112 trial

  2026-02-12

  article
  PublisherDOI

• Radiotherapy With a 12-Gene Expression Assay for Ductal Carcinoma In Situ

  JAMA Oncology · 2025-10-16

  articleOpen access

  Importance: Breast ductal carcinoma in situ (DCIS) requires personalized treatment given its variable natural history. This study reports the first prospective oncologic outcomes of radiotherapy decisions as guided by 12-gene molecular assay, the DCIS score (DS). Objective: To assess surgical outcomes following preoperative breast magnetic resonance imaging (MRI) in women with DCIS and estimate 5-year and 10-year ipsilateral breast event (IBE) rates in participants given DS-based postoperative radiotherapy recommendations after local excision (WLE). Design, Setting, and Participants: Women with screen-detected DCIS who were eligible for WLE were enrolled to a single-arm, multicenter trial conducted at 75 institutions within the Eastern Cooperative Oncology Group-American College of Radiology Imaging Network (March 2015 to April 2016) and received a preoperative breast MRI-guided surgical treatment. Those who underwent successful WLE were advised to omit radiotherapy for low DS (<39) and receive radiotherapy for intermediate/high DS (≥39). Those achieving WLE as final treatment and successful DS assay (n = 171) were included in a prespecified analysis. Participants were followed up every 6 months for DCIS or invasive IBE, and 5-year IBE rates were analyzed from July to November 2023. Intervention: DS-based postoperative radiotherapy recommendations. Main Outcomes and Measures: Five-year IBE rates, with 95% CIs. Results: Among the 339 participants, the mean (SD) age was 59.1 (10.1) years. A total of 171 women (50.4%) received WLE for pure DCIS with free surgical margins and had DS data available. A total of 159 (93.0%) adhered to DS-based radiotherapy recommendations; 7 of 82 patients (8.5%) with a low DS underwent radiotherapy, and 5 of 89 patients (5.6%) with an intermediate/high DS declined radiotherapy. Over median (range) follow-up of 5 (0.5-5.0) years, 8 of 171 women experienced IBEs (4.8%; 95% CI, 2.4%-9.4%). IBE rates were similar for participants with a low DS(5.1%; 95% CI, 1.9%-12.9%) and participants with an intermediate/high DS (4.5%; 95% CI, 1.7%-11.7%). Among the 159 women who had adhered to DS-based radiotherapy recommendations, IBE rates were similar for participants with a low DS (5.5%; 95% CI, 2.1%-14.1%) and intermediate/high DS (4.8%; 95% CI, 1.8%-12.3%). Conclusions and Relevance: The findings of this prespecified analysis of a clinical trial suggest that DS-guided radiotherapy post-WLE for DCIS shows markedly lower 5-year IBE rates (approximately 5%) for intermediate/high DS than previously reported data following WLE alone. Despite a limited sample size, these data potentially provide support for radiotherapy use in patients with intermediate/high DS, and omission when DS is low, although confirmatory studies are needed. Trial Registration: ClinicalTrials.gov Identifier: NCT02352883.

  PublisherOA PDFDOI

• Abstract P1-01-05: Conducting Ancillary Studies during an Active NCTN/NCORP Screening Trial – The TMIST (ECOG-ACRIN EA1151) Experience

  Clinical Cancer Research · 2025-06-13

  article

  Abstract Background: The primary aim of the Tomosynthesis Mammographic Imaging Screening Trial (TMIST) is to determine whether women randomly assigned to be screened through 3-5 rounds with tomosynthesis (TM) have fewer advanced cancers than the population screened with digital mammography (DM) over 3-8 years after entry. In addition, there are 15 secondary aims with data being collected in the areas of imaging assessment, medical physics, breast biology and pathology, long-term follow-up, and health care utilization. Women ages 45 to 74 are eligible to participate. The study will enroll 108,508 women. Participants may also volunteer to contribute blood and/or buccal smears to the TMIST biorepository. Approximately 70% of TMIST participants have agreed to do so. Because of the size of the TMIST study and vast amount of data to be collected, there is an opportunity for investigators to utilize TMIST data to support various research questions not covered in TMIST. Methods:The TMIST study team developed a process where investigators who would like access to the TMIST data can submit a concept while the trial is ongoing for access to data in a protected manner. The process starts with the project investigator reaching out to the TMIST study chair. If the study chair, lead statistician, and ECOG-ACRIN (EA) co-Principal Investigator think the project has promise; a timeline for when the project could take place (either (1) during the TMIST clinical trial or (2) after the end of the trial and publication of the primary paper) is developed. The next steps involve reviews by the TMIST Data Safety and Monitoring Board and the EA Executive Review Committee. All ancillary projects proposed will require an external funding plan and budget before the project moves out of concept review inside of EA. Once the project concept clears all required EA approvals it then goes to the National Cancer Institute (NCI) Division of Cancer Prevention (DCP) for their approval. NCI Central Institutional Review Board (CIRB) approval is also required for the project to start while the trial is still active but is not sought until funding has been received. Two projects have secured external funding, have completed the EA committees’ review processes, and have received NCI approval. One is a case control study assessing short-term breast cancer risk through image-based analysis of screening mammograms (Project PI: Jon Steingrimsson, PhD, Brown University). The second is a case control study to assess the impact of breast compression pressure versus force in screening mammography on the likelihood of developing interval breast cancers (Project PIs: Etta Pisano, MD and Aili Maki, PhD, University of Toronto). Both projects involve analysis of images where software is being applied to TMIST images on computer systems controlled by EA IT personnel. Both projects are expected to be completed in the next year. Two additional projects have been approved for grant submission through the process described above. The PreSCRiB study (PI: Elizabeth Burnside, MD MPH, U of Wisconsin) will utilize Machine Learning applied to TMIST and All of Us data, including genetics, mammograms, social determinates of health and other data to develop individualized screening strategies for women. The second project (PI; Marc Ryser, PhD, Duke University) will utilize TMIST data to validate an algorithm the investigators have developed to assess overdiagnosis. Another project that is in development and will likely be submitted for approval and funding in the next 6-9 months is a collaboration between TMIST and UK-based clinical trial PROSPECTS study teams to compare rates of all cancers and advanced cancers for annual, biennial, and 3-year screening, with analysis by age, race, ethnicity, breast density and other factors. The ongoing TMIST study, as of June 24, 2024, has enrolled 101,394 women. Total enrollment is expected by late 2024 or early 2025. Follow-up on enrolled participants is expected to end in early 2028. Citation Format: Etta Pisano, Constantine Gatsonis, Mitchell Schnall, Melissa Troester, Elodia Cole , Jean Cormack, Jon Steingrimsson, Ilana Gareen, Martin Yaffe, Laura Collins, Amarinthia Curtis, Ruth Carlos, Kathy Miller, Christopher Comstock. Conducting Ancillary Studies during an Active NCTN/NCORP Screening Trial – The TMIST (ECOG-ACRIN EA1151) Experience [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-01-05.

  PublisherDOI

• Statistics in Medicine — What’s in an Estimand?

  New England Journal of Medicine · 2025-12-17 · 3 citations

  article
  PublisherDOI

• MRI-based Radiomic Features for Risk Stratification of Ductal Carcinoma in Situ in a Multicenter Setting (ECOG-ACRIN E4112 Trial)

  Radiology · 2025-04-01 · 2 citations

  articleOpen access

  Models using clinical and MRI-based radiomic features identified from ductal carcinoma in situ lesions improved prediction of disease upstaging at surgery compared with standard clinical information alone.

  PublisherOA PDFDOI

• Erratum for: MRI-based Radiomic Features for Risk Stratification of Ductal Carcinoma in Situ in a Multicenter Setting (ECOG-ACRIN E4112 Trial)

  Radiology · 2025-04-01

  erratum
  PublisherDOI

• Tree-based methods for estimating heterogeneous model performance and model combining

  ArXiv.org · 2025-06-02

  preprintOpen access

  Model performance is frequently reported only for the overall population under consideration. However, due to heterogeneity, overall performance measures often do not accurately represent model performance within specific subgroups. We develop tree-based methods for the data-driven identification of subgroups with differential model performance, where splitting decisions are made to maximize heterogeneity in performance between subgroups. We extend these methods to tree ensembles, including both random forests and gradient boosting. Lastly, we illustrate how these ensembles can be used for model combination. We evaluate the methods through simulations and apply them to lung cancer screening data.

  PublisherOA PDFDOI

• Amyloid PET and clinical management in a diverse, cognitively impaired population: The New IDEAS Study

  Alzheimer s & Dementia · 2025-07-01 · 5 citations

  articleOpen access

  INTRODUCTION: The New Imaging Dementia-Evidence for Amyloid Scanning (IDEAS) study (NCT04426539) evaluated the association between amyloid positron emission tomography (PET) and changes in clinical management among ethnoracially diverse, clinically heterogeneous patients. METHODS: We assessed diagnosis and management plan before and 90 ± 30 days after amyloid PET among Medicare beneficiaries who met 2018 National Institute on Aging-Alzheimer's Association criteria for mild cognitive impairment (MCI) or dementia. We aimed to identify ≥ 30% change in a composite patient management endpoint (CPME; i.e., changes in Alzheimer's disease [AD]/non-AD medications, changes in counseling). RESULTS: Among 5757 participants (median age 75 years; 21.7% Black, 20.3% Latinx, 58.1% all other races/ethnicities [AORE]), a change in CPME occurred in 59.0% (95% confidence interval 57.6%-60.5%) of individuals post PET. Change varied by ethnoracial identity and type of clinical presentation: Black (MCI: 55.3%, dementia: 55.8%), Latinx (MCI: 53.7%, dementia: 61.9%), AORE (MCI: 62.0%, dementia: 58.3%), typical (MCI: 64.8%, dementia: 60.9%), atypical (MCI 45.5%, dementia: 53.6%). DISCUSSION: Amyloid PET is associated with clinical management among diverse, clinically heterogeneous populations. HIGHLIGHTS: Changes in management plan occurred in 59% of patients 90 days after amyloid positron emission tomography. Rates of change in management exceeded the pre-specified goal of > 30% across ethnoracial groups. Rates of change in management also exceeded > 30% among amnestic and non-amnestic Alzheimer's disease presentations.

  PublisherOA PDFDOI

• Data‐driven analysis of 10,361 amyloid‐PET scans from the IDEAS study reveals two primary axes of variation

  Alzheimer s & Dementia · 2024-12-01

  articleOpen access

  Abstract Background The variability in the regional distribution of Aβ‐PET signal and its relation to clinical features is debated. We used data‐driven approaches to uncover heterogeneity in cortical Aβ‐PET signal from a large representative sample collected through the IDEAS study. Methods We analysed cross‐sectional Aβ‐PET collected from 10,361 patients with MCI or mild dementia scanned in 295 PET facilities using one of the 3 FDA‐approved tracers. Central image processing resulted in template‐space SUVR images (reference: whole cerebellum) and centiloid (CL) values. Spatial independent component analysis was used to decompose SUVR volumes into 40 independent components. After excluding noise components, participants’ scores were extracted for each of the remaining 11 grey matter (GM) components describing cortical and subcortical binding. K‐means clustering was used on these GM component scores to assign each participant to different Aβ‐PET clusters based on GM binding (Figure 1). Results Three informative clusters of PET binding were estimated. Cluster 1: Aβ‐(n=4729, CL mean=2±23) with low GM binding, and two Aβ+ clusters; Cluster 2(n=2484, CL mean=76±34) and Cluster 3(n=3148, CL mean=86±32). Subtracting average SUVR of Clusters 2 and 3 showed they differed along a posterior‐anterior gradient with Cluster 2 showing an occipital predominant pattern. Principal component analysis conducted on the GM scores confirmed two dominant axes of variation separated the clusters, a Aβ‐ to Aβ+ axis and, an anterior‐posterior axis (Figure 2). Statistically significant but weak differences were observed between the two Aβ+ Clusters (2 vs. 3); Visual Read (positive: 95% vs. 92%); Clinical Stage (dementia: 47% vs. 41%); Age (76.9±6.4 vs. 75.9±6.2), however, most clinical variables showed no differences (Figure 3a). 48 ADNI participants with Aβ‐PET and post‐mortem neuropathology data (11 Female, Age mean=79.7±7.4, PET‐Death mean=2.3±1.7years; Aβ‐CL mean=71.2±55.5; APOE4(0/1/2)=22/21/5; Diagnosis(CN/MCI/AD)=6/8/32) were applied to the model fit on IDEAS data. Qualitatively, no differences in neuropathology were observed between the two Aβ+ Clusters (Figure 3b). Conclusion Data driven classification of Aβ‐PET reveals two primary axes reflecting Aβ load and anterior‐posterior binding, with the later not clearly related to clinical or pathological variation. Future work will apply new data to this model and investigate if this spatial variation in Aβ‐PET is related to longitudinal changes in pathology.

  PublisherOA PDFDOI

• Clinical Impact of Amyloid PET in Diverse Medicare Beneficiaries with Cognitive Impairment: Preliminary Results from New IDEAS

  Alzheimer s & Dementia · 2024-12-01

  articleOpen access

  Abstract Background The Imaging Dementia—Evidence for Amyloid Scanning (IDEAS) study demonstrated that amyloid PET changes patient management in &gt;60% of Medicare beneficiaries with MCI/atypical dementia. IDEAS had limited racial/ethnic diversity and excluded patients with “typical” amnestic clinical presentations. Here we present preliminary results from the New IDEAS study, which evaluates the clinical impact of amyloid PET in a more racially, ethnically and clinically diverse cohort. Method Launched in December 2020, the New IDEAS study (NCT04426539) is recruiting Medicare beneficiaries with typical (amnestic) and atypical (non‐amnestic) MCI/dementia at 142 dementia specialty clinics across the U.S. Multifaceted, community‐engaged and culturally‐tailored strategies are implemented to enhance the diversity of the cohort. Based on self‐identified race/ethnicity, patients are enrolled into 1 of 3 cohorts: Black/African‐American (BAA), Latino/Hispanic (LAT), or not BAA or LAT (NBL, all other racial/ethnic identities). Patients undergo amyloid PET using an FDA‐approved tracer with local visual read at 127 imaging facilities. Changes in management are measured between the pre‐PET visit (intended management, assuming no access to amyloid PET) and 90‐day post‐PET visit (implemented management, incorporating amyloid PET results). A composite management endpoint captures changes in one or more of the following: AD drug therapy, non‐AD drug therapy, counseling about safety and future planning. Result Out of 5,209 registered participants, 3,328 (63.9%) participants completed amyloid PET, pre‐ and post‐PET visits at time of analysis (median age 74; 55.2% female; 22.6% BAA/17.5% LAT). Demographic and clinical features by cohort are shown in Table 1. Compared to NBL, BAA and LAT cohorts presented with greater clinical impairment, more frequent atypical clinical presentations and lower rates of amyloid PET positivity. Changes in the composite management endpoint occurred in 57.5% of all participants (57.2% MCI, 58.1% dementia), with similar rates across cohorts (Table 2). Following PET, anti‐amyloid antibodies were newly recommended in 7.0% of BAA, 7.4% of LAT and 9.0% of NBL. Changes in management were more frequent in typical than atypical clinical presentations (Table 3). Conclusion These preliminary findings highlight the clinical utility of amyloid PET in a diverse and representative sample of cognitively impaired patients recruited in real‐world practice.

  PublisherOA PDFDOI

Recent grants

• NIH Grant U01CA079778

  NIH · $49.5M · 2013

• ECOG-ACRIN Network Group Statistics and Data Management Center

  NIH · $151.8M · 2014–2026

• NIH Grant U01CA074696

  NIH · $1.3M · 2003

• NIH Grant U10CA079778

  NIH · $2.4M · 2014

Frequent coauthors

• Etta D. Pisano

  University of Pennsylvania

  160 shared

• Bradley S. Snyder

  Brown University

  145 shared

• Arnold M. Epstein

  Brigham and Women's Hospital

  106 shared

• Lucy Hanna

  Providence College

  95 shared

• Suddhasatta Acharyya

  Daiichi Sankyo (United States)

  94 shared

• Eric C. Schneider

  92 shared

• Martin J. Yaffe

  Ontario Institute for Cancer Research

  88 shared

• Mitchell D. Schnall

  University of Pennsylvania

  87 shared

Education

• Ph.D.

  Princeton

• M.S.

  Cornell

Awards & honors

• Fellow of the American Statistical Association