Martin McIntosh, PhD – Fred Hutchinson Cancer Research Center

Beth Karlan, MD – Cedars Sinai Medical Center

Early detection, a key to improved outcomes for women with ovarian cancer, is a major focus of our research program.  This project provides a model for our general strategy in meeting translational research goals, which includes 1) exploitation of emerging molecular technologies to identify biologically relevant genes, proteins and antigens as candidate markers and targets for translation, 2) a systematic approach to prioritizing markers and targets for evaluation, 3) a collaborative approach to evaluating candidate markers and targets that includes evaluation of candidates identified by colleagues at other institutions, and 4) use of novel statistical methods to use markers to predict biologic phenotype.

In the early detection project we have progressed through the first 2 phases of the translational research process.  We are actively engaged in the third phase, and propose to complete the fourth over the next 5-years.

The primary objective of this project is to recommend a set of markers and an algorithm that can be used in a clinical trial for ovarian cancer screening to prevent ovarian cancer from escaping early detection.  Achieving this goal requires the development of a novel statistical model to identify promising markers, as well as the evaluation of several currently existing and novel markers using serum specimens from two currently funded studies for ovarian cancer screening.

To date, project investigators have successfully developed two algorithms that have been described in several publications (see list below).  The first, the Parametric Empirical Bayes (PEB) Method, is used to screen for cancer using markers measured over time.  The second combines theoretical and practical work to characterize (1) the optimal method to combine two markers, and (2) a practical way to estimate this combination.  Both algorithms have been employed to evaluate the performance of several ovarian cancer biomarker candidates.

Dr. McIntosh and colleagues have made great strides in analyzing marker data submitted by study collaborators.  Repository specimens from 160 POCRC participants were identified and sent to collaborators who measured respective markers in the specimens.  Collaborators who responded to the 2002 Project RFA and received Scientific Committee approval were recently sent sub-aliquots of 160 specimens from the POCRC Specimen Repository (Including sera/plasma from women with normal, benign, and malignant ovaries) for biomarker validation assays.  To date, 11 investigators have received POCRC RFA specimens, and 15 biomarkers have been measured.  In addition, the POCRC Repository contains serum/plasma that may be used for additional validation assays.

The most commonly used screening strategy for detecting ovarian cancer examines a woman’s CA 125 level (a tumor marker for ovarian cancer) but ignores her screening history when deciding the result of the screen. It has been shown that considering the screening history in the screening algorithm can improve sensitivity.  Because only about 50% of all early stage ovarian cancer patients have abnormal CA 125 levels at the time of diagnosis, only half of the population can even potentially benefit from its use in screening.

Many putative markers for ovarian and other cancers exist, most of which are based on detection of cancer-related proteins in serum, but none have the sensitivity or specificity necessary for application to widespread screening for ovarian cancer. Several researchers have shown that using many imperfect markers together can achieve better performance than using CA 125 alone, but these results were in the context of a diagnostic test, not screening.  However, even though screening and diagnostic tests have differences, the potential gain in screening when using multiple markers was clearly demonstrated.

The primary goal of our research is to combine these two approaches, and invent procedures that can perform screening with multiple markers in a longitudinal manner.  Although some methods for using a single marker cannot be extended to the multiple marker case, other tools of modern statistics/mathematics and computation can be adapted to this end (model based clustering and variable selection techniques in particular).

The development of the PEB screening algorithm will have practical applications in the clinical setting as it is simple to use, is highly robust, and can detect a wide range of tumor behaviors. It is also particularly useful when screening with a new marker whose behavior in the pre clinical period is not well known. In addition, the recommendation of a panel of markers for ovarian cancer that intends to prevent ovarian cancer from escaping early detection has the potential to contribute to a decline in the number of late stage ovarian cancers diagnosed.