Young Age on Starting Prostate-specific Antigen Testing Is Associated with a Greater Reduction in Prostate Cancer Mortality: 24-Year Follow-up for the Göteborg Randomized Population-based Prostate Cancer Screening Trial

Abstract

Background:

The risk of death from prostate cancer (PC) depends on age, but the age at which to start prostate-specific antigen (PSA) screening remains uncertain.

Objective:

To study the relationship between risk reduction for PC mortality and age at first PSA screening.

Design, setting, and participants:

The randomized Göteborg-1 trial invited men for biennial PSA screening between the ages of 50 and 70 yr (screening, n = 10 000) or no invitation but exposure to opportunistic PSA testing (control, n = 10 000).

Intervention:

Regular versus opportunistic PSA screening.

Outcome measurements and statistical analysis:

We modeled the nonlinear association between starting age and the absolute risk reduction in PC mortality in three settings: (1) intention to screen (randomized arms); (2) historical control (screening group and 1990–1994 registry data); and (3) attendees only (screening attendees and matched controls). We tested whether the effect of screening on PC mortality depends on the age at starting screening by comparing survival models with and without an interaction between trial arm and age (intention-to-screen and attendees only).

Results and limitations:

Younger age on starting PSA testing was associated with a greater reduction in PC mortality. Starting screening at age 55 yr approximately halved the risk of PC death compared to first PSA at age 60 yr. The test of association between starting age and the effect of screening on PC mortality was slightly greater than the conventional level of statistical significance (p = 0.052) for the entire cohort, and statistically significant among attendees (p = 0.002). This study is limited by the low number of disease-specific deaths for men starting screening before age 55 yr and the difficulty in discriminating between the effect of starting age and screening duration.

Conclusions:

Given that prior screening trials included men aged up to 70 yr on starting screening, our results suggest that the effect size reported in prior trials underestimates that of currently recommended programs starting at age 50–55 yr.

Patient summary:

In this study from the Göteborg-1 trial, we looked at the effect of prostate-specific antigen (PSA) screening in reducing men’s risk of dying from prostate cancer given the age at which they begin testing. Starting at a younger age reduced the risk of prostate cancer death by a greater amount. We recommend that PSA screening should start no later than at age 55 yr.

1. Introduction

The risks of incident prostate cancer and death from prostate cancer are strongly dependent on age, but the optimal age at which to initiate prostate-specific antigen (PSA) screening for prostate cancer is unclear. Current guidelines, all of which advocate shared decision-making, differ in their recommendations regarding the age at which to start screening [1]. For example, the European Association of Urology [2] suggests offering PSA testing starting at age 40–50 yr; the Memorial Sloan Kettering Cancer Center recommendations [3] and the National Comprehensive Cancer Network guidelines [4] suggest obtaining baseline PSA at age 45 yr; the American Cancer Society [5] and the American College of Physicians [6] recommend starting at 50 yr; and the US Preventive Services Task Force [7] and the American Urological Association [8] recommend 55 yr as the starting age. The most conservative guideline groups rely on level 1 evidence and suggest starting at age 55 yr because that was the starting age in the core age group in the ERSPC trial [9]. Other guideline groups also include observational studies and computer simulation modeling studies in the evidence synthesis that informs their health policy recommendations, with collective evidence suggesting benefits to starting PSA screening before the age of 55 yr [10–12]. The conflicting guidelines create confusion among primary care physicians and urologists regarding at what age to recommend that men begin PSA screening. Evidence from randomized trials comparing different starting ages is lacking and long-term follow-up from ongoing trials is awaited. In 2010, we reported the first results from the Göteborg-1 randomized population-based prostate cancer screening trial, demonstrating a 44% reduction in prostate cancer mortality at 14 yr of follow-up in favor of PSA screening [13]. With extended follow-up to 22 yr, the mortality reduction is sustained, with a relative risk reduction of 29% [14].

The Göteborg-1 trial is unique in that it started to invite a randomly selected group of 10 000 men aged 50–64 yr inclusive to PSA-based screening in 1995; the men were reinvited to screening every 2 yr up to the age of 70 yr. We sought to study the effects of PSA screening on prostate cancer mortality by age on starting screening, hypothesizing that the effect is greater in younger age groups as more men will be diagnosed at an earlier and curable stage owing to the stage shift induced by screening.

2. Patients and methods

The Göteborg-1 trial (Controlled Clinical Trials number ISRCTN54449243) has been described in detail elsewhere [13]. In brief, the study started in 1995, randomizing 10 000 men aged 50–64 yr to invitation to participate in biennial PSA testing and 10 000 men to a control group. Men who emigrated or died immediately before randomization were excluded. In contrast to previous reports from the Göteborg-1 trial [13], this study includes men with prevalent cancers at randomization, but three men who were younger than 50 yr or older than 64 yr at randomization were excluded, leaving 9972 men in the screening group and 9974 men in the control group. The age at last invitation to screening was 67–71 yr (average 69). The biopsy threshold was initially PSA ≥3.4 ng/ml and later changed to PSA ≥2.54 ng/ml for assay harmonization, as previously reported [13]. Follow-up was complete up to December 31, 2018 (24 yr). Information on prostate cancer incidence, vital status, and emigration was obtained via linkage of the study database to the Swedish Cancer Registry and the Swedish Population Registry. The causes of death for every man diagnosed with prostate cancer was ascertained by an independent committee blinded to the randomization group [15]. The ethical review committee at Göteborg University approved the study in 1994.

2.1. Statistical analysis

We sought to estimate the effect of a biennial PSA screening program (with stopping age of ~70 yr) on reducing the risk of prostate cancer mortality in relation to age at starting screening between the ages of 50 and 64 yr. Although this is a large study with long follow-up, the number of cancer-specific deaths per age is relatively low, in particular among men aged 50–55 yr at study start, who were aged 74–79 yr at 24-yr follow-up. Therefore, an attempt to determine the optimal starting age or to test, for instance, whether it is more beneficial to start screening at age 50 vs 55 yr is not feasible. Instead, our approach was to test the overall hypothesis of an association between starting age and the effect of screening on cancer-specific mortality, and to use nonlinear modeling to obtain a suggested suitable starting age, as stated in a prespecified analysis plan. A detailed description of the statistical methods is provided in the Supplementary material. A brief description follows here.

2.1.1. Hypothesis testing

The primary analysis was performed according to the intention to screen (ITS) principle. To test the effect of age at starting screening on the risk of prostate cancer death, we evaluated whether a survival model with only age on starting screening and group (screening or control) as covariates was improved by adding an interaction term between age and group, where the effect of age was assumed to be log-linear. The same test was performed for attendees. A p value ≤0.05 was considered significant.

2.1.2. Modeling

The survival models have three weaknesses: (1) the age at the end of follow-up depends on the starting age, and hence biases how the “true” risk of dying from prostate cancer depends on the starting age; (2) the risk of dying before randomization also depends on the starting age, and might cause bias; and (3) the assumption of a log-linear effect of the starting age on the hazard ratio. Therefore, exploratory modeling of the association between the risk reduction and the starting age was performed that addressed these issues as follows: (1) follow-up until the same age rather than the same number of years; (2) the risk of dying/having PC before randomization was estimated from registry data and combined with the study data; and (3) nonlinear splines were fitted to the risk differences. The modeling was performed in three different settings:

1. ITS setting (comparison of the randomized arms);

2. Historical control setting (screening group versus 1990–1994 registry data before the PSA era); and

3. Attendee setting (screening attendees versus matched controls).

3. Results

Prostate cancer diagnoses and deaths for the entire cohort and the two subgroups used for the modeling—men randomized at age 50–64 yr (followed to age 74 yr, inclusive) or at age 53–64 yr (followed to age 77 yr, inclusive)—are presented in Table 1. The hypothesis of an association between the starting age and the effect on cancer-specific mortality did not meet the conventional level of statistical significance (p = 0.052) when applied to the entire cohort (ITS), but was statistically significant when restricted to attendees only (p = 0.002).

Table 1 –

Number of men, PC diagnoses, PC deaths, and other-cause deaths at 24 yr of follow-up

Group	Trial arm	Men (n)	PC diagnoses (n)	PC deaths (n)	Other-cause deaths (n)
All men	Control a	9974	1238	187	4123
All men	Screening a	9972	1597	136	4189
Subgroup 1	Control	9974	952	93	2272
Subgroup 1	Screening	9972	1390	64	2331
Subgroup 2	Control	7337	823	109	2174
Subgroup 2	Screening	7371	1067	77	2271

PC = prostate cancer; subgroup 1 = men randomized at age 50–64 yr (followed to age 74 yr); subgroup 2 = men randomized at age 53–64 yr (followed to age 77 yr).

^aAmong the 9974 men in the control group, 27 had a prevalent PC diagnosis at the time of randomization, eight of whom subsequently died of PC and 12 of whom died from other causes. Among the 9972 men in the screening group, 28 men had a prevalent PC diagnosis at the time of randomization, 10 of whom subsequently died of PC and 11 of whom died from other causes.

The ITS modeling setting showed that younger age at the start of screening was associated with a greater reduction in prostate cancer mortality (Fig. 1A,B). To obtain a better effect of screening, Figure 1 suggests that screening should start no later than at age 55 yr, since the effect decreases more rapidly for each year after that. Between the ages of 55 and 60 yr, there is an approximately linear decrease in effect, and after age 60 yr the effect is considerably less.

>Fig. 1 –

Absolute risk reduction in PC mortality by age on starting screening. Crosses represent the absolute risk reduction in each year. Solid lines represent the nonlinear association between age at randomization and the absolute risk reduction in PC mortality using restricted cubic splines with a monotonicity constraint. Dashed lines represent the 95% bootstrap confidence intervals (n = 10 000 bootstrap samples). PC = prostate cancer.

In the historical control setting, comparison of the group invited to screening to historical prostate cancer mortality data from 1990–1994, when there was little exposure to opportunistic PSA testing in this population, revealed a greater reduction in prostate cancer mortality. Figure 1C indicates an advantage by starting screening before the age of 55 yr, though the advantage in starting at 50 versus 55 yr is smaller than that in starting at 55 versus 60 yr (Fig. 1C,D).

Stage, grade, and PSA distributions among men diagnosed with prostate cancer by study arm and age group are shown in Supplementary Table 3 for men diagnosed in the first screening round. For men diagnosed at the first screening invitation, those aged 60–64 yr had more aggressive tumors, whereas the distributions are fairly similar for ages 50–54 and 55–59 yr, suggesting that starting screening after the age of 60 yr may be too late.

In the attendee setting, in which the analysis was restricted to men who attended the first screening round in 1995–1996, there were larger effect sizes from screening in comparison to the ITS setting (Fig. 1E,F).

Absolute risk reduction (aRR) results are shown in Supplementary Table 1. Starting screening at 55 versus 60 yr approximately halves the risk of prostate cancer death, that is, it approximately doubles the effect size of screening in all three settings (eg, aRR 5.0%, 95% confidence interval [CI] −8.1% to −1.5%, vs 2.4%, 95% CI −5.6% to 1.4% for the ITS setting). Starting screening at 50 versus 55 yr increases the effect by nearly a quarter in the historical control setting, while the effect in the ITS and attendee settings is smaller. The number needed to invite to screening and the number needed to diagnose to save one death from prostate cancer before a certain age were estimated using the ITS modeling approach for prostate cancer deaths (Supplementary Figs 1 and 2, Supplementary Table 2).

4. Discussion

We undertook the present study to assess the effects of age at first screen using data from the Göteborg-1 trial. For attendees we saw evidence that starting screening sooner reduced prostate cancer mortality more than starting later (p = 0.002); however, for the entire cohort, the effect did not meet the conventional level of statistical significance (p = 0.052). Such an effect has a clear biological rationale given that prostate cancer is generally a rather slow-growing disease: a proportion of men who start PSA screening at 60 yr will have incurable disease that would have been detectable and curable had they had an earlier PSA test. We also found that increases in effect size with age were nonlinear, with a much bigger difference between 55 and 60 yr—a twofold difference in aRR—compared to the difference between 50 and 55 yr.

A relationship between the starting age and the screening effect size has also been observed in modeling studies. Hendrix et al [16] found that a risk-stratified prostate cancer screening strategy was cost-effective compared to biennial screening starting at 55 yr but not at 45 yr. In a microsimulation study of 230 scenarios, Getaneh et al [17] found the most optimal strategy to be screening with 3-year intervals between the ages of 55 and 64 yr, whereas shifting the starting age to 50–55 yr was less desirable from a cost-effectiveness perspective (fewer life years and quality-adjusted life years gained and lower prostate cancer mortality). Our finding is also indirectly supported by a consecutive series with matched prostate biopsy and radical prostatectomy specimens in which there was no statistically significant difference in the rate of high-risk prostate cancer between men aged ≤55 yr and men aged >55 yr [18]. Our results have important implications for interpretation of current screening trials for prostate cancer, such as the ERSPC and CAP studies. The mean age at screening start was 61 yr in ERSPC [9] and 59 yr in CAP [19]. At 16 yr of follow-up in ERSPC, regular PSA screening every 2–4 yr was associated with a significant reduction in prostate cancer mortality in comparison to the control group (RR 0.80, 95% CI 0.72–0.89; p < 0.001) [20]. After median follow-up of 10 yr in CAP, there was no difference in prostate cancer mortality between men randomized to a single PSA test versus usual care without PSA (RR 0.9, 95% CI 0.85–1.08; p = 0.5) [19]. The question becomes what would the result of these trials have been had screening started at age 55 yr rather than at any age between 50 and 70 yr. Our analysis suggests that starting at age 60 yr reduces the effect of screening on prostate cancer mortality by approximately 50% when compared to starting at age 55 yr (Supplementary Table 1). Extrapolation of these data to the ERSPC study suggests that the relative reduction in prostate cancer mortality at 16 yr would be closer to 40% than 20%. Commenting on the potential effect of an earlier start for screening in CAP is more challenging, as this trial used a one-time PSA screening strategy and the follow-up is still short given the long natural history of prostate cancer.

Furthermore, it is important to consider the size of the benefit for the younger age group. Starting screening at age 50 yr is more beneficial than starting at age 55 yr, although smaller compared to starting at age 55 compared to 60 yr (Supplementary Table 1). This analysis does not take into account changes in life years or quality-adjusted life years when starting population-based screening at an earlier age. Furthermore, because starting screening at an earlier age (between 50 and 55 yr) is not associated with a substantial increase in prostate cancer incidence [21] (Supplementary Fig. 3)—so a cancer found via screening starting at age 50 yr would only not be found at age 55 yr if the patient died in the meantime—this means that starting earlier becomes mainly a concern regarding the number of PSA tests performed, which is a relatively minor concern in comparison to prostate cancer death and overdiagnosis. However, our analysis is associated with some degree of uncertainty. Our estimates of the effects of starting screening at ages earlier than 55 yr may be relatively imprecise given the low number of events; these men are still rather young to be at risk of dying from prostate cancer, even with 24 yr of follow-up. In the historical setting of the pre-PSA era, the effect of starting screening at an earlier age was sustained for men younger than 55 yr. One factor contributing to the difference from the ITS setting is the clustering of prostate cancer deaths in the control group among men who started screening at ages 57–58 yr. This clustering is unexpected and probably due to chance. In the ITS setting, this leads to a large effect of screening for these starting ages, driven by the control group rather than by the screening group, and is therefore not seen in the historical setting. Furthermore, even though the group sizes were large, the number of events (prostate cancer deaths) are relatively few, so it is difficult to establish an optimal starting age.

The effect of screening on prostate cancer mortality was somewhat greater when the screening group was compared to a historical control group instead of the randomized control group. The younger age groups have been exposed to opportunistic screening for a longer time period compared to the older age groups. Prostate cancer mortality has also decreased in Sweden over the past two decades, and more in the younger age groups, but not at all for men older than 85 yr [22]. The decline in prostate cancer mortality is multifactorial and is likely to be explained by a combination of PSA screening and improvements in treatments for both localized and advanced disease.

Similarly, the effect of screening on prostate cancer mortality was greater when comparing attendees to matched controls, probably because the effect was not diluted by men not attending. In a prior study by our group, we found a much greater reduction in mortality when men screened in the Göteborg-1 trial were compared to unscreened men in the Malmö Preventive Project as opposed to the control group in the trial: there was a 71% relative reduction in prostate cancer mortality at 17 yr for men starting screening at ages 50–54 yr [23]. Possible differences in therapies and treatment outcomes over time is another limitation in the comparison with the historical controls.

Prostate cancer incidence up to age 75 yr was similar between all age groups, suggesting no increase in the risk in incidence by starting to invite men at an earlier versus later age (Supplementary Fig. 3). This means that starting screening at an earlier age does not increase the risk of overdiagnosis by a substantial amount, an important consideration for screening programs aiming to balance benefits and risks. Incidence during the last three screening rounds was somewhat higher in the youngest age group (50–52 yr; Supplementary Fig. 3), possibly because opportunistic testing took off around the year 2000, which affected the youngest men most, but could also be due to chance (overlapping confidence intervals).

Our study has several strengths. First, it provides level 1 evidence of a reduction in disease-specific mortality in a large-scale PSA screening trial based on the longest follow-up available to date (24 yr). Second, because the ethical review committee in Göteborg allowed upfront randomization before obtaining consent, our study probably resembles a situation in which a regular screening program is introduced, with invitation of a larger population (effectiveness design). In addition, such randomization minimizes “healthy screenee” biases, which is a problem with trials that perform randomization after obtaining informed consent (efficacy design) [24]. The main limitation of this study is that the age on starting screening and the duration of the screening period cannot be separated. The younger age groups have also been exposed to opportunistic screening for a longer period. Furthermore, the conclusions from this study are only generalizable to the screening protocol of the Göteborg-1 trial: screening of men every 2 yr between the ages of 50 and 70 yr, with biopsy recommended when PSA >2.5–3.4 ng/ml, and with 24 yr of follow-up. Moreover, the control group in the Göteborg-1 trial has become increasingly exposed to PSA testing in more recent years [25,26]. However, the degree of contamination was low at the start of the trial in 1995 [13] and remained less than the PSA testing rate of 86% in the control group at 15 yr in the Prostate, Lung, Colorectal and Ovarian Cancer screening trial in the USA [27,28]. To circumvent this, we used a historical comparison against national registry data in one of the modeling approaches.

5. Conclusions

The effect size for PSA screening reported in randomized trials underestimates that of currently recommended screening programs starting at age 50 or 55 yr.