Building a crowd-sourced challenge using clinical trial data.

Abstract

e16047 Background: Community-based competitions have proven effective at drawing large cross-disciplinary teams to solve complex problems in medicine; however, few competitions have been conducted using cancer clinical trial data due to privacy, legal concerns, and restricted availability of data. Project Data Sphere, LLC (PDS) and Sage Bionetworks/DREAM are launching “Prostate Cancer DREAM Challenge” (Challenge) using historical comparator arm clinical trial data from PDS, an independent initiative of the non-profit CEO Roundtable on Cancer. The goal is to improve understanding of disease progression and treatment toxicity in order to improve clinical management and patient outcomes. Methods: The control arms of four phase 3 clinical trial data sets in metastatic castrate-resistant prostate cancer (mCRPC) - over 2,000 patients - were curated to create training and validation data sets. A team of prostate cancer clinicians and scientists selected 2 Challenge questions: prediction of OS, and prediction of treatment discontinuation. Curation was followed by a “Dry Run”, a mini-simulation of the competition to assess data quality and evaluate baseline predictive models. Challenge organizers set up the legal and data governance framework to facilitate data hosting on Synapse, Sage Bionetworks’ data platform; and promoted the Challenge to a broad and diverse audience. Results: A framework for conducting a crowd-sourced competition using clinical trial data has been established, with an expected launch by Q1 2015. The “Dry Run” implemented the recently published Halabi mCRPC OS model (Halabi, et al., JCO, 2014) and demonstrated robust predictive performance in the data (tAUC = 0.70; P = 6e-07 log rank test). Exploratory (de novo) modeling revealed potential improvements to the model, supporting the concept of an open community Challenge to improve existing state-of-the-art clinical models. Conclusions: The Challenge serves as a proof of concept for a new research paradigm in which historical clinical trial data may be used to crowd-source solutions to clinical questions. By sharing insights from the Challenge we encourage innovative approaches to using such data.

Ähnliche Arbeiten