Baseline Prostate-Specific Antigen Level and Risk of Prostate Cancer and Prostate-Specific Mortality: Diagnosis Is Dependent on the Intensity of Investigation

Prostate cancer is the most commonly diagnosed cancer in men and a leading cause of cancer death in the United States and Europe (1, 2). The incidence of prostate cancer has increased dramatically in the past two decades due in large part to the widespread use of prostate-specific antigen (PSA) testing (1). In 2001, between 45% and 72% of White men in the United States aged ≥50 years underwent PSA testing in the preceding year (3). Annual screening for prostate cancer, using PSA testing and digital rectal examination, has been recommended by the American Cancer Society and the American Urological Association (1, 4). PSA screening is not recommended in the United Kingdom; however, the Prostate Cancer Risk Management Programme aims to provide balanced information to men who are concerned about prostate cancer to allow them to make informed choices regarding PSA testing (5).

A PSA level of 4.0 ng/mL is considered an appropriate cutoff for selecting men for prostate biopsy (6); however, lower cutoff values are often used (7). A recent report from the Prostate Cancer Prevention Trial showed that prostate cancer occurs in men with low PSA levels (8), suggesting that there is no “normal” PSA but rather a continuum of risk of prostate cancer based on an individual's PSA level (9, 10). As a consequence, there is uncertainty relating to the most appropriate level, and how intensively, to investigate low PSA levels. Early invasive investigations may result in many men undergoing unnecessary biopsy with a possible increase in the diagnosis of clinically insignificant cancers, whereas postponing prostate biopsy may result in a missed opportunity to identify and instigate early, perhaps life-saving, treatment in patients with aggressive cancers.

The rate of diagnosis of cancer in men who undergo prostate biopsy at low PSA levels is well documented (7-10). However, for men to make a fully informed choice, they must also be aware of the outcomes associated with not proceeding to immediate prostate biopsy. Although these data are not currently available, two aspects of PSA testing and subsequent investigation of men in Northern Ireland in the mid-1990s may help provide this information. Firstly, all PSA results from men in Northern Ireland as well as prostate cancer diagnoses and prostate cancer deaths are collated by the Northern Ireland Cancer Registry (NICR). Secondly, although PSA testing has been widespread in Northern Ireland since 1994, with ∼15,000 men having their first PSA test annually (6% of all men aged ≥50 years; ref. 11), the incidence of prostate cancer remained relatively unchanged until 2002 (Fig. 1A). Prostate biopsy was also not freely available in Northern Ireland in the mid-1990s, with few prostate biopsies done before 2000 (Fig. 1B). Therefore, many men underwent PSA testing and did not proceed with early invasive investigation.

To better inform clinical decision-making in this complex area, we assessed the risk of prostate cancer and prostate cancer–specific and all-cause mortality by baseline PSA level in a large population-based cohort study in a country where asymptomatic PSA screening is not recommended and during a period where, in general, men with low (<4.0 ng/mL) and moderately raised (4.0-10.0 ng/mL) PSA did not undergo routine invasive investigation.

Northern Ireland has a population of ∼820,000 men of whom 220,000 (26.6%) are over 50 years old (12). The population is stable and homogenous with little migration (0.7% annually) and 99.2% of its inhabitants are White (12). All PSA tests are done in 1 of 10 laboratories using several immunoassays (Abbott Axsym, Abbott Architect, Bayer Immuno 1, Wallac Autodelphia, Roche Elecsys), with few tests done before 1994 (11). The NICR maintains a confidential electronic database of all PSA results from these laboratories for prostate cancer surveillance purposes. Using unique identifiers (name, date of birth, and address), the NICR routinely links these PSA data to their database of incident cancers and the Registrar General's Office (Northern Ireland) database of deaths occurring within the region. The reason for PSA testing is not recorded in the NICR database, but data from an ongoing follow-up study in Northern Ireland suggest that (13), between 1994 and 1997, ∼65% of first PSA tests were done in men with lower urinary tract symptoms, whereas <20% were asymptomatic screening tests, although there is evidence that asymptomatic or opportunistic screening is increasing within the region (11).

Data were included in this study from all men who had their first PSA test between January 1, 1994 and December 31, 1998. Dates of incident prostate cancer diagnoses and dates and causes of death among cohort members, until December 31, 2003, were available from the NICR. For men with a histologic diagnosis of prostate cancer, information on tumor grade was extracted from histopathology reports. A tumor was considered high grade if the Gleason score was ≥7 or, when Gleason scoring was not available, the tumor had any poorly differentiated component. The NICR receives death data from the Registrar General's Office (Northern Ireland), which codes deaths according to internationally agreed criteria (14). Cohort members were considered to have died from prostate cancer only if this was coded as the primary cause of death.

Age at first PSA test was categorized into four bands: <50, 50 to 59, 60 to 69, and ≥70 years. The level of baseline PSA was categorized into six groups as follows: 0 to 0.99, 1.00 to 1.99, 2.00 to 2.99, 3.00 to 3.99, 4.00 to 9.99, and ≥10.0 ng/mL. As tumor grade was not available for 839 (30.4%) cases, two variables were created: one where all tumors with missing grade information were considered to be low grade and one where they were assigned as high grade.

The rates of prostate cancer diagnosis (number of incident cancers divided by person-years of risk) and diagnosis of high-grade prostate cancer were calculated for each age band and level of baseline PSA. Prostate cancer–specific and all-cause mortality rates were calculated in a similar manner. The number of men requiring follow-up for 5 years to diagnose one prostate cancer was calculated from the cancer diagnosis rate. Cox proportional hazards models were used to calculate hazard ratios (HR) for prostate cancer and high-grade prostate cancer risk associated with each category of baseline PSA level; 0 to 0.99 ng/mL was used as the reference category. Separate models were created for each age band. Cox proportional hazards models were also used to age-adjust survival analyses of prostate cancer–specific, and all-cause, mortality (all subjects) by baseline PSA category. The proportional hazards assumption of the Cox models was tested through graphical examination of the log-log plots of the variables used in the models. Internal validity of the model was tested by comparison of observed Kaplan-Meier values with the Cox-predicted estimates. The log-log plots were closely parallel and the predicted values were close to observed, indicating that the proportional hazards assumption was not violated. All statistical analyses were done using Stata (Intercooled Stata 8.0; Stata).

All patient identifiable information was removed before the research team accessed the data used in this study, and no patient contact was made during the study. Therefore, ethical approval was not sought.

In total, 68,354 men had a baseline PSA test between 1994 and 1998 and were included in the original cohort, with 2,827 (4.1%) diagnosed with prostate cancer during the observation period; 418 (0.6%) men without cancer and 10 (0.4%) with cancer were excluded from further analysis due to missing data; 15,212 (22.4%) men died during follow-up, with prostate cancer being the primary cause of death in 783 (5.1% of all deaths). The mean (SD) duration of follow-up was 6.04 (2.5) years (maximum, 10.0 years). Characteristics of the cohort by baseline PSA and age category are presented in Table 1. Mean and median baseline PSA levels were 12.3 and 1.7 ng/mL, respectively. Thirty percent of men had a baseline PSA of <1.0 ng/mL, with 10% having a level of ≥10.0 ng/mL. Mean age at baseline testing was 65.2 years, with the majority of men (66.8%) being older than 60 years and almost 40% over 70 years old. Fifty-nine men who had their first PSA test on the same day as their cancer diagnosis and 35 men who died on the same day as their first PSA test were not included in the Cox proportional hazards analyses of cancer diagnosis and mortality, respectively.

The number of prostate cancers, time to diagnosis, and proportion of high-grade cancers by baseline PSA category are presented in Table 2. The number of cancers diagnosed increased with increasing baseline PSA level. The median time to cancer diagnosis was similar in the PSA ranges of <10.0 ng/mL (3.3-4.6 years) but was markedly lower in those with a PSA of ≥10.0 ng/mL (4 months). The age-adjusted hazard curve for prostate cancer and time to diagnosis by baseline PSA level is illustrated in Fig. 2. Overall, 2,162 (76.7%) men were histologically diagnosed with prostate cancer, with 508 (18.0%) diagnoses based on clinical opinion. Thirty (1.0%) men were diagnosed based on death certificate data only, of whom 26 died from prostate cancer; 34.8% of cancers were considered high grade from histopathology reports, with this increasing to 54.6% when all cases with missing tumor grade were considered high grade (Table 2). There was a similar proportion of high-grade cancer across PSA categories in histologically confirmed cases only (P = 0.37). This pattern continued when only Gleason score 8 to 10 tumors were assessed and when cases with missing grade data were considered to be high grade, except for a baseline PSA of ≥10.0 ng/mL where a higher proportion of high-grade cancer was observed (P < 0.01; Table 2).

Sixty-five percent of men diagnosed with cancer had a baseline PSA of ≥10 ng/mL, whereas <15% had a PSA of <4.0 ng/mL. The rate of diagnosis of prostate cancer was clearly related to the level of baseline PSA (Table 3). When baseline PSA was <4.0 ng/mL, the rate of prostate cancer diagnosis in all men was 1.26 cases per 1,000 person-years [95% confidence interval (95% CI), 1.14-1.39], which equates to following 159 men for 5 years to diagnose one prostate cancer. At any age, a PSA of <2.0 ng/mL was associated with a very low rate of prostate cancer diagnosis (<2 cases per 1,000 person-years). Within age groups, the rate of diagnosis of prostate cancer and the corresponding HR increased substantially with PSA level. This increase was more evident in younger men; the HR for prostate cancer diagnosis in a man aged <50 years with a baseline PSA of 3.0 to 3.99 ng/mL (compared to a level of 0-0.99 ng/mL) was 140, whereas the corresponding ratio in a 60- to 69-year-old man was 12.2.

This pattern was also seen when the diagnosis of high-grade prostate cancer was examined (Table 4). The rate of diagnosis of high-grade cancer is shown, with all cancers in which these data were not available coded firstly as low grade and then as high grade. The actual risk of high-grade cancer will lie between these two estimates. The rate of diagnosis of high-grade prostate cancer at low PSA levels was very low; the overall rate when baseline PSA was <4.0 ng/mL was 0.64 per 1,000 person-years (95% CI, 0.56-0.73). As for all prostate cancer diagnosis, the risk of high-grade cancer increased substantially with increasing baseline PSA level (P for trend < 0.001). This increase was attenuated when cancers with missing data were assigned as high grade; however, the trend remained unchanged.

The Kaplan-Meier survival curves for prostate cancer–specific and all-cause mortality by baseline PSA are presented in Fig. 3. The rate of occurrence of prostate cancer–specific death was low (<2 cases per 1,000 person-years) at all PSA categories of <10.0 ng/mL (Table 5). Prostate cancer–specific mortality increased with increasing PSA, with the most dramatic increase seen for PSA levels of ≥10.0 ng/mL. When baseline PSA was <4.0 ng/mL, the rate of prostate cancer–specific mortality was 0.18 cases per 1,000 person-years, whereas the overall mortality was much higher at 28.71 cases per 1,000 person-years. In all baseline PSA categories, the risk of prostate cancer–specific mortality was much lower than all-cause mortality. The risk of death from all causes, compared with the lowest PSA category, was not raised in any category of baseline PSA, except when PSA was ≥10.0 ng/mL.

This is the first population-based study to investigate the association between baseline PSA levels and the risk of prostate cancer diagnosis in men who did not proceed to early investigation. As stated previously, many men in Northern Ireland in the mid-1990s underwent PSA testing and did not proceed to prostate biopsy (Fig. 1). Further, less than 5% of men in this study with a baseline PSA of 4.0 to 9.99 ng/mL were diagnosed with prostate cancer within a year (Fig. 2). It is probable that men progressed to prostate biopsy only if their PSA increased to ≥10.0 ng/mL, if they had a steadily increasing PSA, or if they became symptomatic. The population studied in this report therefore provides a unique opportunity to study the long-term clinical significance of PSA values without being overly influenced by early invasive investigation and treatment.

As there is no PSA level at which there is no risk of prostate cancer (8-10), men, and their physicians, have a difficult decision to make when presented with a PSA result: Do they proceed to prostate biopsy or adopt a more conservative approach? The results from this study provide data to inform men who have a low baseline PSA of their risk of prostate cancer and prostate cancer–specific mortality (Tables 3 and 5) if they do not undergo early investigation. This is in contrast to their risk of prostate cancer if they do proceed to prostate biopsy. Based on data from Thompson et al. (8), if all men in our population with baseline PSA of <4.0 ng/mL continued to prostate biopsy, a total of 7,818 cancers would have been diagnosed. Alternatively, if they underwent annual PSA screening, proceeding to biopsy when their PSA went above a predetermined cutoff (e.g., 2.5 ng/mL), as suggested by Antenor et al. (7), 3,525 cancers would have been diagnosed. This compares with 402 cancers that were actually diagnosed. Therefore, when the results from these studies were applied to the current population (7, 8), only 5.1% of histologic prostate cancers and 11.4% of screen-detected cancers presented clinically over a 5- to 10-year period, which, although not unexpected given the lead time associated with PSA testing (15), helps to quantify the risk of significant prostate cancer for individual men. Using the prostate cancer risk calculator described by Thompson et al. (10), a 65-year-old man with a PSA of 2.27 ng/mL had a risk of prostate cancer on biopsy of 25%. Based on our data, the risk of clinically diagnosed prostate cancer in this scenario, with a mean follow-up of 6 years, was only 1.5%. Moreover, the risk of prostate-specific mortality was only 0.2%, whereas overall mortality was over 70 times higher at 14.6%. Men who are diagnosed with cancer after delayed investigation may have more advanced disease at presentation, however, given that most will be older than 70 years old, they are more likely to die from other causes than from prostate cancer (16).

This study showed a low risk of clinical diagnosis of prostate cancer, high-grade cancer, and prostate cancer–specific mortality over a 5- to 10-year period at low levels of baseline PSA. The rate of prostate cancer and high-grade cancer diagnosis in men with a baseline PSA of <4.0 ng/mL was <2 and <1 cases per 1,000 person-years, respectively. The rates of prostate cancer diagnosis presented in Table 3 may in fact be an overestimation of clinically diagnosed cases. As shown in Fig. 2, in the 4.0 to 9.9 ng/mL group, ∼3% of men were diagnosed with prostate cancer within 6 months of their baseline PSA. After this time, the cancer diagnosis rate slowly increased, suggesting early investigation and diagnosis in a small proportion of men within this group; this pattern was not apparent in men with low (<4.0 ng/mL) PSA values. Although prostate cancer did occur at low PSA, only 0.8% of all men with a baseline PSA level of <4.0 ng/mL were diagnosed with cancer, which equates to following almost 160 men for 5 years to diagnose one prostate cancer. This compares to a risk of 5.5% to 6.5% in screening studies (7, 17), and a prevalence of 15.2% when all men were biopsied (8). Although the populations under observation differ between these studies, the very different proportions diagnosed with prostate cancer, at similar PSA levels, are mainly due to the differences in timing and intensity of investigation.

When baseline PSA was <4.0 ng/mL, the rate of prostate cancer–specific mortality was very low at <0.2 cases per 1,000 person-years, whereas the overall mortality was much higher at 28.7 cases per 1,000 person-years. The prostate-specific mortality rates observed may actually be an underestimation. As a small proportion of men were investigated and diagnosed with prostate cancer as a result of their PSA test, they may have received treatment and so have an improved survival. However, as there were few radical prostatectomies done in Northern Ireland before 2000 (4 in 1996 and increasing to 43 in 2001; ref. 18), it is unlikely that treatment for localized prostate cancer would have had an effect on prostate-specific mortality, particularly given the relatively short follow-up period. Any improvements in survival are likely to be limited to men with high PSA (≥10.0 ng/mL), who may have received early hormonal therapy.

Other studies have assessed the long-term risk of prostate cancer based on an early PSA value (7, 17, 19-21). Many of these studies have focused on the relative risk of cancer diagnosis associated with baseline PSA (19-21), comparing increasing levels of PSA to a low PSA reference category, rather than the “absolute” rate of cancer detection (7, 17). In a nested case-control study using stored sera, Gann et al. found a steadily increasing relative risk of prostate cancer with increasing baseline PSA (19). With 10 years of clinical follow-up, the relative risks are remarkably similar to the present study, although the current data have the advantage of being population-based and therefore not requiring the selection of suitable control subjects. Fang et al. also assessed the long-term risk of prostate cancer using stored sera (20). Over a 20-year period, they found that, in men aged 40 to 49 and 50 to 59 years, there was a 3.5 times increased relative risk of prostate cancer in those who had a PSA above the median for their age compared with those with an baseline PSA below the median. The present study showed markedly increasing HRs with increasing PSA category. However, as these ratios are based on the very low risk in the reference category of 0 to 0.99 ng/mL, they are not as informative as the “absolute” rates of cancer diagnosis, which remain low across all low PSA values (Table 3). The “absolute” rates of cancer diagnosis in the present study follow a similar pattern to studies in screened populations (7, 17), although the magnitude of cancer detection is much lower in the present study. HRs (or relative risks) also require that an artificial reference category is chosen, which may not be clinically relevant; therefore, the “absolute” rate of cancer detection is much more clinically important.

Overdetection of prostate cancer has been defined as a screen-detected cancer, which would not have caused death if left untreated (22), or detection of a cancer, which would not have been diagnosed within an individual's lifetime in the absence of screening (15, 23). The rates of overdetection due to PSA screening have been estimated to be greater than 86% if death is taken as the endpoint or ∼30% to 50% if prostate cancer diagnosis only is considered. Clinically insignificant cancers have generally been defined using pathologic features (24, 25), with large surgical studies showing a rate of insignificant cancers of <1% to 17% (24-27). Although the remainder was considered “clinically significant,” data from the pre-PSA era showed that less than 1 in 3 men diagnosed with prostate cancer eventually died from the disease (28). This suggests that the majority of “clinically significant” cancers may not cause harm during a man's lifetime and may therefore be considered as overdetected cases. The difficulty with this group is that it is impossible at present to differentiate at an early stage those men who will ultimately have progression of their disease from those who will not, although PSA velocity may have a role in identifying those men who will progress to life-threatening disease (29). The current study found that the majority of men with low PSA levels, although they may have histologic or screen-detectable prostate cancers, will not develop clinically apparent disease over a 5- to 10-year period. Therefore, most of these cancers could be considered as clinically insignificant over this period, which may decrease the value of proceeding with early investigation, particularly in older men.

The strengths of this study include that it is population-based and very large. Also, the Northern Ireland population is stable, and the NICR rigorously collates cancer incidence data from the whole population. It is therefore likely that few prostate cancer diagnoses or deaths among the cohort went unrecorded. There are several limitations worthy of note. The main limitation is the relatively short mean follow-up time of 6 years. Undoubtedly, there are cases of prostate cancer among the cohort that will come to clinical attention in the ensuing years and further deaths from prostate cancer will accrue. Mean age of men included was 65.2 years, which is old compared with populations targeted for PSA testing (1, 4). The results reported here may not be applicable to younger men, who have a life expectancy exceeding the follow-up period of the study. Secondly, it is possible that errors in data collection or data entry or during the matching process may have resulted in missed cancer cases or deaths. Internal validation procedures in the NICR ensure that this is very uncommon and unlikely to significantly affect results (30). Only 49 (0.9%) of all prostate cancers diagnosed in Northern Ireland between 1994 and 2003 have not been matched to the PSA database. Thirdly, different PSA assay platforms were used during the study period, which, given their inherent variability, may not provide results that are directly comparable. However, the inclusion of a cluster term for the laboratory of test origin in relevant analyses did not materially affect results. Fourthly, only a small proportion of men in the cohort may have undergone asymptomatic screening (13). As men undergoing PSA screening have earlier stage and grade disease compared with symptomatic men (15, 26), if uninvestigated, they would be less likely to present clinically or die from prostate cancer over a 5- to 10-year period. Therefore, the rates of clinical cancer detection and prostate-specific mortality in a screened population may be lower than those seen in this study. Finally, the lack of clinical information on the individuals tested prevents inferences being made about how factors, such as reason for PSA testing, family history, medications, rectal examination findings, further investigations or treatments for prostate cancer, etc., influence the patterns of diagnosis and mortality observed. Further, follow-up data, such as rates of biochemical recurrence or metastasis-free survival, are not available.

When deciding on further investigation following a PSA result, men need to be aware not just of the likelihood of a diagnosis of prostate cancer but also of having a cancer that may cause them harm during their lifetime or, more importantly, kill them. This study shows that, when a conservative approach to further investigation is employed, over a 5- to 10-year period, clinical prostate cancer is infrequently diagnosed in men with low PSA levels and death from prostate cancer is rare. Rates of clinically diagnosed prostate cancer and mortality data are presented based on age band and PSA level. These data should be routinely used when clinicians and patients are deciding on further management following PSA testing, particularly in older men.