In:
Cancer Research, American Association for Cancer Research (AACR), Vol. 77, No. 13_Supplement ( 2017-07-01), p. 4725-4725
Abstract:
Introduction: Multiple myeloma (MM) is a cancer of the plasma cells in the bone marrow, and its clinical course depends on a complex interplay of clinical traits and molecular characteristics of the plasma cells. Since risk-adapted therapy is becoming standard of care, there is an urgent need for a precise risk stratification model to assist in therapeutic decision-making. While progress has been made, there remains a significant opportunity to improve patient stratification to optimize treatment and to develop new therapies for high-risk patients. To accelerate the development and evaluation of such risk models in MM, we formed a DREAM Challenge, a crowd-sourced competition that engages large cross-disciplinary teams of experts to address complex problems in biomedicine. Methods and Data: In collaboration with Multiple Myeloma Genome Project (MGP), clinical variables, patient outcomes, genetic, and gene expression data from thousands of samples were curated and harmonized from multiple public and private studies. In preparation for the challenge, a team of data scientists was assembled to evaluate this data, benchmark public high-risk models, and assess established prediction metrics with regard to progression-free survival (PFS) and overall survival (OS), the clinical endpoints of the challenge. Docker containers will be used to validate submitted models on private data that would otherwise not be available and to facilitate the transition of the best performing predictive signature to a clinical application. The MM DREAM challenge is accessible at: synapse.org. Results: The international staging system (ISS) for myeloma was used as a baseline classifier for high-risk patients (PFS & lt; 18mo). We evaluated published high-risk signatures - UAMS-5, UAMS-17, UAMS-70, and EMC92 - as benchmarks and observed that they consistently outperformed the baseline ISS predictor. High-risk prediction scores from these models were moderately correlated, suggesting published classifiers capture non-overlapping determinants of risk. Development of de novo classifiers by our team integrating clinical and molecular data highlighted opportunities for model refinement and supports rationalization of a crowd-sourced challenge to advance the field. Conclusion: Preliminary analysis of the challenge data suggests there is an opportunity to significantly improve risk stratification models in MM. In addition to the robust benchmarking of existing classifiers, we anticipate new, more accurate models will be proposed through a MM challenge given the scale of the combined data sets. We hope to uncover novel clinical and molecular traits that may yield insight into the pathology of MM and provide direction for follow-up studies. Importantly, this challenge will illustrate the advantages of leveraging public data and crowdsourcing to address therapeutically relevant questions in oncology. In addition, this challenge establishes a community resource for future research and benchmarking of novel classifiers. Citation Format: Michael Mason, Michael Amatangelo, Daniel Auclair, Doug Bassett, Hongyue Dai, Andrew Dervan, Erin Flynt, Hartmut Goldschmidt, Dirk Hose, Konstantinos Mavrommatis, Gareth Morgan, Nikhil Munshi, Alex Ratushny, Dan Rozelle, Mehmet Samur, Frank Schmitz, Ken Shain, Anjan Thakurta, Fadi Towfic, Matthew Trotter, Brian Walker, Brian S. White, Thomas Yu, Justin Guinney. Multiple Myeloma DREAM Challenge: A crowd-sourced challenge to improve identification of high-risk patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 4725. doi:10.1158/1538-7445.AM2017-4725
Type of Medium:
Online Resource
ISSN:
0008-5472
,
1538-7445
DOI:
10.1158/1538-7445.AM2017-4725
Language:
English
Publisher:
American Association for Cancer Research (AACR)
Publication Date:
2017
detail.hit.zdb_id:
2036785-5
detail.hit.zdb_id:
1432-1
detail.hit.zdb_id:
410466-3
Permalink