Nomograms play growing role in Pca outcomes

Nomograms are now available to predict the recurrence of prostate and renal cell cancer, among other conditions, and their applications may eventually expand. In this exclusive Urology Times interview, Michael W. Kattan, PhD, discusses the development of these prediction models and their current and future use. Dr. Kattan is currently chairman of the department of biostatistics and epidemiology at the Cleveland Clinic. Much of his work on nomograms was conducted at Memorial Sloan-Kettering Cancer Center, New York, where he was an outcomes research scientist. The interview was conducted by UT Editorial Consultant Robert C. Flanigan, MD, professor and chairman of the department of urology, Loyola University Medical Center, Maywood, IL.

Q Give us a basic understanding of what nomograms are.

A A nomogram is a graphical representation of a prediction model that attempts to take a continuous equation or formula and apply it to the bedside for an individual patient. It uses a points-based system, and it represents the formula graphically rather than in a table, to preserve accuracy.

In a nomogram, your actual PSA level determines your actual number of points, and that is why it's graphical-to preserve continuous variables in a continuous fashion and to preserve the accuracy associated with this. In essence, the goal is accurate prediction. The most accurate method used for prediction is probably a continuous formula or an equation, but that formula needs to be in a practical format; you can't calculate it in your head. A nomogram is a graphical device for that formula-either on paper or in computer software-for use as a prediction tool.

Q What types of nomograms are available for use in urology?

A Our prostate cancer tool features over a dozen prediction models, spanning all clinical states. We also have a renal cell cancer tool that predicts the probability of recurrence based on pathologic features.

Q How do you analyze the value of an individual nomogram, and what tests do you use to determine its reliability?

A The challenge when evaluating a nomogram is to determine how well you think it is going to perform or predict at the individual patient level, and how do you test and measure that.

Let's say a nomogram predicts the probability that a man will not have recurrent prostate cancer following surgery. To determine how well the nomogram works, I need to plug several new patients into it, obtain predictions for them, and then compare this to what actually happened. The problem is that, at the end of that process, I am going to have a prediction for each of these men; some will have recurred and some will not. Of those who did not, the follow-up may be very short for some. If I throw those men out, my proportions get skewed. But if I keep the failures and throw out those who did not fail, I am going to get lopsided odds at the end of the process. The difficulty lies in trying to measure accuracy with this issue of length of follow-up.

Q What can you do to straighten out this confusion?

A The concordance index is the best metric I know of, although it's not perfect. It works like this:

Take all these men that I just obtained predictions for and followed, and pair them up in all possible pairs of patients: man number one with man number two; number one with number three; two with three; and so on, until all possible pairs of men are accounted for.

Now, I am going to just look at those pairs of men in which the man with the shorter follow-up time hit the endpoint, namely PSA recurrence. So in a hypothetical pair of patients, I may have one who had recurrence at year 2, coupled with another guy who made it further than 2 years. In that pair of patients, I definitively know who did worse.