Risk of dying from prostate cancer in men randomized to screening

Differences between attendees and nonattendees


  • Anna Grenabo Bergdahl MD,

    Corresponding author
    1. Department of Urology, Sahlgrenska University Hospital, Goteborg, Sweden
    2. Department of Surgery, Karnsjukhuset, Skovde, Sweden
    • Department of Urology, Sahlgrenska University Hospital, 413 45 Goteborg, Sweden===

    Search for more papers by this author
    • Fax: (011) 46 31 41 56 48

  • Gunnar Aus MD, PhD,

    1. Department of Urology, Sahlgrenska University Hospital, Goteborg, Sweden
    Search for more papers by this author
  • Hans Lilja MD, PhD,

    1. Department of Clinical Laboratories, Memorial Sloan-Kettering Cancer Center, New York, New York
    2. Department of Urology, Memorial Sloan-Kettering Cancer Center, New York, New York
    3. Department of Medicine, Memorial Sloan-Kettering Cancer Center, New York, New York
    4. Department of Laboratory Medicine, Lund University, University Hospital, Malmo, Sweden
    Search for more papers by this author
  • Jonas Hugosson MD, PhD

    1. Department of Urology, Sahlgrenska University Hospital, Goteborg, Sweden
    Search for more papers by this author

  • The current study is part of the European Randomized Study of Screening for Prostate Cancer.



Although the true benefits and disadvantages of prostate cancer screening are still not known, the analysis of fatal cases is important for increasing knowledge of the effects of prostate cancer screening on mortality. Who dies from prostate cancer despite participation in a population-based prostate-specific antigen (PSA) screening program?


From the Goteborg branch of the European Randomized study of Screening for Prostate Cancer, 10,000 men randomly assigned to active PSA-screening every second year formed the basis of the present study. Prostate cancer mortality was attributed to whether the men were attendees in the screening program (attending at least once) or nonattendees.


Thirty-nine men died from prostate cancer during the first 13 years. Both overall (34% vs 13 %; P < .0001) and cancer-specific mortality (0.8% vs 0.3 %; P < .005) were found to be significantly higher among nonattendees compared with attendees. Furthermore, the majority of deaths (12 of 18) among screening attendees were in men diagnosed at first screening (prevalent cases). Only 6 deaths (including 3 interval cases) were noted among men complying with the biennial screening program.


Nonattendees in prostate cancer screening constitute a high-risk group for both death from prostate cancer and death from other causes comparable to that described in other cancer screening programs. Cancer 2009. © 2009 American Cancer Society.

It is now >20 years since prostate-specific antigen (PSA) was introduced into clinical practice.1 With PSA as a tool for the detection and management of prostate cancer, the issue of screening has become widely debated. Despite signs of favorable effects on prostate cancer mortality with extensive PSA use,2 it is still not known who would benefit from regular screening.

Only randomized trials can evaluate the potential benefits and disadvantages associated with screening. The European Randomized Study of Screening for Prostate Cancer is an ongoing multicenter study with a Swedish branch involving 20,000 patients; it was initiated in the early 1990s, and final results are awaited within the coming years.

The success or failure of a screening program depends on several factors. One barrier for success is adherence. People can be noncompliant with regard to screening activities for different reasons, and a large proportion of noncompliance may decrease screening effectiveness. To fully understand the possible effects of population-based screening programs, it is of utmost importance to understand how mortality from prostate cancer is affected by compliance with the screening procedure.

Studies of other cancer screening programs have reported associations between compliance and severity of disease. In breast cancer screening, the majority of cancer deaths occur among nonparticipants in screening. One American study revealed that only 27% of women dying from breast cancer were participants in routine screening in Rhode Island, where 84% receive regular mammograms. Consequently, 73% of women dying from breast cancer were among the 16% who were nonattendees in screening.3 Another breast cancer study demonstrated that the majority (82%) of fatal cases were detected in an unscreened population, and among lethal cases detected at screening, most women were detected at their first screening.4 In cervical cancer screening, a lack of Papanicolaou testing has been implicated as the most common associating factor in the development of invasive cancer.5-7

The objective of the current study was to describe the intermediate-term mortality (from prostate cancer and overall) in men who were randomly selected to biennial PSA screening, comparing responders with nonresponders.


Patient Population and Screening Algorithm

The screening arm of the Swedish branch of the European Randomized Study of Screening for Prostate Cancer forms the basis of the following report. The algorithm of the trial is presented in Figure 1. Of all men in Goteborg born between January 1, 1930 and December 31, 1944 (n = 32,298), 20,000 were randomized to the study as of December 31, 1994. Men with prevalent cancer at time of randomization (n = 55) were excluded. Randomization was conducted before informed consent was obtained (upfront randomization). Therefore, the men randomized to the screening arm (n = 9972) form a representative sample of approximately one-third of the men in the actual age group living in Goteborg at this time.

Figure 1.

A Consolidated Standards Of Reporting Trials (CONSORT) diagram shows the screening algorithm and detected prostate cancer (PC) cases in the screening arm. PSA indicates prostate-specific antigen.

The first screen took place between January 1995 and December 1996. Thereafter, repeated invitations to screening were sent every second year to all men in the screening arm, so that to date, a person could have been invited a maximum of 7 times. A PSA value exceeding the cutoff level was regarded as a positive screening test and led to invitations to further workup, including digital rectal examination, transrectal ultrasound, and laterally directed sextant biopsies. The PSA cutoff was originally set at 3 ng/mL, but because of calibration issues regarding the PSA assay (Prostatus Total/Free PSA-Assay from Perkin-Elmer [Turku, Finland]), the actual cutoff values used differed slightly from the target value. On the basis of the current World Health Organization PSA calibration standard 96/670, the actual PSA cutoff was 3.4 ng/mL during 1995 to 1998 (screening rounds 1-2), 2.9 ng/mL during 1999 to 2004 (rounds 3-5), and 2.5 ng/mL from 2005 onward (rounds 6-7). A PSA value below the cutoff did not lead to any further examination but a reinvitation was extended 2 years later. An upper age limit was used for invitation to screening; during screening round 4, men born between 1930 and 1931 were no longer invited, and during screening round 5, men born between 1932 and 1933 were not invited, and so forth. The average age of the men invited for their last time was 69.0 years.

The current series focused on the screening group only and concerned crude and prostate cancer specific mortality. Comparison with the control group therefore was not performed.

Approximately 4 times yearly, the cohort was matched against the Regional Cancer Register and the National Population Register to identify all cancers diagnosed within as well as outside the screening program and deaths within the cohort. All cancer cases and deaths until December 31, 2007 were included in this analysis.

Tumor stage/grade, treatment received, and mortality data were prospectively registered or retrieved from medical records and death certificates. A separate cause of death committee was appointed to evaluate cases found to be doubtful regarding cause of death.

Grouping of Cancer Cases

All biopsy specimens were analyzed and graded according to the Gleason grading system.8 Tumors were staged and classified according to the 1997 International Union Against Cancer staging system.9 Patients diagnosed with prostate cancer were categorized into 3 risk groups based on the classification presented by D'Amico et al10 (ie, low-risk patients [stage T1c-2a disease, PSA level ≤10 ng/mL, or Gleason score ≤6], intermediate-risk patients [stage T2b disease, PSA level of 10-20 ng/mL, or Gleason score 7], and high-risk patients [stage T2c disease, PSA level >20 ng/mL, or Gleason score ≥8]), but with N1 and M1 cases added to the high-risk group.

All men were consecutively classified into attendees/nonattendees. Men who had attended at least once were regarded as attendees, and those who had never attended were labeled as nonattendees. On the basis of participation and compliance to the recommended diagnostic procedures, further subgrouping was performed (Table 1). Data regarding the deceased men were reported in relation to these subgroups to better describe men who died of prostate cancer despite the finding that they were enrolled in a prostate cancer screening program.

Table 1. Subgroups of Cancers Detected in the Screening Arm of the Study
  1. PSA indicates prostate-specific antigen.

Complete on-protocol attendeeParticipated in the screening program according to the protocol (PSA test at 2-y intervals and biopsy if PSA level >3 ng/mL).
Interval cancerParticipated in the screening program but diagnosed during the 2-y screening interval.
>2-y intervalParticipated at irregular intervals >2 y.
High PSA, no biopsyRefused biopsy despite a PSA level >3 ng/mL but were later diagnosed with prostate cancer.
After-study cancer (attendee)Diagnosed >2 y after last screening visit (after screening had stopped) but had previously attended.
NonattendeeDid not participate at all.
After-study cancer (nonattendee)Diagnosed >2 y after last screening visit (after screening had stopped) and was previously a nonattendee.

Statistical Analysis

Cumulative survival plots were calculated with Kaplan-Meier estimates. Men who died of unrelated causes or were lost to follow-up were censored in the cancer-specific survival analysis. Differences between groups were evaluated with the log-rank test. The Cox regression model was also used to evaluate differences in cumulative risk for prostate cancer death and to calculate the hazards ratio (HR).


Cancer Cases and Survival

During a maximum follow-up of 13 years (mean 12.0 years; range, 0-13.0 years) a total of 1076 prostate cancers were diagnosed in the screening arm of the study. The correlation with participation in the program is depicted in Figure 1. Survival analyses revealed that the risk of dying from any cause among nonattendees (34%; 816 of 2394 men) was close to 3-fold higher compared with that among the attendees (13%; 962 of 7578 men) (P < .0001) (Fig. 2). Prostate cancer-specific mortality was also found to be significantly higher in nonattendees compared with attendees. The cumulative prostate cancer mortality at 13 years was 0.8% among nonattendees compared with 0.3% among attendees, resulting in an HR of 0.375 (95% confidence interval [95% CI], 0.198-0.711; P < .005) (Fig. 3).

Figure 2.

Overall survival among screening attendees (n = 7578) and nonattendees (n = 2394) is shown.

Figure 3.

Cancer-specific survival among screening attendees and nonattendees is shown.

Subgroup Analysis of the Deceased

Of 1778 deaths from all causes within the screening arm, 39 (2.2%) were because of prostate cancer. The baseline characteristics, risk grouping, and survival of those 39 men are presented in Table 2. The majority of men dying from prostate cancer (31 of 39 men) had high-risk disease at the time of diagnosis.

Table 2. Characteristics of Men Dying From Prostate Cancer in the Screening Arm
Complete AttendeeInterval>2-Year IntervalHigh PSA, No BiopsyAfter Study (Attendee)All AttendeesNonattendeeAfter Study (Nonattendee)All Nonattendees
  1. PSA indicates prostate-specific antigen.

Median PSA at diagnosis, ng/mL (range)10.7 (3.6-210)7.6 (5.1-23.6)92.41404 (8.6-2800)15.6 (10-21)10.745 (3.3-2100)720 (340-1100)66.524
Median age at diagnosis, y (range)63 (61-69)68 (67-69)6361 (56-66)72.5 (72-73)6462 (54-68)70.5 (69-72)6363
Median age at time of death, y (range)70 (61-74)68 (68-72)6865.5 (63-68)74.5 (73-76)7067.5 (55-74)74.5 (73-76)6869
Biopsy Gleason score
T classification
 T12114226 (15%)
 T27211115516 (41%)
 T3-4611872917 (44%)
N classification
 N08193312 (31%)
 N133114 (10%)
 NX43112111021223 (59%)
M classification
 M011111144418 (46%)
 M123117821017 (44%)
 MX22224 (10%)
Risk group
 Low213114 (10%)
 Intermediate3144 (10%)
 High103111161321531 (80%)
Median time from diagnosis to death, y7.

Twenty-three of the 39 deaths occurred in attendees to screening. Fifteen of those men had followed the screening program according to the protocol, but 12 had disease detected at the first screen (prevalent cases). Three deaths among attendees were related to interval cancers. Another 3 deaths among attendees occurred in men who had failed to follow the screening protocol; 1 had attended twice but with a screening interval of 4.5 years, and 2 men refused biopsy despite indication. Two attendees who died had fallen out of the study because of age but were diagnosed with aggressive prostate cancer at ages 72 years and 73 years, respectively.

Sixteen prostate cancer deaths occurred in the group of screening nonattendees. Two of those were men who had fallen out of the study because of age after being offered 3 rounds of potential screening opportunities.


Radical retropubic prostatectomy was chosen in 9 of the 39 cases. Eight of those were in the 23 attendees. One attendee was initially followed with watchful waiting. Six patients (3 attendees and 3 nonattendees) received radiotherapy and 2 (1 attendee and 1 nonattendee) were treated primarily with cryotherapy. The remaining patients (21 of 39 patients; 54%) received immediate hormonal therapy as their primary treatment.



The results of the current study demonstrated that failing to attend prostate cancer screening indicates a higher all-cause mortality. This is in keeping with earlier studies demonstrating that people who participate in screening programs are generally more health conscious than nonattendees (healthy screenee effect).11, 12 For example, smoking has been shown to be negatively associated with all types of cancer screening adherence.13 According to a study on mammographic breast cancer screening in Toronto, women with diabetes were less likely to have an examination during a 2-year period than were women without diabetes, after adjusting for age and other covariates.14 In this report, we did not investigate the comorbidity among the men further, but concluded that nonattendees have a higher mortality from all causes, which fits very well with a potentially unsound lifestyle and comorbidity. It also may reflect the simple fact that men who already have been diagnosed with an illness realize that their potential gain from participating in prostate cancer screening is low and thus do not attend.

Nonattendees also had a significantly higher risk of prostate cancer death (Fig. 3). This difference between attendees and nonattendees may be interpreted as a favorable effect of screening, but such an interpretation is not unambiguous. There is an obvious self-selection bias in the screening group in terms of being an attendee or nonattendee. In analogy with the abovementioned finding that nonattendees seek medical attention at later stages, one may assume that this group already had a larger number of more advanced undetected cancers at the time of randomization, resulting in a shorter lead time for those cancers. This explanation is supported by the finding that at 2 years after randomization, this group already had begun to demonstrate a higher risk for prostate cancer death (Fig. 3). It is unlikely that treatment effects will manifest themselves that early. Nonattendees also had more advanced cancers at the time of diagnosis (higher median PSA, 15 of 16 belonged to the high-risk group; see Table 2).

The only way to assess the effect of screening on mortality is through an intention-to-treat analysis of the complete randomized screening study, including both the intervention and the control arm. A large difference in mortality between attendees and nonattendees such as what was observed in this series may carry the risk of underestimating the true screening effect in the individual. The intention-to-treat analysis will reflect the population effects from screening, not the effect for an individual considering PSA testing. To evaluate the efficacy of screening, one may adjust for attendance, as was proposed by Cuzick et al.15

Our finding of higher mortality rate among nonattendees is in keeping with that of Labrie et al, who reported a risk ratio of 0.36 (95% CI, 0.19-0.65) for men attending a screening program compared with nonattendees.16


The aim of the screening program is to prevent deaths from prostate cancer. Therefore, the number of deaths among those men actually screened is of certain interest and reflects the efficacy of the screening program. The subgrouping of attendees (n = 23) revealed that only 15 deaths were among “complete attendees” (as intended in the algorithm), and 3 were among noncompliant subjects, as depicted in Table 2. Twelve of the 15 deaths occurred among complete attendees whose disease was detected at the time of the first screen. Other studies of fatal outcomes in cancer screening have regarded patients presenting with disease at the time of first screening (prevalent cases) as unscreened.17 If applying this rule, the (previously) unscreened patients with cancer detected at screening (n = 12) together with the nonattendees (n = 16) form a group that comprises 72% of all men who died from prostate cancer in the intervention group (ie, active screening). Only 6 men participating in regular screening died from prostate cancer (3 deaths from prostate cancer plus 3 interval cancers), prevalent cases excluded.

The finding that most deaths among attendees occurred in those diagnosed at the time of first screening demonstrates that some of these cancers were already too advanced when they were first detected. The majority (18 of 23) of the attendees who later died from prostate cancer were aged >60 years when they were first invited to PSA testing. In the majority of screening programs, men are recommended to initiate PSA testing at ages 50 to 55 years, and the risk of harboring advanced cancers at that age is most likely much lower. Starting later may, according to the results of the current study, be hazardous, with several men already found to have advanced disease at the time of their first screen. However, to our knowledge, there is still a lack of proof that earlier diagnosis will prevent deaths.

Interval Cancers

The rate of interval cancers was low in this study, as reported earlier.18 In the entire screening arm, there have been 45 interval cancers of a total of 1076 prostate cancers diagnosed at the time of last follow-up (Fig. 1). Interval cancers contributed 3 of the 39 fatal prostate cancer cases (7.7%). This rate is low compared with corresponding figures in breast cancer screening, with a 15% interval cancer rate in women dying from breast cancer.19 In addition, fewer interval cancers in prostate cancer screening are lethal compared with those in breast cancer screening.20 A large proportion of the prostate cancers detected as interval cancers are most likely not diagnosed because of clinical presentation between screening rounds, but rather because of unorganized PSA testing in asymptomatic men outside the program.

Taking into account that during 13 years of follow-up there were only 3 lethal interval cancers diagnosed and another 3 deaths among men who had complete attendance in the screening program and in whom cancer was not detected during the prevalence screen, one might conclude that the efficacy of the screening algorithm used, with biennial PSA testing, appears acceptable.


The results of the current study demonstrate a significantly higher risk of death among nonattendees compared with attendees in the screening arm of the study. Nonattendees are at a higher risk for death from any cause as well as death from prostate cancer.

Among those men who were classified as attendees and still died of prostate cancer, the majority either had their disease detected at the time of the first screening round (prevalent cases) or were noncompliant with the 2-year screening algorithm/biopsy recommendations. Interval cancers contributed few deaths compared with other cancers found on screening, suggesting that a 2-year screening interval sufficiently detects even rapidly growing tumors in the prostate.

Noncompliance is a limitation for screening effectiveness. A large number of nonattendees may reduce the screening efficacy of the current and other screening studies that randomize men upfront before informed consent is obtained.

The finding that only 3 deaths from prostate cancer (plus 3 interval cancer deaths) occurred after 13 years among men who had undergone repeated PSA screening is encouraging, but only results from the randomized study can determine whether screening can reduce prostate cancer mortality.


We thank Helén Ahlgren and Maria Nyberg for providing assistance in data management for the study.

Conflict of Interest Disclosures

Supported by grants from the Swedish Cancer Society (Grant 3792-B96-01XAB) and the Research Council (Medicine) in Sweden (Grant 20,095). The study was also supported by grants from Wallac Oy, Turku, Finland; Schering Plough, Sweden; and Abbott Pharmaceuticals, Sweden. None of the funding sources has had access to the data or been involved in the data collection, data management, or writing of the article.

Dr. Lilja holds the patent for free prostate-specific antigen and human kallikrein 2 assays.