Fax: (301) 480-0465
Prostate-specific antigen velocity and prostate cancer gleason grade and stage†
Article first published online: 1 MAR 2007
Published 2007 American Cancer Society
Volume 109, Issue 8, pages 1689–1695, 15 April 2007
How to Cite
Pinsky, P. F., Andriole, G., Crawford, E. D., Chia, D., Kramer, B. S., Grubb, R., Greenlee, R. and Gohagan, J. K. (2007), Prostate-specific antigen velocity and prostate cancer gleason grade and stage. Cancer, 109: 1689–1695. doi: 10.1002/cncr.22558
This article is a US Government work and, as such, is in the public domain in the United States of America.
- Issue published online: 4 APR 2007
- Article first published online: 1 MAR 2007
- Manuscript Accepted: 20 DEC 2006
- Manuscript Revised: 13 DEC 2006
- Manuscript Received: 19 OCT 2006
- National Cancer Institute
- prostate-specific antigen;
- prostate-specific antigen velocity;
- Gleason score;
- pathologic stage;
- prostate cancer
Increased preoperative prostate-specific antigen (PSA) velocity (PSAV) has been associated with increased prostate cancer mortality and higher Gleason scores. The authors evaluated the relation between PSAV, biopsy Gleason score, and pathologic stage in men who were enrolled in a prostate cancer screening trial.
Data were analyzed from 1441 men who were enrolled in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial who received ≥2 PSA screens and were diagnosed with prostate cancer within 1 year of the last screen. PSAV was estimated by using all screening PSA values within 6 years prediagnosis.
Both PSA and PSAV were related to biopsy Gleason score. The multivariable odds ratios (OR), controlling for PSA and demographics, for having a Gleason score of 7 to 10 were 1.3 (95% confidence interval [95% CI], 0.9–1.9), 2.2 (95% CI, 1.5–3.3), and 2.3 (95% CI, 1.4–3.9) for men with PSAV values from 0.5 to 1 ng/mL per year, from 1 to 2 ng/mL per year, and >2 ng/mL per year, respectively, compared with men who had PSAV values <0.5 ng/mL per year. The median PSAV was 0.60 ng/mL per year for men with Gleason scores from 2 to 6 versus 0.84 ng/mL per year for men with Gleason scores from 7 to 10 (P < .0001). Among 658 men who underwent prostatectomy, both PSA and PSAV were associated with advanced pathologic stage in univariate analyses; however, when the analysis controlled for clinical stage and biopsy Gleason score, the associations of PSA and PSAV were no longer statistically significant.
PSAV and PSA levels were associated independently with biopsy Gleason score. Among men who underwent prostatectomy, PSAV and PSA were not predictive of advanced pathologic stage when the analysis was controlled for biopsy Gleason score and clinical stage. It cannot be determined yet whether PSAV is predictive of long-term prostate cancer outcome in this cohort. Cancer 2007. Published 2007 by the American Cancer Society.
Currently, the majority of men in the United States who are diagnosed with prostate cancer come to medical attention because of elevated or rising levels of prostate-specific antigen (PSA).1 In addition to being used for the early detection of prostate cancer, PSA also has been evaluated as a prognostic marker for prostate cancer outcome and as a predictor of advanced prostate cancer stage or grade. The rate of change in PSA levels, known as PSA velocity (PSAV), also been evaluated as a predictor for prostate cancer stage and grade as well as for prostate biopsy outcome.
Recently, in a cohort of men who underwent radical prostatectomy, D'Amico et al. demonstrated that high preoperative PSAV, independent of PSA level and biopsy Gleason score, was a risk factor for death from prostate cancer.2 That study also indicated that men with high PSAV were more likely to have an elevated Gleason score and to have advanced pathologic stage. In addition, a recent analysis of the Baltimore Longitudinal Study of Aging cohort showed that PSAV measured 10 to 15 years before diagnosis distinguished men who developed nonfatal prostate cancer from men who developed fatal prostate cancer.3 Other studies also have demonstrated that PSAV at diagnosis is related to markers of prostate cancer aggressiveness.4–6
A number of population-based studies simultaneously examined the relation between PSAV and both prostate cancer diagnosis and Gleason score or cancer stage and produced mixed findings. An analysis of the placebo arm of the Prostate Cancer Prevention Trial indicated that PSAV was associated significantly with positive biopsy on univariate analysis but not in a multivariate analysis that controlled for PSA level.7 Similarly, PSAV was associated significantly with Gleason score in a univariate analysis but not in a multivariate analysis that controlled for PSA level. Data from the Rotterdam Center of the European Randomised Study of Screening for Prostate Cancer indicated that there was no significant association on univariate analysis between PSAV and a positive biopsy, and there was only a modest univariate association between PSAV and prostate cancer aggressiveness as measured by Gleason score and clinical stage.8 In a longitudinal study in Austria, Berger et al. reported a strong association between PSAV and both prostate cancer diagnosis and Gleason score.9
The prostate component of the ongoing multicenter Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial was designed to test whether screening with PSA and digital rectal examination (DRE) reduces prostate cancer mortality.10 For this report, we examined PSA levels and PSAV in a group of men who were diagnosed with prostate cancer and who previously had undergone multiple screening tests in the PLCO Cancer Screening Trial. This cohort included both men who did and men who did not undergo radical prostatectomy as therapy for their prostate cancer. Specifically, we examined PSA levels and PSAV as risk factors for an elevated biopsy Gleason score. In addition, in the cohort of men who underwent prostatectomy, we examined whether PSAV and/or PSA level, when combined with clinical predictors (clinical tumor classification [T-stage] and biopsy Gleason score), had better predictive ability for advanced pathologic stage compared with clinical predictors alone.
MATERIALS AND METHODS
The PLCO Cancer Screening Trial is a multicenter, randomized, controlled trial that was designed to test the efficacy of screening for 4 types of cancer in individuals aged 55 years to 74 years at baseline.10, 11 Randomization to a screened or control arm took place between November 1993 and July 2001, and almost 155,000 individuals were randomized at 10 different screening centers. Men in the screening arm received PSA and DRE at baseline (Year 0) and then annually through Year 3, and they received PSA without DRE at Years 4 and 5. Men in the screening arm also received flexible sigmoidoscopy and chest x-ray. Exclusion criteria included history of a prostate, lung, colorectal, or ovarian cancer; surgical removal of the entire prostate; having taken finasteride in the past 6 months; and, starting in 1995, having had ≥1 PSA blood test in the past 3 years. At the time of randomization, individuals filled out a self-administered demographic and medical/screening history questionnaire. Informed consent was obtained from all participants in the PLCO Cancer Screening Trial.
All PSA tests were performed at a single laboratory using the same Hybritech assay. A PSA result of >4 ng/mL was considered positive. Men were referred to their private physicians or health plans for follow-up of a positive PSA or DRE screen. Thus, prostate cancer diagnosis and pathology were carried out at many different institutions and by many different pathologists across the country, and there was no centralized pathology review. Prostate cancer cases were ascertained through routine follow-up of men with positive screens and through the use of an annual study update form that asked about any cancer diagnoses. Certified tumor registrars at each screening center used a standardized protocol to abstract clinical and pathologic (if available) tumor (T), lymph node (N), and metastases (M) characteristics; biopsy Gleason score; and Gleason score from radical prostatectomy (if performed). Quality control of abstracted stage and Gleason score data was performed by the PLCO Cancer Screening Trial Coordinating Center.
The cohort that was used for this analysis consisted of all men who had a confirmed prostate cancer diagnosis by December 31, 2004 and ≥2 study PSA screens, with the last screen occurring within 1 year of diagnosis. PSAV was estimated as the regression slope of all annual study screening PSA values (up to 6) before diagnosis; this estimate is denoted as PSAV. For a sensitivity analysis, PSAV also was calculated using only the most recent 2 annual screening PSA levels (denoted as PSAV). The PSA level was defined as the last screening PSA level before diagnosis.
The risk of advanced biopsy Gleason score (7–10), as a function of PSAV, PSA, and other covariates, was modeled by using logistic regression. Multivariate models included PSA (<4 ng/mL, 4–7 ng/mL, 7–10 ng/mL, and ≥10 ng/mL), age at diagnosis (<70 years vs ≥70 years), calendar year of diagnosis (1993–1998 vs 1999–2004), and race (black vs other), in addition to PSAV (<0.5 ng/mL per year, 0.5–1 ng/mL per year, 1–2 ng/mL per year, or >2 ng/mL per year). Values for linear trend for the ordinal variables (eg, PSAV, PSA) were calculated by coding each as a single quantitative variable with values corresponding to the median level in each ordinal category. A similar model was used to assess the risk of advanced pathologic stage (TNM stages III and IV). This model was run on the cohort of men with clinical T1 or T2, Nx/N0, and Mx/M0 disease who underwent radical prostatectomy. Also included as covariates in this model were clinical T-stage (T1, T2A, T2B/T2C) and biopsy Gleason scores (2–7 vs 7–10).
The predictive ability of the logistic models was assessed quantitatively by using the area under the receiver operator characteristic (ROC) curve. To examine how PSA and PSAV each contributed to predictive ability, ROC areas were compared for the full model (with both variables), for the full model without PSA, and for the full model without PSAV. The statistical significance of differences in ROC area were assessed by using the method of Hanley and McNeil, which takes into account correlations in ROC areas between models fit to common data.12
Table 1 displays the demographics, screening characteristics, and clinical characteristics of the men in this analysis. Most men were white, and the median age at diagnosis was 67 years. Greater than 70% of men had at least 3 study PSA screens. In total, 26% of men had Gleason scores from 7 to 10, and 46% of men underwent prostatectomy. The median PSAV (PSAV) was 0.66 ng/mL per year with an interquartile range from 0.29 ng/mL per year to 1.22 ng/mL per year. When PSAV was computed using only the last 2 PSA values, ie, PSAV, the median velocity was greater (1.02 ng/mL per year), and the interquartile range was wider (0.31–2.01 ng/mL per year). The PSAV and PSA levels were highly correlated (Spearman correlation, 0.64). In all, 54% of men with PSAV levels >2 ng/mL per year had PSA levels >10 ng/mL, compared with 8% of men with PSAV levels of 1 to 2 ng/mL per year and <3% of men with PSAV values <1 ng/mL per year with PSA levels >10 ng/mL.
|Total no. of patients||1441|
|Age at diagnosis, y|
|No. of PSA screens, %|
|PSA, ng/mL||4.1, 5.2, 7.1|
|PSAV, ng/mL/y*||0.29, 0.66, 1.22|
|PSAV, ng/mL/y†||0.31, 1.02, 2.01|
|Mean interval from last PSA to diagnosis, mo||4.2|
|Biopsy Gleason score, %|
Table 2 presents a univariate analysis of biopsy Gleason score by PSAV and PSA levels. The proportion of men with a Gleason score of 7 versus Gleason scores of 8 to 10 increased significantly with PSAV. In total, 16% and 3.6% of men with PSAV <0.5 ng/mL per year had a Gleason score of 7 and Gleason scores of 8 to10, respectively, compared with 28% and 9.1% of men, respectively, with PSAV >2 ng/mL per year. For PSA, 14% and 5.1% of men with PSA levels >4 ng/mL had a Gleason score of 7 and Gleason scores of 8 to 10, respectively, compared with 30% and 11.6% of men, respectively, with PSA levels >10 ng/mL. PSAV was significantly greater for men with Gleason scores of 7 to 10 than for men with Gleason scores of 2 to 6 (median, 0.84 ng/mL per year vs 0.60 ng/mL per year; 95% confidence interval [95% CI] for difference in medians, 0.13–0.36 ng/mL per year). Similarly, the median PSA level of 5.7 ng/mL for men with Gleason scores of 7 to 10 was significantly greater than the median PSA level of 5.1 ng/mL for men with Gleason scores of 2 to 6 (95% CI for difference, 0.24–0.95 ng/mL).
|Variable||No. of patients||Percentage of men|
|Gleason score 7||Gleason score 8–10|
|PSA level, ng/mL|
Table 3 presents the results of the multivariable logistic regression model for advanced biopsy Gleason score (7–10). The ORs associated with PSAV were 1.3, 2.2, and 2.3 for PSAV values of 0.5 to 1 ng/mL per year, 1 to 2 ng/mL per year, and >2 ng/mL per year, respectively, and the latter 2 ORs were elevated significantly. The OR, as a linear variable, for a 1 ng/mL per year increase in PSAV was 1.23 (95% CI, 1.11–1.35). The ORs associated with PSA were 1.2, 1.3, and 1.9 for PSA levels from 4 to 7 ng/mL, from 7 to 10 ng/mL, and >10 ng/mL, respectively (P value for linear trend = .03). A positive DRE result, calendar year 1999 to 2004, and age ≥70 years also were associated significantly with advanced Gleason score, with ORs of 1.8, 1.7, and 1.4, respectively; black race had a nonsignificant OR of 1.5. The area under the ROC curve for the full model was 0.646. Removing PSAV from the full model resulted in a significant decrease in ROC area to 0.626 (P = .03), whereas removing PSA from the model resulted in a nonsignificant change in ROC area to 0.642.
|Variable||OR (95% CI)|
|All men||PSA < 10 ng/mL||PSA ≥ 10 ng/mL|
|<0.5||1 (Ref)||1 (Ref)||1 (Ref)|
|0.5–1||1.3 (0.9–1.9)||1.3 (0.9–1.9)||1.2 (0.2–8.7)|
|1–2||2.2 (1.5–3.3)||2.4 (1.6–3.6)||1.1 (0.2–6.7)|
|≥2||2.3 (1.4–3.9)||1.5 (0.8–2.9)||5.1 (1.1–23)|
|1 ng/mL/y increase, linear†||1.23 (1.11–1.35)||1.18 (1–1.4)||1.30 (1.1–1.5)|
|P value for linear trend||.003||.13||.001|
|PSA level, ng/mL|
|<4||1 (Ref)||1 (Ref)||—|
|7–10||1.3 (0.8–2.1)||1.4 (0.8–2.3)||—|
|P value for linear trend||.03||.2||—|
|Positive DRE||1.8 (1.3–2.5)||1.7 (1.2–2.4)||2.2 (0.9–5.5)|
|Age ≥70 y||1.4 (1.1–1.8)||1.3 (1.02–1.8)||1.4 (0.7–2.9)|
|Black race||1.5 (0.9–2.4)||1.3 (0.7–2.1)||3.1 (0.8–11.8)|
|Years 1999–2004||1.7 (1.2–2.2)||1.5 (1.1–2.1)||3.4 (1.4–8.2)|
We observed a significant interaction (P = .03) between PSA level >10 ng/mL and PSAV. Therefore, we ran the logistic model separately for men with PSA levels <10 ng/mL and PSA levels ≥10 ng/mL (Table 3). In the PSA <10 ng/mL group, the OR decreased from a PSAV of 1 to 2 ng/mL per year (OR, 2.4) to a PSAV >2 ng/mL per year (OR, 1.5); however, the 2 ORs were not statistically significantly different. In contrast, for men in the PSA ≥10 ng/mL group, the OR for a PSAV of 1 to 2 ng/mL per year was not elevated (OR, 1.1), whereas the OR for a PSAV >2 ng/mL per year was elevated significantly (OR, 5.1).
To examine the effect of different estimates of PSAV, we also ran the overall model using PSAV derived from only the 2 most recent PSA levels (PSAV). The ORs associated with PSAV were muted compared with the ORs based on PSAV; specifically, ORs for PSAV values of 0.5 to 1 ng/mL per year, 1 to 2 ng/mL per year, and >2 ng/mL per year were 0.96, 1.1, and 1.5, respectively (P value for linear trend = .02). The area under the ROC curve for the full model with PSAV was 0.635 compared with the ROC area of 0.646 noted above for the model with PSAV; however, this difference was not statistically significant.
Table 4 shows the results from the analysis of advanced pathologic stage among the 658 men with nonadvanced clinical stage disease who underwent prostatectomy. The proportion of men with advanced pathologic stage increased significantly and in a similar fashion both with increasing PSAV and with increasing PSA. Rates of advanced stage disease increased from 18% to 32% for PSAV levels from <0.5 ng/mL per year to >2 ng/mL per year and increased from 17% to 32% for PSA levels from <4 ng/mL to >10 ng/mL. Rates of advanced-stage disease also increased with PSAV, but less markedly than with PSAV (Table 4). Clinical T-stage and biopsy Gleason scores also were associated significantly with advanced pathologic stage on univariate analysis. The results of the multivariate model indicated that both clinical T-stage and biopsy Gleason scores were associated significantly with advanced pathologic stage, with an OR of 3.5 for biopsy Gleason scores of 7 to 10 (compared to with biopsy Gleason scores of 2 to 6) and an OR of 2 for clinical T2B/T2C tumors (compared with T1 tumors). Neither PSA, nor PSAV (or PSAV), nor any of the demographic variables was associated significantly with advanced stage in the full model. The area under the ROC curve for the full model was 0.710. Removing PSAV, PSA, or both PSA and PSAV from the full model resulted in only a slight and nonsignificant decrease in the ROC area; the ROC area with neither variable in the model was 0.701 (P = .3 for comparison with the full model).
|Variable||No. of patients||Percentage with advanced pathologic stage||Multivariate OR (95% CI)†|
|PSA level, ng/mL|
|Biopsy Gleason score|
|Clinical tumor classification|
In this investigation, we demonstrated that, in a cohort of men who received multiple screening tests with PSA, high PSAV, even when controlling for PSA level and other covariates, is a risk factor for advanced biopsy Gleason score. Removing PSA level from the full predictive model resulted in minimal decreased predictive ability for advanced Gleason score, as reflected in the area under the ROC curve, which decreased only from 0.646 to 0.642. In contrast, removing PSAV from the full model resulted in a statistically significant decrease in the ROC area from 0.646 to 0.626. However, even the full model, which included PSA, and PSAV, and DRE, had relatively low predictive ability, as evidenced by an ROC area of only 0.646. Among the men who underwent prostatectomy, we observed that both PSA and PSAV had statistically significant associations with advanced pathologic stage on univariate analysis; however, neither was associated significantly with advanced stage in a multivariable analysis that controlled for clinical stage and biopsy Gleason score.
Estimates of PSAV depend on the number and timing of PSA samples. Calculating PSAV using only the last 2 annual PSA levels (PSAV) resulted in a velocity estimate that had both a greater median and greater variability than the estimate that utilized all screening PSA levels (PSAV). In men who had >2 screens, the median PSAV that was calculated using the first 2 PSA levels was 0.3 ng/mL per year, compared with medians of 0.6 ng/mL per year using all PSA levels and 1 ng/mL per year using the last 2 PSA levels. One explanation for the lower slope at earlier time points is that some men had a precancerous stage or very early-stage cancer at the earlier screens and, thus, had lower PSAV levels. In this sense, PSAV may be a more accurate (less biased) estimate of PSAV at diagnosis than PSAV. Conversely, there also may be an issue of selection bias, because diagnosis is more likely at the time of a high spike in PSA; thus, in some men, PSAV may reflect an artificially elevated level of PSAV. In addition, because PSAV is estimated by using more time points, it is a more stable estimate of PSAV; this is reflected in its lower variability. Statistically, choosing between PSAV and PSV as the best estimator of PSAV at diagnosis may be a question of a trade-off between bias and variance. Here, with respect to the predictive ability of PSAV for biopsy Gleason score, the resulting ORs using PSAV were muted compared with the ORs that were obtained with PSAV, and the ROC area was lower, although the difference was not statistically significant (0.646 vs 0.635). For patients who have multiple (3 or 4) PSA levels available within 1 year of diagnosis, using only those values probably would provide the best estimate of PSAV at diagnosis. However, when PSA levels are available only at annual intervals, the increased noise inherent in a slope estimate based on only 2 time points may indicate that the use of additional PSA levels, of necessity further back in time, is warranted.
D'Amico et al. recently reported that, among men who underwent radical prostatectomy for clinically localized prostate cancer, patients with PSAV levels >2 ng/mL per year had significantly greater advanced pathologic stage and Gleason scores than patients with PSAV levels <2 ng/mL per year.2 Specifically, among men who had PSAV levels <2 ng/mL per year, 26% had prostatectomy Gleason scores of 7 to 10, and 25% had advanced pathologic stage disease. By comparison, among men who had PSAV levels >2 ng/mL per year, 31% had Gleason scores of 7 to 10, and 30% had advanced-stage disease. We observed a somewhat greater magnitude of the effect of PSAV on stage and Gleason score among men in the PLCO Cancer Screening Trial who underwent prostatectomy: Men who had PSAV levels <2 ng/mL per year had a 37% prevalence of prostatectomy Gleason scores of 7 to 10 and an 18% prevalence of advanced-stage disease compared with men who had PSAV levels >2 ng/mL per year, who had a 47% prevalence of Gleason scores of 7 to 10 and a 28% prevalence of advanced-stage disease. D'Amico et al. also reported that PSAV was an independent predictor of death from prostate cancer when controlling for biopsy Gleason score and clinical stage; furthermore, this effect persisted even when the analysis was controlled for pathologic findings at radical prostatectomy. In our study, the addition of PSAV (or PSA) to a model that contained biopsy Gleason score and clinical stage did not significantly improve the model's ability to predict advanced pathologic stage. However, whether PSAV in this cohort is predictive of prostate cancer death, independent of Gleason score and clinical stage, remains unknown. Because the PLCO Cancer Screening Trial is a randomized clinical trial with a prostate cancer mortality endpoint, prostate cancer death outcomes are not yet available for analysis. However, this issue will be examined in the future as the endpoint results of the PLCO Cancer Screening Trial are released. In the Baltimore Longitudinal Study of Aging, Carter et al.3 recently also demonstrated that PSAV distinguished men who were dying from prostate cancer from men who were surviving with prostate cancer.
Thiel et al. reported a higher average PSAV in 38 patients with nonorgan-confined disease (1.88 ng/mL per year) than in 43 patients with organ-confined disease (1.12 ng/mL per year), although the difference was not statistically significant.6 Goluboff et al. demonstrated a correlation with PSAV (specifically, PSA doubling time) and pathologic stage but not with Gleason score in a group of men who underwent prostatectomy for clinically localized disease.5 Patel et al., in a group of 202 men who underwent radical prostatectomy, observed that men with PSAV levels >2 ng/mL per year were more likely to have pathologic T3 disease and positive surgical margins than men with PSAV <2 ng/mL per year.4 Berger et al. observed a significant relation between PSAV and both Gleason score and pathologic stage in a group of men who underwent regular PSA screening.9 These results generally are consistent with our current findings. In contrast, Thompson et al., in the placebo arm of the Prostate Cancer Prevention Trial (PCPT), observed no significant relation between PSAV, which they defined as the slope of log PSA over the 3-year period prior to diagnosis, and Gleason scores ≥7 when the analysis was controlled for PSA level; however, those authors did observed an elevated risk associated with PSAV on univariate analysis.7 A major difference between the PCPT and the PLCO Cancer Screening Trial was that most of men in the PCPT were diagnosed based on an end-of-study biopsy and not based on elevated PSA or suspicious DRE, like most men in the PLCO cohort. Consequently, the distribution of PSA was much lower in the PCPT cohort than in the PLCO cohort; 63% of patients in the PCPT, compared with only 15% of patients in the PLCO, had PSA levels ≤3 ng/mL at diagnosis.
A limitation of the current study is that PSA levels were examined prospectively as part of a screening study, and clinical decisions were made based on their values; thus, these levels influenced the probability and timing of cancer diagnosis. This contrasts to a situation in which blood samples are obtained and frozen, and PSA (or other marker) levels are determined only after diagnosis. In the PLCO Cancer Screening Trial, a man with high PSA (and concurrent prostate cancer) at baseline likely would have been diagnosed at that time and would not receive further screens; thus, he would not be eligible for inclusion in this PSAV cohort, which requires multiple PSA screens. Therefore, our current findings may not be generalizable to men who are not undergoing regular screening. Another limitation of the study is that factors on which the PLCO Cancer Screening Study does not collect data and for which we could not control in our analyses, such as urinary tract infections, recent sexual activity, and recent colonoscopy, may increase PSA levels transiently and, thus, may affect our estimates of PSAV.13, 14 Such transient changes would tend to dilute the observed correlation between PSAV and prostate cancer grade and stage.
PSAV and PSA both had a significant, independent association with biopsy Gleason score in a cohort of annually screened men who were diagnosed with prostate cancer. Neither PSAV nor PSA was associated with advanced pathologic stage when the analysis was controlled for clinical variables.
Supported by contracts from the National Cancer Institute.