|Home | About | Journals | Submit | Contact Us | Français|
The National Comprehensive Cancer Network and American Urological Association guidelines on early detection of prostate cancer recommend biopsy on the basis of high prostate-specific antigen (PSA) velocity, even in the absence of other indications such as an elevated PSA or a positive digital rectal exam (DRE).
To evaluate the current guideline, we compared the area under the curve of a multivariable model for prostate cancer including age, PSA, DRE, family history, and prior biopsy, with and without PSA velocity, in 5519 men undergoing biopsy, regardless of clinical indication, in the control arm of the Prostate Cancer Prevention Trial. We also evaluated the clinical implications of using PSA velocity cut points to determine biopsy in men with low PSA and negative DRE in terms of additional cancers found and unnecessary biopsies conducted. All statistical tests were two-sided.
Incorporation of PSA velocity led to a very small increase in area under the curve from 0.702 to 0.709. Improvements in predictive accuracy were smaller for the endpoints of high-grade cancer (Gleason score of 7 or greater) and clinically significant cancer (Epstein criteria). Biopsying men with high PSA velocity but no other indication would lead to a large number of additional biopsies, with close to one in seven men being biopsied. PSA cut points with a comparable specificity to PSA velocity cut points had a higher sensitivity (23% vs 19%), particularly for high-grade (41% vs 25%) and clinically significant (32% vs 22%) disease. These findings were robust to the method of calculating PSA velocity.
We found no evidence to support the recommendation that men with high PSA velocity should be biopsied in the absence of other indications; this measure should not be included in practice guidelines.
Some guidelines for early detection of prostate cancer recommend biopsy on the basis of high prostate-specific antigen (PSA) velocity (rate of change of PSA level), even in the absence of elevated PSA levels or positive digital rectal exam (DRE). However, the clinical value of PSA velocity for prostate cancer detection in men without other indications for biopsy is unknown.
The predictive value of PSA velocity, adjusting for age, PSA levels, DRE, family history, and prior biopsy, was assessed for 5519 men in the placebo arm of the Prostate Cancer Prevention Trial, in which men received an end-of-study biopsy regardless of clinical indication.
PSA velocity added little to the predictive accuracy of high PSA levels or positive DRE and would substantially increase the number of men recommended for a biopsy. The authors found no evidence to support prostate biopsy in men with high PSA velocity in the absence of other indications.
Implementation of PSA velocity as a guideline would be unlikely to improve patient outcomes and would lead to a large number of unnecessary biopsies.
There might be better methods to calculate PSA velocity that could improve predictive accuracy. The cohort was not followed until death and some differences may emerge on very long-term follow-up.
From the Editors
Practice guidelines produced by expert committees are generally presumed to constitute distillations of the best published evidence. There is good reason to believe, however, that many guidelines are not as evidence based as they might be. Burgers et al. (1) reviewed 32 oncology guidelines using a validated guidelines assessment instrument. The mean score given to the guidelines was only 42 of a total of 100 for “rigor of development.” The means for individual items in the rigor of development domain, for example, “systematic methods used to search for evidence,” “explicit link between the recommendations and the supporting evidence,” and “clear description of criteria used to select evidence,” were below 50% for all but one item (1).
Close examination of prostate cancer guidelines, such as those by the National Comprehensive Cancer Network (NCCN) (2), reveals that many aspects of these guidelines are based on evidence. The recent results of the European Randomized Trial of Prostate Cancer Screening (3) give qualified support for prostate cancer screening; there is also clear evidence supporting the use of prostate-specific antigen (PSA) (4) and free PSA (5) to identify cancer, the exclusion of younger (6) and older men (7,8) from screening, and differential recommendations for African Americans and those with a family history of prostate cancer (9).
However, one particular aspect of the guidelines stands out with respect to its evidentiary base, which is the use of PSA velocity. The NCCN guidelines state that men with a high PSA velocity (rate of change of PSA level)—greater than 0.35 ng mL−1 y−1—should consider biopsy even if absolute level of PSA is very low. The guidelines cite Carter et al. (10), who reported an association between PSA velocity and a diagnosis of fatal prostate cancer approximately 10–15 years subsequently. However, it is unclear why a marker that predicts aggressive prostate cancer many years in the future should be used to suggest immediate biopsy to patients. Moreover, PSA velocity was not demonstrated to add predictive accuracy to PSA alone. As such, the apparent predictive value of PSA velocity might simply reflect that PSA and PSA velocity are highly collinear—reported correlations are as high as 0.93 (11)—and that PSA itself is highly predictive of advanced prostate cancer (12). In addition, the cut point of 0.35 ng mL−1 y−1 appears to have been based on visual inspection of the receiver operating characteristics (ROC) curve: “[it] could be one reasonable choice—among others—to balance sensitivity and specificity for detection of life-threatening cancer” (10). PSA velocity is also included in the American Urological Association (AUA) PSA Best Practice Statement (13), albeit with a disclaimer regarding whether it adds predictive value to PSA alone (13). Similarly to the NCCN guidelines, the AUA states that a PSA velocity threshold of 0.4 ng mL−1 y−1 may improve prostate cancer detection for men with low PSA levels.
The problem with evaluating a recommendation to biopsy men with low PSA on the basis of a high PSA velocity is that of finding a suitable dataset, because only men with a high PSA are usually biopsied. The obvious exception is the Prostate Cancer Prevention Trial (PCPT), in which men received an end-of-study biopsy irrespective of PSA. Because men in the PCPT received regular PSA tests, excellent PSA data are available. As such, the PCPT provides a perfect test case for PSA velocity. In a previous report, we briefly described analyses of PSA velocity in the PCPT (9); herein we report an extensive set of analyses to investigate the clinical value of PSA velocity for prostate cancer detection in men otherwise without an indication for biopsy.
The methods and results of the PCPT have been previously published (14,15). In brief, men aged 55 years and older, with no previous prostate cancer diagnosis, normal digital rectal exam (DRE), and baseline PSA of 3.0 ng/mL or less were randomly assigned to finasteride or placebo for 7 years. Men were followed with yearly PSA tests, with biopsy recommended for men with PSA higher than 4.0 ng/mL (adjusted in the finasteride arm to take account of treatment-related reductions in PSA). After 7 years of therapy, all men who were not diagnosed with prostate cancer were asked to consent to an end-of-study biopsy. This dataset includes 5652 men from the placebo arm of the PCPT who underwent biopsy and had a PSA available and DRE performed within the year before the biopsy. There were missing data on one or more predictors for 133 men, leaving 5519 participants for analysis. For men who had more than one biopsy during the PCPT, only the last biopsy was used here.
Our initial set of hypotheses concerned the marginal predictive accuracy of PSA velocity. We first created a logistic regression model using standard predictors for prostate cancer on biopsy, including age (modeled continuously), log PSA at biopsy (modeled continuously), family history of prostate cancer (yes/no), DRE (yes/no), and whether the patient had a prior prostate biopsy (yes/no). We used ROC curve analyses and calculated the area under the ROC curve (AUC) using fourfold cross-validation. AUC is a measure of the predictive accuracy of a marker or model and is scaled from 0.5 (no better than chance) to 1 (a perfect predictor). We then compared the AUC value with that of a statistical model including these predictors plus PSA velocity, calculated by linear regression of time on all log PSA values before biopsy. We also compared a model with only log PSA with one with both log PSA and PSA velocity. On the grounds that PSA velocity might differentially improve the detection of more aggressive cancers, we repeated this analysis defining a biopsy as positive only if the Gleason score is 7–10. Because some critics have suggested that cancers found on the PCPT end-of-study biopsy were small and not clinically significant, we also modeled “clinically significant” prostate cancer as an outcome, based on the “Epstein” criteria (16). Cancer was defined as clinically significant if any of the following applied: clinical stage greater than T1c, PSA density at least 0.15 ng mL−1 g−1, Gleason score of at least 7, tumor in three or more cores, or at least one core with more than 50% cancer involvement. For the analysis of high-grade disease, low-grade tumors were categorized along with no cancer; comparably, in the analysis of clinically significant disease, cancers not meeting the criteria were placed in the same category as benign biopsy.
PSA velocity can be calculated in various ways, with differences as to the method of calculation (eg, log transformed or untransformed PSA values), the PSA values entered (eg, all values or just those within a certain time frame), and the patients eligible (eg, all patients or only those with a certain minimum number of PSA values over a certain minimum period of time). To be as comprehensive as possible, we repeated analyses with alternate methods of calculating PSA velocity.
As a direct evaluation of the NCCN and AUA guidelines, we evaluated the positive predictive value of various PSA velocity cut points in a subset of men with low PSA. To be comprehensive, we used three different PSA velocity cut points (0.35, 0.5, and 0.75 ng mL−1 y−1) and two different definitions of low PSA (<4 and <2.5 ng/mL). Given that PSA velocity is strongly associated with PSA values, and PSA is associated with cancer risk, we thought it likely that the risk of cancer in men with low PSA would be higher above particular PSA velocity thresholds than below. Accordingly, we chose some reasonable PSA cut points to compare with PSA velocity cut points. If a PSA cut point could provide comparable sensitivity and specificity to a PSA velocity cut point in the subset of men with low PSA, there would be little rationale for what can be a time-consuming and complex calculation of PSA velocity. For this analysis, we calculated PSA velocity using linear regression on untransformed PSA values 2 years before biopsy. Our rationale for restricting the analyses to the most recent PSA values was twofold. First, restricting the analyses in this way reflects the NCCN guidelines themselves, which state that “measurement should be made on at least three consecutive specimens drawn over at least an 18–24 month interval” and that “longer time periods … decrease predictive power” (2). A similar statement is made in the AUA guidelines. Second, if the period were not restricted, very few men would meet the criteria for high PSA velocity but low PSA. Almost 90% of the men in our study cohort had a biopsy after 7 years on study. To have a PSA velocity of 0.35 ng mL−1 y−1 or more for 7 years but a final PSA of at least 2.5 ng/mL, participants would have to have a starting PSA of 0.05 ng/mL, which is extremely rare in men of screening age. All statistical tests were two-sided. All analyses were conducted using SAS version 9.
Men with cancer were slightly older than those with negative biopsies, were more likely to be African American, and to have a family history of prostate cancer. They were also more likely to have been biopsied in an earlier round of screening (Table 1). Although most cancers had a Gleason score 6 or below (78%), only a minority (19%) met the Epstein criteria (16) for indolent disease (Table 2). Approximately one quarter of cancers could not be categorized by the Epstein criteria (16) because of inadequate pathological information. There is strong evidence of an association between PSA velocity and biopsy outcome (P < .001 by χ2).
When PSA velocity was added to a multivariable prediction model for prostate biopsy outcome, the strong associations seen on univariate analyses were reduced. Odds ratios were moderate, and only a minority of PSA definitions were independently predictive (Table 3). All analyses were repeated using raw PSA values (ie, without log transformation), but this led to decreases in predictive value.
There was little evidence that PSA velocity adds an important level of predictive accuracy to either standard predictors or to PSA alone (Table 4). Using linear regression on all log PSA values to calculate PSA velocity (other definitions of PSA led to smaller increments in AUC), the AUC for prediction of any cancer using the multivariable model increased by 0.007, from 0.702 to 0.709, when PSA velocity was added to the model. AUC increased by 0.010, from 0.682 to 0.692, comparing PSA alone to PSA plus PSA velocity (Table 4). The value of velocity was less for predicting clinically significant or high-grade cancers with increases in AUC of 0.005 and 0.001, respectively, compared with the standard predictors alone (Table 4).
When we investigated the guidelines on PSA velocity explicitly, we did see that PSA velocity predicted cancer in men without a conventional indication for biopsy, that is, men with low PSA and a normal DRE (Table 5). For example, the risk of cancer in men with a PSA velocity above 0.35 ng mL−1 y−1 is higher than in men with PSA velocity less than this cut point (21% vs 15%) (Table 5). However, superior risk stratification can be achieved simply by choosing a different PSA cut point, especially for the endpoints of high-grade cancer or clinically significant cancer. Following the NCCN guidelines and biopsying men with normal DRE and PSA less than 4 ng/mL if they had a PSA velocity above 0.35 ng mL−1 y−1 would lead to 115 additional cancers being identified but 433 unnecessary biopsies; a 2.5 ng/mL PSA threshold would result in a very similar number of unnecessary biopsies (n = 436) but would find 24 more cancers. In other words, the PSA threshold of 2.5 ng/mL has the same specificity as the PSA velocity threshold 0.35 ng mL−1 y−1 (87%) but a higher sensitivity (23% vs 19% for any cancer, 41% vs 25% for high-grade tumors, 32% vs 22% for clinically significant disease). Of particular note, use of PSA velocity criterion as per NCCN guidelines would lead to a large number of biopsies, with biopsy recommended in one in seven men without a conventional biopsy indication. Only a small fraction of men with very low PSA, 2.5 ng/mL or less, had a high PSA velocity, and there was no difference in risk of cancer in men above and below PSA velocity thresholds (Table 5).
These negative findings prompted us to conduct additional unplanned analyses. Our aim was to see whether methods of formulating changes in PSA in terms other than PSA velocity might have some clinical role in aiding decisions about prostate biopsy. We first examined a more recently published algorithm of PSA velocity, the risk count method (17), which involves determining the number of times that PSA velocity increases above a certain threshold during the course of a patient’s PSA history. Although there was some suggestion that patients who experienced several large rises in PSA had a lower chance of prostate cancer, most results failed to reach statistical significance after adjustment for PSA (Table 6). We also explored whether the percentage change in PSA might be of benefit. Results here were more promising. After adjusting for standard clinical predictors, men who had more than 50% increase in PSA in the year before biopsy, such as from 3 to 4.6 ng/mL, had a reduced risk of a positive biopsy (odds ratio = 0.53; 95% confidence interval = 0.34 to 0.84; P = .01). However, very few men met this criterion (170, 3%), limiting its clinical value.
We have analyzed perhaps the only dataset available that can evaluate the guidelines concerning PSA velocity. We found no evidence to support prostate biopsy in men with high PSA velocity in the absence of other indications, such as a positive DRE or high PSA. Overall, PSA velocity did not importantly add predictive accuracy to a standard predictive model or to just PSA alone and, more specifically, PSA velocity cut points had inferior risk separation compared with PSA cut points in men with low PSA and negative DRE. In other words, if a clinician feels that the current PSA thresholds are insufficiently sensitive, he or she would be better off identifying patients to biopsy by using low PSA thresholds than by adding PSA velocity as a criterion for biopsy.
Our findings appear to contradict a body of evidence apparently supporting the relationship between PSA velocity and prostate cancer, but this contradiction is more apparent than real. First, like other authors, we found strong evidence for a univariate association between PSA velocity and positive biopsy (P < .001). In general, this lead to higher risks of cancer above specific PSA velocity cut points than below. However, we also found that PSA velocity does not add important predictive value to PSA and other standard predictors; in other words, we did not find that the use of a PSA velocity criterion for biopsy would improve clinical decision making. To our knowledge, these questions have not been addressed by prior authors (18). That PSA velocity is strongly associated with biopsy outcome on univariate, but not multivariable, analysis, is easily explained by collinearity. The correlation between PSA and PSA velocity was close to 0.9 when analyzing all PSA values before biopsy, and it is naturally difficult for a marker to add value to a predictor with which it has a strong correlation.
Our findings are consistent with several other studies. Investigators from the Rotterdam center of the European Randomized trial of Screening for Prostate Cancer (ERSPC) have reported that PSA velocity does not help predict biopsy outcome (19) or clinically significant prostate cancer (20). Eggener et al. (21) used a more sophisticated modeling approach to find a very small increment in predictive accuracy associated with PSA velocity, largely as a result of a small number of men with very high PSA velocities and a reduced risk of cancer. This negative association was likely attributable to high PSA velocity being associated with benign inflammatory conditions (21). Eggener’s result was replicated almost exactly by our own recent study (22) of two ERSPC cohorts, in which we similarly found a very small increment in predictive accuracy associated with PSA velocity, again explained largely by a minority of men with reduced risk at high PSA velocities (22). Together with these prior reports, our findings suggest that men with a sudden large rise in PSA should be carefully evaluated for benign disease, possibly including a repeat PSA, before referral for prostate biopsy.
One possible argument against our conclusions might be that, although PSA velocity does not help find prostate cancer, it does help detect the aggressive cancers most likely to shorten a man’s life; that is, PSA velocity is of value for prognostication if not detection. Supporters of this argument might point to Carter’s findings (10) that a PSA velocity of 0.35 ng mL−1 y−1 is associated with aggressive cancers diagnosed many years later (10), or D’Amico's oft-cited finding (23) that a PSA velocity of 2.0 ng at the time of definitive treatment is a predictor of cancer-specific death (23). Yet, we found no evidence that PSA velocity helps to detect more aggressive cancers. Indeed, the small increment in predictive accuracy associated with PSA velocity was reduced when we restricted our analyses to high-grade cancers or those that met the Epstein definition of clinical significance.
Moreover, doubts can be raised about whether PSA velocity is indeed of value for prognostication. Carter (10) and D’Amico's (23) articles were based on a small number of events (20 and 27, respectively) and neither examined whether PSA velocity was of incremental benefit, for example, by comparing AUC with and without PSA velocity. Indeed, in a systematic review of the literature (18), we found a near complete absence of evidence that PSA velocity adds predictive value to PSA. Our own group has evaluated all 22 definitions of PSA velocity and PSA doubling time on a radical prostatectomy dataset (24). We found that most definitions did not add predictive accuracy to PSA alone for the end point of either recurrence or metastasis; a small number of definitions added a small amount to the AUC, although not importantly more than would be expected by chance. Furthermore, no definition added to the prediction of both recurrence and metastasis, and the previously cited “red flag” of 2 ng mL−1 y−1 did not help predict either outcome.
This work has several potential limitations. First, it is possible that there might be better schedules to measure PSA and better methods to calculate PSA velocity. Yet, patients in the PCPT were scheduled by protocol to have yearly PSA testing, in line with typical guidelines on PSA screening, including those of the NCCN. Moreover, we analyzed our findings using a large number of different approaches to the calculation of PSA velocity and did not find important differences in our results. Second, we have previously been criticized as inappropriately focusing on accuracy rather than reclassification (25). We believe that both are important when evaluating a marker. It is hard to see how a marker can lead to important reclassification if it does not materially add to prediction and therefore report both in this study. Most pertinently, our study is a direct evaluation of a published guideline, and the appropriate methods are self-evident. Third, it might also be suggested that the cancers detected in PCPT were “clinically irrelevant” because of the inclusion of the end-of-study biopsy for men with low PSAs. Yet, our results were similar when analyses were restricted to high-grade cancers or to clinically significant cancers; furthermore, our study methods are able to replicate exactly those recommended in the NCCN and AUA guidelines, biopsy for men with low PSA but high PSA velocity.
It is impossible to prove a negative and it is not inconceivable that PSA velocity could have a role in prostate cancer detection. For example, the men in our study with low PSA and negative DRE were screened for 7 years; therefore, conceivably, PSA velocity could be of value for men with either greater or fewer years of screening. Similarly, although we found no evidence that PSA velocity differentially detected aggressive cancers, our cohort has not been followed until death, and it is conceivable that some differences may emerge on very long-term follow-up. That said, guidelines should generally be based on the presence of positive evidence, not on the absence of negative evidence. It is difficult to support the inclusion of PSA velocity in a guideline on the grounds that it might conceivably be shown to be of benefit in some future study.
In conclusion, we have formally evaluated published guidelines on PSA velocity for prostate cancer detection, examining several definitions of cancer and numerous different methods of calculating PSA velocity. We found no reason to believe that implementation of the guideline would improve patient outcomes; indeed, its use would lead to a large number of unnecessary biopsies. We therefore recommend that organizations issuing policy statements related to PSA and prostate cancer detection remove references to PSA velocity.
Dr. Hans Lilja holds a patent for free PSA assays.
The funders did not have any involvement in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication. The initial analyses were designed by A. J. Vickers and H. Lilja in New York and implemented by C. Till and C. M. Tangen in Seattle. I. M. Thompson, who also had access to the data, advised on the interpretation of the results and suggested additional analyses. All authors were involved in the writing and revision of the manuscript.
Supported in part by funds from David H. Koch provided through the Prostate Cancer Foundation, the Sidney Kimmel Center for Prostate and Urologic Cancers and P50-CA92629 SPORE grant from the National Cancer Institute to Dr. P. T. Scardino, chair of the Department of Surgery at Memorial Sloan-Kettering Cancer Center. Additional support was obtained from National Institutes of Health (U01 CA086402 to I.M.T.; PO1 CA108964 to C.T., C.M.T., and I.M.T.; U01 CA032102 to C.M.T. and I.M.T.; and P30 CA054174 to I.M.T.).