Racial disparities in prostate cancer mortality: a model-based decomposition of contributing factors

Abstract

To investigate the relative contributions of natural history and clinical interventions to racial disparities in prostate cancer mortality in the United States, we extended a model that was previously calibrated to Surveillance, Epidemiology, and End Results (SEER) incidence rates for the general population and for Black men. The extended model integrated SEER data on curative treatment frequencies and cancer-specific survival. Starting with the model for all men, we replaced up to 9 components with corresponding components for Black men, projecting age-standardized mortality rates for ages 40-84 years at each step. Based on projections in 2019, the increased frequency of developing disease, more aggressive tumor features, and worse cancer-specific survival in Black men diagnosed at local-regional and distant stages explained 38%, 34%, 22%, and 8% of the modeled disparity in mortality. Our results point to intensified screening and improved care in Black men as priority areas to achieve greater equity.

Prostate cancer has the largest racial disparities of any cancer in the United States. Black men have a risk of diagnosis that is 1.67 times (180 vs 108 per 100 000) and a risk of cancer death that is 2.06 times (36 vs 17 per 100 000) that for White men (1). The causes of these disparities are multifactorial and result from the relationships between the social and structural factors that define race, health-care delivery, and cancer biology.

Studies have shown that disparities in prostate cancer incidence reflect some increased risk that can be attributed to germline genetics (2-3). At least one study suggested that increased genetic risk of prostate cancer among Black men in the United States may reflect genetic bottlenecking that resulted from the transatlantic slave trade (4). However, roughly 70% of prostate cancer diagnoses in the United States occur sporadically. Disparities in prostate cancer incidence may also reflect the influence of structural determinants of equity (eg, structural and systemic racism, environmental racism, and economic and public policies), which drive harmful exposures (eg, violence, stress, poverty, pollution, and unhealthy food sources) that have measurable adverse biological effects in a wide range of diseases (5). Structural determinants of equity may influence other relevant health factors in cancer development and detection, including social, health-care, and biological determinants of health (2-3).

Disparities in mortality also reflect the impact of structural and social inequity, which are more likely mediated through health factors including access to, utilization of, and quality of care [ie, less frequent curative treatment (6-7) and lower quality care (8) for Black patients]. Several studies have demonstrated that Black patients have similar or better oncologic outcomes when they receive care in equal-access settings (9-12). It is important to note that these studies match patients not only by cancer characteristics but also by social factors (eg, eligibility criteria for Department of Veterans Affairs, academic center, or clinical trial populations).

Despite demonstrated inequities in care utilization and outcomes, few national guidelines provide specific recommendations, such as the age to initiate discussions about screening, for Black men (13-14). Racial equity in prostate cancer outcomes requires an acknowledgement that prostate cancer has a different natural history and worse outcomes among Black men. However, tailoring policies to address racial inequities and disparities in prostate cancer requires an understanding of the driving factors that contribute most to the excess mortality in Black men so that policies and interventions that focus on the most influential drivers can be prioritized.

Modeling is an established tool for linking cancer screening and treatment patterns to disease incidence and mortality outcomes. The Cancer Intervention and Surveillance Modeling Network (CISNET) has developed models of prostate cancer for the general population and for Black men in the United States. The models account for disease natural history, including rates of developing disease, tumor progression, and diagnosis. The model parameters were estimated so that incidence patterns projected by the models under population screening rates match those observed in the Surveillance, Epidemiology, and End Results (SEER) registry. A study involving 3 CISNET prostate cancer models estimated that, compared with the general population, Black men have 1.28-1.56 times the risk of developing prostate cancer and, once developed, 1.44-1.75 times the risk of diagnosis with distant stage disease in the absence of screening (15). These models have not yet been used to evaluate the extent to which these differences (as opposed to differences in care) explain disparities in prostate cancer mortality.

In this study, we quantify the relative contributions of differences in disease natural history, historical screening and treatment utilization, and cancer-specific survival to racial disparities in prostate cancer mortality. We extend a CISNET model of prostate cancer incidence to project mortality rates for the general population and for Black men in the United States. We then deconstruct the mortality differences to quantify how much each factor explains the modeled disparities. The results are intended to help guide investments in health interventions that can create greater equity in outcomes.

Methods

Overview

We extend a CISNET model of prostate cancer incidence that was previously estimated for the general population and for Black men in the United States. The extended model projects prostate cancer mortality using simulated cancer-specific survival times that reflect historical patterns of early detection and curative treatment. Relative contributions of differences in disease natural history, screening trends, frequencies of curative treatment, and cancer-specific survival to modeled disparities in mortality are calculated by starting with model projections for the general population and then systematically updating these one at a time using the inputs for Black men. This study was determined to be exempt by the Fred Hutchinson Cancer Center institutional review board.

Modeling incidence

The Fred Hutchinson Cancer Center prostate cancer model has been described in detail previously (16-17). Briefly, the model consists of 2 linked submodels. One submodel simulates individual prostate-specific antigen (PSA) levels over a man’s lifetime (Supplementary Figure 1, available online). On average, PSA grows exponentially with a higher growth rate after onset of preclinical (ie, asymptomatic, biopsy detectable) low- or intermediate-grade (Gleason score ≤7) cancer and an even higher growth rate after onset of preclinical high-grade (Gleason score ≥8) cancer. The PSA growth submodel was estimated using linear mixed models fit to longitudinal PSA test results in the placebo arm of the Prostate Cancer Prevention Trial using end-of-study prostate biopsies to identify cases and noncases.

The second submodel represents disease onset, progression from local-regional to distant stage disease, and progression from preclinical to clinical disease (Supplementary Figure 2, available online). The hazard of disease onset increases with age following a Weibull distribution (Supplementary Figure 3, available online). The probability of high-grade disease increases with age at onset following a generalized logistic growth curve (Supplementary Figure 4, available online). After onset, hazards of metastasis and clinical diagnosis increase with individual PSA levels (Supplementary Figures 5 and 6, available online).

To estimate the natural history submodel for the general population, we simulated men over the period 1975-2000. We relied on a previous reconstruction of historical PSA screening trends in the United States that was based on responses from the National Interview Health Survey and the SEER–Medicare linked database (Supplementary Figures 7, available online) (18). We assumed the frequency of receiving biopsy after a test result with PSA of at least 4.0 ng/mL increased with the PSA level and decreased with age as observed in the Prostate, Lung, Colorectal, and Ovarian cancer screening trial (16). Sensitivity of biopsies to detect occult cancer increased over calendar years to approximate the dissemination of more modern biopsy schemes (19-20). Then, conditional on the PSA growth submodel, we identified parameters in the natural history submodel that most closely reproduced observed incidence in the SEER program by age, year, stage, and grade at diagnosis (15-16, 20-21). In a subsequent analysis, we re-estimated the natural history submodel for Black men accounting for their PSA utilization patterns and assuming no differences in the PSA growth submodel or in frequencies of prostate biopsy compared with the (predominantly White) participants of the Prostate Cancer Prevention Trial and Prostate, Lung, Colorectal, and Ovarian trial, respectively (22-23).

After estimating models for the general population and for Black men, PSA screening rates were updated to account for effects of US Preventive Services Task Force recommendations against screening for ages 75 years and older in 2008 and for all ages in 2012. Experimental parameters that filtered the reconstructed screening rates for these age groups starting in these years were identified so that the model approximately matched observed incidence rates (24). In the present study, the previously identified filter parameters were assumed not to differ by race (25).

Modeling mortality

We previously estimated receipt of curative treatment (surgery or radiation) within 12 months of prostate cancer diagnosis among local-regional stage patients by fitting a multinomial regression model to SEER data by age and grade at diagnosis for all races combined (26). For this study, the regression model was extended to estimate receipt of curative treatment for Black patients. We also previously estimated receipt of concurrent androgen deprivation therapy by fitting a logistic regression model to data from the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) registry for all races combined (26). For this study, we assumed receipt of adjuvant androgen deprivation therapy in Black patients was similar to that in the CaPSURE registry.

We previously estimated prostate cancer survival among untreated patients diagnosed in SEER in 1980-1986 by age, stage, grade, and race (Supplementary Figure 8, available online) (27-28). In this study, differences in baseline survival by race reflect differences in the effectiveness of care not explicitly captured by the benefits of screening and curative treatment, which are described next.

For local-regional stage patients, baseline prostate cancer survival was improved to reflect the efficacy of curative treatment based on results for surgery reported by the Scandinavian Prostate Cancer Group 4 randomized controlled trial (29). Similar to a previous study for all races (26), we used age-specific improvements for surgery or for radiation with concurrent androgen deprivation therapy (hazard ratio [HR] = 0.50 for ages younger than 65 years, HR = 0.63 for ages 65-74 years, and HR = 0.90 for ages 75 years and older) and less favorable improvements for radiation monotherapy (HR = 0.63 for ages younger than 65 years and HR = 0.90 for ages 65 years and older). In this study, we assumed that efficacy of curative treatment in Black patients was similar to that in the (predominantly White) participants of the Scandinavian Prostate Cancer Group 4.

We previously modeled the benefit of PSA screening using a cure fraction that increases with lead time (30). Specifically, screen-detected patients who were simulated to die of prostate cancer in the absence of screening were reassigned to die of other causes with probability $1-\text{exp}\hspace{0.17em}\left(\theta \times L\right)$, where $L$ is the individual lead time and $\theta$ is an exponential rate parameter that we estimated using mortality data from the European Randomized Study of Screening for Prostate Cancer (ERSPC) (31). In this study, to account for higher baseline survival in the ERSPC, which began enrollment in 1993, than was observed in SEER in 1980-1986, we assumed the improvement manifested linearly over the period 1990-1995 and specified a hazard ratio with form $1-\left(1-\varphi \times A\right)\left(Y-1990\right)/5$, where $A$ and $Y$ are age and calendar year at diagnosis. We simultaneously estimated $\theta$ and $\varphi$ to match published prostate cancer mortality results in the ERSPC (30). Confidence intervals (CIs) were obtained conditional on all other model parameters (32). We assumed that improvements in baseline cancer-specific survival did not differ by race and that efficacy of PSA screening in Black men was similar to that in the (predominantly White) participants of the ERSPC.

Combining model extensions, we simulated populations of all men and Black men following age distributions in SEER in 1975 (Supplementary Figure 9, available online). All-cause mortality was generated from US life tables by birth cohort and race conditional on survival to the starting age in 1975 (Supplementary Figure 10, available online). Prostate cancer diagnoses and deaths were tracked over time, and prostate cancer incidence and mortality rates were projected over the period 1975-2020. Note that, because mortality was projected only for men diagnosed during this period, our projected mortality rates are actually incidence-based mortality rates (33-34). However, because incidence-based mortality rates for prostate cancer converge to total mortality rates within 15 years of follow-up (35), we examined these projections starting in 1990 and refer to them as mortality rates without this qualification.

Analysis

Beginning with the model for all men (model 0), we systematically replaced components for all men with corresponding components for Black men and projected prostate cancer incidence and mortality rates per 100 000 men ages 40-84 years. We first replaced the starting age distribution in 1975 (model 1), then also all-cause mortality (model 2), then also natural history parameters (first for developing preclinical disease [model 3], then for tumor stage and grade [model 4]), then also clinical diagnosis (model 5), then also receipt of PSA screening (model 6), then also receipt of curative treatment (model 7), then also baseline survival for men diagnosed with local-regional stage disease (model 8), and finally then also baseline survival for men diagnosed with distant stage disease (model 9).

Relative contributions of each component to modeled disparities in incidence and mortality were calculated as the ratio of 1) differences in projections because of the replaced component ([model $X+1$]–[model $X$] for incidence components $X=0,...,5$ and mortality components $X=0,...,8$) to 2) differences between projections for Black men and for all men ([model 6]−[model 0] for incidence components and [model 9]−[model 0] for mortality components). Quantitative results are presented for the year 2019, the most recent year for which observed data on both incidence and mortality rates are available.

Results

Estimated model components for Black men

In the PSA era, Black men diagnosed with local-regional stage prostate cancer were estimated to have a lower frequency of undergoing surgery and (for younger ages) slightly higher frequency of receiving radiation, independent of patient age or tumor grade, compared with the general population (Supplementary Figure 11, available online).

We estimated the PSA benefit parameter to be $\theta =0.19$ (95% CI = 0.06 to 0.37) and the age-dependent improvement in baseline cancer-specific survival to be $\varphi =0.0105$ (95% CI = 0.0097 to 0.0111) (Supplementary Figure 12, available online). Thus, a man who would have died of prostate cancer in the absence of screening but whose cancer was detected 5 years early by screening has a 62% probability that early diagnosis and curative treatment will result in cure, thereby extending his life until death from other causes. If his lead time is instead only 1 year, this probability is only 18%. Baseline cancer survival in the PSA era is higher than in the pre-PSA era before accounting for the potential benefits of early detection and curative treatment. This improvement is greater for younger men, corresponding in 1995 and thereafter to HR = 0.50 for men diagnosed at age 50 years, HR = 0.60 for men diagnosed at age 60 years, and so on.

Estimated contributions to disparities

Observed and projected age-standardized incidence and mortality rates for Black men and for all men are shown in Figure 1. Projected incidence rates for both groups reasonably reproduce observed rates. Projected mortality rates for all men underestimate observed rates in 1990, but the discrepancy shrinks in later years. Projected mortality rates for Black men produce a premature peak in 1990 and overproject observed rates starting in 2002 with a growing discrepancy in later years.

Figure 1.

A) Age-standardized prostate cancer incidence rates per 100 000 for all men and for Black men aged 40-84 years at diagnosis from the Surveillance, Epidemiology, and End Results (SEER) program (data from the core 9 registries for 1975-2018 and from the core 8 registries after Detroit was removed for 2019) and corresponding model-projected incidence rates after sequentially replacing components for all men with components for Black men. B) Age-standardized prostate cancer mortality rates per 100 000 for all men and for Black men aged 40-84 years at death from the National Center for Health Statistics (NCHS) and corresponding model-projected mortality rates after sequentially replacing components for all men with components for Black men.

The relative magnitudes of each component’s contribution to disparities in incidence in 2019 are summarized in Table 1. Starting with the model projections for the general population (model 0), replacing the age distribution input with the one for Black men (model 1) had a negligible effect on projected incidence rates. Also using the all-cause mortality rates for Black men (model 2) similarly had a negligible effect. However, also using the risk of developing prostate cancer for Black men (model 3) dramatically increased projected incidence rates, accounting for 89% of the modeled disparity in 2019. Also using the probability of high-grade disease and the risk of metastasis for Black men (model 4) further increased projected incidence rates, accounting for another 10% of the modeled disparity. In contrast, also using the risk of clinical diagnosis and screening rates for Black men (models 5 and 6) had modest effects (−2% and 4%, respectively).

Table 1.

Age-standardized prostate cancer incidence rates per 100 000 for all men and for Black men aged 40-84 years at diagnosis from the Surveillance, Epidemiology, and End Results (SEER) program in 2019, age-standardized prostate cancer mortality rates per 100 000 for all men and for Black men aged 40-84 years at death from the National Center for Health Statistics (NCHS) in 2019, corresponding model-projected incidence and mortality rates after sequentially replacing components for all men with components for Black men, and relative contributions to modeled disparities attributed to each component

Source	Incidence			Mortality
	Rate	Δa	%Δb	Rate	Δa	%Δb
Observed rates for all men (from SEER or NCHS)	419.4	–	–	44.7	–	–
Model 0: base model for all men	431.8	–	–	46.2	–	–
Model 1: model 0 + age distribution for Black men	433.3	1.5	0%	46.1	−0.1	0%
Model 2: model 1 + all-cause mortality for Black men	430.2	−3.1	−1%	46.5	0.4	1%
Model 3: model 2 + disease onset for Black men	704.3	274.2	89%	76.1	29.5	38%
Model 4: model 3 + disease features for Black men	736.4	32.1	10%	102.3	26.2	34%
Model 5: model 4 + clinical diagnosis for Black men	729.2	−7.1	−2%	100.1	−2.2	−3%
Model 6: model 5 + screening for Black men	740.9	11.6	4%	99.0	−1.1	−1%
Model 7: model 6 + treatment for Black men	–	–	–	100.0	1.0	1%
Model 8: model 7 + localized survival for Black men	–	–	–	117.4	17.4	22%
Model 9: model 8 + distant survival for Black men	–	–	–	124.0	6.6	8%
Observed rates for Black men (from SEER or NCHS)	657.9	–	–	95.1	–	–

^aDifference in modeled rate due to replacing the indicated component for all men with the one for Black men.  – = not applicable; NCHS = National Center for Health Statistics; SEER = Surveillance, Epidemiology, and End Results.

^bDifference expressed as a percentage relative to the total difference in modeled rates for Black men and all men.

The relative magnitudes of each component’s contribution to disparities in mortality in 2019 are also summarized in Table 1. Replacing the age distribution and all-cause mortality rates for Black men (models 1 and 2) also had negligible effects on projected mortality rates. Replacing the risk of developing prostate cancer for Black men (model 3) explained 38% of the modeled disparity. Replacing the probability of high-grade disease and the risk of metastasis for Black men (model 4) explained another 34% of the modeled disparity. Replacing baseline cancer-specific survival for Black men diagnosed with local-regional stage disease (model 8) explained another 22%. Replacing baseline cancer-specific survival for Black men diagnosed with distant stage disease (model 9) explained another 8%. Replacing the risk of clinical diagnosis, screening rates, and the frequencies of curative treatment (models 5, 6, and 7) had minor effects (−3%, −1%, and 1%, respectively).

Discussion

Understanding which factors contribute most to racial disparities in prostate cancer mortality is a challenging research topic. A quantitative decomposition of the contributing factors that includes disease natural history and clinical interventions requires a model of that natural history and a synthesis of data on utilizations and efficacies of those interventions. In this study, we extended a previously developed model of prostate cancer natural history and added data on curative treatment frequencies and cancer-specific survival for the general population and for Black men.

Synthesizing available information for the present analysis included estimating curative treatment utilization for Black patients, for which we used data from SEER and CaPSURE. These large surveillance databases allowed estimating frequencies of standard prostate cancer therapies within granular subgroups. Our finding that Black patients have lower frequencies of surgery compared with White patients after adjusting for age and cancer grade, for example, is consistent with prior studies (36). Our estimate of improved baseline cancer-specific survival in the PSA era compared with the pre-PSA era is also consistent with prior studies (30, 37-38). We assumed the same efficacies of screening and curative treatment for Black men as for the predominantly White participants in landmark randomized trials. These assumptions are supported by studies that demonstrated that Black and White patients matched by tumor characteristics have similar benefits of treatment in various clinical settings (11-12), and to the best of our knowledge, there is no credible justification for assuming otherwise.

Our results identify the key drivers of the excess incidence and mortality in Black men and should aid in prioritizing interventions to address persistent disparities. We found that the increased frequency of preclinical disease in Black men, which explained 89% of the modeled disparity in incidence, explained 38% of the modeled disparity in mortality. Although we do not yet have a direction for narrowing the gap in incidence, an analysis involving 3 CISNET models suggested beginning screening earlier in Black men to provide them with an equal opportunity to benefit from early detection (15).

Another 34% of the modeled disparity in mortality was attributed to more aggressive tumor features (stage and grade) at diagnosis in Black men. The substantial role attributed to this factor underscores the potential value of intensified screening (ie, screening earlier and more often) in Black men. The contribution of advanced cancer stage and grade among Black men may reflect more aggressive disease, delays in diagnosis, or both and is important to consider in the context of tailored screening recommendations for Black men. Two CISNET models projected that more frequent screening (annual) over a targeted age range (45-69 years) would reduce prostate cancer mortality for Black men by a greater relative amount compared with historical (approximately every 2 years) screening while also controlling overdiagnosis (39).

Still, another 30% of the modeled disparity was attributed to worse baseline cancer-specific survival for Black patients. This baseline survival represents a catchall measure of the effectiveness of care not captured by effects of screening and curative treatment utilization. In fact, racial differences in historical screening and curative treatment utilization given patient age and tumor characteristics accounted for relatively small contributions in the modeled disparities in mortality. The large contribution of differences in baseline survival, which implicates a nonspecific deficiency in the effectiveness of care Black patients receive, is consistent with previous proposals to educate patients and providers about barriers to care, increase trial participation, and undertake other measures designed to provide equitable access to quality care (36). Differences in baseline survival could also be related to apparently worse outcomes for Black patients diagnosed with low-risk disease and followed with active surveillance (40), possibly because of underascertainment of high-grade disease (41). Interestingly, with this baseline survival held fixed at the 1995 level, the model overprojects mortality rates for Black men by a magnitude close to the amount attributed to differences in baseline survival, suggesting that further research is needed into whether the racial disparity in the quantity and quality of effective care has narrowed since the mid-1990s.

This study has several limitations. The results depend entirely on a model that is necessarily a simplification of complex biological processes, clinical interventions, and their interaction with each other and with race—a social and historical construct. We assumed that Black men have similar PSA levels and biopsy frequencies and receive biopsies with similar diagnostic accuracy. The model projections do not account for systematic differences in access to quality care, trust in the health system, understanding of disease risk, socioeconomics, rural–urban geographic characteristics, or other relevant social determinants of health. Estimated contributions of several factors (eg, differences in screening rates by race) were small and numerically unreliable. This limitation, coupled with the model’s overprojected mortality rates for Black men in recent years, supports a focus on the qualitative takeaways of this analysis, particularly which factors appear to be the major drivers of modeled disparities. Finally, the relative contributions examined in this study may shift in response to future changes in clinical practice, for example, if screening rates among younger Black men increase.

Despite these limitations, the insights of this study are informative. They underscore the potentially critical importance of targeted and intensified screening for Black men to find and treat prostate cancers at an early stage. Although our analysis suggests that historical differences in curative treatment utilization played a limited role, it highlights that less effective care, whether due to lower quality or uptake of care among Black patients, and more aggressive disease and/or delays in diagnosis appear to be major contributors to existing disparities in outcomes. Studies are needed to better understand and evaluate the benefits, harms, and costs associated with addressing these priority areas.

Data Availability

The data underlying this article were derived from a combination of public sources—population counts, prostate cancer incidence rates, initial treatment frequencies, and cancer-specific survival rates from the Surveillance, Epidemiology, and End Results program (https://seer.cancer.gov/); prostate cancer mortality rates from the National Center for Health Statistics (https://www.cdc.gov/nchs/); U.S. life tables from the Human Mortality Database (https://www.mortality.org/); and published results from prostate cancer screening and treatment trials—and data use agreements about prostate-specific antigen test results and biopsy outcomes with Prostate Cancer Prevention Trial researchers (https://www.cancer.gov/types/prostate/research/prostate-cancer-prevention-trial-qa) and about androgen deprivation therapy utilization with the Cancer of the Prostate Strategic Urologic Research Endeavor (https://urology.ucsf.edu/research/cancer/capsure). Additional information about the Fred Hutchinson Cancer Center prostate cancer microsimulation model, parameter estimates, and output is available upon request from the corresponding author.

Author contributions

Roman Gulati, MS (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Software; Validation; Visualization; Writing—original draft; Writing—review & editing), Yaw A Nyame, MD (Writing—original draft; Writing—review & editing), Jane M Lange, PhD (Data curation; Formal analysis), Jonathan E Shoag, MD (Writing—review & editing), Alex Tsodikov, PhD (Data curation; Formal analysis; Methodology), Ruth Etzioni, PhD (Conceptualization; Funding acquisition; Supervision; Writing—original draft; Writing—review & editing).

Funding

This work was supported by the National Cancer Institute (U01CA199338, U01CA253915, L60CA274879, P50CA97186, and R50CA221836). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. JES was supported by the Damon Runyon Cancer Research Foundation and the Bristol Myers Squibb Foundation. YAN was supported by the Andy Hill Cancer Research Endowment (CARE) Fund (2021-DR-01) and the Department of Defense (W81XWH2110531).

Monograph sponsorship

This article appears as part of the monograph “Reducing Disparities to Achieve Cancer Health Equity: Using Simulation Modeling to Inform Policy and Practice Change,” sponsored by the National Cancer Institute, National Institutes of Health ([Modeling Precision Interventions for Prostate Cancer Control; 5 U01 CA253915-02]).

Conflicts of interest

YAN is a scientific advisor for Ortho-Clinical Diagnostic and Immunity Bio Inc. JES served as a key opinion leader for Fortec Medical and received grant support from Bristol Myers Squibb Foundation. The other authors declared no potential conflicts of interest.