Prostate cancer is the most common cancer among men in most industrialised countries.1 Large randomised studies have been launched to assess the effectiveness of prostate cancer screening with serum prostate-specific antigen (PSA).2, 3 Despite the widespread use of PSA for prostate cancer screening, its sensitivity has not been evaluated properly, i.e., using a longitudinal approach in the context of a randomised trial. The sensitivity of a screening program provides a key measure of screening performance; i.e., it is an essential intermediate indicator in the absence of mortality results.
Sensitivity is the ability to detect disease cases, determined by the frequency of false-negative test results. In the screening setting, cross-sectional evaluation of sensitivity (as used for evaluation of diagnostic tests) based on results obtained by simultaneously performed tests is not sufficient because screening is intended to detect cases that would clinically surface in the future (above all, cases that would do so before the next screening round). Furthermore, diagnostic assessment is generally limited to subjects with a positive screening test, and it is prone to overdiagnosis, i.e., detection of cancers that would not have surfaced clinically in the absence of screening. This approach yields inflated estimates of sensitivity.4, 5 This is particularly evident for prostate cancer, which has a high prevalence of indolent cancers at autopsy. A more valid approach for evaluation of the sensitivity of screening is the incidence method, which is based on the reduction in disease incidence following screening (i.e., a longitudinal approach). Ideally, sensitivity is estimated by comparison between screening and control groups formed by randomisation, i.e., in an experimental setting. With this approach, test sensitivity (sensitivity as a characteristic of a test) is defined on the basis of the difference in disease incidence following a negative test, relative to an unscreened control population. Thus, e.g., an 80% difference in incidence corresponds to sensitivity of 80%.
Screening for prostate cancer is very common in many countries, but few studies have evaluated the sensitivity of PSA in prostate cancer screening. Here, we report estimates of sensitivity of the PSA test in prostate cancer screening based on interval cancer incidence in a large, population-based randomised trial.
MATERIAL AND METHODS
The Finnish prostate cancer screening trial was started in May 1996, and it is the largest component of the European Randomised Study of Prostate Cancer Screening.6 The study population consists of men born 1929–1944 and resident in the metropolitan areas of Helsinki and Tampere. In the beginning of each calendar year 1996–1999, 8,000 men aged 55, 59, 63 or 67 years were randomly allocated to the screening arm, with a total of 32,000 men in 4 years. Of these, 412 men were excluded because of death (n = 249), diagnosis of prostate cancer (n = 137) or moving abroad (n = 26) before start of follow-up. The remaining 48,458 men formed the control arm of the trial and were not contacted. From the control arm, 625 men were excluded because of death (n = 413), diagnosis of prostate cancer (n = 179) or moving abroad (n = 33) before start of follow-up.
Invitations to participate in the study were mailed to the men in the screening arm in 4 batches each calendar year. After informed consent, a blood sample was drawn and sent for analysis to a central laboratory. PSA determinations were performed with both the Tandem-E (Beckman-Hybritech, San Diego, CA) and Delfia (Wallac, Turku, Finland) assays. Screen-negative men were defined as those having either a serum PSA concentration <3.0 ng/ml or <4 ng/ml with a negative ancillary test in the PSA range 3.0–3.9 ng/ml: either a benign digital rectal examination (DRE) finding (in 1996–1998) or free-to-total PSA ratio ≥0.16 (in 1999).
Interval cancers were identified through record linkage with the Finnish Cancer Registry, which is a nationwide, population-based cancer registry with practically complete coverage of cancer cases in Finland.7 To ensure the completeness of case ascertainment for the most recent years, a record linkage of the study population with the discharge database of hospitals in the study area was also conducted.
The start of follow-up for screen-negative men was the date of screening attendance if PSA was <3 ng/ml or the date of the ancillary test if PSA was 3.0–3.9 ng/ml. For the control arm, follow-up started 181 days after randomisation, which was the average time elapsed between randomisation and start of follow-up in the screening arm. End of follow-up was the date of death (from any cause), emigration, prostate cancer diagnosis, time of second screening round for the screening arm and 4 years from start of follow-up for the control arm or the common closing date (31 December 2000). Hence, cases detected in the second screening round were excluded. Information on vital status was obtained for all subjects by record linkage with the Population Registry.
The study protocol was approved by the ethical committees of the participating hospitals, the National Research and Development Centre for Welfare and Health as well as the National Authority for Medico-Legal Affairs.
Test sensitivity, ST, was estimated based on the incidence method by comparing the observed incidence of interval cancer with cancer incidence expected in the absence of screening:
where II is cancer incidence among screen-negative men and IC is cancer incidence in the control arm.4, 5
Sensitivity was adjusted for incomplete (i.e., <4 years) follow-up by calculating the mean annual incidence rates by follow-up year.
The confidence interval (CI) for ST was calculated using 10,000 computer simulations, assuming that the numbers of prostate cancer cases in both the screening and control arms are random variables that follow a Poisson distribution.
A total of 32,000 men were randomised to the screening arm; 30,195 were invited and 20,790 (69%) participated (Fig. 1). Among the screened men, 1,823 (8.8%) had a serum PSA concentration ≥4.0 ng/ml and 513 prostate cancers were detected at screening, corresponding to an overall detection rate of 2.5%. In addition, 30 cancers were detected among 140 (13%) of 1,071 men with PSA of 3.0–3.9 ng/ml who had either a suspicious DRE finding or a free-to-total PSA ratio of ≤0.16. Hence, the overall detection rate was 2.6%.
During the follow-up of the 17,897 men with PSA <3 ng/ml, 49,492 person-years were accrued and 19 prostate cancers were detected within the 4-year interscreening interval (Table I). In addition, among the 811 men with PSA of 3.0–3.9 ng/ml and a negative ancillary test, 5 cases were detected. Among 47,833 men in the control arm, 539 cancers were diagnosed.
Table I. Numbers of Men, Person-Years and Prostate Cancers With Incidence Rate by Trial Arm, Screening Participation and Serum PSA Concentration
he mean interval cancer incidence among men with PSA <3 ng/ml based on a mean follow-up of 2.8 years was 48/100,000 person-years and that for all screen-negative men (including those with PSA 3.0–3.9 combined with a benign DRE or free/total PSA ratio <0.16) was 55/100,000 (Table I). The mean incidence of prostate cancer in the control arm was 425/100,000 person-years.
Test sensitivity was calculated by comparing prostate cancer incidence among screen-negative men (interval cancer incidence) with the incidence in the control arm. For PSA <3 ng/ml, test sensitivity was estimated to be 0.89 (1–48/425, 95% CI 0.84–0.93). Test sensitivity for the combination of serum PSA <4 ng/ml and ancillary test in the range 3.0–3.9 ng/ml was estimated to be 0.87 (1–55/425, 95% CI 0.82–0.92). In other words, the PSA test was able to identify 87–89% of interval cancers. Test sensitivity for PSA <3 ng/ml was 1.00, 0.86, 0.87 and 0.84 for follow-up years 1–4, respectively. The corresponding figures for PSA <4 ng/ml (with ancillary test for men with PSA 3.0–3.9) were 0.99, 0.85, 0.84 and 0.85.
Screening for prostate cancer with the PSA test is controversial, and practices vary from annual testing to no routine screening. No valid data are available on the effect of PSA screening on mortality. Here, we report estimates of sensitivity of the PSA test for prostate cancer screening in a population-based trial. We used longitudinal evaluation, with test sensitivity defined as reduction in cancer incidence in test-negative men compared to unscreened controls. Our results show that PSA as a screening test can detect almost 9 of 10 (87–89%) cases that would have surfaced clinically during the 4-year rescreening interval.
We defined test sensitivity as the proportion of men with a positive test among all men with prostate cancer and estimated it by means of interval cancer incidence after a negative screen (PSA <3 ng/ml or <4 ng/ml combined with a negative ancillary test in the PSA range 3.0–3.9 ng/ml). As men with preclinical disease among screen-negative subjects cannot be identified, interval cancers (cases surfacing clinically during the first screening interval) were used as an indicator of cancers missed at screening.
Information on cancer incidence was obtained from the Finnish Cancer Registry, coverage of which is virtually complete for solid tumors.7 Cancer registry data were supplemented by active follow-up from databases of hospitals in the study area. Case ascertainment through hospitals may have been incomplete for the latest follow-up years, due to the fact that some of the subjects have moved outside the study area. This is, however, probably nonselective, i.e., similar in both trial arms, and hence unlikely to affect the results.
Because of overdiagnosis of prostate cancer at screening, we estimated sensitivity based on the incidence method, i.e., by comparing interval cancer incidence in the screening and the control arms.4, 5 This approach avoids the artificial inflation of sensitivity due to detection of indolent cases that cannot be addressed when using yield of screening. However, we did not exclude interval cases that were likely to be indolent, e.g., detected at cystoprostatectomy or at stage T1a in either trial arm. Assuming that the randomised screening arms do not differ in this respect, this may cause some nondifferential misclassification that is likely to underestimate sensitivity.
Generally, test sensitivity is calculated by comparison of men actually screened with those in the control arm. This comparison is asymmetric in the sense that nonparticipants are excluded from the screening arm. To evaluate this, we also conducted an alternative analysis with correction for the difference between attendees and nonattendees, but results were essentially unchanged because the incidence among nonattendees was similar to that in the reference group.
Test sensitivity of PSA in the screening context was 0.87–0.89. It decreased slowly with time since screening, but the interval cancer incidence remained well below the rates in the control arm throughout the 4-year screening interval. Our findings are comparable with a sensitivity estimate of 81% for a cut-off level of 3 ng/ml obtained in a cross-sectional study of 1,002 men aged 45–80.8 A slightly higher estimate was obtained in the American Cancer Society prostate cancer detection project.9 Another analysis with a similar design but using patients from primary care rather than a referral centre, gave a sensitivity of 86% for PSA with a cut-off of 4 ng/ml.10 However, the earlier analyses had important limitations as they were based on a cross-sectional design with biopsies of all men. Hence, they probably included substantial numbers of indolent cases and excluded cancers detected subsequently among biopsy-negative men. Results from serum bank studies are probably less prone to bias and have shown very high test sensitivity based on elevated serum PSA levels 5–10 years prior to clinical diagnosis of prostate cancer.11, 12, 13, 14 A Finnish serum bank study showed a sensitivity of 85% for the 5 years following sample drawing.12 In a simulation study based on the Dutch volunteer-based screening trial, it was estimated that screening with a 4-year interval would have 70% sensitivity.15 Yet, the sensitivity measure was not specified and likely referred to episode sensitivity.
The ability of a test to predict prostate cancer risk obviously decreases with time, and sensitivity is highest for short screening intervals. Our findings suggest that acceptable sensitivity could be achieved even with a rescreening interval longer than 4 years. However, the optimal screening interval remains unknown as it also depends on whether the cases detected at subsequent screening are still curable.
In our screening protocol, men with a PSA of 3.0–3.9 were subjected to an ancillary test (either DRE or free/total PSA ratio). Only 5 interval cancers were detected among 811 men with a PSA of 3.0–3.9 ng/ml and a negative ancillary test compared to 19 cases among almost 18,000 with PSA <3 ng/ml. The fact that we were able to achieve almost similar sensitivity for both cut-off levels (0.87 vs. 0.89) is very encouraging as it suggests minimal loss of prostate cancer cases with PSA levels of 3.0–3.9 ng/ml due to patient selection for diagnostic examinations on the basis of free-to-total PSA ratio.
A possible reduction in prostate cancer mortality achieved through screening requires early detection and successful treatment of those screen-detected cancers at a curable stage that in the absence of screening would develop into lethal disease. Our results indicate that only a small minority of cancers are missed at screening as only 11–13% of the expected cancers were false-negative at screening. However, definitive evaluation of prostate cancer screening must be based on analysis of mortality in the screening and control arms. Yet, as the median interval from PSA increase to death has been as long as 17 years,9 meaningful results may not be available before 10–15 years of follow-up.
In conclusion, early results from the Finnish prostate cancer screening trial indicate that PSA is a sensitive tumour marker and a valid screening test in the context of population-based prostate cancer screening.
We thank Dr. T. Nummi for statistical consultations, the clinics of the Cancer Society of Pirkanmaa and the Cancer Society of Finland as well as Aleksin Lääkäriasema for blood sample drawing, Mr. P. Hiedanpää and Mr. J. Koivuniemi for programming and data management, Ms. H. Brugnoli for data entry and Ms. M. Heikkilä for secretarial assistance.