Prediction Tools for Early-Stage Prostate Cancer

Development of Models for Outcome Prediction in Early-Stage Prostare Cancer Using Molecular and Clinical Variables

Introduction

The natural history of localized prostate cancer (PCA) is enigmatic leading to significant controversies concerning proper management. Most early stage PCA are curable with local therapy, however many are relatively indolent such that some men do not require aggressive therapy. On the other hand, between 20 and 40 percent of men undergoing supposedly curative therapy for early stage disease, and incurring significant morbidity, will nonetheless relapse. Therefore, physicians and patients face the difficult task of determining the clinical significance of an individual patient's disease and selecting the most appropriate of multiple potential therapies.

A critical challenge is to develop means to distinguish indolent cancers from those that are potentially lethal so that therapeutic procedures can be tailored to an individual patient. Recent findings hold promise that a comprehensive molecular assessment of cancers affords the greatest opportunity for improved, clinically meaningful classification. In previous studies, we have determined that the molecular features of PCA can improve outcome prediction accuracy. In this proposal, we aim to continue our program to develop, refine, and evaluate the clinical utility of methods combining molecular and clinical variables for accurate prediction of outcome in patients with early stage PCA. The ultimate goal is to provide patients and clinicians with the most accurate outcome prediction possible, allowing them to make informed decisions for primary and adjuvant treatment.

Research in the Initial Four Years of the SPORE, Future Aims

In the first four years of SPORE funding, we have identified and abstracted clinical data and follow-up for 2,380 men in Great Britain that followed a course of watchful waiting. Based on this data, we developed an outcome prediction nomogram model for the probability that a man will survive his PCA for 120 months if he does not treat it immediately. The model accurately predicts 120 month survival with a concordance index of 0.70. We also collected tissue blocks from 922 of these patients and prepared multi-tissue arrays. In prior work, we have developed predictive nomograms for patients treated by radical prostatectomy and also prepared multi tissue blocks from 1,051 Memorial Sloan-Kettering patients with prostatectomy and long term follow-up. We are currently using these data and tissue resources for large scale evaluation of molecular biomarkers that have potential for outcome prediction and to add prognostic precision to the nomogram models. These analyses will be statistically sound and provide prognostic information that is therapy specific.

[Enlarge Image] Expression levels for the 5 most informative genes predictive of outcomes in prostatectomy cases

We have also shown that models of outcome prediction can be developed based on molecular variables and that a combined model of molecular and clinical variables provides a significant improvement in accuracy. However, it is apparent that optimal model development will require large sample size and benefit from the use of clinically significant endpoints. The next step in further development of the predictive model will involve analysis of a large consecutive series of well annotated localized PCAs using high throughput molecular characterization that includes genome-wide transcript profiling, comprehensive gene copy number analysis, and sequence analysis of PCA associated genes. Predictive models that use clinical variables alone, molecular variables alone and models combining both clinical and molecular variables will be developed. The best performing predictive algorithms will be determined based on accuracy and concordance index in training and test set evaluation.

Specifically, this project aims to:

1. Conduct a large scale evaluation of promising molecular biomarkers that are potentially useful for outcome prediction using routinely processed prostate cancer tissue samples from both conservatively treated British watchful waiting patients and aggressively treated Memorial Sloan-Kettering patients.
2. Develop an outcome nomogram for early stage prostate cancer based on comprehensive molecular analysis of a large series of well-annotated prostate cancer samples with clinically significant endpoints.
3. Develop robust assays for determination of the predictive signature in the clinical setting.
4. Test the final predictive algorithm in a wide spectrum of prostate cancer patients to evaluate accuracy and establish the general utility of the clinical test. 

Work on this project is strongly supported by Memorial Sloan-Kettering's CORE facility in Pathology.

Project Leaders

• Principle Investigator: William Gerald, MD, PhD
• Co-Principle Investigator: James Eastham, MD

Collaborators

• Victor Reuter, MD 
• Peter T. Scardino, MD 
• Georg Bartsch, MD (Innsbruck Medical University)
• Michael Ittman, MD, PhD (Baylor College of Medicine)
• Thomas Wheeler, MD (Baylor College of Medicine)
• Jack Cuzik, PhD (Cancer Research UK)
• Christopher Foster, PhD (Liverpool University)
• Michael Kattan, PhD (Cleveland Clinic Foundation)
• Richard Macchia, MD (SUNY Downtate Medical School)
• Gerald Soff, MD (Feinberg School of Medicine)