To elucidate the role of prostate-specific antigen (PSA) doubling time (PSAdt) as a progression criterion in patients with low-risk prostate cancer managed by active surveillance (AS).
To assess the correlation between PSAdt during AS and final histopathology after radical prostatectomy (RP) in patients meeting predefined progression criteria.
Patients and Methods
A total of 258 consecutive patients on an AS programme were included in the study.
The PSAdt was calculated in patients with two or more PSA values, and 95% confidence intervals (CIs) were calculated in patients with four or more PSA values.
Progression risk groups were defined as follows: high-risk: PSAdt <3 years, rebiopsy Gleason score (GS) ≥4 + 3, more than three positive biopsy cores, and/or bilateral tumour or cT ≥2c disease; intermediate-risk: PSAdt 3–5 years, GS = 3 + 4 or cT2b disease; and low-risk: PSAdt >5 years, without histopathological or clinical progression.
Definitive treatment was recommended for patients in the high-risk group and treatment options were discussed with those in the intermediate-risk group.
A total of 2291 PSA values obtained during AS were available, of which 2071 were considered valid in the 258 patients.
PSAdt values with 95% CIs were calculated in 221 patients based on a median of 8 PSA values.
The 95% CIs for PSAdt overlapped considerably and in up to 91% of the patients, the 95% CIs overlapped among the risk group definitions.
A total of 26% (68/258 patients) underwent RP after meeting the progression criteria.
There was no association between preoperative PSAdt and final histopathology (P = 0.87).
The uncertainty of calculated PSAdt during AS leads to a significant risk of patients being misclassified in terms of risk of progression, which limits the use of PSAdt in the management of patients on AS.
The incidence of prostate cancer (PCa) has increased worldwide during recent decades . As a result of early detection programmes, nearly 50% of newly diagnosed patients are diagnosed with low-risk disease and could, as such, be candidates for therapy with curative intent [2-4]; however, the treatment of patients with low-risk PCa is controversial, and the benefit of radical prostatectomy (RP) over conservative treatment in these patients has been questioned . Active surveillance (AS) has been introduced as a means of reducing overtreatment without compromising long-term cancer-specific survival in patients with low-risk PCa . A crucial element of AS is the ability to identify those patients who are likely to progress while cure is still possible. Most AS programmes use a combination of repeated biopsies (rebiopsy), PSA kinetics and clinical tumour category (cT) to assess patients . A recent study has questioned the ability of PSA doubling time (PSAdt) to identify patients with clinically significant PCa, as it found no correlation between prediagnostic PSAdt and PCa-specific mortality in patients treated conservatively .
The primary aim of the present study was to calculate PSAdt values with 95% CIs to assess the probability of misclassifying the risk of progression when using PSAdt in the monitoring of patients undergoing AS. As a secondary objective, we evaluated the correlation between PSAdt during AS and final histopathology in patients who eventually underwent RP after meeting predefined progression criteria.
Patients and Methods
Between 2002 and January 2012, 258 consecutive patients entered an AS programme at our institution. Patient data were prospectively registered in a database approved by the Danish Data Protection Agency (file#2006-41-6256). The median (range) follow-up was 3 (0.3–9.5) years. Results from this AS cohort have been published previously . Briefly, patients with an expected low tumour burden, typically patients with a combination of PSA ≤10 ng/mL, Gleason score (GS) ≤6 and cT≤2a disease, were considered candidates for AS. No strict criteria regarding age were used, but patients aged <65 years were considered less suitable for AS. Patients deviating from the inclusion criteria outlined were accepted within the programme if they had a strong wish for AS instead of immediate curative therapy.
Patients were followed with PSA testing and DRE every 3 months after starting the AS programme and a rebiopsy was performed after 1 year. After the first year patients were classified according to their risk of progression as follows: high risk: PSAdt <3 years, progression on rebiopsy with GS ≥4 + 3, more than three positive biopsy cores, and/or bilateral tumour, or cT≥2c disease; intermediate risk: PSAdt 3–5 years, progression to GS = 3 + 4 or cT2b disease; and low-risk: PSAdt >5 years (including ‘negative’ PSAdt) and without histopathological or clinical progression. After the first year, curative therapy was recommended for patients categorized as high risk. Treatment options were discussed with patients in the intermediate-risk group. Patients in the intermediate-risk group who opted for continued AS had a closer follow-up with DRE and PSA tests every 3 months and additional biopsies after 2 years on AS. Patients categorized as low risk had DRE and PSA tests every 6 months and additional biopsies were only performed if patients changed risk category (based on PSAdt and/or cT category). After 3–5 years on AS, and if none of the progression criteria was fulfilled, patients were followed up with annual PSA tests.
The PSA value at the start of the AS programme was defined as the PSA value immediately before the diagnostic biopsy but, for patients diagnosed by TURP, only PSA values measured after TURP were included. PSA values measured during AS were only considered valid in patients who did not receive PSA suppressant therapies. All PSA values after TURP, or after commencing 5-α reductase inhibitors to relieve BOO while on AS, were censored. Furthermore, all outliers were reviewed carefully and PSA values that were assumed to be falsely elevated (i.e. after rebiopsy, UTI, or urinary retention) were excluded. PSAdt was calculated according to the Memorial Sloan-Kettering Cancer Center guidelines (http://nomograms.mskcc.org/Prostate/PsaDoublingTime.aspx) defined as the ln(2) divided by the estimated slope of the linear regression with log of PSA as the dependent variable and time of PSA test as the explanatory variable. All available PSA values were used for PSAdt calculation. The 95% CI values for PSAdt were calculated only for patients with four or more valid PSA values. For patients with a negative PSAdt (<0 years, PSA decreasing during follow-up) only the lower 95% CI could be calculated. The linear regression models were assessed using conventional diagnostics. Exact 95% CIs were used for binary data. The hypothetical probability of being misclassified as a high-risk patient (i.e. PSAdt <3 years vs PSAdt ≥3 years) was considered for different sampling patterns. The probability of misclassification was calculated assuming that the regression coefficient (i.e. the slope of PSA on the log scale regressed on time) was Student-t distributed and, using the median of the root mean square errors, estimated from the data. Statistical analysis was performed using sas software (v9.2, SAS Institute, Cary, N.C. USA).
We stratified the final histopathology of the surgical specimen into three prognostic histopathology groups based on pathological tumour category (pT), GS and lymph node category (N) as follows: poor (pT≥3, GS ≥8 or N1 disease); intermediate (pT≤2c, GS ≤7, and N0/x or pT≤2c disease, GS ≤6, >10% tumour volume, and N0/x); and good (pT≤2c, GS ≤6, ≤10% tumour volume and N0/x disease).
Baseline characteristics are shown in Table 1. The median (interquartile range [IQR]) age at the start of AS was 66 (63.5–68.5) years. The majority of the study population (78%) had low-risk PCa according to the D'Amico risk classification .
Table 1. Baseline characteristics of the 258 patients at entry into the AS programme
The median (IQR) diagnostic PSA used as the baseline PSA value (at the start of the AS programme) for the 258 patients was 6.5 (2.7–10.3) ng/mL. A total of 2291 PSA values were available during AS, of which 2071 were considered valid. For these 2071 PSA values, the median (IQR) value was 6.7 (4.8–9.1) ng/mL. A total of 221 patients had four or more valid PSA values and the calculations of PSAdt and 95% CIs were based on a median of 8 (4–29) PSA values. A goodness-of-fit test could not detect a lack of linearity. During follow-up, based on calculated PSAdt, 30 patients were classified as high risk (PSAdt <3 years), 34 as intermediate risk (PSAdt 3–5 years) and 157 as low risk (PSAdt >5 years), respectively.
In total, 67 patients had a negative PSAdt. For those 154 patients with a positive PSAdt, Figure 1 shows the calculated PSAdt values with 95% CIs. The 95% CIs showed that there was a considerable overlap between risk groups for all 221 patients with four or more valid PSA values (Table 2). For patients classified as high risk based on a PSAdt <3 years, the upper limit of the 95% CI overlapped with the intermediate- and low-risk range in 69 and 39% of the cases, respectively, while for 76% of the patients classified as intermediate risk based on a PSAdt of 3–5 years the lower limit of the 95% CI overlapped with the high-risk definition. For the 157 patients classified as low risk, the lower limit of the 95% CI overlapped with the intermediate- and high-risk range of PSAdt for 40 and 13% of patients, respectively.
Table 2. Overlap of 95% CIs for the calculated PSAdt values between the risk group classifications
PSAdt <3 years, %
PSAdt 3–5 years, %
PSAdt >5 years, %
After a median 1.2 years, 68 of the 258 patients (26%) who were on AS eventually underwent RP after meeting the progression criteria. Based on surgical final histopathology, 22% were found to be in the poor histopathology group, 47% in the intermediate histopathology group, and 31% in the good histopathology group, according to the criteria defined above. Table 3 lists the type and frequency of progression criteria applied in the 68 patients, stratified by histopathology group. PSAdt and progression on DRE did not correlate with final histopathology. In fact, out of the 23 patients who met the PSAdt progression criterion, only three had poor final histopathology. Not surprisingly, because of the definition of the histopathology groups, findings on re-biopsy correlated well with final pathology at RP.
Table 3. Correlations between the progression criteria and the three prognostic groups, based on final histopathology in the surgical specimen
Table 4 shows the correlations between the three histopathology groups based on the surgical specimen and the baseline characteristics when the 68 patients who underwent RP began the AS programme. No significant differences in the distribution of characteristics were found among the three groups.
Table 4. Correlations of baseline characteristics correlated with the three prognostic groups based on final histopathology in the surgical specimen
To investigate whether PSAdt during AS correlated with final histopathology, the PSAdt values were compared among patients in the three histopathology groups (Table 5). In the 68 patients PSAdt during AS was calculated using a median of 6 (2–16) PSA values; two patients had no valid PSA values as a result of PSA suppressant treatment. Of the 68 patients, 27 had a PSAdt of <3 years, 11 had a PSAdt of 3–5 years, and 28 had a PSAdt of >5 years. No statistical association between PSAdt and histopathology at RP could be established (P = 0.9).
Table 5. Correlation between analysed PSAdt (calculated after excluding outlying PSA values and values taken during PSA suppressant treatment) in the three prognostic groups, based on final histopathology in the surgical specimen
The role of PSA kinetics in the management of PCa has been investigated in numerous situations. PSAdt before diagnosis has been shown not to predict PCa mortality  or progression on re-biopsy in patients managed by AS ; however, PSAdt values obtained before RP have been shown to predict PCa-specific mortality and survival in patients with castrate-resistant PCa [12, 13]. In patients managed by watchful waiting in the SPCG-4 study, a short PSAdt was associated with an increased relative risk of lethal PCa , but among men with a PSAdt of 3–10 years there was a cumulative incidence of lethal PCa after 6 years of 7–9%. In that study the calculated PSAdt was performed after 2 years of follow-up, and in 14.6% of the patients, it was based on three or more PSA values. The authors conclude that their ‘findings raise the question of whether early PSA characteristics are suitable and safe as decision tools for therapeutic intervention among low-risk patients managed with active monitoring’; however, that study, and other studies to our knowledge, did not address the uncertainty of calculated PSAdt values and therefore it is unclear if their results reflect the actual performance of PSAdt or the uncertainty of the estimates.
Today patients who undergo serial PSA testing after diagnosis are mostly managed by AS. In the present study we analyse the uncertainty and limitations of calculated PSAdt as a progression criterion in this setting. In the present study, a goodness-of-fit test confirmed that PSAdt reliably reflects the change in PSA over time, but the wide CI of the calculated PSAdt led to a significant overlap between the risk-group definitions and, consequently, a significant risk of misclassification according to the progression criteria used in our AS programme. Furthermore, in the 68 patients who eventually underwent RP in our AS cohort, the preoperatively calculated PSAdt did not correlate with final histopathology.
The use of PSAdt as a progression criterion in patients with presumed low-risk PCa managed by AS has recently been introduced [15, 16]. The original threshold used was 2 years but as <20% of the AS cohort fulfilled this criterion the threshold was subsequently changed to 3 years . Klotz et al. found that a PSAdt <3 years in univariate analysis correlated with an 8.5-fold increased risk of biochemical recurrence after definitive therapy, thereby supporting the use of PSAdt as a marker for aggressive PCa [16, 18]; however, this interpretation has limitations. Firstly, the patients undergoing definitive therapy received different treatments with different definitions of biochemical recurrence and, secondly, the tumour characteristics of the patients were not reported and hence not adjusted for in multivariate analysis.
PSA is not a cancer-specific marker [19-21], which may significantly limit its use in monitoring patients with small tumour volumes. As a result of selection, patients on AS are likely to have small tumour volumes in prostates also affected by varying but significant amounts of BPH tissue. Although morphometric studies have shown that PSA levels increase with ∼3 ng/mL/g PCa tissue, which is approximately 10 times the increase found based on BPH alone [22, 23], the PCa contribution to the total PSA level may be obscured by the significant contribution from benign tissue. Thus, the use of PSAdt might also have considerable limitations in the presence of large components of benign tissue in the prostate gland.
Although the linearity assumption for PSAdt calculation was clearly fulfilled (goodness-of-fit test), the variation and number of PSA tests resulted in a wide CI, which led to overlap in the CIs between the definitions of risk of progression in a significant number of patients (Figure 1). We calculated the probability of a true PSAdt ≥3 years being misclassified as a PSAdt <3 years (high-risk group) or vice versa. The results are shown in Figure 2; the x-axis shows the true PSAdt and the y-axis the predicted probability of misclassification. The probabilities for three different PSA testing scenarios are shown. Pattern 1 a PSA test every 3 months for 1 year (a total of five PSA values, degrees of freedom (DF) = 3); pattern 2 a PSA test every 3 months for 1 year and thereafter at 6-month intervals up to 2 years (a total of seven PSA values, DF = 5); and pattern 3 as (2) but up to 3 years (a total of nine PSA values, DF = 7). The mean square error used for these calculations was 0.045 (the median mean square error for patients with four or more PSA tests with PSA log-transformed). We chose to analyse the probability of being falsely classified as belonging to the high risk of progression group, because patients categorized as such are normally advised to undergo curative therapy, i.e. a high-risk classification has obvious clinical consequences. Figure 2 shows that, after 1 year (red line) on AS, the risk of being misclassified is substantial. One in five patients with a true PSAdt of 5 years and one in 10 with a true PSAdt of 9 years will be recommended curative therapy because of misclassification. With more PSA tests taken over a longer period of time, the PSAdt estimate becomes more accurate with less risk of misclassification. After 2 years (green line) only one in 20 with a true PSAdt of 5 years and one in 100 with a PSAdt of 9 years will be misclassified. Although it seems intuitive that an increasing number of PSA values would enhance the accuracy of the calculated PSAdt, none of the currently available online calculators provide 95% CIs, nor do they estimate the risk of misclassification.
PSAdt are currently being used as a trigger for curative treatment and rebiopsy in a number of AS cohorts [9, 16, 18]. The duration of follow-up and number of PSA values used to calculate PSAdt varies between the studies. The Toronto group requires at least three PSA values, obtained over minimum of 6 months . In the Prostate Cancer Research International: Active Surveillance (PRIAS) study and in our own series, decisions are based on a minimum of five PSA values obtained over 1 year of follow-up [9, 18]. Thus, all programmes base their calculation of PSAdt on a short follow-up and a limited number of PSA values. As shown by Figure 2, such a follow-up regimen (represented by the red line) to calculate PSAdt is associated with a significant probability of misclassification. While a longer follow-up and more PSA values may decrease the probability of misclassification, a dilemma exists because a PSAdt based on longer follow-up may underestimate the most current disease status. This concern is reflected in the recommendations from the Prostate-Specific Antigen Working Group's Guidelines on PSAdt. The group recommends that only PSA tests obtained during the last 12 months should be used in the calculation so as to reflect the patient's present and most relevant disease .
Retrospective studies have shown that a relatively high percentage of patients with low-risk disease and potential candidates for AS are upgraded and/or upstaged in their final histopathology when they instead undergo immediate RP [26, 27]. Acknowledging, therefore, that the inclusion criteria for AS have a limited accuracy when it comes to selecting patients with true low-risk PCa, it becomes even more important that the progression criteria used when monitoring patients on AS on the one hand are able to identify patients with poor histopathology sufficiently early for curative therapy to be possible, and on the other hand are specific enough to spare patients with true low-risk disease from curative treatment.
Only patients who met the predefined progression criteria in the programme were available for a comparison of the final histopathology after RP and the PSAdt during AS. Logically, this comparison has significant limitations since PSAdt was one of the criteria used to select patients for RP. In our comparison, we were not able to establish a correlation between PSAdt and final histopathology; however, had the entire AS cohort eventually undergone RP, a correlation might have existed. Sengupta et al.  found that PCa-specific mortality was greater in patients with a preoperative PSAdt <1.5 years compared to those with longer PSAdt. In a recently published study, O'Brien et al.  reported that there was a trend for prediagnostic PSAdt, calculated according to the Memorial Sloan-Kettering Cancer Center guidelines, towards predicting biochemical recurrence and metastatic disease in almost 3000 patients after RP, but these authors did not find PSAdt useful in patients with a PSA <10 ng/mL and GS ≤6, i.e. candidates for AS, which supports the present findings. Also consistent with our findings, Maruyama et al.  recently reported that preoperative PSAdt in 47 patients eligible for AS, who underwent RP instead, was not associated with final histopathology or biochemical recurrence.
In conclusion, whilst acknowledging the limitations of the study design, the wide CI of the calculated PSAdt leads to a significant number of patients on AS being misclassified, especially after a short observational period. Furthermore, PSAdt during AS showed no correlation with final histopathology. Our results emphasize that this variable is unsuitable for use as a progression criterion in the management of patients on AS.
Frederik B. Thomsen's Ph.d. is sponsored by IMK Almene Fond.