Gilligan T, Wang PS, Levin R, Kantoff PW, Avorn J. Racial Differences in Screening for Prostate Cancer in the Elderly. Arch Intern Med. 2004;164(17):1858-1864. doi:10.1001/archinte.164.17.1858
Black men are more likely than white men to be diagnosed as having advanced prostate cancer, and their prostate cancer mortality rates are more than twice as high. Low screening rates among black men may contribute to these disparities, but there are few data on racial differences in prostate cancer screening.
To present a case-control study of racial differences in the use of prostate-specific antigen (PSA) as a screening test among Medicare beneficiaries in New Jersey and to assess the degree to which race is associated with prostate cancer screening.
The study used a statewide database of claims data from Medicare Parts A and B, Medicaid, and the state's Pharmaceutical Assistance for the Aged and Disabled program. Prevalent cases of prostate cancer were excluded using the state's cancer registry. Of 139 672 men who underwent PSA screening, 34 984 were randomly selected along with an identical number of controls matched by month and year of birth. After men with International Classification of Diseases, Ninth Revision, Clinical Modification,or Current Procedural Terminology codes indicative of prostate cancer were excluded, 33 463 case patients and 33 782 control subjects remained.
The use of PSA screening was strongly and inversely associated with black race (odds ratio [OR] = 0.50; P<.001), poverty (OR = 0.33; P<.001), and near poverty (OR = 0.69; P<.001). Multivariate logistic regression analysis after age, socioeconomic status, comorbidity, and use of health care services were controlled for revealed that black race remained a strong predictor of not undergoing PSA screening (OR = 0.65; 95% confidence interval, 0.60-0.70).
Elderly blacks are substantially less likely to undergo PSA screening than elderly whites. Differences in socioeconomic status and comorbid conditions explain only a small part of the racial differences in screening rates.
Major racial differences persist in prostate cancer incidence and mortality figures in the United States.1,2 Black men are approximately 60% more likely to be diagnosed as having prostate cancer than white men and twice as likely to die of it. Prostate cancer in black men, compared with white men, has been associated with younger age and more advanced stage and higher grade of disease at the time of diagnosis.3- 5 Numerous studies have investigated the patterns and causes of these discrepancies, but the subject remains contentious, particularly as to whether there are racial biological differences that might play a role.6- 15 If relevant biological differences exist, they have not been clearly identified, although differences in serum testosterone levels have been reported.16- 22 Several studies have reported higher serum testosterone levels in adolescent and young adult black males and in pregnant black women than in whites of the same age and sex.18,20- 22 Testosterone is believed to be a key factor in the development of prostate cancer, and if it is true that black men have, on average, higher serum androgen levels, then it is a reasonable hypothesis that that difference could contribute to the higher incidence of prostate cancer seen among black men. However, studies of the relationship between serum androgen levels and prostate cancer have produced conflicting results.23- 26
Several socioeconomic and behavioral differences are believed to account for at least part of the higher prostate cancer incidence and mortality rates seen in black men. Alterable risk factors have yet to be clearly identified, although there is some evidence that dietary differences may be relevant.27,28 Diets high in animal fats are thought to contribute to the development of prostate cancer, but the results of studies of whether blacks have a higher intake of these lipids have been contradictory.27,29,30 With regard to variations in racial mortality rates, differences in treatment have been clearly demonstrated: black men with known prostate cancer are 16% to 30% less likely than whites to undergo prostatectomy.31- 34 These percentages parallel the findings of lower surgical resection rates in black patients with early-stage lung cancer35 and colorectal cancer36 as well as the findings of substantial racial differences in the treatment of ischemic heart disease,37 end-stage renal disease,38 and a variety of other ailments.
The association of black race with more advanced-stage disease at diagnosis has led to the hypothesis that the higher prostate cancer mortality rates in black men might be reduced by more aggressive efforts at early detection, although it remains to be established whether early diagnosis through screening results in lower mortality rates. Only limited published data exist describing prostate cancer screening rates among blacks and whites.39- 41 Studies using telephone surveys have reported lower prostate cancer screening rates among blacks, but such data depend on patient recall. Patients' self-reports about whether or not they have undergone prostate cancer screening have been demonstrated to be highly inaccurate.42 Etzioni et al43 used the combined Medicare-SEER database to evaluate prostate cancer screening using serum prostate-specific antigen (PSA) testing. Blacks were 25% less likely than whites to undergo PSA screening. However, their study did not control for socioeconomic status (SES) or comorbid conditions, which are very important confounders in studies of racial health care differences.
We report the results of a statewide population-based case-control study of differences in PSA screening among blacks and whites aged 65 years and older who participated in the Medicare and/or Medicaid program in the state of New Jersey. The analysis is based on Medicare and Medicaid claims data and controls for age, SES, and comorbid conditions. The aim of this study was to assess the degree to which race is associated with prostate cancer screening.
Medicare claims data for beneficiaries living in New Jersey were available for years 1991 through 1996. During each year of the study period, the Medicare eligibility file for the state contained more than 1 million enrollees aged 65 years or older. Among men older than 65 years, 97% to 98% are beneficiaries of Medicare Part A (services received during hospitalizations and nursing home stays), and 95% of beneficiaries also subscribe to Part B (outpatient medical services and procedures).44,45 Approximately 9% of Medicare beneficiaries participate in Medicare health maintenance organizations and do not have claims in the Medicare database.45 The database used for this study therefore includes information on approximately 90% of the eligible population. It also includes all hospitalizations and outpatient medical services for each enrollee. Information included data from Medicare Parts A and B. The records include up to 10 diagnostic and 5 procedure codes per hospitalization or outpatient encounter as well as the dates services were provided and demographic information, including the patient's age, sex, self-described race, and ZIP code. A unique claim number is assigned to each Medicare beneficiary, so individuals can be tracked longitudinally. The PSA testing was assigned unique Current Procedural Terminology (CPT) codes beginning in 1993 (code 84153 for total PSA, code 84154 for free PSA, and code G0103 for PSA screening).
Approximately 500 000 individuals were enrolled in the NJ Medicaid program during each year of the study period, 40% of whom were male and nearly 12% of whom were at least 65 years old. Blacks made up more than one third of the NJ Medicaid population. Medicaid claims data include demographic information; inpatient, outpatient, and nursing home charges; dates of services; and the diagnoses and procedures associated with the claims. During the years of this study, the program had no deductibles or maximum benefit and charged no co-payments. The indigent status of Medicaid enrollees results in essentially no out-of-pocket (ie, out-of-system) health care use.
The NJ Pharmaceutical Assistance to the Aged and Disabled (PAAD) program has income-eligibility criteria that are more generous than those of Medicaid, so PAAD includes a less impoverished population. During the study period, PAAD eligibility required an annual income below $15 700 for single persons and $19 520 for married persons. These figures at the time represented the highest income ceiling of any such program nationally, producing a beneficiary population that was both large and further from extreme poverty. During the study period, approximately 200 000 persons were enrolled in the NJ PAAD program each year, with the vast majority being at least 65 years of age.
The NJ Cancer Registry aims to record all cancers diagnosed among state residents. Since 1978, state law has mandated that hospitals, physicians, dentists, and clinical laboratories report all new cancer cases with 6 months to the registry. A standardized reporting system, International Classification of Diseases for Oncology (ICD-02), is used to classify pathologic, histologic, and staging information. Attempts are made to include information on NJ residents diagnosed as having cancer at institutions outside the state through information provided according to agreements with neighboring states. Existing data in the registry are periodically matched to information from reporting hospitals and physicians; state death, motor vehicle, and income tax records; and federal databases such as the National Death Index. The registry data used in this study, which were last updated in 2001, included cancers diagnosed between 1980 and 1998.
The eligible population for this study (both case patients and control subjects) consisted of male Medicare or Medicaid beneficiaries at least 65 years of age who were enrolled in Medicare and/or Medicaid between January 1, 1993, and December 31, 1996. To be eligible for our analysis, individuals must have used either program for a nondrug service during the 12 months before their index date as well as during the 12- to 24-month period before their index date.
To protect the privacy of patients, the data on the beneficiaries of Medicare, Medicaid, and PAAD had previously been made anonymous. All person-specific identifiers were connected to coded study numbers. These data were assembled on a person-specific basis into a relational database using a commercially available software package (Sybase Inc, Dublin, Calif). This relational database contains information on all filled prescriptions, procedures, physician encounters, hospitalizations, and long-term care for anyone who was reimbursed by one of these programs. Records of any service, hospitalization, or prescription for an individual in the relational database were then linked to all prostate cancer information on that individual in the registry. Approval for this study was obtained from the institutional review boards of Brigham and Women's Hospital, Boston, Mass, and from the New Jersey Cancer Registry, Trenton.
The PSA users were defined as men in the study population who had undergone PSA testing between January 1, 1994, and December 31, 1996. The index date was defined as the date of the first PSA test within this period. The PSA tests were considered to be routine screening tests unless there were any additional PSA tests performed within the previous 9 months. Men who had undergone PSA testing within the 9 months before their index date were excluded because frequent PSA testing would likely indicate that the testing was performed as a result of symptoms or abnormalities found on examination or medical tests rather than for routine screening purposes. For each case, a male control subject was identified. The control subjects had no PSA testing claims anytime prior to or within 1 year after their index date. The PSA users and controls were matched by month and year of birth and by year of index date. Index dates within the study period were randomly assigned to controls.
Men were excluded if they had a diagnosis of prostate cancer in the NJ Cancer Registry on or before their index date. Men were also excluded if (1) their health care use data included CPT codes from before their index date indicating that they had undergone prior radical prostatectomy, radical cystectomy (cystoprostatectomy), bilateral orchiectomy, prostate biopsy, or brachytherapy or (2) their use data prior to their index date included ICD, Ninth Revision, Clinical Modification (ICD-9-CM) codes suggestive of prostate cancer as well as either ICD-9-CM or CPT codes for procedures commonly used in a diagnostic or staging workup for prostate cancer. In the primary analysis, we did not exclude men whose only evidence of prostate cancer was a claim that included ICD-9-CM code 185 because claims data have been shown to have positive predictive values (PPVs) as low as 36% for identifying cancer cases.46 The low accuracy of code 185 most likely results from physicians billing PSA screening tests under code 185 using the justification that they are ruling out prostate cancer. Because our decision to include some men whose claims data included code 185 would be expected to increase the risk of including prevalent cases in our study population, we conducted a secondary analysis in which all men with ICD-9-CM code 185 prior to their index date were excluded.
χ2 Analyses were performed to evaluate whether PSA testing was associated with race and/or other variables, including age, SES, and comorbidities. Socioeconomic status was estimated on the basis of insurance status. Enrollment in the Medicaid and PAAD programs was a marker for poverty and near poverty, respectively. Men in Medicare only were considered nonpoor. Multivariate logistic regression was performed to determine whether an independent association between race and PSA testing was present when SES, age, and comorbidities were controlled for. Associations are reported as odds ratios (ORs). Comorbidities were assessed in several ways. For our primary analysis, we used both the Charlson Comorbidity Index score (Charlson score)47 as adapted by Deyo et al48 and the following measures of the individual's use of services: days hospitalized, days in a nursing home, and ambulatory physician visits.49 We also performed secondary analyses controlling for all the specific comorbidities listed in Table 1.
In the population studied, 139 672 men met the initial eligibility criteria for PSA users (not including ICD-9-CM and CPT code exclusion criteria). A random sample cohort of 34 984 PSA users was selected, and an equal number of matched controls were identified. Per study design, none of these men had a prior diagnosis of prostate cancer in the NJ Cancer Registry. From the 69 968, we excluded 2723 men because of ICD-9-CM or CPT codes that were indicative of a diagnosis of prostate cancer. The study population thus consisted of 67 245 men who were at least 65 years old, including 33 463 PSA users and 33 782 controls. Among the study population, 5.7% of the individuals were black and 87.6% were white. A slight majority (57.9%) were 65 to 74 years old, and only 7.3% were older than 84 years. Medicaid participants represented 5.6% of the study population, while 11.6% were PAAD beneficiaries. Characteristics of the population are listed in Table 1.
In univariate analysis, black men were only half as likely as white men to undergo PSA testing (OR = 0.50; P<.001). Other variables associated with a substantially lower likelihood of PSA testing included other race, Medicaid enrollment, PAAD enrollment, age older than 84 years, and a number of specific comorbidities, including end-stage renal disease, heart failure, stroke, myocardial infarction, and depression (Table 1). Hypertension was associated with a greater likelihood of PSA testing. Men with cancers other than prostate cancer were slightly less likely to undergo PSA screening. (Patients with prostate cancer were excluded from the study population.)
The association between black race and a lower likelihood of undergoing PSA testing persisted in multivariate analysis in which age, SES, and comorbid conditions were controlled for (Table 2). The OR for black men undergoing PSA screening was 0.65. Poverty and near poverty were also associated with a lower likelihood of undergoing PSA screening tests. The association between black race and PSA testing remained relatively stable whether we controlled for individual comorbid conditions (OR = 0.56), for the Charlson score alone (OR = 0.59), for medical care use alone (outpatient visits, days hospitalized, and days in nursing homes) (OR
= 0.62), or for all of these variables (OR = 0.63). The association between poverty and PSA testing also remained relatively stable, with an OR ranging from 0.42 to 0.51 in the different analyses. Table 3 presents the ORs for the variables assessed that were associated with PSA screening test rates. When we excluded all men with a claim listing a diagnosis of prostate cancer (ICD-9-CM code 185), 3020 of the 67 245 subjects dropped out, including 1737 PSA users and 1283 controls. The OR for PSA screening associated with black race remained 0.50 (P<.001) on univariate analysis and 0.65 (95% confidence interval, 0.60-0.70) on multivariate analysis.
Other predictors of significantly lower use of PSA screening included end-stage renal disease, other race, dementia, chronic liver disease, congestive heart failure, myocardial infarction, delirium, psychosis, stroke, obesity, and nonprostate cancer. Only end-stage renal disease was associated with a lower likelihood of undergoing PSA testing than poverty; these 2 characteristics were the only ones identified with a lower likelihood of PSA screening than black race or other nonwhite race. The only variables associated with a significantly increased likelihood of undergoing prostate cancer screening were a diagnosis of hypertension and a greater number of ambulatory visits.
We analyzed PSA screening tests in a population-based statewide database of elderly men in New Jersey who had no prior diagnosis of prostate cancer. The results indicate that black men are substantially less likely than white men to undergo PSA screening. This finding held up when poverty and near poverty as well as comorbid conditions and health care use were controlled for. Similar results were found for men categorized as "other race." We focused on black men because they have the highest incidence and mortality rates of prostate cancer and because the category of other race is a lumping together of several diverse groups with very different rates of prostate cancer (Asians, native Americans, Latinos, and Pacific Islanders). It is worth emphasizing that nonwhite race was a stronger predictor of not undergoing PSA screening than near poverty (PAAD participation) or any comorbid condition except end-stage renal disease, chronic liver disease, and dementia. Only poverty (Medicaid) and end-stage renal disease were stronger predictors than race.
The major limitation of assessing screening rates by analyzing claims data is that there is no way to know why a given test was ordered. Prostate-specific antigen tests may be sent to screen for prostate cancer, to monitor prevalent cases of prostate cancer, or to work up signs and symptoms for which prostate cancer is on the differential diagnosis, such as obstructive urinary symptoms associated with benign prostatic hyperplasia. Excluding prevalent cases is thus essential. We have previously assessed the sensitivity of the NJ Cancer Registry for detecting breast cancer and found that 78% of cases were identified.50 This figure increased to 91% when registry data were combined with an algorithm using Medicaid-Medicare surgical claims data. We used a similar strategy based on both registry and claims data to identify prevalent cases in the present study. In a secondary analysis aimed at maximizing sensitivity at the risk of including more false-positive results, we excluded all men with a prior claim that included ICD-9-CM code 185. As reported above, the strength of the association between race and PSA screening did not change. The stability of our results in this secondary analysis is not surprising because we would not expect the inclusion of a small number of prevalent cases to affect PSA testing rates among blacks and whites in a roughly similar manner given the higher incidence of prostate cancer among blacks and the tendency of whites with prostate cancer to be treated more aggressively.
Our decision to exclude from our analysis men who underwent PSA testing in the 9 months before their index date reflects the priority we placed on excluding PSA testing that was not performed for screening purposes. Some men may have undergone screening PSA tests at intervals of less than 9 months, and they would be inappropriately excluded from our analysis. We believe that this risk is justified by the benefit of excluding men whose PSA tests were ordered as a result of symptoms of prostatism or a previous diagnosis of prostate cancer. We also may have missed men who underwent screening if they had PSA tests performed at Veterans Affairs hospitals, because such tests would not be billed to Medicare.
Previous reports have assessed the validity of using Medicare claims data to estimate cancer incidence and screening rates.43,45,50 While sensitivities of 84% to 88% for identifying cases of prostate cancer (and up to 97% for other malignancies) have been reported, false-positive results have been problematic, with the result of low PPVs. Warren et al46 used the Surveillance, Epidemiology, and End Results (SEER)-Medicare database to assess the utility of claims data to estimate breast cancer incidence and found that a diagnosis of breast cancer in physicians' claims had a PPV of only 17% (after prevalent cases were excluded). The reason for the significant number of false-positive breast cancer diagnoses in physician claims data was unclear, but the authors postulated that they resulted from claims filed for services used to rule out the diagnosis when breast cancer was suspected. Combining inpatient and outpatient claims data raised the PPV to 36%, while inpatient claims data alone had a PPV of 88%. Few data are available regarding the PPV of a diagnosis of prostate cancer in claims data in the PSA era (ie, after 1990). Fisher et al51 assessed the accuracy of hospital discharge claims data from 1984 through 1985 and used individual chart review as their "gold standard." They reported a sensitivity of 84% and a PPV of 88%, but it is important to note that outpatient claims were not included and that PSA was not being used as a screening test at that time. It is unclear whether these results are applicable to the current era. Using more recent data, Cooper et al52 reported that Medicare claims data had a sensitivity of 88% for identifying cases of prostate cancer (using the SEER database as a gold standard), but their analysis was unable to assess specificity or PPV.
In light of the published studies discussed above and the stability of the results in our different analyses, we believe that we successfully excluded the vast majority of prevalent cases of prostate cancer and that it is highly unlikely that the unintentional inclusion of a small number of prevalent cases has significantly confounded our results. A more difficult source of confounding is our inability to know from claims data why a PSA test was ordered. The PSA tests that are ordered as a result of signs or symptoms of benign prostatic hyperplasia or prostate cancer are not generally considered screening tests. Regarding signs or symptoms of obstruction, there are no known substantial black-white racial differences in the incidence of benign prostatic hyperplasia, so we would not expect additional PSA testing among whites to result from differences in the obstructive symptoms.53 A PSA test that is ordered as a result of abnormal results on a digital rectal examination, on the other hand, would not be a screening test in the strictest sense of the word, but would still have resulted only as a consequence of a screening strategy. The PSA tests that are ordered in men who present with metastatic prostate cancer or metastatic cancer of unknown primary origin are clearly not screening tests, but, in the PSA era, these men represent a small fraction of patients with prostate cancer and a much smaller fraction of men who undergo PSA screening.2,43
Several other studies support our findings of substantial racial differences in PSA screening. Etzioni et al43 evaluated the SEER-Medicare database and reported that, by the end of 1996, 33% of white men and 25% of black men were undergoing PSA testing at least twice a year. In whites, 83% of prostate cancer diagnoses were preceded by a PSA test, compared with 77% in blacks. However, the extent to which this difference may have been related to differences in SES or comorbid conditions was not addressed. Steele et al39 reported the results of 2 telephone surveys asking men about whether they had undergone PSA testing or a digital rectal examination. Relative to white men, the OR for black men who reported undergoing screening was 0.3 after annual income was controlled for (>$25 000 vs < $25 000). The generalizability of these results is limited by the requirement that subjects have working telephone service as well as the time and the willingness to complete the survey. Furthermore, the inaccuracy of men's self-reports regarding whether they have undergone screening limits the reliability of these results. Jordan et al42 compared 276 men's self-reports of PSA screening and digital rectal examinations with their medical records. Sensitivity and specificity were 74% and 65% for PSA testing and 82% and 45% for digital rectal examinations, respectively.
The present study is unique in that it presents an analysis of racial differences in the use of PSA screening tests that is based on population-based data and thats controls for SES and comorbid conditions at the level of the individual. Studies using claims data generally have had to estimate SES from the median income of the individual's census tract or ZIP code. Such aggregate assessments of SES are less accurate as well as less powerful predictors of outcomes than individual-level measures.54 As measures of SES, Medicaid and PAAD claims have the advantage of high specificity because of the strict eligibility criteria for participation in these programs. However, the sensitivity is lower because some individuals who are eligible do not apply or enroll.54 If eligible blacks were less likely to enroll than eligible whites, the ability of our study to control for confounding as a result of racial differences in SES would be reduced.
Our study does not shed light on the causes of racial differences in the use of PSA screening tests or on the impact of these differences. Potential contributing factors to differential screening rates include racial differences in access to care, in education, in patient preference, and in providers' recommendations and medical decisions. Previous work has shown that differences in patient decision making and education are relevant. Tingen et al55 recruited 249 men to participate in a prostate cancer education program and then offered free prostate cancer screening by the physician of their choice. Only 38% of blacks chose to undergo screening compared with 56% of whites, and race was a stronger predictor of screening than education or annual income. Demark-Wahnefried et al40 surveyed 1500 men who reported for screening at National Prostate Cancer Awareness Week screening events and noted that blacks were less likely than whites to have a regular physician, to have undergone prior screening, or to believe that prostate cancer could be asymptomatic or could be cured. Similarly, in their survey study of 944 men undergoing screening, Barber et al56 reported that, compared with whites, blacks had lower levels of knowledge about prostate cancer. If mass screening for prostate cancer is ever endorsed, achieving high rates of compliance among black men will likely require an wide spread and culturally appropriate educational outreach effort. Because prostate cancer incidence is particularly high among blacks, they would have the most to gain from screening and would represent a high-priority population.
The impact of fewer blacks undergoing PSA screening tests will be unclear until it is established whether prostate cancer screening prolongs lives. However, even in the absence of persuasive evidence defining the impact of screening on mortality, our findings are important. Prostate cancer screening may serve as a marker of surveillance and intensity of care. Our findings thus serve to reinforce previous reports of substantial and persistent racial disparities in the delivery of medical care in the United States. Moreover, current screening guidelines from a variety of major medical organizations oppose mass screening but recommend that physicians discuss prostate cancer screening with their male patients of appropriate age and offer these men the opportunity to undergo screening if they so choose.57- 60 In the current climate of uncertainty about the benefits of screening, men should have access to screening information and to screening services regardless of race so that they can make informed decisions. If prostate cancer screening proves to be of benefit in the future, reducing this racial disparity will be a high priority given the much higher incidence of prostate cancer among blacks.
Correspondence: Timothy Gilligan, MD, Dana-Farber Cancer Institute, 44 Binney St, D-1230, Boston, MA 02115 (Timothy_Gilligan@dfci.harvard.edu).
Accepted for publication November 28, 2003.
We thank the State of New Jersey Department of Health and Senior Services and the New Jersey Cancer Registry for providing access to the cancer registry data used in this study.