Fax: (011) 31 10 70 35315
Selective detection of aggressive prostate cancer†
Article first published online: 15 JUL 2011
Copyright © 2011 American Cancer Society
Volume 118, Issue 10, pages 2568–2570, 15 May 2012
How to Cite
Schröder, F. H. and Roobol, M. J. (2012), Selective detection of aggressive prostate cancer. Cancer, 118: 2568–2570. doi: 10.1002/cncr.26385
See referenced original article on pages 2651-8, this issue.
- Issue published online: 3 MAY 2012
- Article first published online: 15 JUL 2011
- Manuscript Accepted: 27 JUN 2011
- Manuscript Received: 14 JUN 2011
Screening for prostate cancer remains controversial despite 3 reports regarding randomized controlled trials of screening using serum prostate-specific antigen (PSA) as the main parameter for detection, all of which demonstrated a significant decrease in prostate cancer mortality.1-3 The numbers needed to screen (NNS) and the numbers needed to treat (NNT) to avoid 1 man dying of his disease are strongly dependent on the absolute risk reduction and the difference in incidence between the screening and control arms evaluated in a cumulative fashion. In the European Randomized Study of Screening for Prostate Cancer (ERSPC) study as a whole, with a median 9-year follow-up, the NNS and NNT amounted to 1410 men and 48 men, respectively, to save 1 man from prostate cancer death. The Göteborg trial demonstrated the effect of an increased follow-up by reporting an NNS of 293 and an NNT of 12 based on the intent-to-screen analysis.
The authors of the 2 ERSPC-related studies1, 3 agree in their conclusion that PSA-driven screening as performed within the ERSPC protocol leads to an unacceptably large degree of overdiagnosis and potential overtreatment. There is no doubt that the high rates of overdiagnosis reported with screening,4, 5 which amount to 44% to 66% depending on the underlying population and screening methodology, are a main factor impacting the quality of life associated with PSA-driven screening. Data from prospective studies have indicated that the majority of cancers that are potentially overdiagnosed because of their nonaggressive character can be found within the very low PSA ranges.6 However, there appears to be no PSA value that allows us to completely exclude the presence of potentially aggressive prostate cancer as determined by the Gleason score.7 Therefore, it is with good reason that health care providers demand changes in screening methodologies to avoid or decrease overdiagnosis before considering the introduction of population-based screening, despite clear evidence of its effectiveness.
Overdiagnosis is defined as the diagnosis of cancers that will not surface during a person's lifetime or kill their carriers. The definition of over diagnosis includes tumor characteristics and a man's life expectancy. Several studies have used empirical or model-based estimates of the prevalence of potentially overdiagnosed cancers in clinical and screen-detected populations of men. Kattan et al8 designed a nomogram for the identification of potentially “indolent” disease using 2 large radical prostatectomy series. This nomogram was validated on ERSPC data in the study by Steyerberg et al.9 The nomogram identified approximately 30% of cancers as being potentially “indolent” in 2 validation studies.10, 11 Using biopsy evidence and radical prostatectomy specimens, Epstein et al developed a set of predictive parameters of insignificant disease based on cancers identified in men with no evidence of cancer on digitial rectal examination (DRE) (T1c disease).12, 13 Their system remains arbitrary because it was validated in radical prostatectomy series of patients but never in a longitudinal fashion in relation to predictors of the natural history of prostate cancer.
Based on what has been stated above, overdiagnosis appears to be the main issue that needs to be resolved to make screening for prostate cancer acceptable. Available nomograms help to identify such disease. Formally, it also appears necessary to differentiate between “indolent” disease identified by the nomograms developed by Kattan et al8 or Steyerberg et al9 and the term “insignificant” disease introduced by Epstein et al,12, 13 terms that do not describe identical groups of cancers. For the time being, there is no other choice than to attempt to identify potentially overdiagnosed cancers and to accept a potential overestimation of their prevalence by considering the remainder as aggressive tumors. This is in fact also the goal of the application of the Steyerberg nomogram9 and the criteria of Epstein et al.12, 13
In the study by Williams et al,14 a clinical population of 635 men participating in a multicenter study of the Early Detection Research Network (EDRN) of the National Cancer Institute in the United States was used to develop a model for predicting histologically aggressive prostate cancer. Because the authors dichotomized the prognosticators of the criteria by Epstein et al12, 13 into aggressive and insignificant disease, the percentage of insignificant (or indolent?) cases was identified simultaneously. The authors concluded that on the basis of their findings, 24.6% of prostate biopsies could be avoided. This estimate is close to the estimate of 33% of avoidable biopsies determined using the Prostate Cancer Research Foundation (SWOP)/ERSPC Prostate Risk calculator based on ERSPC data if a probability cutoff for a positive biopsy of 12.5% is used.15 If one accepts, as in the study by Williams et al,14 that predictors designed for the identification of “insignificant disease” are used in a dichotomous way, the question of whether one should primarily study aggressive or indolent disease becomes irrelevant.
The EDRN study group of 6 independent centers included a total of 901 men at risk for prostate cancer but had to exclude 19% for various reasons to arrive at a total study population of 635 men. Cancer was detected in 274 men (43%). Of these men, 88 (14%) and 186 (29%), respectively, were classified as having indolent and aggressive prostate cancer. A predictive model for aggressive cancer was developed. In multivariate modeling, age, body mass index, family history of prostate cancer, an abnormal DRE, and the log of PSA density (PSA divided by prostatic volume) were found to be significant predictors of histological aggressiveness as defined by the criteria of Epstein et al12, 13 used in a dichotomous fashion. Slight modifications of these criteria were applied and described in detail.
A total of 3833 of the 4734 participants in the control arm of the Prostate Cancer Prevention Trial (PCPT) were used to validate the model.16 A total of 901 patients (19%) were excluded because of missing data. A sensitivity analysis using the imputation technique for missing data was performed. The validity of the created model was assessed by receiver operating characteristic (ROC) analysis.
The authors are to be congratulated for the development of another predictive model with which to determine the presence of indolent or aggressive prostate cancer. However, several constructive comments appear to be warranted.
Is the PCPT Data Set Adequate for the Validation of This New Predictive Model?
The EDRN data are based on clinical cases with a relatively high risk of prostate cancer. The population of men randomized to the control group of the PCPT trial were characterized by having a PSA level ≤ 3.0 ng/mL and a negative DRE at the time of study entry. Obviously, men with prostate cancer were excluded before randomization by questionnaire. Positive biopsies occurred during a 7-year period if clinical indications warranted or after 7 years in all men who were not biopsied during this follow-up period.
The validation of the newly developed prediction model by means of ROC analysis demonstrates a very impressive result (area under the curve [AUC], 0.78), which might be surprising when examining the differences between the 2 study cohorts, confirmed by the overprediction of the model.
Conversely, the results might not be that surprising if we study the performance characteristics of PSA alone in the placebo arm of the PCPT trial. In the study by Thompson et al,17 the sensitivity and AUC of PSA for the detection of all prostate cancers, those with a Gleason score ≥ 7, and those with a Gleason score ≥ 8 were calculated in both the treatment arm receiving finasteride and the placebo arm. The results of these analyses in the placebo arm, comprised of 5112 men with 1100 cases of prostate cancer detected with Gleason grading available were actually comparable to the newly developed model. The AUC for PSA in discriminating prostate cancers with a Gleason score ≥ 7 versus those with a Gleason score of ≤ 6 or no cancer was 0.781. With a sensitivity of 92.1%, the specificity was 37.3%. Making the contrast even more obvious was that comparing men with prostate cancers with a Gleason score of ≥ 8 versus all other men resulted in an AUC of 0.824. The newly developed model also had an AUC of 0.78 in the PCPT validation cohort. Related sensitivity and specificity values were 90% and 38%, respectively. PSA appears to be a very powerful discriminator within this particular data set. These data question the additional value of the newly developed multivariable model. It must be noted that the PCPT placebo arm cohort in the study by Thompson et al17 and the current study were not exactly identical because of some exclusions and minor differences with regard to the definition of potentially aggressive prostate cancer. A validation in a less “artificial” cohort such as the PCPT study cohort is warranted.
The use of PSA density in the model combines prostatic volume and PSA levels. This limits the reader's understanding of the individual weight of these 2 parameters. In addition, the study by Williams et al14 did not describe the technique of volume determinations in the EDRN and the PCPT trial, which makes comparability doubtful considering the well-known individual and technical variations of the measuring technique. Furthermore, the biopsy indications in the EDRN sample and the PCPT trial were not identical.14, 16 These differences are likely to influence the model outcomes. The finding that the EDRN-based model overestimated the outcomes of the validation set confirms that predictive models should not be applied to different populations without an awareness of the setting from which the models were derived.
Some other small comments are warranted. The creation of the model was based on a relatively small sample size. The authors reported an accurate identification of men at low risk (< 10%) for the detection of aggressive prostate cancer.14 These conclusions are based on just 1 biopsy. It is well understood that biopsies are often not representative of the Gleason scores noted in radical prostatectomy specimens, which can be considered as “true Gleason scores.” When viewed in light of this, it is questionable whether potentially missing 10% of aggressive prostate cancers is acceptable.
The high rates of overdiagnosis associated with PSA-driven screening can be considered the main reason why PSA-based screening of the male population at risk is unlikely to be introduced by health care providers. The article by Williams et al14 makes a contribution to the issue of the selective detection of nonaggressive versus potentially aggressive prostate cancer. Despite some critical remarks, it appears that the results, the avoidance of 24.6% of biopsies identified as unnecessary, at the cost of missing 10% of biopsy detectable, potentially aggressive prostate cancer is in line with the findings of others.15 Another validation of the EDRN-based model in a clinical set of men considered for prostate biopsy is recommended.
As a final conclusion, we strongly support the statement by Williams et al that PSA as a sole parameter should not be used in the early detection of prostate cancer.14
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.
- 14Selective detection of histologically aggressive prostate cancer: an Early Detection Research Network (EDRN) prediction model to reduce unnecessary prostate biopsies with validation in the Prostate Cancer Prevention Trial (PCPT). Cancer. 2011; 118: 2651-2658., , , et al.