Improving prognostic modeling of prostate cancer through a crowdsourcing competition
University of Turku researchers developed a winning solution for predicting overall survival of patients with metastatic castrate resistant prostate cancer (mCRPC), treated with first-line anti-mitotic chemotherapy. The primary benefit of this Challenge will be to establish new quantitative benchmarks for prognostic modeling in mCRPC, with a potential impact for clinical decision making and ultimately improving the design of future clinical trials.
The joint FIMM-UTU team headed by Prof. Tero Aittokallio participated in the Prostate Cancer DREAM Challenge, which is one of the community-based crowd-sourcing competitions for highly demanding scientific problems.
The participating teams were asked to submit predictive models based on over 150 clinical variables from the comparator arms of four phase III clinical trials with over 2,000 mCRPC patients treated with first-line docetaxel chemotherapy. The comparator arm of the clinical trials was being used to evaluate the effectiveness of the therapy.
The winning team combined researchers with various expertise from University of Turku, Institute for Molecular Medicine Finland (FIMM), and Helsinki University Hospital. This multi-disciplinary team put their heads together and brought in their own expertise to improve the collaborative solution. The team will be given the opportunity to present its predictive model at the DREAM Conference in Philadelphia.
The best solution will be published in a top journal, as part of the Challenge overview paper.
– The datasets provided by the organizers involved many challenging factors, including integrating data from various sources, controlling potential batch effects, and choosing the most important features which may be partly non-overlapping between the studies, says joint UTU-FIMM doctoral student Teemu Daniel Laajala, who coordinated the multi-site team, and continues
– Obviously we are excited to have performed best in the challenge and hope that the mathematical models not only perform well in terms of a statistical metric, but also eventually provide an actual clinical benefit to the patients.
– From the machine leaning perspective, we often read papers that evaluate methods’ performance in simplified datasets, but without this kind of independent and open competitions, it is very difficult to understand the relative benefits and limitations of each method in practice. The DREAM challenges enable us to not only to compare our performance against other teams in a fully-blinded manner, but also to explore the good and bad choices the others have made and therefore learn the best practices in the field, explains Tapio Pahikkala from the Department of Information Technology.
Metastatic CRPC accounts for one third of all patients with metastatic disease. Although a number of options exist for treatment of mCRPC, the impact of these treatments has been modest in terms of overall or disease-specific mortality in the past 20 years. Innovative research approaches are therefore urgently needed for improving the treatment of patients with mCRPC. The top performing models have the potential to become standards in the field as winning models will be promoted for further vetting by the American Joint Committee on Cancer (AJCC).
More information:
>> The Prostate Cancer DREAM Challenge
>> The winning team
>> The DREAM Conference in Philadelphia
>> Press release by SAGE
EH
Image: Prostate Cancer DREAM Challenge