Federated learning enables big data for rare cancer boundary detection

Nature Communications · 2022 · 326 citations

• Computer Science
• Computer Science
• Machine Learning

Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

Outcomes of Observation vs Stereotactic Ablative Radiation for Oligometastatic Prostate Cancer

JAMA Oncology · 2020 · 1120 citations

• Medicine
• Oncology
• Internal medicine

Importance: Complete metastatic ablation of oligometastatic prostate cancer may provide an alternative to early initiation of androgen deprivation therapy (ADT). Objective: To determine if stereotactic ablative radiotherapy (SABR) improves oncologic outcomes in men with oligometastatic prostate cancer. Design, Setting, and Participants: The Observation vs Stereotactic Ablative Radiation for Oligometastatic Prostate Cancer (ORIOLE) phase 2 randomized study accrued participants from 3 US radiation treatment facilities affiliated with a university hospital from May 2016 to March 2018 with a data cutoff date of May 20, 2019, for analysis. Of 80 men screened, 54 men with recurrent hormone-sensitive prostate cancer and 1 to 3 metastases detectable by conventional imaging who had not received ADT within 6 months of enrollment or 3 or more years total were randomized. Interventions: Patients were randomized in a 2:1 ratio to receive SABR or observation. Main Outcomes and Measures: The primary outcome was progression at 6 months by prostate-specific antigen level increase, progression detected by conventional imaging, symptomatic progression, ADT initiation for any reason, or death. Predefined secondary outcomes were toxic effects of SABR, local control at 6 months with SABR, progression-free survival, Brief Pain Inventory (Short Form)-measured quality of life, and concordance between conventional imaging and prostate-specific membrane antigen (PSMA)-targeted positron emission tomography in the identification of metastatic disease. Results: In the 54 men randomized, the median (range) age was 68 (61-70) years for patients allocated to SABR and 68 (64-76) years for those allocated to observation. Progression at 6 months occurred in 7 of 36 patients (19%) receiving SABR and 11 of 18 patients (61%) undergoing observation (P = .005). Treatment with SABR improved median progression-free survival (not reached vs 5.8 months; hazard ratio, 0.30; 95% CI, 0.11-0.81; P = .002). Total consolidation of PSMA radiotracer-avid disease decreased the risk of new lesions at 6 months (16% vs 63%; P = .006). No toxic effects of grade 3 or greater were observed. T-cell receptor sequencing identified significant increased clonotypic expansion following SABR and correlation between baseline clonality and progression with SABR only (0.082085 vs 0.026051; P = .03). Conclusions and Relevance: Treatment with SABR for oligometastatic prostate cancer improved outcomes and was enhanced by total consolidation of disease identified by PSMA-targeted positron emission tomography. SABR induced a systemic immune response, and baseline immune phenotype and tumor mutation status may predict the benefit from SABR. These results underline the importance of prospective randomized investigation of the oligometastatic state with integrated imaging and biological correlates. Trial Registration: ClinicalTrials.gov Identifier: NCT02680587.

Implementation of Germline Testing for Prostate Cancer: Philadelphia Prostate Cancer Consensus Conference 2019

Journal of Clinical Oncology · 2020 · 277 citations

• Medicine
• Family medicine
• Gynecology

PURPOSE: Germline testing (GT) is a central feature of prostate cancer (PCA) treatment, management, and hereditary cancer assessment. Critical needs include optimized multigene testing strategies that incorporate evolving genetic data, consistency in GT indications and management, and alternate genetic evaluation models that address the rising demand for genetic services. METHODS: A multidisciplinary consensus conference that included experts, stakeholders, and national organization leaders was convened in response to current practice challenges and to develop a genetic implementation framework. Evidence review informed questions using the modified Delphi model. The final framework included criteria with strong (> 75%) agreement (Recommend) or moderate (50% to 74%) agreement (Consider). RESULTS: , and mismatch repair carriers. Collaborative (point-of-care) evaluation models between health care and genetic providers was endorsed to address the genetic counseling shortage. The genetic evaluation framework included optimal pretest informed consent, post-test discussion, cascade testing, and technology-based approaches. CONCLUSION: This multidisciplinary, consensus-driven PCA genetic implementation framework provides novel guidance to clinicians and patients tailored to the precision era. Multiple research, education, and policy needs remain of importance.

Provider Engagement in Radiation Oncology Data Science: Workshop Report

JCO Clinical Cancer Informatics · 2020 · 4 citations

• Computer Science
• Data science
• Medical physics