Kirsten Howard, Alex Barratt, Graham J. Mann, Manish I. Patel. A Model of Prostate-Specific Antigen Screening Outcomes for Low- to High-Risk MenInformation to Support Informed Choices. Arch Intern Med. 2009;169(17):1603–1610. doi:10.1001/archinternmed.2009.282
Information is needed to aid individual decision making about prostate-specific antigen (PSA) screening.
We aimed to provide such information for men aged 40, 50, 60, and 70 years at low, moderate, and high risk for prostate cancer. A Markov model compared patients with vs without annual PSA screening using a 20% relative risk (RR) reduction (RR = 0.8) in prostate cancer mortality as a best-case scenario. The model estimated numbers of biopsies, prostate cancers, and deaths from prostate cancer per 1000 men over 10 years and cumulated to age 85 years.
Benefits and harms vary substantially with age and familial risk. Using 60-year-old men with low risk as an example, of 1000 men screened annually, we estimate that 115 men will undergo biopsy triggered by an abnormal PSA screen result and that 53 men will be diagnosed as having prostate cancer over 10 years compared with 23 men diagnosed as having prostate cancer among 1000 unscreened men. Among screened men, 3.5 will die of prostate cancer over 10 years compared with 4.4 deaths in unscreened men. For 1000 men screened from 40 to 69 years of age, there will be 27.9 prostate cancer deaths and 639.5 deaths overall by age 85 years compared with 29.9 prostate cancer deaths and 640.4 deaths overall in unscreened men. Higher-risk men have more prostate cancer deaths averted but also more prostate cancers diagnosed and related harms.
Men should be informed of the likely benefits and harms of PSA screening. These estimates can be used to support individual decision making.
Prostate-specific antigen (PSA) screening for prostate cancer causes uncertainty and concern for men and their physicians. Findings from 2 large randomized trials1,2 that were supposed to provide answers and clarity may have worsened the situation. The trials report evidence of benefit but also evidence of substantial harm. Adding to community uncertainty are conflicting recommendations from health authorities and clinical guidelines. For example, the American Cancer Society “believes that health care professionals should discuss the potential benefits and limitations . . . discussion should include an offer for testing with the PSA blood test and digital rectal exam (DRE) yearly, beginning at age 50, to men who are at average risk of prostate cancer and have at least a 10-year life expectancy.”3 The American Urological Association likewise considers that the PSA test is a “valuable screening tool” that “should be appropriately offered to men” 40 years and older.4,5 In contrast, the US Preventive Services Task Force has assigned an “I” (“insufficient evidence”) rating to PSA screening, noting that PSA screening is “associated with important harms, including frequent false-positive results and unnecessary anxiety, biopsies, and potential complications of treatment of some cancers that may never have affected a patient's health.”6
Although these organizations, and the clinical guidelines they have released, do not endorse “population screening,” they do recommend that men be informed about the pros and cons of screening. However, this may increase confusion as men and their physicians struggle to access balanced, evidence-based information for individual decision making. They may find competing claims about the likely benefits and the potential for harm7 and opposing information produced by individuals or groups that may have a vested interest in promoting screening.
Decision aids prepare patients for complex decisions by providing information about different options and by helping people clarify their personal values regarding different options and outcomes.8 Decision aids are effective in increasing knowledge, creating more realistic expectations about outcomes, and reducing decisional conflict.9 They present balanced, evidence-based information about outcomes in a way that meets standards for information to be used for informed decision making.10,11 For prostate cancer screening, decision aids should present the chances of all relevant outcomes occurring (biopsies, prostate cancer diagnoses, and prostate cancer deaths) for men who choose vs those who decline screening. Risks should be presented as event rates for specified populations (eg, men who choose vs those who decline screening), and these event rates should use consistent denominators (eg, per 1000 or per 10 000 men) over comparable time frames.8,10 Risk presentation should allow patients to view event rates specific to their own situation, for example, based on age or family history. Despite the obvious need, decision support materials for men regarding PSA screening do not include such information.12
To address this information gap, we developed a model of outcomes of annual PSA screening for men at varying prostate cancer risk defined by family history (low, moderate, and high risk) over time frames appropriate for informed decision making rather than simply per screening test. We used the same approach previously used for breast cancer screening,13 which meets information standards for patient decision aids.10,11 Outcomes of PSA screening are presented over 10 years per 1000 men aged 40, 50, 60, and 70 years at different risk levels. Thus, we have taken the best available evidence from population studies and translated it into information that is useful for individual decision making. Data presented herein can be readily incorporated into patient decision aids to prepare men to make choices about PSA screening or can be used to inform clinical consultations with men considering screening.
A Markov model was constructed for 2 hypothetical cohorts that either participate in or decline annual PSA screening. The model assumes 100% participation in the screening cohort and 0% participation in the nonscreening cohort, generating outcomes for men who were screened regularly vs those who were not screened. The model was constructed for men at low, moderate, and high risk, defined by family history, using published estimates of the numeric increase in risk for differing types and numbers of affected relatives.14- 16 Low-risk men are those with no first-degree relatives affected by prostate cancer. Men with 1 affected first-degree relative are considered to be at moderate risk (relative risk [RR], 2.5 vs no family history), and men with 2 or more affected first-degree relatives are considered to be at high risk (RR, 5.0 vs men with no family history). Calculations were conducted separately for all 3 risk categories. The model is based on annual screening commencing at age 40 years, consistent with American Urological Association recommendations.4,5 It generates outcomes over 4 × 10-year time frames: from age 40 years, from age 50 years, from age 60 years, and from age 70 years (Table 1) for men who start screening at age 40 years and continue to be screened, compared with men who decline screening at all ages. Sensitivity analyses were conducted on the mortality benefit and by increasing the starting age of low-risk men to 50 years (American Cancer Society3). All rates were converted to annual probabilities. Modeling was conducted using a spreadsheet program (Microsoft Excel; Microsoft Corp, Redmond, Washington). Data sources and assumptions underlying the model are outlined in the following subsections and in Table 2.1,14- 20
The incidence in unscreened men was based on Australian age-specific incidence rates from prescreening years 1982 to 1988.24 (In Australia, PSA screening began in 1989, when the test was publicly funded.) These rates were linearly extrapolated to 2005 based on the temporal trend observed from 1982 to 1988. Proportions of localized and nonlocalized cancers in unscreened men were taken from the European Randomized Study of Screening for Prostate Cancer (ERSPC), based on Dutch cancer registry data before commencement of the trial.18 These Dutch data provide comparability with other data from the ERSPC used in the model. Prostate cancer incidence and mortality rates were similar in Australia24 and the Netherlands in the 1990s21,25; thus, we assumed that the proportions of localized and nonlocalized cancer would be similar too.
Age-specific prostate cancer mortality rates in 2005 for Australian men were used.24 Because these data combine prostate cancer mortality for screened and unscreened men, we adjusted it using estimates of the age-specific proportion of men undergoing a PSA test (eSupplementhttp://www.archinternmed.com). The PSA screening practices have changed over time, and prostate cancer is a slow-growing cancer, so we used PSA test utilization data from 10 years earlier by obtaining estimates of the age-specific proportion of men who had a PSA test in 1995 from Medicare data. (Medicare is the Australian national health care insurer that covers all Australian permanent residents and citizens.)
The numbers of prostate cancers detected were based on the relative number of cancers in the screened vs unscreened trial arms of the ERSPC.22 These ERSPC data were applied to the incidence in unscreened Australian men to calculate the number of cancers detected in screened men. When the ERSPC commenced, the criteria for a prostate biopsy were a PSA level of 4.0 ng/mL or greater (to convert to micrograms per liter, multiply by 1.0) and/or an abnormal rectal examination or abnormal transrectal ultrasonographic findings. As of May 1997 (midway through screening round 1), the PSA cutoff value for recommending a biopsy was decreased to 3.0 ng/mL, and rectal examination and transrectal ultrasound were discontinued.26 Overall test positivity and the subsequent number of biopsies by age were estimated from the positive predictive value reported in the ERSPC.1 All men with a positive PSA screen were assumed to have a biopsy. We assumed that PSA testing has the same test characteristics for all 3 risk groups (in the absence of data providing sensitivity, specificity, and positive predictive value of PSA testing by risk group). Interval cancers were estimated in 2 ways, depending on age: (1) for ages 40 to 59 years, we used annual age-specific incidence rates for unscreened men and (2) for ages 60 to 79 years, we used the interval rate reported in the ERSPC.1
The model further assumes that men diagnosed as having prostate cancer will accept effective treatment. Because the aim of the model was to estimate the main outcomes, we have not explicitly modeled individual treatments because choice of treatment is likely to vary between individual men, depending on their physician and treatment center and on the men's personal preferences regarding potential benefits and adverse effects. In the base case, we assumed that screening and subsequent effective treatment confers a mortality benefit.
Data from the ERSPC1 suggest a 20% relative reduction in prostate cancer mortality in screened vs unscreened men (RR of prostate cancer death, 0.8; 95% confidence interval, 0.65-0.98). We use an RR of prostate cancer mortality of 0.8 as a best case and test, in sensitivity analyses, an RR of 1.0, consistent with ERSPC results1 and a recent Cochrane systematic review.19
Because prostate cancer is slow growing and because the mortality benefit of screening is not immediate, we assumed that the mortality benefit accumulates over time. Based on ERSPC results (Figure 2 in the study by Schröder et al1), we assumed that there is no mortality benefit in the first 7 years and that the 20% RR reduction is achieved by 9 years (median follow-up of the ERSPC). We assumed that mortality benefit declines linearly during the 10 years after screening ceases. The RR was applied to the age-specific prostate cancer mortality rate for unscreened men to derive the age-specific prostate cancer mortality rate for screened men for each risk group.
Australian Bureau of Statistics data were used to provide estimates of age-specific all-cause mortality rates.20 Age-specific all-cause mortality rates were split into prostate cancer–specific mortality and mortality due to causes other than prostate cancer and then were applied to screened and unscreened cohorts. The total number of deaths in each year was calculated by summing deaths due to prostate cancer and deaths due to causes other than prostate cancer.
Results for all age groups are provided in Table 3 (low familial risk) and in Table 4 (moderate and high familial risk) using an RR of prostate cancer mortality of 0.8.1 Results of sensitivity analyses on the mortality benefit using an RR of 1.0 are given in the eTablehttp://www.archinternmed.com.
Using 60-year-old men as an example, of 1000 men aged 60 years who are screened and who have annual screening with PSA during the next 10 years, 115 will have a prostate biopsy triggered by an abnormal PSA screen result, and, of these, 87 will not have prostate cancer. This represents an 11.5% chance of having a prostate biopsy due to PSA screening and an 8.7% chance of having a biopsy and not finding prostate cancer (experiencing a false alarm). Approximately 53 men will be diagnosed as having prostate cancer during the 10 years (28 detected by screening plus 25 interval [clinically diagnosed] cancers) compared with 23 men diagnosed as having prostate cancer among 1000 unscreened men. Of 1000 screened men, 3.5 will die of prostate cancer during the 10-year time frame compared with 4.4 unscreened men. This is in the context of 116 deaths from all causes in the screened group and 117 deaths from all causes in the unscreened group.
Similar interpretations apply to the other age groups. In younger men, the numbers of prostate cancer diagnoses, deaths from prostate cancer, and deaths from all causes are lower. In older men, the numbers of these events are higher, with increasing age-specific prostate cancer incidence and mortality, but also in the context of increasing all-cause mortality.
We also examined the impact of screening to age 69 years on prostate cancer and all-cause mortality until age 85 years (the life expectancy of Australian men) (Table 3). For men aged 40 years who screen to age 69 years, deaths were cumulated for 45 years. For men aged 50 years who screen to age 69 years, deaths were cumulated for 35 years. For men aged 60 years screening to age 69 years, deaths were cumulated for 25 years. Men 70 years and older may continue to be screened, depending on life expectancy; so, for these men, we assumed that they continued to be screened until age 79 years and cumulated deaths over 15 years. Under the best case, for 1000 men who are screened from ages 40 to 69 years, with deaths cumulated to age 85 years, there were 27.9 prostate cancer deaths and 639.5 deaths overall in men who screen compared with 29.9 prostate cancer deaths and 640.4 deaths overall in men who do not screen.
Compared with low-risk men, the overall pattern of events is similar for moderate- and high-risk men, although the absolute chance of events is higher (Table 4). Men with a moderate or high familial risk had a higher chance of having cancer diagnosed, a higher chance of undergoing a prostate biopsy, and a higher chance of experiencing a false alarm. This is balanced against a greater absolute mortality benefit in terms of prostate cancer deaths over 10 years and up to age 85 years (Table 4).
The sensitivity analysis that varied the RR of prostate cancer mortality to the null (RR = 1.0)19 is given in the eTable. This sensitivity analysis highlights the relatively high chance of experiencing a false alarm or of having screen-detected prostate cancer for men who choose to be screened over several decades, which needs to be balanced against no improvement in mortality rates. If there is no mortality benefit of PSA screening, the detrimental impact of screening on quality of life would be substantial.
Changing the age at which low-risk men start screening from 40 to 50 years (and, therefore, delaying the mortality benefit) has little effect on the number of prostate cancer deaths. During the 10 years from ages 50 to 59 years, there are 0.45 deaths in men previously screened from age 40 years compared with 0.51 deaths in men who commenced screening at age 50 years. Accumulating prostate cancer deaths to age 85 years, there are 28.39 deaths in 1000 men who commence screening at age 40 years compared with 28.45 deaths in 1000 men who commence screening at age 50 years.
This article presents comprehensive risk- and age-specific estimates of the benefits and harms of PSA screening that can be used by men and their physicians for individual decision making about undergoing PSA screening. The estimates should help men and physicians achieve the goal of full discussion of the benefits and risks of PSA screening, as recommended in clinical practice guidelines and advice from health agencies.4,6
Although other models of PSA screening exist,27- 29 this is the first use of a balance sheet approach comparing outcomes for men who choose to be screened with those for men who decline screening and is an approach consistent with best practice for individual decision making.10 This model is unique in modeling screening outcomes across time frames that support informed decision making, that is, cumulated over 10 years and accumulated during a lifetime of screening participation to age 85 years. We deliberately present the outcomes for men who choose to participate in screening regularly (ie, 100% participation in screening) and who, therefore, can expect to obtain the full mortality benefits of PSA screening. The comparator is men who do not participate at all. This is the comparison that is needed for decision making at an individual level. However, note that this is different from a policy-level approach if one wanted to estimate outcomes, including cost-effectiveness, for a population with less than 100% participation in screening.
Despite the recently published results of the ERSPC,1 there is still considerable uncertainty regarding the mortality rate reductions that may be attainable with population-based PSA screening, as demonstrated by the results of the recently published Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.2 Participants in randomized trials may be more adherent, or the screening interval of the trials may not be clinically relevant. For example, the ERSPC reports on 2 screening rounds with a screening interval of 4 years, whereas the guidelines that advocate screening recommend annual screening. It is also possible that men who screen annually may be inherently different from men who do not screen, for example, in terms of cancer risk, comorbidities, or other risk factors, that, in a population screening context, may all mean that the mortality benefit from clinical trials may not be fully realized. We used an RR of 0.8 for prostate cancer mortality1 and tested an RR of 1.0 in sensitivity analysis. Even using the RR of 0.8, it is striking that the absolute mortality benefit is small even in men at the highest levels of familial risk and even when prostate cancer deaths are cumulated to 85 years of age. This reflects the relatively small proportion of overall deaths due to prostate cancer and the fact that deaths from other causes partly offset those averted by screening.
On the harm side of the balance sheet, the present estimates provide a sobering illustration of the frequency of harms men are likely to experience if they participate in PSA screening. The risk of having a false alarm rises strongly with age and with increasing familial risk. Up to 21.1% of low-risk men will experience a false-positive result during 10 years of screening. Prostate cancer biopsies carry risks such as hemorrhage and infection.
In moderate- and high-risk men, mortality reductions are more sizable. However, the increased risk of prostate cancer in these men means that harms are also more sizable, with very high proportions of men experiencing false-positive results and having screen-detected cancers.
More important, participation in screening considerably increases the likelihood of having prostate cancer diagnosed; there is approximately a 2- to 4-fold increase in the risk of having prostate cancer diagnosed compared with men who do not participate in screening. Yet, few of these men die of prostate cancer, and death rates (from prostate cancer and from all causes) are similar in screened and unscreened men. The only interpretation is that many men with screen-detected prostate cancer are having cancer therapies for clinically insignificant cancers. Because we cannot predict accurately which screen-detected prostate cancers will progress to life-threatening disease, all these men will most likely be offered cancer treatment, which may include surgery, radiotherapy, and endocrine therapy.
Hospital admissions for a principal diagnosis of prostate cancer doubled between 2000-2001 and 2005-2006, and prostatectomies in men with a principal diagnosis of prostate cancer increased by 56% during the same period.30 The risks of these treatments in terms of adverse effects on quality of life are substantial.23 Twenty percent to 70% of men experience erectile dysfunction or impotence after treatment, and 15% to 50% experience long-lasting urinary incontinence. Although the risks are less after radiotherapy compared with prostatectomy, the mortality benefits may not be as great, the risks of reduced potency and incontinence are still considerable, and up to 25% of patients experience bowel problems.23 An undebated consequence of guidelines that recommend informed decision making and consumer choice is the potentially substantial financial implications of those choices that will be borne by the health system.
A variety of areas of uncertainty exist in the available data, and, therefore, a range of assumptions were made. First, there is no organized PSA screening program in Australia, and so we had to rely on data from multiple sources to estimate screening outcomes such as test positivity rates, positive predictive values, and interval cancer rates. By assuming that every positive PSA screening result is investigated by means of biopsy, we may have overestimated the number of biopsies; it is possible that, once a man has had an abnormal PSA test result and a subsequently negative biopsy result, the next positive PSA test result may not be investigated using another biopsy. Second, because data are simply not available, we had to model the incidence for unscreened men by extrapolation of prescreening incidence, assuming that incidence has not been greatly affected by changes in prostate cancer risk factors during the past 20 years; we also assumed that the proportional increase in incidence in screened compared with unscreened men and the mortality benefit attributed to screening are consistent across men who are at low, moderate, and high risk. We believe that these assumptions are reasonable in the absence of data to the contrary. This means that, although there is clearly some uncertainty regarding each estimate (in Tables 3 and 4 and the eTable), the general picture of the results is likely to be correct, and the relative frequency of benefits to harms is broadly representative of what could be expected with participation in PSA screening. We modeled annual screening because that reflects most guidelines and common clinical practice; however, the balance of harms to benefits may change with a longer screening interval. Estimation of interval cancers was difficult in the absence of data from a population screening program. We applied the best and most applicable data by using ERSPC results for older men (≥60 years) and the estimated incidence in unscreened men for younger men. It is possible this may have led to overestimation of interval cancers; however, the effect of this on the overall picture of benefits vs harms of PSA screening is small.
In addition, we limited this analysis to consideration of mortality rather than explicitly considering the effect on quality of life. The aim of this study was to provide men and their physicians with enough information about the broad potential outcomes to help inform a choice about prostate cancer screening. As yet, data on quality of life in screened and unscreened men are unavailable; prostate cancer treatment may lead to improvements in quality of life because of a reduction in local recurrence or metastatic disease; however, just as likely, it may lead to decreases in quality of life as a result of long-term adverse effects, such as impotence and incontinence. Thus, the overall effect on an individual's quality of life is likely to depend on how that individual views the trade-off between potential benefits and harms; only each individual can decide what an acceptable level of benefit to harm trade-off is.
The information in Tables 3 and 4 translates the results of the latest evidence from population studies into information for individual men and their physicians to use when considering the pros and cons of PSA screening. Physicians who want to support their patients in this decision can use the information to provide estimates of the likely outcomes (particularly the risk of a screen-detected cancer and the risk of death from prostate cancer) if the man chooses to screen or not to screen. These data are tailored by age and by risk status based on family history. The information presented herein is suitable for inclusion in patient decision aids for men considering PSA screening. Australian patient decision aids using these data are currently under evaluation. For researchers who would like to test decision aids in their setting, the model is available on request.
In conclusion, before undergoing PSA screening, men should be aware of the possible benefits and harms and of their chances of these benefits and harms occurring. Even under optimistic assumptions, the net mortality benefit is small, even when prostate cancer deaths are cumulated to 85 years of age. These quantitative estimates can be used to support the goal of individual informed choices about PSA screening.
Correspondence: Kirsten Howard, BSc(Hons), MAppSc, MPH, MHealthEcon, PhD, Health Economics, Screening and Test Evaluation Program, School of Public Health, University of Sydney, Edward Ford Bldg (A27), Sydney, New South Wales, Australia.
Accepted for Publication: June 16, 2009.
Author Contributions: Dr Howard had full access to all of the data in the study and takes full responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Howard, Barratt, Mann, and Patel. Acquisition of data: Howard, Barratt, Mann, and Patel. Analysis and interpretation of data: Howard, Barratt, and Mann. Drafting of the manuscript: Howard and Barratt. Critical revision of the manuscript for important intellectual content: Howard, Barratt, Mann, and Patel. Statistical analysis: Howard. Administrative, technical, and material support: Howard and Barratt. Study supervision: Howard and Barratt.
Financial Disclosure: None reported.
Funding/Support: This work was undertaken as part of the Screening and Test Evaluation Program funded by the National Health and Medical Research Council of Australia under program grant 402764.
Role of the Sponsor: The funding agency had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.