Statistical Methods for Evaluating Predictive Cancer Biomarkers

Statistical Methods for Evaluating Predictive Cancer Biomarkers


Study Objective

This project develops statistical methods to measure the benefits of treatment and preventive interventions and variations in treatment benefits associated with molecular biomarkers (such as somatic mutations) in clinical and epidemiological studies.

Who is conducting the study?

This project is collaboration between Drs. Jaya Satagopan, Alexia Iasonos and Sean Devlin of the Biostatistics Service.


We refer to biomarkers associated with variation in treatment benefits as predictive biomarkers. This project has provided unique insights into the definition of statistical interactions for identifying predictive cancer biomarkers. Some key publications are listed below:

  • JM Satagopan, A Iasonos, Q Zhou (2015). Prognostic and predictive values and statistical interactions in the era of targeted treatment. Genetic Epidemiology, 39: 509-517. PMCID: 4784265.
  • A Iasonos, P Chapman, JM Satagopan (2016). Quantifying treatment benefit in molecular subgroups to assess a predictive biomarker. Clinical Cancer Research, 22: 2114-2120. PMCID: 48563220.
  • JM Satagopan, A Iasonos (2017). Measuring differential treatment benefit across marker specific subgroups. Contemporary Clinical Trials (In Press). PMID: 28254404.
  • SM Devlin, JM Satagopan (2016). Statistical Interactions from a Growth Curve Perspective. Human Heredity. 82 (1-2): 21-36. PMID: 2874310
  • JM Satagopan, A Iasonos, JG Kanik (2017). A reconstructed melanoma data set for evaluating differential treatment benefit according to biomarker subgroups. Data Brief, 12:667-675.  PMCID: PMC5435579


Since patient-level clinical data are not available from several phase III clinical trials of treatment and putative predictive biomarkers, this project used digital data extraction techniques to approximate patient-level data from three published cancer studies given below. These are approximate data to facilitate clinical research and should not be taken as original patient-level data to make conclusive statements about the role of treatment and biomarkers on cancer outcomes. Click here to download digitally extracted approximate data.


Computer programs written in the R programming language for digital data extraction and for measuring differential treatment benefit across biomarker subgroups are available here.

Project Members