Receiving a cancer diagnosis is scary and accompanied by many questions. Will it be treatable? What is the prognosis for survival? What are my options?
The answers a doctor gives to these questions are based on years of research studying cancer. But even given the knowledge physicians and cancer researchers have gathered over decades, predicting individual patient outcomes is a significant challenge.
One of the ways by which researchers study cancer is by studying large populations of patients with a specific type of cancer and evaluating whether any factors or characteristics within the population, such as tumor size or stage, are more likely than others to be associated with a particular disease outcome like survival.
While we’ve identified several factors based on these population-level studies that predict better survival, predicting whether an individual patient will or won’t respond to a certain therapy or predicting individual chances of survival is still very difficult. This is largely due to the incredible heterogeneity of cancer itself. By this I mean that every tumor is different; one might be more aggressive than another, or may respond more readily to available therapies. Predicting which tumors will behave in which certain ways is one of the major focuses of cancer research today.
For clinicians, the ability to predict survival and treatment outcomes is extraordinarily important. These kinds of predictions are developed by studying large numbers of people with and without a certain type of cancer. But how are researchers able to glean any meaningful information from what is certainly a staggering amount of data? In recent years, researchers have been using the power of computer modeling to do what we as people cannot – find important trends in large populations to understand how we might make prognostic predictions.
In a study published last week, David Thurtle and colleagues at the University of Cambridge show us how they can do just that.
Prostate cancer is the most common cancer in men, affecting 1 in every 9 men. Although a common cancer, only about 3% of men with prostate cancer will die of their disease. Treatment of prostate cancer can entail radical prostatectomy, or removal of all or part of the prostate. For men with less aggressive disease, a less aggressive treatment may be sufficient, allowing them to avoid radical treatment options that are associated with a number of risks, including incontinence and erectile dysfunction.
Therefore, predicting which patients are at risk of more aggressive disease is an important factor in deciding treatment options.
Depending on how far a tumor has progressed by the time it is discovered, and how aggressive or fast-growing it is, it may take years for it to grow and cause any symptoms. To test for cancer that may be present but not yet affecting our health, as we get older, it is recommended that we undergo cancer screening regularly. For prostate cancer screening, clinicians today use a blood test, which tests for elevated levels of a molecule called prostate-specific antigen, or PSA. Although there is no established normal PSA level, higher levels can indicate the presence of prostate cancer.
However, PSA screening is not perfect and can turn out false negative and false positive results. A false negative PSA result would show that PSA levels were normal in the presence of prostate cancer. In contrast, a false positive result would reveal elevated PSA levels in the absence of a prostate tumor. To confirm the PSA screening test, a clinician will recommend a biopsy, where a piece of the prostate is removed and examined for the presence of cancer.
Although the PSA test is a helpful tool, it has limited prognostic value, which means that it cannot determine whether a given prostate tumor is more aggressive than another. This in turn leads to uncertainty regarding the best treatment course.
Prostate cancer treatments can vary from conservative surveillance, often called “watchful waiting”, to radical prostatectomy, or removal of the prostate. The lack of the ability to predict the outcomes of individual cancers means that some men with aggressive cancer may not get the aggressive therapy they need, while men with less aggressive cancer may get a radical therapy they may not need. Although some prognostic prediction models for prostate cancer do exist, they are not easily applied at the individual level.
Therefore, a better prognostic method to predict which men would benefit from radical therapy and which would be better candidates for conservative observation is urgently needed.
PREDICT Prostate
In their latest study, Thurtle et al. set out to develop a model that could predict—on an individualized basis—overall survival outcomes and the most effective treatment options for newly diagnosed patients with non-metastatic disease. To glean clinically meaningful information from a large heterogeneous population, the authors used computational modeling, which involves using computers to build complex models to make clinical predictions about patient disease.
To develop their model, the group used a cohort of 7,063 men, 842 of whom had died from prostate cancer. They then validated the model to ensure it worked using a second group of 3,026 men, 360 of whom had died from prostate cancer. Finally, the authors measured the accuracy of the model on a third cohort of 2,546 men, 133 of whom died due to prostate cancer. The variables that were taken into account by the computer model were: age, PSA level, T-stage (estimated disease extent), histological grade (estimate of cancer “aggressiveness”), ethnicity, comorbidity and primary treatment type, information normally gleaned by clinicians during patient and disease assessment.
Using their model and these three population cohorts, the researchers found that their model, termed PREDICT Prostate, could predict prostate cancer-specific outcomes and mortality using the disease variables described above. Using statistical and computational methods, the model assessed the relative risks associated with these variables, and with this information generated a prognostic prediction for individual patients.
For example, during the 10-year follow-up time in the third population cohort used for validation of the model, the authors found that PREDICT Prostate predicted 89 deaths due to prostate cancer, 236 deaths due to causes other than prostate cancer, and 325 deaths overall. The actual death rate in this cohort during this time was: 105 due to prostate cancer, 225 due to other causes, and 330 overall.
Although the model was unable to measure treatment changes after 12 months and had a relatively small validation cohort, PREDICT Prostate is a promising new method that can be used to inform treatment decisions for patients with newly diagnosed disease, which is one of the biggest challenges that clinicians face today. Importantly, PREDICT Prostate is easily accessible on a web interface.
According to the authors, “the model incorporates variables available for almost any man diagnosed around the world and has wide potential applications in informing patient, clinician, and multidisciplinary team decision-making to reduce both over- and undertreatment”.
With the help of computer modeling and cloud computing, we may now have the power to develop more informative ways to evaluate large populations of people and establish more innovative and reliable ways to predict disease outcomes, allowing clinicians to have better answers to the scary questions.
So what can we do about prostate cancer?
Get screened!
For prostate cancer, the American Cancer Society recommends PSA screening at age 50 for men of average risk, age 45 for men of high risk, and age 40 for men at even higher risk who have one or more first-degree relatives with prostate cancer.
And stay tuned! Here on What’s the Deal, I will be running a cancer screening series, discussing ways we can protect ourselves from cancer by being screened regularly.
