If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
To determine the distribution of longitudinal changes in serum prostate-specific antigen (PSA) levels from a population-based sample of men.
Patients and Methods
In this prospective cohort study, a random sample of Olmsted County, Minnesota, men aged 40 to 79 years in 1990 were followed up biennially from January 1, 1990, through August 29, 2007. Serum PSA levels were determined at each examination, and men were censored for follow-up with a diagnosis of prostate cancer or treatment for benign prostatic hyperplasia. The empirical distributions of annual percent change and annual absolute change in serum PSA level were calculated and tabulated, including the median and 75th and 95th percentiles.
For men with PSA measurements 2 years apart, the median annual percent change in serum PSA level was 4.83% and the 95th percentile was about 49.76%. The variability in estimated annual change decreased with increasing time between assessments, with a 95th percentile of 21.82% after 8 or more years between assessments. Although the median absolute change per year increased with increasing age, the median percent change per year was fairly consistent across age groups.
These data demonstrate that, with shorter intervals between assessments, greater variability should be expected. These distributions should prove helpful to patients and clinicians in interpreting changes in serum PSA levels observed in typical clinical practices.
The measurement of serum prostate-specific antigen (PSA) levels for the prediction of prostate cancer is one of the most widely used cancer screening tools. When PSA screening was introduced, abnormal levels were determined using a single cut point.
This was due, in part, to the recognition that elevated levels at a single point in time might reflect stable levels or very slowly rising levels that would more likely reflect benign or indolent disease as opposed to aggressive disease.
In response to these concerns, several investigators have suggested using the change in serum PSA levels over time as an indication of prostate cancer. One of the first studies to this end was based on the Baltimore Longitudinal Study of Aging.
These investigators evaluated thawed serial samples from men who developed prostate cancer and a comparison group of men who did not. They used mixed-effects regression models to determine a cut point for change in serum PSA velocity (slope). On the basis of a comparison of these smoothed estimates of slope, they suggested a single cut point for absolute change in serum PSA level (0.75 ng/mL per year; equivalent to 0.75 μg/L per year) for an upper limit of normal. More recently, others have suggested that the amount of normal change depends on the starting serum PSA level,
There are, however, a number of limitations to the studies done to date. First, the use of mixed-effects regression models to estimate PSA velocity tends to smooth out the variability in serum PSA levels encountered in real-life use. Second, different methods of modeling have been used,
it is useful to step back and examine changes in serum PSA levels in a normal, unselected population. This provides a more realistic picture of the distribution of changes in serum PSA measurements that might be of greater use to clinicians in interpreting results encountered in their practices. To this end, we used information from the Olmsted County Study of Urinary Symptoms and Health Status Among Men, with up to 16 years of follow-up with biennial measurement of serum PSA levels, to describe rates of change in PSA levels in the community.
Patients and Methods
The Mayo Clinic Institutional Review Board and the Olmsted Medical Center Institutional Review Board approved this study.
Specific information regarding the study has been detailed previously.
To summarize, in 1990, white men between 40 and 79 years of age residing in Olmsted County, Minnesota, were identified through the Rochester Epidemiology Project. Potential participants were excluded on the basis of having previous prostate surgery or prostate cancer diagnosis or specific urologic diagnoses that affect voiding (other than benign prostatic hyperplasia). A total of 2115 (55%) of the 3874 men identified as eligible for the study participated and completed the study protocol. In a separate study, medical records of participants and nonparticipants were compared, and few differences were found. Participants were slightly more likely to report a urologic diagnosis.
From the cohort of 2115 men, a random sample of 537 men (25%) was selected. These men completed a comprehensive urologic examination, including measurement of serum PSA level. A total of 476 men (89%) participated and completed the detailed urologic examination.
Participants were followed up biennially from January 1, 1990, to August 29, 2007, for a maximum of 16 years of follow-up. In the second and third biennial examinations, men who dropped out of the study were replaced with men randomly chosen from the Olmsted County population (n=158). Observations after diagnoses of prostate or bladder cancer or treatment for benign prostatic hyperplasia were censored at the last examination before the diagnosis. After censoring, 18 men with no remaining observations were removed from analyses.
All serum PSA concentrations were determined in the same laboratory. In the baseline, fifth, and sixth rounds, the Tandem-R monoclonal immunoradiometric PSA assay (Hybritech, Inc., San Diego, CA) was used to measure samples, as specified by the manufacturer. The Abbott IMx assay (Abbott Diagnostics, Abbott Park, IL) was used to measure samples at the first and second follow-up (second and third rounds). For the fourth and final rounds, samples were tested with the Tandem-E PSA assay (Hybritech, Inc., San Diego, CA). During this entire time period, the coefficient of variation averaged 3% to 4% (G. Klee, MD, PhD, oral communication, March, 1995). Serum samples were collected before the digital rectal examination and transrectal ultrasonography.
The empirical distribution of annual percent change in serum PSA level was calculated by interval between examinations by dividing the difference in PSA level by the initial PSA measurement and the number of years intervening between the measurements and multiplying by 100. Given the biennial examinations, annual percent changes were calculated at 2-year intervals, and the distributions of these changes were summarized and presented as box plots. The distributions of these changes were tabulated, including the median and 75th and 95th percentiles. Similar calculations were made for absolute changes in serum PSA level as well. In addition, 2-stage and mixed-effects regression models were constructed to corroborate the change in serum PSA level for the cohort overall.
Because serum PSA levels follow a log-normal distribution, regression analyses were based on natural log-transformed serum PSA levels. A nomogram depicting the distribution of serum PSA levels observed at the 2-year follow-up as a function of the baseline serum PSA level was developed on the basis of the empirical distributions, highlighting the 75th and 95th percentiles of the distribution.
Serum PSA levels at baseline increased with increasing decade of age. The median serum PSA levels were 0.7, 0.9, 1.4, and 2.1 ng/mL for men in their 40s, 50s, 60s, and 70s, respectively, and were similar to those used in the development of previously reported age-specific reference ranges,
but also include the participants used as replacements in follow-up.
In Table 1 and Figure 1, the annual percent change in serum PSA level for the entire cohort is presented, stratified by intervening time. For 2-year intervals, the median percent change was 4.83%, with some attenuation in younger men. The 95th percentile for annual percent change, however, was much greater at 49.76%. This 95th percentile of percent change was fairly consistent across all ages, ranging from 47.67% to 52.38%. With increasing time between serum PSA level measurements, the estimated median change ranged between 3% and 4%. Notably, however, the dispersion about the median decreased with increasing intervening time interval (Figure 1). This leveled out after about 8 years. For 8-year intervening intervals and longer, the 95th percentile was 22% with little variation across all ages, ranging from 20% per year to 28% per year at 8 years (Figure 1; Table 1).
TABLE 1Pairwise Annualized Percent Change in Serum Prostate-Specific Antigen Level, by Intervening Time Interval: Olmsted County Study of Urinary Symptoms and Health Status Among Men, 1990-2007
In Table 2, the distribution of 2-year annual differences in absolute serum PSA levels is presented. As would be expected given the age-related increase in serum PSA level at baseline, the absolute change in serum PSA level increases with age. The median change increased from 0.02 ng/mL for men in their 40s to 0.16 ng/mL for men in their 70s. This age-related increase was observed for the 75th and 95th percentiles as well. As baseline PSA level increases with age
Two-staged and mixed-model analysis using all 16 years of data corroborated these results (data not shown). Modeled results were, on average, similar for both the 2-staged and mixed models; however, less variability was observed in the mixed-model results. In general, there was a fairly constant slope across age groups for men 50 years of age or older (median range, 4.0% per year to 4.9% per year), with a slight attenuation among 40- to 49-year-olds (median, 2.8% per year).
The nomogram presented in Figure 2 is based on the observed distribution of 2-year annual percentage differences in serum PSA determination as a function of baseline serum PSA level. The tops of the blue and gold areas represent the 75th and 95th percentiles of what would be expected on the basis of the empirical distribution of 2-year differences. Thus, if one assumes that this distribution reflects that of serum PSA levels in men without prostate cancer, the top of the yellow area corresponds to 95% specificity for 2-year change in serum PSA level for identifying men without prostate cancer. These data may be useful for referring men for further evaluation because men with the most rapid changes in PSA levels may be at greatest risk for prostate cancer.
In this report we describe distributions of changes in serum PSA levels that may help physicians and patients make sense of changes they observe in real life. These normative values give a sense of an upper limit of normal, providing insight into the specificity of change in PSA level in the detection of prostate cancer (ie, level at which the change correctly identifies men without cancer). We demonstrate that change is best described in terms of a percent change rather than an absolute change because the absolute change increased with increasing baseline values.
The use of absolute change in PSA velocity with a single cut point irrespective of PSA level could contribute to the reported lack of utility in detecting prostate cancer reported by some investigators.
Because absolute change in serum PSA levels over time increases with increasing PSA level, a single cut point in absolute change will be less specific in older age groups, potentially leading to more frequent, unnecessary biopsies. Thus, absolute change, particularly without age-specific cut points, may give a false impression that PSA velocity adds nothing over use of PSA alone in detecting prostate cancer. However, as percent change is relatively consistent across all ages, correlation with baseline serum PSA level and the need for multiple cut points become less of a concern. Although few studies have assessed the utility of percent change per year in detecting prostate cancer, Vickers et al
found that it provided some benefit over standard clinical predictors.
One obvious drawback of using percent change per year is that it is slightly more difficult to calculate than absolute changes in PSA levels. Therefore, we provide a nomogram that might help patients and physicians in deciding how to interpret changes in serum PSA levels and aid them in deciding whether or not to pursue a prostatic biopsy. This pictorial representation gives a sense of how unusual a change might be compared with what would be expected in a general population. It is important to note, however, that this nomogram has not been tested as a management tool.
These observations build on the work from the Baltimore Longitudinal Study of Aging.
In that study, the investigators observed changes in serum PSA values from samples that had been banked from their cohort. On the basis of mixed-effects regression models, they determined that the upper limit of normal would be a change of 0.75 ng/mL. A number of studies have since examined PSA velocity,
Although these findings may prove helpful in clinical practice, several potential limitations should be kept in mind. First, the interval between examinations for this study was 2 years. Thus, the data from the biennial examinations may not provide necessary insight for annual changes, which are often observed in clinical practice. On the basis of the observations of a decrement in variability across increasing intervening intervals, however, the estimates from these biennial examinations probably represent a minimum estimate for changes 1 year apart. In addition, there were a number of interim treatments within the cohort. It is well known that treatment with 5α-reductase inhibitors is associated with a decrease in serum PSA levels.
Moreover, surgical treatment may reduce serum PSA levels owing to a decrease in prostatic epithelial mass. Both of these factors should have been accounted for in our censoring, however.
A third potential limitation is the change in assay over the course of the study. Although the coefficient of variation remained fairly constant during the study period, the change in assay may have introduced some additional variability. We have previously shown comparability among the assays,
at least at the lower PSA levels seen in this cohort. This extra variability, however, is probably overshadowed in comparison with the variability introduced by use of different assays and laboratories,
as would be observed in typical clinical care. Finally, in this study, we did not examine the changes in serum PSA levels as related to prostate cancer. Because our goal was to determine normative values, however, this was an appropriate approach. Biologically, it makes sense that there is a lag between disease onset and clinical diagnosis in which a rise may be accelerated.
Thus, the inclusion of these men would add “noise” to our estimates and potentially systematically inflate the upper end of the distribution. In real life, however, one cannot know in advance who will eventually be given a diagnosis of prostate cancer, and with the retrospective analysis of data from this cohort, the censoring on future events would likely introduce additional bias. This is particularly problematic and counter to the goal of establishing normative values.
In addition to these potential limitations, the data must be interpreted with the representativeness of the study cohort in mind. While the Olmsted County cohort affords many advantages, baseline nonparticipation and dropout during the course of the study might introduce additional biases. An examination of the baseline characteristics of the cohort indicated that, if there was any participation bias, it was likely very small.
Moreover, this study included only whites. Extrapolating findings to men of different race and ethnicity should be done with caution.
We determined the normal variability in a serum PSA level in a cohort of white men who were studied systematically at 2-year intervals. These data may prove helpful to patients and clinicians in interpreting changes in serum PSA levels observed in typical practices. Men with rapidly increasing PSA levels should be evaluated more closely to determine whether the unusually rapid increase could be due to a benign acute condition or whether a prostatic biopsy should be scheduled.
The authors thank Ms Tina Condon for assistance in the preparation of the submitted manuscript and Ms Marcia Goodmanson for providing laboratory oversight and running many of the serum PSA assays.
Grant Support: This study was supported in part, by grants from the National Institutes of Health ( DK058859 , AG034676 , AR030582 , and RR000585 ) and Merck Research Laboratories .
Potential Competing Interests: Dr Girman is an employee of and a shareholder in Merck Research Laboratories. Dr Klee has received research grants and royalties for unrelated technologies from Beckman Coulter, Inc . Dr Jacobsen has received research grants from Beckman Coulter, Inc .
I would like to comment on the study by Jacobsen et al1 published in the January 2012 issue of Mayo Clinic Proceedings that documented the changes in serum prostate-specific antigen (PSA) values in a large group of men. The authors determined that the median annual change in PSA was about 4.8%, while the 95th percentile for PSA increase was about 50%. Interestingly, while the baseline PSA values and the absolute increases in PSA values increased with age, the increases in PSA were relatively constant across all ages when expressed as a percentage of the baseline value.
We thank Dr Keller for his comments regarding our study of longitudinal changes in serum PSA levels published in the January 2012 issue of Mayo Clinic Proceedings.1 As Dr Keller notes, our article shows that using changes in serum PSA levels expressed as percent change per year yields more stable findings across different ages and also provides a nomogram to aid clinicians in interpreting changes in serum PSA levels observed in a normal clinical practice. Much controversy currently surrounds the use of serum PSA measurements, and while PSA is not a perfect test, at this time it is still the only widely available option for screening for prostate cancer.