Screening Men at Increased Risk for Prostate Cancer Diagnosis: Model Estimates of Benefits and Harms

Among men in economically developed countries, prostate cancer is the most frequently diagnosed tumor and the third most common cause of cancer death (1). PSA screening can advance the point of diagnosis and, when coupled with early treatment, can likely reduce the risk of cancer-specific death (2). However, PSA screening can also detect cancer that would never have presented clinically (3). Current guidelines either recommend against routine PSA-based screening (4, 5) or advocate shared and informed decision making (6–8) in men with average risk.

Certain subgroups have increased risk for prostate cancer diagnosis. African American men historically had prostate cancer incidence that is 1.6 times that in Caucasian peers. Family history of prostate cancer has also long been recognized as a risk factor (9). Recently, the Identification of Men with a genetic predisposition to ProstAte Cancer: Targeted screening in BRCA1/2 mutation carriers and controls (IMPACT) study found higher prostate cancer incidence among BRCA1/2 carriers relative to noncarriers (10).

Higher observed incidence may be due to greater risk of disease onset or more aggressive cancer. Although still controversial, many believe that men of African descent are more likely to be diagnosed with advanced prostate cancer and to die of the disease (11). Studies have indicated that BRCA1/2 carriers may have higher risk of metastasis and worse cancer-specific survival than noncarriers (12, 13).

A common reaction to identifying a subgroup with increased risk is to consider screening that subgroup more intensively than average-risk individuals. A recent editorial stated that as male germline BRCA2 mutation carriers have higher risks of early onset, aggressive disease, and higher prostate cancer mortality, “screening intervals should be shorter and the PSA thresholds lower for them than for men in the general population” (14). However, connections between the ways that risk may be increased and appropriate screening approaches have not been systematically studied. Different types of increased risk, including higher frequency of onset of latent disease, more rapid progression to an advanced state, and shorter survival (e.g., due to poor response to standard therapies), may have different implications for targeted screening.

In this article, we investigate targeted screening in men with increased risk of prostate cancer diagnosis. We adapt an established model of prostate cancer natural history, detection, and survival to implement possible mechanisms of increased risk and use the resulting models to examine harm–benefit tradeoffs of different strategies. We demonstrate general lessons by considering screening decisions for men with germline BRCA mutations.

To investigate targeted screening in subgroups with average and increased prostate cancer risks, we used the Fred Hutchinson Cancer Research Center (FHCRC) model (15, 16). The model projects PSA growth before and after onset of latent disease using data from the placebo arm of the Prostate Cancer Prevention Trial (17). The risk of onset increases with age, and risks of progression to metastasis and to diagnosis increase with PSA levels. These risks were estimated so the model simulates prostate cancer incidence under U.S. screening patterns that matches incidence in the Surveillance, Epidemiology, and End Results (SEER) program.

To implement U.S. screening patterns, we used a reconstruction (18) that combined data from the National Health Interview Survey and the SEER-Medicare database. Following common practice in the United States, we assumed men with PSA >4.0 ng/mL were referred to biopsy. Receipt of biopsy was based on frequencies in the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial (19), which involved patient–physician decision making and reflects real-world practice. Finally, we assumed biopsy sensitivity to detect latent cancer increased with calendar year to reflect dissemination of extended-core biopsy schemes (15).

In the absence of treatment, cancer-specific survival was derived from SEER-untreated cases diagnosed in 1983 to 1986, just before PSA screening began. Early detection improves survival by detecting would-be metastatic cases while they are still localized and reassigns survival accordingly. This “stage-shift” effect of screening is consistent with the mortality reduction in the European Randomized Study of Screening for Prostate Cancer (16) and is not contradicted by PLCO results (20). Frequencies of initial treatment were derived from SEER cases diagnosed in 2010 by age and tumor stage and grade. Efficacy of radical prostatectomy was based on the Scandinavian trial of prostatectomy versus watchful waiting (HR = 0.56; ref. 21), and we assumed similar efficacy for contemporary radiotherapy (22, 23). Noncancer survival was based on U.S. life tables (24), and overall survival was the earlier of cancer-specific and noncancer survival.

The model allowed us to project disease prevalence, incidence, and prostate cancer mortality in the average-risk population. Under a specified screening strategy, we estimated the lifetime probability of overdiagnosis as the frequency of screen-detected cancers that would not present clinically before noncancer death and the lifetime probability of life saved as the difference between model-projected frequencies of prostate cancer death with and without screening (16).

To represent hypothetical increased risks, we applied a multiplicative factor (HR = 2.0) to risks of latent onset, progression to a symptomatic or metastatic state, and/or prostate cancer death following diagnosis. For each type of increased risk, we simulated outcomes for 10 million men under all combinations of starting age (40, 45, or 50 years), ending age (64 or 69 years), frequency (annual or biennial), and PSA threshold for biopsy (1.0 or 3.0 ng/mL). We calculated lifetime probabilities of overdiagnosis and life saved, and we compared approaches using the additional number needed to diagnose (NND) to prevent one prostate cancer death, calculated as the ratio of these probabilities. We defined preferred strategies as those with NND ≤5 as an example of a policy target; in specific settings, other criteria could be used.

To investigate how risk might be increased for men with germline BRCA mutations, we estimated the risk of latent onset corresponding to BRCA status using published results from the first screening round of the IMPACT study (10). IMPACT investigators recruited undiagnosed men ages 40 to 69 years from families with known or suspected BRCA1 or BRCA2 mutations in 2005 to 2013 and assigned carrier status based on genetic testing. Participants were offered annual (biennial in the Netherlands) PSA screening with referral to biopsy for PSA >3.0 ng/mL.

We simulated BRCA1 and BRCA2 cohorts in the IMPACT study and identified increased risks of onset relative to the general population that best matched study results. Before entering the study, we assumed participants were screened according to patterns in the United States before 1995 (18) with biopsy frequency and sensitivity as for average-risk men. After recruitment, we assumed 80% of men with PSA >3.0 ng/mL received a biopsy to match study results (10), and a biopsy had 90% sensitivity to detect cancer for consistency with the number of cores in the study protocol. We simulated each cohort 10,000 times over a range of values for the HR for onset; estimation was based on maximizing a Poisson likelihood for observed and projected cancers diagnosed, and 95% confidence intervals (CI) were derived using the profile likelihood method (25).

Figure 1 illustrates model-projected Kaplan–Meier curves for the specified endpoints among men at average and increased (doubled) risks of latent onset, diagnosis, and cancer-specific death. The risk of diagnosis is among men with onset in their lifetimes, and the risk of cancer death is among men who present clinically. All endpoints are in the absence of screening, treatment, or competing death.

Figure 2 illustrates projected lifetime probabilities of overdiagnosis and life saved by screening for men with average and all combinations of increased risks. Dashed lines show NND = 5; strategies above this line have fewer than 5 overdiagnoses per life saved.

In the average-risk population (top left), biennial screening to age 64 years with biopsy referral when PSA >3.0 ng/mL had the smallest NNDs (range across starting ages, 4.1–4.4), changing to annual screening raised NNDs closer to 5, and screening to age 69 years pushed NNDs above 5. All strategies with biopsy referral for PSA >1.0 ng/mL implied substantially higher NNDs (range, 7.8–9.9) due to more overdiagnosis with few additional lives saved.

In men with increased risk of onset, the NNDs of all strategies had similar rank order as in average-risk men, but probabilities of overdiagnosis and life saved were both higher. In men with increased risk of progression, the probability of overdiagnosis was lower and the probability of life saved was higher compared with average-risk men under the same strategies. In men with increased risk of cancer death, the probability of overdiagnosis was similar and the probability of life saved was higher compared with average-risk men. These results match expectations.

In men with increased risks of onset and progression, a greater number of screening strategies yielded NND ≤5 compared with average-risk men. Thus, if our goal was to achieve NND ≤5 while maximizing benefit, we could consider more intensive strategies in men at higher risk. For example, annual screening to age 69 years with PSA threshold 3.0 ng/mL could be considered in men with increased risks of onset and progression but not in average-risk men or in men with only increased risk of onset. This example shows that preferred strategies may differ across subgroups with different mechanisms of increased risk. In no subgroup did lowering the PSA threshold from 3.0 to 1.0 ng/mL improve outcomes, indicating that under our model assumptions, these strategies would not be preferred.

Finally, we examine men with germline BRCA1/2 mutations. Our assumptions about screening before the IMPACT study implied 35% of simulated participants received at least one PSA test, similar to the reported 37% of participants (10). Comparing observed and projected cancers diagnosed during a single round of screening, we estimated nonsignificantly increased risk of onset in BRCA1 carriers (HR = 1.05; 95% CI, 0.63–1.59) and significantly increased risk of onset in BRCA2 carriers (HR = 1.81; 95% CI, 1.14–2.78). Consequently, our general results about screening men at increased risk of onset may apply to BRCA2 mutation carriers. In particular, a wider range of screening ages may be acceptable, but lowering the PSA threshold for biopsy referral is likely to increase overdiagnosis with few additional lives saved.

Risk stratification promises to reduce the costs and harms of screening by focusing resources on the subgroups most likely to benefit. However, realizing its potential requires that we understand the mechanisms by which the subgroups have increased risk. Is the increased risk due to earlier and more frequent disease onset, greater likelihood of progression to symptoms or metastasis, shorter survival, or some combination of these?

This article investigates ways in which the risk of prostate cancer diagnosis might be increased and links the mechanism of increased risk to targeted screening strategies. Although it may seem intuitive that higher risk strata should be screened more intensively than those with average risk, we note that in certain cases, more intensive screening may achieve greater benefit only at the expense of much greater harm. A key conclusion of our investigation is that the way in which risk is increased should factor into the decision about whether and how to recommend screening.

We also offer the following specific conclusions: (i) individuals with increased risk of onset are both more likely to be overdiagnosed and more likely to benefit from early detection, and thus, appropriate screening ages and intervals depend on the trade-offs deemed acceptable; (ii) individuals with increased risk of progression are less likely to be overdiagnosed, and a shorter interval between screens may be recommended; and (iii) individuals with increased risk of cancer death are more likely to benefit from early detection, and a shorter interval between screens may also be recommended provided the survival benefit is greater with earlier detection (as assumed in this study). However, in general, we find that a lower PSA threshold for biopsy disproportionately increases overdiagnoses relative to additional lives saved. Therefore, we do not recommend this unless the increased chance of overdiagnosis is considered to be acceptable given the additional lives saved.

It might seem surprising that men with increased risk of onset can be more likely to be overdiagnosed. This association is a direct consequence of representing this increased risk using an HR: An HR greater than 1 implies not only onset at an earlier age, on average, but also a higher cumulative prevalence of latent disease at every age, relative to the general population. In other words, under this mechanism of increased risk, the high-risk subgroup always has a larger pool of latent disease, and consequently greater absolute magnitudes of screening harms and benefits. However, the higher chance of overdiagnosis due to increased risk of onset may be partially or completely offset by increased risk of progression, so the net effect on the chance of overdiagnosis depends on the relative magnitudes of these increased risks.

This work was motivated by research linking the frequency of prostate cancer, and particularly metastatic prostate cancer, with germline mutations in BRCA2 and other DNA repair genes (26, 27). As genetic screening appears set to expand in these patients, counseling unaffected family members may become increasingly common. Within our model, we estimated that BRCA2 germline mutation carriers in the IMPACT study have 81% higher risk of onset than the general population. If BRCA2 is only associated with increased risk of onset, we would recommend a wider range of screening ages than in the average-risk population. However, there is accumulating evidence that men with germline BRCA1/2 mutations also exhibit varying degrees of more aggressive tumor features and worse baseline survival than average-risk men (12). A recent study estimated HR = 2.36 for the risk of metastasis after treatment and HR = 2.17 for cause-specific survival after treatment among all (BRCA1 and BRCA2) mutation carriers combined (13). These results suggest that BRCA2 carriers may resemble the highest risk subgroup in Fig. 2 and motivates considering an expanded range of screening ages and more frequent screen tests in these men. However, based on our analysis, lowering the PSA threshold for biopsy (e.g., using a PSA threshold of 1.0 ng/mL over a more accepted threshold like 3.0 ng/mL) would not be well advised.

The problem of tailoring screening to subgroups with differing genetic risk has been recognized previously. In breast cancer, Kurian and colleagues considered women with BRCA1 and BRCA2 mutations and developed a tool to project disease incidence and mortality under different prevention and screening strategies (28, 29). Heijnsdijk and colleagues examined natural history in these women and evaluated mammography screening with and without MRI (30). In prostate cancer, Yen and colleagues considered risk stratification using SNPs and presented suggestions for more intensive screening in men with higher risk to reduce mortality by a similar magnitude to that achievable in the average-risk population (31). Pashayan and colleagues projected implications of screening based on polygenic risk rather than age in breast and prostate cancer (32). These studies also reflect the importance of understanding mechanisms and magnitudes of increased risk before developing targeted screening strategies.

In practice, developing an appropriate screening strategy in a particular subgroup requires prospectively collected serial screening and incidence data and a model that can estimate average and increased risks of disease onset and sojourn times (33). For prostate cancer and other cancers, natural history has already been studied for the average-risk population and certain increased risk subgroups (34). Once preferred screening strategies are determined in particular high-risk subgroups, further research is needed to investigate programs to identify the high-risk strata (e.g., genetic screening).

Like all modeling applications, ours is subject to limitations. The FHCRC model is a simplification of PSA growth and cancer progression designed to approximate the general male population. Because different components of the model were estimated using different data sources, its absolute projections may be less reliable than relative comparisons. Furthermore, although it closely approximates U.S. prostate cancer incidence trends with and without PSA screening by age, stage, and grade, the model abstracts a set of complex biological processes and interventions, and it may not generalize to particular subgroups. For example, increased risks may not be simple multiples of average risks. Our main data source for germline BRCA mutation carriers, the IMPACT study, is also limited. Importantly, carriers in this study may not be representative of the general carrier population. Furthermore, it involves only one round of screening and no metastatic cases. Thus, we cannot use this study to determine precisely which mechanism of increased risk (onset or progression) explains the observed incidence in that study. Consequently, we also consider other studies that bear on the mechanism of increased risk among BRCA1/2 mutation carriers.

In conclusion, different mechanisms and magnitudes of increased prostate cancer risks produce different levels of harms and benefits for a given screening approach. Consequently, screening guidelines for men with increased risk should take evidence about these factors into account. In general, although an efficacious screening test can produce greater benefit in subgroups with increased risk than in the average-risk population, more intensive screening does not always improve harm–benefit trade-offs.

No potential conflicts of interest were disclosed.

The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NCI.

Conception and design: R. Gulati, H.H. Cheng, P.H. Lange, P.S. Nelson, R. Etzioni

Development of methodology: R. Gulati, R. Etzioni

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Gulati

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Gulati, R. Etzioni

Writing, review, and/or revision of the manuscript: R. Gulati, H.H. Cheng, P.H. Lange, P.S. Nelson, R. Etzioni

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.H. Lange

Study supervision: R. Etzioni

This work was supported by grant numbers U01 CA199338 (to R. Gulati and R. Etzioni) and P50 CA097186 (to all authors) from the NCI. H.H. Cheng received a 2015 Prostate Cancer Foundation Young Investigator Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.