2 Clarke Drive
Suite 100
Cranbury, NJ 08512
© 2024 MJH Life Sciences™ and OncLive - Clinical Oncology News, Cancer Expert Insights. All rights reserved.
With more than 240,000 new cases expected in the United States in 2012, prostate cancer is among the nation's most common tumor types.
Scanning electron color-enhanced image depicts prostate cancer cells.
Photo courtesy Annie Cavanagh/Cell Image Library
1
With more than 240,000 new cases expected in the United States in 2012, prostate cancer is among the nation’s most common tumor types. And yet despite its prevalence, treatment decisions remain fraught with controversy. From solo oncology practices to technology-filled cancer centers, physicians and their patients continue to grapple with wrenching and fateful decisions over the disease’s management.
A 2009 National Cancer Institute (NCI) study2 showing that annual screening for prostate cancer led to more diagnoses but no fewer deaths starkly illustrates the conflicts. Some patients benefit little from the surgery, radiation, and other treatments prompted by their diagnoses, the NCI noted, while suffering debilitating adverse events such as incontinence and impotence.
Faced with these high-stakes choices, a growing number of physicians are turning to predictive tools called nomograms to help them gauge more precisely how a patient’s cancer is likely to progress and to better assess the chance of a cure through a given treatment.
James Mohler, MD
Derived from institutional databases that include clinical and outcomes data for anywhere from hundreds to thousands of patients, these formulas consider multiple factors such as prostate-specific antigen (PSA) levels and Gleason scores—weighted to reflect their relative contribution to a particular outcome—to calculate a patient’s pathologic stage, chance of recurrence, and likelihood of a cure, among other results.
“We use nomograms as an important part of the education of every patient,” said James Mohler, MD, associate director and senior vice president for Translational Research and chairman of the Department of Urology at Roswell Park Cancer Institute in Buffalo, New York. “They really help men assess their pathologic stage and cure rate.”
Now widely available on the Internet, nomograms enable patients weighing treatment options to plug in their own clinical data and receive the probability of an outcome, such as the likelihood that their cancer is organ-confined.
The widening use of nomograms, however, has drawn the scrutiny of some skeptics, who question their predictive power and urge patients and physicians to closely examine the data from which they are derived. Those who embrace them insist that their ability to harness the predictive power of data represents a huge advance over anecdotal evidence.
Alan W. Partin, MD, PhD
Alan W. Partin, MD, PhD, director of the Brady Urological Institute at Johns Hopkins Medicine in Baltimore, Maryland, and a pioneer in the field, recounted his patients’ hunger for more precise assessments of their cancer and their frustration over the lack of it.
“When we would talk about risk with patients, they would ask us, ‘How good is good? Are we talking about 10% good or 90% good?’ “ recalled Partin, who 2 decades ago began to examine information from a large group of men “with similar blood tests, exams, and pathology reports” who had undergone radical prostatectomies at Johns Hopkins “to see how the cancer had spread.”
In 1993, he and Patrick C. Walsh, MD, then the institute’s urologist-in-chief, developed probability tables known as Partin tables by correlating 3 key pieces of information—PSA level, Gleason score, and estimated clinical stage—with patients’ actual pathologic stage. The tables perform what the institute calls a “virtual surgery” by allowing physicians to predict what would be found if the prostate were removed surgically and examined by a pathologist.
Flow chart shows broad categories in The Partin Tables. (James Buchanan Brady Urological Institute website, http://urology.jhu.edu/prostate/partintables.php).
Updated periodically to reflect current data, and based on the results of more than 5000 men who underwent surgery at Johns Hopkins, they have been shown to predict with 95% accuracy a man’s likelihood of being cured by treatment.
“I use them with every patient. I tell patients that they have come in with three tests and that we will compare them to other people with similar data,” Partin said in an interview. “If options exist, that’s when we use nomograms, typically right after diagnosis, to determine how far the cancer has spread and how many years a patient will survive disease-free after a particular treatment. Nomograms are very useful in predicting when cancer is organ-confined and help us make decisions about whether to perform surgery, for example.”
He added, “The healthcare arena, and particularly the people who are paying for treatment, will likely rely on these metrics more and more to decrease the chance of unnecessary treatments being offered.”
Michael Kattan, PhD
Since the Partin tables were developed, researchers have devised predictive models that incorporate even more variables, while exploring more endpoints and outcomes, including the probability of treatment success for salvage radiation therapy for men who have experienced prostate cancer recurrence after a radical prostatectomy.
“What nomograms do is strive for more precision,” said Michael Kattan, PhD, chairman of the Department of Quantitative Health Sciences at the Cleveland Clinic in Ohio, who has developed some of the most widely used nomograms, including models that predict the probability that a cancer will be indolent.
Kattan said it was his own brush with oncology’s lack of predictive capability that inspired his interest.
“I was in a PhD program in business school and was diagnosed with Hodgkin disease. As I faced questions about treatment—whether to choose radiation on top of chemotherapy—I looked for predictions to help me and found they were not there. In the world of business, predictions are everywhere, but in medicine, the field is still immature,” said Kattan.
In developing nomograms, he added, “I wanted to be able to plug in actual patient values.”
He accepted a postdoctoral fellowship in 1993 in medical informatics at Baylor College of Medicine in Houston, Texas, and later that year teamed up with Peter T. Scardino, MD, who is now the chairman of the Department of Surgery at Memorial Sloan-Kettering Cancer Center in New York City, to create new models.
“We’ve come a long way toward recognizing the need to incorporate individual patients’ preferences into decisions. The one-size-fits-all model in prostate cancer doesn’t work for individual patients,” Kattan said.
“
Nomograms are most helpful when the decision-making is most difficult.” ”
—Michael Kattan, PhD
“Nomograms are most helpful when the decisionmaking is most difficult, when there are trade-offs, for example, between quality of life and quantity of life or when comparing one side effect to another,” he said. “They are not that useful if there is only one treatment for a disease.”
He noted that in prostate cancer alone there are “a couple hundred now and they predict further into the future and incorporate new biomarkers. Data such as a man’s outcome two years after treatment is one refinement. Endpoints have been added that go beyond just recurrence, such as quality of life, including the effects of toxicity.”
Andrew K. Lee, MD, MPH
Some oncologists view nomograms with caution, however, particularly as they have proliferated.
“They’re very seductive. One can go to a Website and plug in a few numbers and a percentage regarding outcomes comes out. It seems so simple and accurate, but there is no way to predict with 100% certainty how well a given patient will do. For example, a more advancedstage patient may do well despite the low percentages given by the nomogram,” said Andrew K. Lee, MD, MPH, associate professor in the Department of Radiation Oncology at the University of Texas MD Anderson Cancer Center in Houston and director of its Proton Therapy Center.
Some nomograms, he added, have wide confidence intervals that undermine their accuracy.
“Say the nomogram predicts a risk of 76%, that number could actually have a range of 65% to 89%,” Lee said. “These models may also perform differently in terms of where within the nomogram a patient falls. The variables you put in for your patient may turn out to be a constellation of numbers that never existed in that data set. The nomogram may work well for more typical numbers, but may not perform as well with outliers. It tries to make a prediction by extrapolating from the data it has.”
In the Journal of the National Comprehensive Cancer Network, Lee described a nomogram for brachytherapy that had underperformed.3
He said in an interview that he was more broadly concerned that the models are based on “retrospective data that are prone to selection biases we can’t always discern.”
“We cannot know, for example, why a given patient opted for surgery over radiation. There may have been other factors that influenced the choice. Unknowns like that can taint retrospective data. It’s not the same as running a randomized trial, which accounts for known and unknown confounding variables,” Lee said.
Lee recommended that oncologists examine nomograms in the context of their clinical experiences. “A physician may not have the time to validate them, but they should be tested for how well they perform within one’s own practice,” he said.
Micrograph shows Gleason pattern of a prostatic acinar adenocarcinoma.
Kattan, Mohler, and others agree that nomograms are only as good as their databases.
“We must be careful about nomogram quality. If confidence intervals are wide, such as in unusual clinical situations where there are not many people like you, or when nomograms are constructed from a small database, they don’t function as well and may not be accurate enough to really provide much help to an individual patient,” observed Mohler, who chairs the National Comprehensive Cancer Network (NCCN) Prostate Cancer Treatment Guidelines Panel.
The data, Kattan said, must be carefully collected.
“Say you need to plug in the patient’s presurgery PSA value, but there are several in the chart collected over time. If you’re doing this in a hurry, you may just pick one of them and it may not be the right one,” he said. “Also, you need to follow patients after surgery, but they don’t always come back.”
Kattan acknowledged that the availability of nomograms has complicated treatment discussions for some physicians.
“The proliferation of nomograms has put pressure on doctors. It’s their decision whether to trust them or not. Sometimes publications are linked to them and so the data can be evaluated,” he said, adding, “Not all community oncologists use them. Some do and some don’t. It makes visits longer and adds variables to the decision making.”
But nomograms become more reliable as data collection, performed mostly by high-volume institutions, researchers who organize clinical trials, and occasionally pharmaceutical companies, continues to improve as interest in risk assessment grows, he said.
“The desire to predict more outcomes will demand more and better collection of data,” said Kattan. “What will probably help us do a better job are electronic health records that allow us to collect data more efficiently. If we use them wisely and meaningfully, these databases are useful for research as well as patient care. We need to change the way we think. Data collection is the biggest hurdle, not the lack of wizard prediction models or supercomputers.”
The National Comprehensive Cancer Network recommends that the risk stratification scheme in its guidelines be used to gauge a patient's risk level. Nomograms should be used to "provide additional and more individualized information."
Here are several of the better known nomograms used in prostate cancer treatment:
Description A
Description B
Website
Cleveland Clinic
7 prostate cancer nomograms are among 19 calculators for oncologic and other diseases
http://www.lerner.ccf.org/qhs/risk_calculator
James Buchanan Brady Urological Institute at Johns Hopkins Medicine
The Partin Tables and the Han Tables are available
http://urology.jhu.edu/prostate/partintables.php
http://urology.jhu.edu/prostate/hanTables.php
Memorial Sloan- Kettering Cancer Center
7 calculators for prostate cancer patients, including a tool that analyzes average male life expectancy, are offered
http://nomograms.mskcc.org/Prostate/index.aspx
He added, however, “It takes a lot of time and money to do this accurately, and it’s hard to get funding for it. These funds are not coming from the government or insurance companies, so the task is falling on the backs of academically oriented researchers. It’s an unfunded mandate.”
Partin has already updated his own tables 2 times. “The disease has evolved because of early detection and so the probability distributions have changed,” he explained.
Mohler noted that the first NCCN prostate cancer treatment guideline, published in 1997, referenced the Partin tables in the manuscript section. By 2003, their use was common, and many academic researchers thought they should be added to the guidelines to encourage and facilitate ease of use. The guidelines now reference 11 nomograms, but strongly encourage the use of only the Partin and Kattan nomograms at this time, he said.
“We include nomograms in guidelines if the data have been collected well,” said Mohler. “Typically, data sets are from individual institutes’ experience. They are tested and evaluated elsewhere. Memorial Sloan- Kettering has historical and current versions of its nomograms. The historical nomogram has been vetted by many institutions. When I counsel patients, I give the current and historical versions. We’re running both through the Roswell Park population to test how well they apply to our particular patients.”
Institutions test a nomogram’s accuracy by running their database through it to see how well it predicts the outcomes of their patients.
“If it doesn’t do well, you test other available nomograms or develop your own,” Mohler said, adding, “I use radiation nomograms more casually because they’re not as mature as some others. I’m just starting to use nomograms for salvage radiation therapy. We want to deliver our patients more information but these nomograms have not been vetted by as many institutions, and we all know that we need better databases that will in turn result in better nomograms.”
While urging caution, Lee notes that the careful use of data and nomograms represents an advance over “medicine that relied in the past on personal experience and anecdotes.”
“If you are able to look at the experience of patients with the same factors within the same frame of reference, it is a way to start a conversation with a patient. But I’d never make a clinical decision solely on the basis of a nomogram,” he said.
“I start with a basic risk stratification schema and add in my own enhancements that we have studied in our own data sets,” Lee continued. “For example, we may look at other factors such as the prior PSA pattern, the findings from imaging studies, more detailed findings from biopsies, the age of the patient, and their overall health. While some nomograms incorporate some of these elements, the overall interpretation of these factors and how they are used to make treatment recommendations still requires careful consideration.”
Mohler noted that discussions of clinical treatment must begin with a consideration of life expectancy and then be balanced with “some more objective form of analysis that considers the threat to that life expectancy posed by the cancer as well as the chance of cure and the risks one accepts for the chance of cure.”
“One could argue that patients and the healthcare system would all benefit from the availability of more and better objective predictions of treatment results,” he said.
“Are insurers using guidelines?,” he added. “I don’t see it happening yet. And the policy of Medicare and the federal government is that guidelines are guidelines, and not rules.”