Anna Lokshin

PhD

University of Pittsburgh

Professor

PLCO (Learn more about this study)

PLCO-395

Sep 4, 2018

Time to Event Prediction Based on Risk Calculation from Biomarkers

Pancreatic cancer (PC) is an extremely aggressive malignancy with 5-year overall survival (OS) rate of <8%. While identification of patients at a resectable stage (i.e. stage I and II) results in 4-fold increase in the 5-year survival rate, early diagnosis is hindered by the asymptomatic nature of PC. Similarly, ovarian cancer (OC) is often a silent disease, showing no obvious signs until late in its development. The 5-year survival rate for advanced cancer is only 15-20%, with most tumors ultimately becoming resistant to treatment. Generic symptoms, late presentation, poor assay sensitivity and specificity along with the lack of early stage cases are major barrier to development of biomarkers for PC. Further, there is lack of prognostic markers to follow up response to chemotherapy and surgery. Efforts have been made to develop methodologies for diagnostic tests, but there is also a need for biomarkers to enable improved prediction of OS pre-diagnosis as well as post-diagnosis following resection.
The PLCO set contains high-quality, prospectively collected, biomarkers, and provides an ideal platform for evaluating the potential of promising biomarkers for early detection and prediction of OS. The PLCO trial was a randomized multicenter trial in the U.S., which was aimed at evaluating the impact of early detection procedures for prostate, lung, colorectal, and ovarian cancer on disease-specific mortality. PC cases were identified by self-report in annual mail-in surveys, state cancer registries, death certificates, physician referrals, and reports from next of kin for deceased individuals, and verified by trained medical record specialists. Incident cases of primary adenocarcinoma of the exocrine pancreas (ICD codes C250– C259) were identified. In this study we aim to identify biomarkers that are predictive of OS separately in PC and OC.

Aim 1. Describe the distribution of biomarkers in PC and OC. Evaluate their utility individually in predicting OS using Cox regression and Kaplan Meier methods with log rank test to compare biomarker quartiles.

Aim 2. Build a predictive model of OS in PC and OC with Cox regression. Candidate biomarkers will be selected from aim 1. Resampling methods will be used to ensure model is not over-fit. Variable selection techniques will be explored, such as backward selection, lasso, and ridge regression.