SEARCH

SEARCH BY CITATION

Keywords:

  • black men;
  • white men;
  • prostate-specific antigen;
  • PSA testing patterns;
  • Medicare;
  • National Health Interview Survey

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

BACKGROUND.

Frequencies of prostate-specific antigen (PSA) test administration were not actively monitored on a national level during the first decade of PSA testing. The objectives of this article were to reconstruct patterns of PSA testing between black and white men in the US and to determine the extent of any racial disparity in PSA use.

METHODS.

Data from the 2000 National Health Interview Survey were used to model the adoption of PSA and to estimate the distribution of age at first test. Longitudinal Medicare claims data were used to estimate the distribution of intervals between tests. The rates of initial and subsequent tests were then combined by simulation to reconstruct individual screening histories. Results are from the reconstructed model.

RESULTS.

Overall, 45% of white men and 43% of black men within ages 40–84 years had at least 1 PSA test by the year 2000. The authors found that among older men, whites adopted PSA screening earlier than blacks, whereas among younger men, this trend was reversed, with blacks adopting screening earlier than whites. Annual testing frequencies generated by the simulation model were higher for white men aged ≥60 years and higher for black men aged <60 years.

CONCLUSIONS.

Findings indicated fairly similar patterns overall of PSA testing for blacks and whites. These similarities indicated that racial disparity in PSA testing is probably not a major factor behind current racial differences in prostate cancer mortality rates and declines. Knowledge of patterns of screening is important to an understanding of the impact of population screening on cancer incidence and mortality, but retrospective data sources have significant limitations when used to estimate these patterns of care. Cancer 2007. © 2007 American Cancer Society.

The prostate-specific antigen (PSA) test was approved by the US Food and Drug Administration in 1986 for monitoring disease status in men with prostate cancer and in 1992 for diagnosis of the disease. Once approved, the test was performed on men with urological symptoms as well as on those who were asymptomatic in an effort to diagnose prostate cancer early and to affect the mortality rate. Use of the test was associated with a dramatic rise in prostate cancer incidence in the early 1990s,1 followed by a subsequent decline. The incidence of distant-stage disease, which had been relatively flat, started to decline dramatically in the early 1990s.2 Prostate cancer mortality also began to decline in the early 1990s, and the decline has continued.3 The role of PSA screening in explaining mortality declines has been questioned by many, but to date, there is no conclusive evidence of screening efficacy from randomized trials.4, 5 Moreover, the use of the PSA test as a screening test in asymptomatic men has been questioned because of the large reservoir of indolent disease in the population and the associated problems of overdiagnosis and overtreatment.6

Black American men are known to have higher incidence and mortality from prostate cancer than white American men. Racial differences have been noted in both pre-PSA and post-PSA eras.3 After a dramatic increase in the early 1990s, incidence rates declined and have recently resumed the pre-PSA trend for white men but not for black men. Whether racial/ethnic differences in prostate cancer mortality trends result from disparities in screening and treatment practices or other (biologic) differences is a hotly debated topic. It remains challenging to disentangle the influence of different domains of inequality in order to understand their separate contributions to prostate cancer disparities.

To make quantitative inferences about the role of PSA in explaining the decline in prostate cancer mortality as well as racial differences in incidence, it is important to understand PSA screening behavior and to quantify how many and how often men have been screened each year in the US since the start of the PSA era. Tracking of screening rates during the early years of the PSA era was not done in real time. Some reconstruction of the diffusion process has been preformed by using Medicare claims linked to Surveillance, Epidemiology, and End Results (SEER) program data,7 but these results pertain only to men older than 65 years of age.8

This article combines cross-sectional data from the 2000 National Health Interview Survey (NHIS) and longitudinal claims information from a more updated linkage of Medicare claims to SEER data to develop a comprehensive model for dissemination and use of the PSA screening test in the US during the years 1988–2000. The 2000 NHIS data on PSA testing include men aged ≥40 years; 3 questions from these data are used to reconstruct the initial screening experience. The SEER-Medicare data includes men aged ≥65 years from 1991 through 2000, and these data are used to reconstruct the serial testing experience after initial PSA testing. Results of these 2 analyses are then combined in a stochastic simulation procedure that produces individual PSA testing histories given age and race. By simulating PSA histories for men born between 1900 and 1960, we were able to provide a complete reconstruction of the utilization of PSA in the United States. We then used the generated histories to summarize the cumulative screening experience between blacks and whites aged ≥40 in the US. We validated these histories against data from NHIS and SEER-Medicare that were not used to develop the simulation model. The NHIS question on the number of PSA tests in the last 5 years, which is not used in the model estimation, was used for validation. Information on frequency of men who had at least 1 test each year from SEER-Medicare was not used in estimation but was used to validate the model reconstruction. Only the time intervals between consecutive PSA tests from SEER-Medicare claims data were used to inform the model on the periodicity of PSA tests.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Estimating Use of Initial PSA Tests

The National Health Interview Survey (NHIS) 20009 data consists of face-to-face interviews on health and other characteristics of each member of a nationally representative sample of households. We used 3 questions from the cancer supplement to estimate the dissemination of first PSA, namely: 1) Have you ever heard of PSA?, 2) Have you ever had a PSA (if you have heard of PSA)?, 3) At which age did you have a first PSA (for those who declared having a PSA)? Age at first PSA was reported in 5-year age groups (<40, 40–44, 45–49, …, 65–69, and ≥70). The question on the number of PSA tests in the last 5 years for men who had heard of PSA and the time since their last PSA were used exclusively to validate results from the model. The study population comprised 6937 men (6084 white and 853 black) ages 40 to 84 years who had replied to the first question. Those who replied no to the question were included in the analysis and considered to be men who had never had a PSA test.

Although the PSA test was approved only for screening in 1992, it was approved in 1986 for monitoring disease progression, and we follow others in assuming 1988 to be the first year in which it was used for screening in the population.8 Among respondents to the NHIS survey, 6% (387 white and 55 black) declared having had a PSA at an age that corresponded to a calendar period prior to 1988. For these cases, we have assumed that they had a PSA test between 1988 and 2000. Percentages and proportions were calculated by using the NHIS sample weights to reflect the US population from which the sample was drawn.

We considered age/year at first PSA test as a time-to-event variable, and we used survival analysis techniques to estimate the distribution of age at first PSA by birth cohort. Specifically, we used the SAS LIFEREG procedure (SAS Institute, Cary, NC) to fit a log-logistic survival model, which incorporates interval censoring and allows extrapolation of estimates beyond the range of data. A separate model was fit to white and black men; each model included 6 birth cohorts (1916–1919, 1920–1924, 1925–1934, 1935–1944, 1945–1954, and 1955–1960). Because age at first PSA was reported in 5-year age intervals, the models were smoothed and could not capture the rapid increase of PSA testing that was observed between 1988 and 1991.8 We therefore adjusted our results to more closely match patterns of first PSA tests published previously by Legler et al.8 The Legler estimates reflect PSA testing frequencies in a cohort of men, aged ≥65 years in 1988. We computed adjustment factors that, when applied to the same cohort in our dataset, yielded estimates of first PSA testing rates that were similar to those of Legler et al. We then applied these adjustment factors across the age groups in our data. The adjustment is described in detail below.

Adjustment of cumulative probability of having a first PSA

To adjust the NHIS cumulative probabilities of having a first PSA to reflect the rapid increase between 1988 and 1991, we used numbers from Legler et al., which estimated PSA testing frequencies from 1988 to 1994 in a cohort of Medicare men born before 1923 (aged ≥65 years in 1988). We compared the Legler et al. estimate with men in the birth cohort 1920–1924 from the NHIS data. The reason for the comparison with a younger cohort of men is because NHIS self-reported data on older men were less reliable. For men born before 1920, there was a large proportion of men with inconsistencies on age at first PSA showing a lower data accuracy.

Let Ly and Py be respectively the race-specific cumulative probabilities of having a first PSA in year y, for men born before 1923 as estimated in Legler et al.8 and for men in the birth cohort 1920–1924 as estimated in this study from the NHIS data.

The adjustment for years y = 1988, …, 1994 is calculated as equation image.

To fill the gap between 1994 and 2000, we fit a logistic curve equation image through 2 points (1994, c1994) and (2000, 1). The adjustments for y = 1988 to 2000 were then applied to the cohort-specific cumulative curves Pi,y, where i denotes birth cohort and y denotes calendar year, to yield:

  • equation image

where P*i,y is the adjusted cumulative probabilities of having a first PSA, reflecting the rapid increase in PSA from 1988 to 1991.

Estimating repeat screening frequencies

The SEER-Medicare data reflect the linkage of the Surveillance, Epidemiology, and End Results (SEER) program data on clinical, demographic, and cause of death information for persons with cancer and Medicare claims.10 The SEER-Medicare data includes a file with Medicare claims for SEER cases diagnosed between 1991 and 2000 from 13 geographic areas representing approximately 14% of the US population10 and another file containing Medicare claims for a 5% random sample of the cancer-free population residing in the SEER areas (controls). In this study, both cohorts of patients were combined to provide population-based estimates of PSA screening frequencies after initial testing. In the analysis, a weight of 20 was applied to noncancer cases from the 5% random sample. Because Medicare is a health insurance program for persons ≥65 years of age, only men aged between 65 and 84 years of age, entitled to both Medicare Part A and B, and not health maintenance organization (HMO) enrollees, for the whole period between the year of their first PSA and 2000 were included in the analysis.

The Healthcare Common Procedure Coding System (HCPCS) current procedural terminology (CPT)-4 codes provided with Medicare claims data was used to identify PSA tests (CPT codes 86316 and 84153). The SEER-Medicare data do not have information on the reason for a PSA test. PSA tests conducted after prostate cancer diagnosis were eliminated. To eliminate follow-up PSA tests conducted in response to suspicious results, we deleted all PSA tests occurring within 3 months of a previous PSA test.

A proportional hazards survival model with a frailty parameter was used to estimate the distribution of intervals between consecutive PSA tests (gap times). Time-to-event is the time to the next PSA test, or to the end of follow-up (December 12, 2000), or to death depending on which occurred first. The frailty parameter allowed us to capture the correlation between PSA testing intervals within an individual. A frailty is an unobservable random effect that acts multiplicatively on hazard rates for the individual-specific time between PSA tests. Men with larger frailties will have larger hazards of subsequent PSA tests and more frequent screening. As is standard in frailty modeling, we assumed that the frailties come from a population that follows a gamma distribution with mean 1, and we estimated the variance of this distribution. A similar model was used in Cronin et al.11 to estimate repeat mammography behavior.

Covariates included in the proportional hazards model were age at the beginning of the gap time (categorized as 65–69, 70–74, 75–79, and 80–84 years of age), race, and an indicator variable for whether or not the year at the beginning of the gap time was before or after 1995. This last dummy variable splits the period into 2 halves, and it was included in the model to investigate behavioral changes in PSA screening in a later phase (1995 and thereafter) compared with an initial phase (1990–1994). The model was fit to include data for both races, and it included interactions of race with age and year at start of the interval. The reason for jointly estimating parameters for both races was to have comparative parameter estimates. The hazard of having a PSA test t months after the last PSA test is given as Equation 1:

  • equation image(1)

where μ is the frailty, assumed to have a gamma distribution with mean 1, and variance σ unknown and to be estimated from the data, and h0(t) is the baseline hazard for white men 70–74 years of age for calendar years prior to 1995, with the mean frailty μ = 1. The parameters exp(βi) give the relative risk, with respect to white men ages 70–74 years prior to 1995. For example, for a black man aged 83 years in 1996, his risk of having a PSA test in 12 months after the previous one is:

  • equation image(2)

and exp(β7black age 80–84 + β9black year ≥ 1995) gives the extra risk of having a next PSA compared with the risk of white men ages 70–74 years before 1995. Model parameters {β1, β2, …, h0(t),σ} were estimated by using an SAS macro12 that modifies the estimation-maximization (EM) algorithm to handle data on large numbers of individuals.

Combining first and repeated PSA screening tests to produce representative PSA histories

We developed a simulation procedure that combines results of 2 estimation models, ie, the distribution of calendar year (or age) at initial PSA and the distribution of interscreening intervals, to generate individual PSA screening histories that reflect the population screening experience. The SEER-Medicare data do not contain information on men younger than 65 years of age. Thus, for men ages 40 and 64 years, we have assumed that the distribution of the interscreening intervals is the same as was estimated for the youngest age group in the Medicare data, ie, men ages 65 and 69 years. We later validate this assumption by using responses to Question 3 (see above) from the NHIS data.

The simulation program, which can be obtained from http://cisnet.cancer.gov/interfaces/ comprises the following steps. For a single man, conditional on his year of birth and race:

  • Step 1. Generate date of first PSA test based on the cumulative distribution of time to first PSA test.

  • Step 2. Generate a frailty parameter from the gamma distribution that a man will keep throughout his lifetime.

  • Step 3. Generate a gap time to the next PSA test, conditional on the men's frailty parameter, age at the previous PSA test, race, and whether or not the previous PSA test occurred before 1995.

  • Step 4. Repeat step 3 until year 2000 or age 85 years, whichever comes first.

The main purposes of this simulation model are to generate PSA testing histories from 1988 to 2000 and to reconstruct the use of PSA testing patterns in the US. The simulation program does not model the time of diagnosis of prostate cancer or time of death. In applications that evaluate the effectiveness of PSA testing, prostate cancer diagnosis and death can be modeled externally and incorporated in microsimulation models that use PSA histories as input.

Model validation using NHIS 2000 and Medicare data

We validated estimates by comparing projections of screening frequencies based on the simulation program with observed data from both the NHIS and SEER-Medicare. PSA histories that were generated for 10,000 black and 10,000 white men born in each year from 1904 to 1960. On the basis of simulated testing histories, we empirically computed the distribution of the number of tests per individual between 1996 and 2000 and compared them with results from the NHIS 2000 question on the number of PSA tests in the last 5 years. The proportion of men ages 65 to 84 years who had at least 1 test each year from 1988 to 2000 calculated from the simulated screening histories was compared with respective frequencies of those in the SEER-Medicare data.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Use of Initial Testing as Estimated From NHIS Data

Figure 1 shows the modeled cumulative proportion of men born between 1920 and 1924 who had at least 1 PSA test by race and calendar year, without (dashed curves) and with (plain curves) the adjustment to better describe the rapid increase after 1991. Figure 2 shows the modeled cumulative proportion of men who had at least 1 PSA test with adjustments for all birth cohorts. The numbers at the end of the curves represent the percentages of men who had at least 1 PSA test by 2000, and they are the same for adjusted or unadjusted curves (The unadjusted curves are not shown in this figure.). Overall, we estimate 2723 (45%) white and 366 (43%) black men ages 40 and 84 years had at least 1 PSA screening test before 2000. The curves for whites compared with the ones for blacks are higher for men born before 1934, similar for men born between 1935 and 1944, and lower for men born after 1944. The model indicates that among men born before 1940, whites initiated PSA testing before blacks, whereas among men born after 1940, blacks initiated PSA testing before whites.

thumbnail image

Figure 1. Adjusted (solid) and unadjusted (dashed) cumulative percentages of men with a history of at least 1 PSA test by calendar year for white and black men born 1920–1924. Dashed curves are estimated from NHIS 2000 data, and plain curves are adjusted, using numbers from Legler et al.,8 to represent the rapid increase of PSA use between 1988 and 1991. The difference between the curves represents the adjustment made. Source: Modeled NHIS 2000 data.

Download figure to PowerPoint

thumbnail image

Figure 2. Cumulative percentages of men with a history of at least 1 PSA test by calendar year for white and black men from different birth cohorts. Curves are estimated from NHIS 2000 data and adjusted to represent the rapid increase of PSA use between 1988 and 1991. Numbers at the end of the curves represent the percentage of men having at least 1 PSA by the year 2000. Source: Modeled cross-sectional NHIS 2000 data.

Download figure to PowerPoint

Variability in PSA Testing Intervals Across Individuals

Frailty is a multiplicative factor on the hazard of having a subsequent PSA; a larger frailty implies more frequent PSA testing. Table 1 displays parameter estimates for the frailty survival model as a risk of having a subsequent PSA test relative to the baseline group (white men ages 70–74 years before 1995). The smaller the relative risk, the less frequently one is likely to have a subsequent PSA test compared with the baseline group. For example, before 1995, for black men ages 70–74 years, the relative risk (RR) is 0.85, which means that before 1995, black men ages 70–74 years were 15% less likely to have a subsequent PSA at any given point in time compared with white men ages 70–74 years. Before 1995, white men ages 70–79 years were the most frequent users of PSA. After 1995, testing frequency increased among black men (RR = 1.03) and decreased among white men (RR = 0.93).

Table 1. Frailty Model Parameter Estimates in Terms of Relative Risk of Having a Subsequent PSA Test by Race, Age Group, and Year of PSA Test
Parameter estimates as relative risk (Standard error)
RaceNo. of menTotal no. PSAAge at PSAYear PSA
65–6970–7475–7980–84≥1995
  • The baseline group is white men ages 70–74 years before 1995. Relative risk represents the risk of having a subsequent PSA at time t compared with the baseline group. For example, black men ages 70–74 years before 1995 had a 15% less probability of having a subsequent PSA at time t than white men of same age. Source: Modeled Medicare SEER data (1991–2000).

  • *

    Relative risks for calendar year 1995 or later are calculated by multiplying the age/race relative risk by the Year PSA ≥ 1995/race relative risk as given by Equation 1 in the text.

Blacks798022,0560.880.850.810.821.03
(0.006)(0.005)(0.006)(0.007)(0.006)
Whites90,790300,7930.9110.990.860.93
(0.002)(0.002)(0.002)(0.001)
Frailty (Variance)    0.75  
Relative risk calculations* of a next PSA test by age after 1995
Age at PSA65–6970–7475–7980–84
Blacks0.900.870.840.84
Whites0.850.930.920.80

The population of frailties follows a gamma distribution with an estimated variance of 0.75 (Table 1). Figure 3A shows the distribution of the frailties and its 25%, 50%, and 75% quartiles, respectively. Figure 3B displays corresponding survival curves (time to next PSA) for white men ages 70–74 years before 1995 (baseline group) with frailties corresponding to each quartile. Among the 25% of men tested least frequently, at least 51% had tests that were more than 2 years apart. Among the 25% of men tested most frequently, at least 53% had PSA tests every year (or more often). Figures 3C and 3D display survival times to the next PSA test for white and black men ages 70–74 years before and after 1995 and survival time to next PSA for white men before 1995 by age at PSA test. These figures indicate that race, age, and calendar year account for only a small fraction of the between-individual heterogeneity in interscreening intervals and that the bulk of this variability is due to other factors or to individual-specific preferences.

thumbnail image

Figure 3. Summary description of the frailty survival model of gap times between consecutive PSA tests. Source: Modeled SEER-Medicare gap times. (A) Gamma distribution of frailties and 25%, 50%, and 75% quartiles. (B) “Survival” time to next PSA for men with different frailties corresponding to quartiles. (C) “Survival” time to next PSA for white and black men aged 70–74 before and after 1995. (D) “Survival” time to next PSA for white men before 1995 by age at PSA test.

Download figure to PowerPoint

Validation of Estimates Based on Results of the Simulation Model

Direct measures from Medicare and NHIS that were not used to construct the simulation model were used to validate the model results. Figures 4 and 5 compare the frequency of PSA tests from 1996 to 2000 calculated from the simulation model with answers to the NHIS question, How many PSA tests did you have in the last 5 years? This question was not used for estimation purposes. There is a reasonable match between the model-projected and observed frequencies, with a slight overestimation by the model of the number of men with no PSA test. Results for younger men (ages 55–64 years) are shown, because the interscreening intervals for this group of men were assumed to be the same as the ones estimated for men ages 65–59 years. The match between the model projections and the observed NHIS data is similar for older men (data not shown).

thumbnail image

Figure 4. Distribution of the number of PSA screening tests received over the 5-year period 1996–2000 among younger (ages 45–60 years) black men. Comparison of simulation model results with the 2000 NHIS question on number of tests received that was not used in the model estimation. Source: Simulation model and NHIS 2000. equation image, NHIS 2000 (not modeled); equation image, Simulation model.

Download figure to PowerPoint

thumbnail image

Figure 5. Distribution of the number of PSA screening tests received over the 5-year period 1996–2000 among younger (ages 45–64 years) white men. Comparison of simulation model results with the 2000 NHIS question on number of tests received that was not used in the model estimation. Source: Simulation model and NHIS 2000. equation image, NHIS 2000 (not modeled); equation image, Simulation model.

Download figure to PowerPoint

Figure 6 compares the model-projected proportion of older men (ages 65 to 84 years) with at least 1 PSA test in a given year against values directly observed from SEER-Medicare data. The simulated and observed results are similar for ages 65–74 years. However, for ages 75–84 years, the simulated proportions are somewhat lower than those observed in the SEER-Medicare data.

thumbnail image

Figure 6. Annual percentage of men who had at least 1 PSA test among the older age groups. Comparison of simulation model results (plain lines) with observed percentages from Medicare claims (dashed lines). Source: Simulation model and SEER-Medicare frequencies.

Download figure to PowerPoint

Figure 7 shows the model-projected proportion of black and white younger men (ages 40 to 64 years) who had at least 1 PSA test in a given year. To our knowledge, this is the first time that the yearly proportion of men who had at least 1 PSA test in this age group has been estimated. The figure shows that among younger men (younger than age 60 years), more black men received testing each year, whereas among older men (ages 60 to 64 years) more white men were tested in earlier years. In this older age group, testing frequencies between white and black men were quite similar in later years.

thumbnail image

Figure 7. Annual percentage of men who had at least 1 PSA test among the younger age groups. Information not previously available and reconstructed by using the simulation model. Source: Simulation model.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

The present article combines information from 2 data sources in an attempt to produce a comprehensive description of the patterns of PSA screening testing between black and white men in the US. Previous estimates of PSA dissemination have been based on Medicare claims,8, 13 refer to specific calendar intervals that are subsets of the ones considered here, and to men aged ≥65 years. Here, we expand the description to the calendar period 1988–2000 and to younger age groups. This allowed us to construct a comprehensive description of how PSA screening disseminated and stabilized among men of different ages and races in the US.

Throughout, the results suggested a reverse in the patterns of PSA use by white and black men in the earlier calendar years compared with later years, especially among younger men. Figure 2 shows that among older men, whites have a higher percentage of men with a history of at least 1 PSA test, whereas among younger men, blacks have a higher likelihood of having had a PSA test. Thus, although white men adopted PSA earlier than black men, black men did not lag too much behind whites in adopting PSA screening, and in more recent years, black men were, in fact, being tested at earlier ages, and, according to the SEER-Medicare data, more frequently compared with before 1995. Figure 7 represents the annual percentage of men who had at least 1 PSA test among younger age groups (ages 40 to 64 years); this information was not previously available. This figure shows that younger black men, ages 40–49 years, are tested more frequently than white men. For older age groups, initially blacks were not tested as frequently as whites, but in the most recent years, the percentage of black men who received at least 1 PSA test is slightly higher or the same as white men. These results clearly suggest a greater activity and awareness on the part of the black community to promote PSA screening in the mid 1990s. In 1997, the American Cancer Society revised guidelines and recommended that men in high-risk groups, including black American men, should begin testing at age 45 years, instead of at age 50 years. This change was preceded by a large literature on racial differences in prostate cancer that showed higher risk of developing prostate cancer and worse prognosis among black American men.

We used what we believe to be the most representative data sources on PSA testing in the US; however, there are several limitations to these data. First, the SEER-Medicare resource only provides information on interscreening times among men aged ≥65 years. For men younger than 65 years, we assumed that PSA screening test frequencies estimated for the youngest group in the SEER-Medicare data, ie, men ages 65 to 69 years, would extrapolate to younger men. The validation (Figs. 4 and 5) comparing the observed and model-projected numbers of PSA tests from 1996–2000 among men 45–64 years of age showed a good correspondence, suggesting that this assumption was reasonable. Second, the NHIS resource only provides information on age at first PSA test within a 5-year interval. This led to overly “smoothed” estimates of the dissemination of first PSA tests. To capture the rapid increase of initial PSA testing between 1991 and 1994, we had to adjust our estimates of first PSA use by using published estimates for a specific age cohort8 and assume that this adjustment applied across all of our age groups. Third, the NHIS data contains information from face-to-face interviews and is subject to the limitations of self-reported data. For example, questions of the type “Age at first PSA test,” “How many PSA tests did you have in the last 5 years?”, and “Time since last PSA test” have been found14 to contain a particular form of error called “forward telescoping”, ie, the reporting or dating of events as more recent than they actually were. Also, men institutionalized or deceased during the 1988–2000 period are excluded from NHIS data. Thus, proportions of first PSA use for older men may be underestimated, and this is, indeed, consistent with our validation presented in Figure 6.

Our validation presented in Figure 6 shows that the model underestimates the annual frequencies of at least 1 PSA test. However, we note that the model developed has been designed specifically to reflect patterns of screening use. For example, because we were interested only in screening PSA tests, the data used to generate the model excluded PSA tests conducted within 3 months of a previous test in order to eliminate follow-up or diagnostic PSA tests. Because the Medicare data contain PSA tests conducted for reasons other than screening, observed frequencies of PSA testing from Medicare may overestimate the use of PSA solely for screening purposes as projected by the model. Again, this is consistent with our validation presented in Figure 6, in which the model-projected screening patterns underestimated the observed, annual testing frequencies in the Medicare population.

In conclusion, our results provide the most comprehensive description to date of PSA usage among men ages 40–84 years since the start of the PSA era. Based on these results, we can conclude with a high degree of confidence that PSA screening use is probably not able to explain any of the observed racial disparities in population prostate cancer trends among whites compared with blacks. Preliminary results from the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial indicate that although diagnostic procedures following a positive PSA or positive digital rectal examination are less intensive among black men, they are also not statistically significantly different from those for white men.4

Accurate knowledge of how screening has disseminated in the population is critical when evaluating the larger question of estimating the contribution of PSA screening to the recent, substantial declines in prostate cancer mortality. This question is currently being addressed by a consortium of modelers participating in the Cancer Intervention and Surveillance Modeling Network sponsored by the National Cancer Institute (CISNET; http://cisnet.cancer.gov/). Surveillance models, such as those developed within the CISNET consortium, can use our results in models designed to determine whether PSA screening patterns can be quantitatively linked to overall mortality declines. These models will also have the capability of incorporating trends in treatment, particularly trends in neoadjuvant and/or adjuvant hormone suppression therapies that have shown some differences in use between whites and blacks,15 and ultimately these models should be able to quantify the roles of screening versus treatment in explaining observed mortality declines.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES
  • 1
    Brawley OW,Knopf K,Merrill R. The epidemiology of prostate cancer part I: descriptive epidemiology. Semin Urol Oncol. 1998; 16: 187192.
  • 2
    Feuer EJ,Mariotto A,Merrill R. Modeling the impact of the decline in distant stage disease on prostate carcinoma mortality rates. Cancer. 2002; 95: 870880.
  • 3
    Ries LAG,Harkins D,Krapcho M, et al. SEER Cancer Statistics Review, 1975–2003. Bethesda, Md: National Cancer Institute; 2006. Available at: http://seer.cancer.gov/csr/1975_2003.
  • 4
    Pinsky PF,Andriole GL,Kramer BS, et al. Prostate biopsy following a positive screen in the prostate, lung, colorectal and ovarian cancer screening trial. J Urol. 2005; 173: 746750; discussion 750–751.
  • 5
    de Koning HJ,Auvinen A,Berenguer Sanchez A, et al. Large-scale randomized prostate cancer screening trials: program performances in the European Randomized Screening for Prostate Cancer trial and the Prostate, Lung, Colorectal and Ovary cancer trial. Int J Cancer. 2002; 97: 237244.
  • 6
    Etzioni R,Penson DF,Legler JM, et al. Overdiagnosis due to prostate-specific antigen screening: lessons from U.S. prostate cancer incidence trends. J Natl Cancer Inst. 2002; 94: 981990.
  • 7
    Surveillance, Epidemiology, and End Results (SEER) Program – SEER 9 Registries, Nov 2002 Submission. 1973–2000. National Cancer Institute, released April 2003, based on the November 2002 submission. Available at: www.seer.cancer.gov.
  • 8
    Legler JM,Feuer EJ,Potosky AL,Merrill RM,Kramer BS. The role of prostate-specific antigen (PSA) testing patterns in the recent prostate cancer incidence decline in the United States. Cancer Causes Control. 1998; 9: 519527.
  • 9
    Centers for Disease Control. National Health Interview Survey (NHIS) 2000. Hyattsville, Md: National Center for Health Statistics; 2002.
  • 10
    Warren JL,Klabunde CN,Schrag D,Bach PB,Riley GF. Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care. 2002; 40(8 suppl): IV-318.
  • 11
    Cronin KA,Yu B,Krapcho M, et al. Modeling the dissemination of mammography in the United States. Cancer Causes Control. 2005; 16: 701712.
  • 12
    Yu B. Estimation of shared gamma frailty models by a modified EM algorithm. Comput Stat Data Anal. 2006; 50: 463474.
  • 13
    Etzioni R,Berry KM,Legler JM,Shaw P. Prostate-specific antigen testing in black and white men: an analysis of Medicare claims from 1991–1998. Urology. 2002; 59: 251255.
  • 14
    Prohaska V,Brown NR,Belli RF. Forward telescoping: the question matters. Memory. 1998; 6: 455465.
  • 15
    Zeliadt SB,Potosky AL,Etzioni R,Ramsey SD,Penson DF. Racial disparity in primary and adjuvant treatment for nonmetastatic prostate cancer: SEER-Medicare trends 1991 to 1999. Urology. 2004; 64: 11711176.