e-Science tools evaluate new population based prostate cancer screening tests
The prostate-specific antigen (PSA) test has a high rate of false positive results leading to unnecessary treatment with life-long complications such as impotence or incontinence.
Early results from the Cancer Risk Prediction Center (CRisP) led the Stockholm County Council to fund STHLM3 to collect orders of magnitude more data; the trial recruits 50,000 healthy men aged 50-69 years to assess the test characteristics of a panel of other biomarkers and calculate cost-effectiveness.
STHLM3 will also assess re-screening protocols using a randomized design. However, it quickly became a challenge how to handle all data, how to perform image analysis, and in particular how to estimate effects of hypothetical screening scenarios. This led to SeRC establishing a project for “e-Science for Prevention and Control” to develop tools for this type of research, in particular STHLM3.
The group identified candidate markers and used machine learning to develop the prediction algorithm at the heart of STHLM3. The SeRC team provided input to the design of the trial, including the selection of the paired design and re-screening protocols. We also developed a “microsimulation” framework to compare screening scenarios (Fig. 1). This involves the simulation of many individuals, including their cancer history, and screening. It brings together methods for discrete event simulation, random numbers, and reporting that we used to implement the prostate cancer model. The simulations are independent, so hundreds of cores can be used to simulate millions of individuals. This established e-Science as a core part of cancer research at KI, with SeRC researchers participating in many different projects. The focus of the team has been to plan and carry out data integration, modeling, inference using quantitative tools, and enable the use of computational infrastructure.
Unifying clinical medicine, preclinical research, informatics, and computer science
The e-Science team includes epidemiologists, bioinformaticians, image analysts and computer scientists working on cancer screening. STHLM3 is an even broader collaboration that also involves the County Council, clinicians (oncologists, urologists, pathologists), the biomedical industry, biobank facilities, and laboratories. The CRisP steering group has representatives from KI and physicians from the Karolinska Hospital. SeRC is represented by SeRC Faculty Keith Humphreys, and two SeRC members are STHLM3 investigators. The collaboration was organized with meetings in the steering group at least ever 3 months, and additional frequent project meetings between the principal investigator and the SeRC team to refine study design and analysis. The steering group specified requirements, with the SeRC project
leaders Martin Eklund and Mark Clements managing execution.
SeRC has contributed over 2MSEK for the e-Science research of STHLM3, while the County Council and CRisP covered the remainder. Overall, SeRC has invested a total of 11 MSEK in e- Science health research, which has made it possible to (i) develop tools for the design of STHLM3, (ii) develop image analysis and statistical methods to predict breast cancer risk, (iii) simulate HPV vaccination and cervical cancer, and (iv) investigate computational methods to integrate health data. There has also been support from the Swedish Research Council and KI.
A SeRC student is the first KI PhD to have e-Science in his thesis title, focused on simulating cost-effectiveness of tests. He is supervised by a SeRC Faculty member, the PI for the STHLM3 trial and the PDC Director. SeRC members are working with colleagues from other universities to extend a course on simulation methods for population health, given for the first time in 2015.
The STHLM3 trial, in combination with e-Science and a unique set of Swedish data resources, will ascertain whether Stockholm County Council should move to organized prostate cancer screening. For the selected biomarkers, we had an industrial partner (Thermo Fisher Scientific) develop a custom chip, and SeRC researchers developed the algorithm and quality assurance methodology. The research has a strong operational focus: how can we change clinical practice to reduce costs from high levels of prostate biopsies, while reducing the possible harms from over-diagnosis and maintaining the mortality benefit from screening?
Complex clinical trials
STHLM3 has broad interest in the international scientific community. If cost-effective, the study will change international clinical practice. The baseline results are due for publication in March 2015. Tentative results from the pilot study suggest that the biomarker panel will reduce the number of prostate biopsies by 20% and will be cost-effective. The integration of e-Science in complex clinical trial design has already had large local impact, and the medical e-Science component (including simulation of outcomes, Fig. 2) is developing into an international strength at KI.
Short-term and long-term impact
One challenge was reaching common definitions; although our research is data- intensive, we do not always work with “Big Data”, and the e-Science aspects include new computational methods even for moderate-sized data. A second challenge was the tension between pragmatism and scientific advances. Clinicians were looking for rapid answers to technical questions, while the best answers required new developments that were less timely. One sign of the collaboration being successful was that the clinicians understood, and enthusiastically supported, that e-Science was essential to the success of STHLM3.
Eeles RA, et al. (SeRC investigators Wiklund F, Grönberg H). Identification of 23 new prostate cancer susceptibility loci using the iCOGS custom genotyping array. Nature Genetics. 2013 Apr;45(4):385-91. doi: 10.1038/ng.2560
Nordström T, Aly M, Clements MS, Weibull CE, Adolfsson J, Grönberg H. Prostate-specific antigen (PSA) testing is prevalent and increasing in Stockholm County, Sweden, Despite no recommendations for PSA screening: results from a population-based study, 2003-2011. European Urology. 2013 Mar;63(3):419-25. doi: 10.1016/j.eururo.2012.10.001
Karlsson R, Aly M, Clements M, Zheng L, Adolfsson J, Xu J, Grönberg H, Wiklund F. A population-based assessment of germline HOXB13 G84E mutation and prostate cancer risk. European Urology. 2014 Jan;65(1):169-76. doi: 10.1016/j.eururo.2012.07.027
Software: microsimulation package for R/C++. GPL-3. Available at GitHub.