The Science Behind Cancer Prognoses

When Jon Matthews was diagnosed with mesothelioma, a particularly deadly form of asbestos-caused lung cancer, his prognosis was dire: he was told not to expect to survive more than nine months. Still alive a year and a half later, Matthews decided to make a bet with his bookie, at 50–1 odds, that he would live past the 25-month median survival for people with his diagnosis. He won that bet in 2008. Last month he won a second, similar bet for more than $8,000, and went back to his bookie with a third wager: that he would survive until at least the middle of 2010.

Stories like this play to an archetypal, against-all-odds medical narrative, and stoke the imagination. But despite the recent media chatter about well-known people living with dire cancer diagnoses—such as Apple CEO Steve Jobs, Sen. Ted Kennedy, and actor Patrick Swayze—there has been little discussion of the science behind prognoses.

At its core, a prognosis, which means "to know beforehand," is a prediction about how a disease will progress, and what impact different treatments will have on survival. The modern concept of a cancer prognosis is based on a simple premise: by knowing how a type of cancer has advanced in past patients, doctors are able to make predictions about the progression of the same disease in current patients.

In the first half of the 20th century, the collection of cancer data shifted from individual teams of surgeons to government agencies and other centralized organizations. Brenda Edwards, associate director of the Surveillance Research Program at the National Cancer Institute (NCI), says that the NCI has continuously collected cancer data from specific geographical regions of the country since 1973.

Every case of cancer diagnosed in the study areas, which currently cover 26 percent of the U.S. population, is anonymously reported to, and tracked by, the NCI cancer registry. From this information, statisticians are able to follow changing survival statistics over time, and the effectiveness of current and past treatments. They also provide oncologists with data for making prognoses. One of the limitations of this approach, says Edwards, is that gathering data takes time, and it's possible for accurate prognoses to lag behind the newest developments in treating, or even diagnosing, cancer.

A prognosis is only as good as the data that go into it. Andrew Vickers, an associate attending research methodologist at Memorial Sloan-Kettering Cancer Center, cautions that "huge numbers of decisions in cancer medicine"—often including prognoses—"are based upon a single variable, your stage, which is a crude risk categorization." Staging, which is primarily determined by the size and spread of tumors, is one of the most useful predictors of how a disease will progress, but doesn't cover all the factors that determine cancer survival. Those factors include, but are not limited to: histology, or grade, of the tumors (the characteristics of the cancer at the cellular level); how the tumors respond to hormones; whether the cancer is detected early or late; what treatments have already been attempted; and the age, general health, and lifestyle of the patient. It's important, says Vickers, for a patient to ask his oncologist what factors are being used to determine his prognosis.

Understanding the limitations of accurately grouping people for the purpose of making individual prognoses is only one of the pitfalls for a cancer patient. Even if cancer data are broken down into very specific subgroups, patients are confronted with making sense of the statistics.

Take median survival times, an example that the late historian and scientist Stephen Jay Gould wrote about more than two decades ago in an essay called "The Median Is Not the Message." Diagnosed with a type of cancer, peritoneal mesothelioma, that at the time had a median survival of eight months, Gould realized that instead of having eight months left to live, the median meant that 50 percent of patients survived eight months, and 50 percent lived longer—some significantly so.

Rather than be depressed about the long odds, Gould, who was comfortable with statistics, was aware that he might be among the small group of patients who lived well past the median survival. Buoyed by his optimism, he enrolled in an experimental treatment program—and lived for 20 more years. Though most of the evidence suggests that having a positive attitude does not correlate with longer survival, having a realistic expectation of what is to come may improve one's quality of life.

The flip side to underestimating survival is overestimating it, and Dr. Ethan Basch, a clinical oncologist at Memorial Sloan-Kettering Cancer Center, says that oncologists who focus on the upper expected limit of survival may be doing their patients a disservice. "The risk for an individual patient is that we give an overly optimistic prediction and he is surprised when it doesn't go well for him," says Basch.

The common issue here is that a useful prognosis can't be reduced to a single number. As Gould noted, variability is one of the biological adaptations coded for by evolution—resulting in a population where one person may respond favorably to a treatment that doesn't work for another. At the very least, it would be useful for doctors to talk with their patients about the probabilities associated with surviving over a range of times.

Oncologists can clarify this point by presenting more of the data behind a prognosis, and doing a better job explaining what the data mean. Edwards suggests that instead rattling off statistics, oncologists might show a graph of population-based survival times for a specific type of cancer, explain what percentage of patients live for how long, and discuss the potential treatments with the patient.

Basch says that he sometimes uses nomograms—computer models of survival, based on the broadest possible sets of data from cancer registries and clinical trials—a tool Vickers says more oncologists should be discussing with their patients. A patient or her doctor can enter the patient's information directly into a computer loaded with the nomogram to get individually tailored predictions about survival, the effectiveness of different treatments, and even the relative risks of treatment side effects.

While nomograms and other computer models will continue to improve the ability of oncologists to provide accurate and specific predictions, there is no substitute for communication between the patient and his medical team, says Basch. Knowing the potential outcome of a disease can have an impact on what type of treatment a patient decides on, what lifestyle choices a patient makes, and whether declining treatment might be the best use of one's time.

To some, the information contained in a prognosis causes too much anxiety, or creates what feels like an expectation of survival, says Basch. The biggest pitfall, then, may actually be the first a patient confronts: whether he wants to know his prognosis at all.