Cumulative number of cancers in a screened cohort: For screen-detected cancers in volunteers, the largest source of age-specific data is that of Richie et al. and it is on these data that the analysis is based. They reported on the outcome of an initial screening in 6630 volunteers in the USA, all of whom had no previous history of prostate cancer, prostatitis or urinary tract infection (UTI). Individuals with either an elevated serum PSA value (> 4 ng/mL) or suspicious DRE underwent biopsy guided by transrectal ultrasound (TRUS) imaging, although a suspicious finding on TRUS itself was not used to initiate biopsy. For the age categories 50–59, 60–69, 70–79 and ≥ 80 years the percentage of individuals with cancer detected were 2.9%, 6.0%, 10.7%, and 14%, respectively. If the criteria for biopsy were solely on the basis of the elevated PSA value the corresponding detection rates would have been 2.0%, 4.9%, 9.0% and 10.1%, respectively, and 4.6% over all ages (mean age 62.8 years).
Several other studies reported the cancer detection rates in volunteers. Rietbergen et al. reported on 10 449 volunteers in the Netherlands (median age 63.0 years) where biopsy was initiated at a PSA of ≥ 4.0 ng/mL, or a suspicious DRE, or suspicious TRUS. The cancer detection rate was 4.8%. Labrie et al. reported on 1002 volunteers in Canada, 4.1% of whom had positive biopsies after a PSA value of ≥ 4.0 ng/mL. Määttänen et al. reported on 5053 screened volunteers in Finland and those with a PSA openface> 4.0 ng/mL had a detection rate of 2.1%. Horninger et al. reported on 1704 Austrian blood donors in the age range 50–65 years and, on the basis of a PSA openface> 4.0 mg/mL or a suspicious DRE, had a cancer detection rate of 3.4%. Clearly there are differences among these results and these may be a consequence of differences in the recruitment procedure of subjects, their age, the efficiency of the biopsy procedure, and the prevalence in different countries.
Now consider a cohort of 100 000 55-year-old men, initially asymptomatic, who are repeatedly screened from a start date in 1991–92 (to accord with the date of the detection rates reported by Richie et al.). Although there are some published data  on the results of serial PSA measurements, these only extend to 4 years and are insufficient for the present purposes. In Table 1, rows A, B and C assume that cases accrue through conventional presentation, i.e. without the benefit of screening of asymptomatic men, and are derived from estimated incidence rates for 1992. Incidence rates for prostate cancer since 1989 have increased dramatically, an effect commonly attributed to the inclusion of extra cases detected by PSA and other screening . Before this the incidence of prostate cancer in the USA was increasing at a rate of ≈ 4% per year over 1973–88 . For use in the construction of Table 1 the 1992 age-specific rates of incidence of prostate cancer under conventional presentation were assumed to be equal to those of 1988, a year essentially unaffected by the inclusion of screen-detected cases, together with a 15% increase to allow for an assumed continuing trend. In Table 1 prostate cancer deaths in the cohort would occur an average of 20 years after the start date. The risk of death from prostate cancer has been steadily increasing (≈ 1%/year over 1973–92), and so accordingly we calculated the risks from the 1992 USA age-specific prostate cancer mortality rates  together with an increase of 20%. The age-specific mortality rates for all other causes of death were taken from the 1992 values with no adjustment, and were assumed to apply equally to cases and non-cases (see later).
Next, the cumulative number of screen-detected cases was estimated, starting with the cancer detection rates (above) of Richie et al.. We consider first the fate of the cohort under the (unlikely) scenario that both the screening process and its results were secret, so that such prostate cancer cases that do occur only do so as a result of conventional clinical presentation.
From the results of Richie et al., 2.9% of screened men (i.e. 2900 of 100 000) are found to have prostate cancer at the age of 55 years. At the age of 65 imagine a second screening is undertaken on all individuals who are still asymptomatic. Remembering that this is done regardless of the results of earlier screening, a cancer detection rate of 6.0% applies, yielding 5068 cancers. However, this is an underestimate. If screening had been undertaken continuously in the years between ages 55 and 65 years it is reasonable to assume that the 2059 cases which occurred by conventional presentation would all have had yielded positive screening results. Of those who died from causes other than prostate cancer some would also have yielded positive results. These can be crudely estimated by assuming that death occurred at the mid-point of the 10-year interval, so that a mean detection rate of (2.9% + 6.0%)/2 should apply to 13467 individuals, giving 599 cancers. Thus we estimate that continuous screening would detect a total 5068 + 2059 + 599 = 7726 cases (alive and dead) by the age of 65 years; similar reasoning applies to subsequent ages.
Although the values for the total number of cases were detected under the hypothesis of secret screening, this was a purely logical step to allow the use of the published data, which relates to initial screening, to be extrapolated to continuous screening. In Table 1, row F gives the cumulative number of cases detected by continuous screening and may be compared with the numbers detected by conventional presentation, i.e. the sum of the entries in rows A, B and C. At ages 65, 75 and 85 years the ratios are 3.75, 2.10 and 1.71, respectively.
Stage and grade distribution of screen-detected disease
Table 2[6,8,13,14] shows results reported from various sources for screen-detected disease in volunteers. Two of the reports give grading by Gleason scores, two use the WHO classification system, and one classifies using both systems. Commonly, Gleason scores of < 5, 5–7 and> 7 are regarded as equivalent to well, moderate and poorly differentiated tumours, and this system is used in Table 2. However, comparison of the results, particularly those of Rietbergen et al., shows that this may be a questionable assumption.
Table 2. The stage and grade distribution of screen-detected disease
|Clinically localized, %|| 78||97|| 96||83|
|Gleason score, %|
|< 5||3.5||25|| 37||–|
|5–7|| 85||67|| 49–63||–|
|> 7||11.6||8|| 0–14||–|
Survival patterns in the era before PSA screening
For cases of prostate cancer in the era before PSA screening we require the risk of death, both from prostate cancer and from other causes, and derive this from the SEER database for cases diagnosed in the years 1983–88; 48741 cases aged ≥ 50 years were recorded and only 58 of these were lost in 9 years of follow-up. Of these cases the numbers classified as localized, regional, distant and unstaged were 28 868, 7111, 8245 and 4517; 3277 cases with extraprostatic disease had a lymph node dissection documented and we assume these were originally cases with clinically localized disease (see Methods) giving a total of 32 145, i.e. 73% of those in whom staging information is available.
The mean (se) ‘non-prostate-cancer-specific’ survival at 10 years for all 48 741 cases was 50.7 (0.3)%, as against an ‘expected’ value of 49.3% for a matched group drawn from the general population, and based on the USA life tables. The latter value includes a small contribution from deaths from prostate cancer and if corrected for, the observed and expected values are very similar. This was also the case when the analysis was stratified by age, stage or grade of disease. Thus for an individual with prostate cancer, the risk of dying from causes other than prostate cancer is essentially the same as that of a matched individual with no prostate cancer; this was assumed in constructing Table 1 and in the subsequent analysis.
The prostate-cancer-specific survival at 10 years of all 48 741 cases was 66.8 (0.25)%; for those aged 50–69 (18 986) and ≥ 70 (29 755) the values were 71.3 (0.36)% and 62.8 (0.36)%. Analysis of the SEER database for prostate cancer-specific survival for the 28 868 cases actually recorded as localized disease yielded a 10-year survival of 81.4 (0.28)%. When the extra 3277 deemed to have been clinically localized (see above) were added to give 32 145 cases, the 10 year prostate cancer-specific survival was 80.1 (0.27)%. After excluding those in whom tumour grade was unknown, this group was further analysed, stratified by grade and age of diagnosis (Table 3). After further excluding those in whom surgery or radiation treatment was recorded as unknown, the results are also given in Table 3 for subsets receiving radical or conservative treatment. Radical treatment was deemed to include any form of radiation treatment, radical prostatectomy, cystectomy, cystoprostatectomy or pelvic exenteration; otherwise treatment was deemed to be conservative.
Table 3. The number of cases (N) and 10-year prostate cancer-specific survival for clinically localized disease
|Treatment/age range||N, mean (se), % survival for Grade: |
|50–69||4936, 95.0 (0.34)||4957, 84.9 (0.56)||1995, 60.7 (1.2)|
|≥ 70||7009, 88.8 (0.50)||6993, 74.5 (0.69)||3471, 55.0 (1.1)|
|All||11945, 91.9 (0.30)||11950, 79.7 (0.44)||5466, 56.1 (0.83)|
|50–69||2504, 96.2 (0.43)||953, 82.8 (1.4)||404, 36.5 (3.6)|
|≥ 70||4992, 89.4 (0.62)||3460, 70.0 (1.2)||1832, 44.5 (1.8)|
|All||7496, 92.3 (0.39)||4413, 73.9 (0.91)||2236, 42.2 (1.5)|
|50–69||1748, 95.4 (0.54)||3176, 86.6 (0.65)||1292, 68.6 (1.4)|
|≥ 70||1095, 90.6 (1.0)||2182, 80.3 (1.0)||1056, 60.8 (1.8)|
|All||2843, 93.7 (0.51)||5358, 84.3 (0.55)||2348, 65.5 (1.1)|
Prostate-cancer-specific mortality patterns in the screened cohort
The survival times of screen-detected cases, unlike those detected by conventional presentation, are unknown. Therefore, for a cohort of 55-year-old men who are subjected to repeated screening, the mortality pattern (i.e. the cumulative number of deaths from prostate cancer over subsequent ages) is hypothetical. In the following sections the mortality pattern was estimated in two different ways.
(a) The baseline mortality pattern calculated from population mortality rates, assuming no screening information: The prostate cancer-specific population mortality rates for 1992, the year assumed as the starting date for the cohort, are derived almost entirely from cases detected conventionally. Prostate cancer deaths will occur on average ≈ 20 years after the screening start date, and some adjustment for trend is necessary. Rates of prostate cancer mortality rose steadily, by ≈ 1%/year over 1973–92, and so the 1992 age-specific rates  were used together with an overall increase of 20%. The mortality rates for all other causes of death were taken from the 1992 values  with no adjustment. The resulting mortality pattern for the cohort is that summarized in row B of Table 1 and is plotted as line A in Fig. 1a,b. This represents the best estimate of the fate of the cohort in the absence of screening information. Equally it represents the fate of the cohort if the use of screening information has no life-saving effect. As such we term it the ‘baseline’ mortality pattern. If the annual 1% rise over 1973–92 of prostate cancer mortality is a reflection of diagnostic improvements, the effect may be expected to plateau. In this case the baseline mortality pattern (A) would be an overestimate.
Figure 1. Mortality patterns (cumulative number of deaths from prostate cancer in the cohort against age) calculated in various ways. a, The baseline pattern (A) is derived from a projection of current population mortality rates. (B) and (C) are patterns calculated when screen-detected cases from the cohort are assumed to be subject to survival times derived from conventionally detected cases of all stages and grades. For (B) no lead time is assumed. For (C) a lead time distribution having a mean of 9 years is assumed. b, The baseline pattern (A) is as in a; other patterns are calculated when screen-detected cases from the cohort are assumed to be subject to survival times derived from conventionally detected cases with clinically localized disease. Survival times used were those for: all treatments (B), conservative treatment (C), radical treatment (D).
(b) The mortality pattern calculated from screening information and conventional survival data:
In this and the following subsections (i) and (ii) the starting point is the cumulative number of cases detected in the cohort by continuous screening, as summarized in row F of Table 1. The risk of death to cases from causes other than prostate cancer were assumed, as before, to be the same as that of the general male population in 1992. For the risk of death from prostate cancer consider the cases to be subject to survival data derived from conventionally detected cases as summarized in Table 3. The survival curves derived from the SEER database only extended to 15 years, and for the present purpose the curves were extrapolated by an exponential decay curve fitted to the last 10 years of data. The outcome of the analysis is not unduly sensitive to the shape of the survival curves beyond 15 years.
(i) Applying survival data from all stages of disease: When the screen-detected cases in the cohort are considered to be subject to the survival times from all stage disease, stratified by age (Table 3) the calculated mortality pattern is plotted as B in Fig. 1a. The resulting number of deaths would clearly exceed, by factors of 5.7, 3.1 and 2.1 at ages 65, 75 and 85 years, respectively, those from the baseline mortality pattern. Compared with conventional detection, screening generates 2–3 times as many cases as conventional detection (see earlier) and these predominantly have early disease. The assumption of survival characteristics the same as those of conventional cases results, not surprisingly, in numbers of deaths (line B) increased by roughly the same factor compared to baseline.
The lead time can be defined in various ways but for the present its operation may be considered as follows. Assume continuous screening takes place but the information is kept secret. Case detection will therefore only occur later by conventional diagnosis after a time interval defined here as the lead time. Immediately after expiry of the lead time, a case is assumed to have the prostate cancer-specific survival times of a man of the same age who has just been diagnosed conventionally. When this assumption was incorporated, a lead time having a normal distribution with a mean of 9 and sd of 6 years resulted in a mortality pattern coinciding most nearly with that of the baseline pattern, and this is plotted as C in Fig. 1a.
(ii) Applying survival data from clinically localized disease: Because nearly all screen-detected cases are of localized disease, in the absence of other information the patient is commonly advised on the basis of survival times derived from equivalent conventionally detected cases. For screen-detected cases examples of the distribution of tumour grade are given in Table 2 and for the following calculations values of 25%, 67%, and 8% are assumed for well, moderate and poorly differentiated disease, respectively.
The screen-detected cases in the cohort were first considered to be subject to conventional prostate-cancer-specific survival data for clinically localized disease (Table 3) for all treatments and with the above mix of tumour grade. The resulting mortality pattern for the cohort is plotted in Fig. 1b as line B; this clearly predicts far more deaths than the baseline mortality pattern. It may be argued that the mix of conservative and radical treatments in Table 3 is historical, derived from treatments in the 1983–88 period, and may not be appropriate for the treatment of screen-detected cases. Two additional mortality patterns are therefore plotted, based on the assumptions that the screen-detected cases are subject to the conventional survival times for clinically localized disease (Table 3) either with conservative treatment (curve C in Fig. 1b) or radical treatment (line D).
It is clear from these that no assumption of the mix of conservative and radical treatments generates a mortality pattern as low as the baseline mortality pattern (A) (i.e. that expected in the absence of screening information). This means that the prostate cancer-specific survival times measured from the date of screen-detection must be considerably longer than those of equivalent conventionally detected cases, even assuming that screening has no life-saving effect.
Thus it is tempting, following the example of the previous subsection (i), to include an allowance for lead time. However, the assumption of a lead time for localized disease only would be inappropriate as it would have no counterpart in the sequence of biological events. Cases will progress in stage and grade during the lead time. Screen-detected cases with localized disease and a particular distribution of tumour grade cannot be regarded as having a survival prospect equivalent to the expiry of a lead time followed by the survival times applicable to conventionally detected cases with localized disease and the same distribution of grade. This objection does not apply to the previous subsection as the analysis covered all stages and grades.