Crowdsourced model could help select prostate cancer trial candidates

Doctor patient
Model could also assist in the design of clinical trials

A predictive model for prostate cancer can help clinicians decide whether patients should receive standard treatment or be offered a chance to enter a clinical trial.

The data-mining approach—described in the journal Lancet Oncology—is the result of a remarkable 'crowdsourcing' initiative which brought together insights from 550 international researchers and drew on 50 computational models for predicting prognosis in patients with metastatic castration-resistant prostate cancer.

Senior author James Costello of the University of Colorado Cancer Centre said the project embraces the call in Vice President Joe Biden's Cancer Moonshot initiative for greater collaboration and openness in cancer R&D.

"Scientists like me who mine open data have been called 'research parasites'," said Costello, who insists that leveraging existing data to gain new insights is a very important part of modern biomedical research. "This project shows the power of the parasites," he added.

Aside from potentially improving patient recruitment into trials—a perennial problem holding back clinical research—the predictive model could also assist in the design of clinical trials by avoiding variation in trial populations, say the authors. That means clinical trial questions will be able to be answered more rapidly because smaller sample sizes will be needed.

The project was overseen as a collaborative effort between 16 institutions, with teams challenged to link clinical measurements to patient survival based on data from five published clinical trials.  

Among the findings was that in addition to established predictive biomarkers such as prostate-specific antigen (PSA) and lactate dehydrogenase (LDH), blood levels of an enzyme called asparate aminotransferease (AST)—an indirect measure of liver function—is an important predictor of survival. A key element was to explore which interactions between measurements were most predictive.

The most successful of the 50 models was submitted by a team led by Tero Aittokallio of the Institute for Molecular Medicine Finland (FIMM), and was found to significantly outperform all other methods, including a reference model widely cited in the literature.

One of the limitations of the model is that it relies on data from published trials that represented the standards of care when the initiative first got underway in 2010, according to the researchers, who point out it would be even more powerful if data on newer drugs were included.

The intention now is to make the model publicly accessible through an online tool with an eye towards clinical application, and according to the scientists the National Cancer Institute (NCI) has already contracted the winning team to set that up.

"This study has shown the benefits of open data access at a time when clinicians, researchers, and the public are advocating for improved platforms and policies that encourage sharing of clinical trial data," they conclude.