investigator_user investigator user funding collaborators pending menu bell message arrow_up arrow_down filter layers globe marker add arrow close download edit facebook info linkedin minus plus save share search sort twitter remove user-plus user-minus
  • Project leads
  • Collaborators

Prospective Validation of a Multi-Marker Prostate Cancer Prediction Model

Hans Lilja

5 Collaborator(s)

Funding source

National Cancer Institute (NIH)
Approximately one million biopsies for prostate cancer are conducted each year in the US. The majority are unnecessary: the most common reason for a prostate biopsy is an elevated level of prostate-specific antigen (PSA) in the blood, but most men with elevated PSA do not have prostate cancer. In seven separate studies, involving over 7500 men and 2250 cancers, we have shown that a statistical model based on measuring isoforms of PSA, and kallikrein-related peptidase 2 (hK2), is a highly accurate predictor of prostate biopsy outcome in men with elevated PSA. In our primary study, the area-under-the-curve of the model was applied to an independent validation set was 0.76, far higher than PSA alone (0.64). We have also conducted decision analyses demonstrating that use of the statistical model to determine referral for prostate biopsy would reduce the number of unnecessary biopsies by about half, but miss only a small number of cancers, almost all of which would be the sort of low grade and stage cancers typically thought to constitute overdiagnosis. All of our prior studies were retrospectively conducted on European populations using frozen archived samples analyzed in a single research laboratory. In this proposal, we will first seek to evaluate the statistical model when applied retrospectively to a US cohort. We will then test whether independent clinical laboratories can measure the panel of four kallikreins accurately using control samples. We will then go on to prospectively collect research blood from patients before a scheduled biopsy. This sample will be analyzed locally, in real time, although the scheduled biopsy will continue irrespective of marker results, with biopsy outcome compared with the prediction from the statistical model. Finally, we will explore how implementation of the model would affect clinical practice using decision-analytic simulation and a vignette study.

Related projects