In this article, the authors have done an excellent job in reviewing
recent findings regarding prostate-
specific antigen (PSA) and other methods for the early detection of prostate cancer. This is a fast-moving field, with new results being reported on a weekly basis. Indeed, it is an exciting time to be conducting research in prostate cancer. At the same time, however, it is far too easy to lose sight of some of the basic principles by which we should judge evidence to make research or clinical decisions. Specifically, there are hard-learned epidemiologic lessons about which we need to constantly remind ourselves.
Source Population Variables
The first of these is to keep in mind the patient population to which we are trying to draw inference. Results can vary from study to study depending on the source population. As a consequence, comparing results from studies with markedly different source populations may be likened to comparing apples and oranges. For example, studies of the characteristics of serum PSA determinations conducted in urology practices can provide important inferences about similar populations. Nevertheless, it is likely that the proportion of men with benign prostatic hyperplasia (BPH) or prostatitis would be greater in these practices than in the general population. Because serum PSA levels are elevated with either of these noncancerous conditions, the distribution of serum PSA levels would be different in studies based on these practices than in the general population. As a consequence, the specificity of this test for any given serum PSA level would most likely be much worse in a urology practice than in a screening program conducted in the general population. Thus, the results of any study evaluating the diagnostic characteristics of serum PSA determi nations, free PSA determinations, PSA messenger RNA, human kallikrein 2, or any other marker must take into account the appropriate target population.
The second point I wish to raise concerns the predictive power of a positive or negative test. The predictive power is a function of not only the diagnostic characteristics of the test (sensitivity and specificity) but also the prevalence of the undiagnosed condition (prostate cancer) in the source population. Consequently, for a given sensitivity and specificity, the predictive value of a positive test will be greater in populations with a higher prevalence or lower in populations with a lower prevalence.
This becomes more complicated when the diagnostic characteristics of a test change as a function of the population setting. As noted above, the specificity of serum PSA determinations decreases in a urology practice when those determinations are compared to the general population. Likewise, the prevalence of prostate cancer will differ between these two groups. Consequently, judgmental statements about levels of predictive value of a positive test become extremely difficult to interpret without knowledge of the prevalence of the disease in the target population.
This latter aspect was highlighted in our recent publication describing the test characteristics of serum PSA determinations in the community setting. In this paper, we discuss the trade-off in sensitivity and specificity for serum PSA determinations in the diagnosis of prostate cancer in population-based cases of prostate cancer from Olmsted County vs controls from the Olmsted County Study of Urinary Symptoms and Health Status among Men. The receiver-operator-receiver-operator-characteristic (ROC) curve that summarizes the predictive power of the test demonstrated an area under the curve of 0.94. This is much greater than the area under the curve for practice-based studies (Figure 1). This shows why an evaluation of PSA based on a urology practice may indicate a need for better markers; whereas from a population perspective, total serum PSA levels are powerful indicators in the diagnosis of prostate cancer in the community setting. In fact, there is little room left for improvement with the additional knowledge gained by measuring free PSA. Unfortunately, many of the studies that have reported the ROC curves for the addition of free PSA did not provide a summary curve that included all the information (a negative test for total PSA levels less than 2.5 ng/mL or positive test for serum PSA level greater than 10 ng/mL) given by the testing cascade. Nevertheless, these results suggest that we already have extremely powerful tools for detecting prostate cancer in the population.
Finally, while it is clear that we can detect prostate cancer, it is not so clear that we should be detecting all cases of prostate cancer. Some of those opposed to general screening point to the disparity between incidence and mortality rates. For example, for 1997, the American Cancer Society projects that 334,500 men will be diagnosed with prostate cancer and 41,800 will die from the disease. Those opposed to screening point to the known probability of side effects in some proportion of the men for whom no treatment should have been undertaken. Consequently, it is disappointing to see that efforts are still being directed towards finding all prostate cancers. Rather, it seems that it would be more important to identify those men with prostate cancer in whom the lack of treatment would result in death or a decrement in quality of life and not identify those for whom treatment is unnecessary or futile.
In summary, these are exciting times in prostate cancer research. The pace at which we are accruing knowledge is fantastic. We do, however, need to remain mindful of how to interpret and use this information as it becomes available.