Objectifying Risk in Patients with Localized Prostate Cancer
Some three quarters of American men will have a PSA level checked at some point in their lives, and as many as 50% will undergo regular PSA screening.1
As such, the modern urologist is charged with managing the most commonly diagnosed male malignancy.2 Heterogeneity in the biology of prostate cancer (PCa) has led to much publicized controversy regarding benefits of not only PCa screening but also its treatment.3
Furthermore, the multitude of treatment options available to patients with localized PCa can make routine clinical decision-making extremely complex. As such, counseling of patients with localized prostate malignancy requires a thoughtful balance of risks.
Yet, in the absence of Level I evidence, many routine decisions remain qualitative and largely depend on patient and physician preferences. A quantitative and informed framework for these decisions is possible nevertheless with the help of a number of predictive models that have been published in the literature over the last decade. The models range from prediction of prostate biopsy results to post-treatment risk stratification.
Several excellent reviews recently have been published summarizing the extensive literature on predictive tools for patients with prostatic adenocarcinoma.4, 5 From the multitude of predictive models that have been published, a few standout and can be quite helpful in objectifying decision-making. These tools are highlighted below.
In a busy practice, clinical realities often make it prohibitive for a hassled physician to (a) find the time to search the literature for most relevant predictive tools and (b) to perform often tedious individualized risk-calculations off of printed nomograms.
Thus, many clinicians use clinical judgment, experience, or even “gestalt” to inform critical decision-making even with our most complex patients. At Fox Chase Cancer Center in Philadelphia, we recently launched a new web-based portal that helps lower the barriers to using existing predictive models in routine clinical practice. The website organizes and operationalizes over 40 interactive predictive models for genitourinary malignancies. The site allows quick and simple individualized point-of-care risk calculation for patients with prostate cancer using selected modern predictive models.
Predicting the Chances of Harboring Prostate Cancer
A large number of predictive models has been developed that can help guide patient counseling regarding outcomes of prostate needle biopsies. One of the most robust models was published by Nam et al.6 The model predicts the chance of finding malignancy and high-grade malignancy in patients undergoing prostate needle biopsy.
The nomogram is based on a sample of over 3,000 men who underwent prostate biopsy at the Sunnybrook Health Sciences Centre in Toronto. The model integrates covariates such as age, race, family history, American Urological Association (AUA) Symptom Score, PSA, total-PSA, and digital rectal examination (DRE). It can predict the likelihood of a patient harboring prostatic adenocarcinoma of any grade and also the chances of finding high-grade disease.6
The area under the curve (AUC)—or for some statistical methods the concordance index (c-index)—for a given model indicates that model's accuracy. An AUC of 0.5 indicates probability equivalent to that of a coin flip, while an AUC of 1.0 designates perfect predictive capabilities. The models in this nomogram possess an AUC of 0.74 for probability of finding any cancer and an AUC of 0.77 finding a high-grade cancer. Although these predictions are imperfect, they exceed the prognostic powers of PSA and DRE alone.6
A similar tool is available from data collected through the Prostate Cancer Prevention Trial7 and is available on a separate website. Another model published by Karakiewicz et al8 only uses age, DRE, PSA, and free PSA only if available.
This model can be quite handy in a busy clinic when the variables needed to execute a more complete model are not accessible. The AUC for this model is 0.69 when only PSA is used and rises to 0.77 when a free-PSA is integrated.8 The models are operationalized for easy use here.