Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Clinical Research

# Health disparities in clinical practice patterns for prostate cancer screening by geographic regions in the United States: a multilevel modeling analysis

## Abstract

### Background:

To our knowledge, no previous study has examined state-level geographic variability and its predictors in clinical practice patterns to screen for prostate cancer in the United States.

### Methods:

We used the Behavioral Risk Factor Surveillance System 2010 data set to analyze geographic variability (by state) and its associated predictors in receiving a PSA test and/or a digital rectal examination (DRE). The study population consisted of men aged 50 years who responded as yes/no when asked about having a PSA test or DRE performed during the last year. We build two multilevel logistic regression models, differing in dependent variables, that is, (1) any prostate cancer screening (PCS) (either PSA and/or DRE), and (2) PCS based on PSA testing (PSAT). Individual characteristics (age, education, employment, marriage, income, race/ethnicity, self-reported health status, obesity, alcohol consumption, smoking status, personal physician presence, and health insurance coverage) were treated as level-1 variables and state characteristics (number of doctors per 100 000 persons per state, US regions and metropolitan statistical area (MSA) codes) were treated as level-2 variables.

### Results:

We found significant geographic variability in receiving PCS and PSAT screening in the United States. For PCS, MSA code was an independent predictor, with men living in urban areas having lower odds of screening (odds ratio (OR)=0.8, 95% confidence interval (CI)=0.7–0.9). In PSAT, the number of doctors per 100 000 persons per state was an independent predictor, with lowest quartile states (0–25% quartile) having lower odds of PSA-based screening (OR=0.78, 95% CI=063–0.94). In both models, all level-1 variables were independent predictors (P<0.05) of PCS, except self-reported health status.

### Conclusions:

Men living in urban areas and states with lower prevalence of doctors have lower odds of screening for prostate cancer and PSAT, respectively, after adjusting for individual variables. Future studies should examine the reasons for these health disparities.

## Introduction

Prostate cancer is the second leading cause of cancer related death among men in the United States, with an estimate of 28 170 deaths during 2012.1 Approximately 2.3 million men had a history of prostate cancer in the United States in 2008.2 Prostate cancer occurs mainly in older men, with more than one-third of cases diagnosed in men aged 65 years or older.3 The economic burden of prostate cancer is substantial with the total spending of $9.86 billion on its related care during 2006 in the United States. These costs included average annual costs per patient of$10 612 in the initial phase after diagnosis, $2 134 for continuing care and$33 691 in the last year of life.4 These statistics show that prostate cancer is a significant public health problem in the United States.

PSA test and digital rectal examination (DRE) are two common methods to screen for prostate cancer. The effectiveness of PSA testing (PSAT) is currently debatable.5, 6, 7 One large clinical trial reported no benefit in prostate cancer related mortality from combined screening with the PSA test and DRE during a median follow-up of 13 years.8 Whereas, another large clinical trial reported that the PSA test alone was associated with a 21% relative reduction in the death rate from prostate cancer at a median follow-up of 11 years.9 Nonetheless, >75% of people aged 50 years or more get screened for prostate cancer at least once during their lifetime.10 Interestingly, 87% of male physicians aged 50 years or above also reported having had a PSA test.11 Although the effectiveness of PSA test is controversial, both the PSA test and DRE are widely used to screen for prostate cancer in the United States.

Previous studies have found evidence of significant geographic variability and health disparities in clinical practice patterns for many cancer preventive services in the United States. For example, one such study found that, after adjusting for individual-level sociodemographic characteristics, area level poverty rate is independently associated with never having had a mammogram (odds ratio (OR)=1.28, 95% confidence interval (CI)=1.03–1.37), clinical breast examination (OR=1.28, 95% CI=1.11–1.48), colonoscopy/sigmoidoscopy (OR=1.10, 95% CI=1.01–1.19) and a fecal occult blood test (OR=1.19, 95% CI=1.12–1.27).12 Other studies with similar findings have been published as well.13, 14 However, to our knowledge, no previous study has examined the state-level health disparities in screening for prostate cancer (PSA and/or DRE based) in the United States. Exclusion of such significant geographic variability from the analysis can lead to atomistic fallacy, that is, false inferences about groups (for example, US population) from the individual-level data (for example, people residing in different states of the United States).15

The purpose of this study is to examine the state-level health disparities and its associated predictors in clinical practice patterns for utilization of prostate cancer screening (PCS) services (PSA test and/or DRE) in the United States, after adjusting for the individuals’ sociodemographic characteristics through the multilevel modeling approach. These sociodemographics are age, education, marital status, employment status, smoking status, health insurance coverage, presence of personal physician, self-reported health status, alcohol drinking, income level, body mass index (BMI) and race/ethnicity.

## Materials and methods

### Research design and study population

This study is a quasi-experimental cross-sectional analysis utilizing the 2010 Behavioral Risk Factor Surveillance System (BRFSS) data set. BRFSS is an annual telephonic survey conducted by the Centers for Disease Control and Prevention, ongoing since 1984 in all the 50 states of the United States, the District of Columbia and the three territories of Puerto Rico, the US Virgin Islands and Guam.16 BRFSS data is representative of the national noninstitutionalized US adult population. The design, methodology, validity and reliability of the BRFSS are ascertained by many studies.17, 18, 19

Eligible population comprised of men aged 50 years or more, as per the PCS recommendations of the American Cancer Society.20 Study population was limited to men whose responses were recorded as either yes or no when asked about having a PSA test or DRE during the last 1 year. Based on the residential status recorded during the BRFSS survey, each individual was assigned to one of the 50 states of the United States, the District of Columbia or the three territories of Puerto Rico, the US Virgin Islands and Guam.

### Measurement and steps

#### Individual-level variables

Individuals residing within each state were treated as level-1 units. The following variables were examined for being predictors of receiving PCS: age, education level, employment status, marital status, income, race/ethnicity, self-reported health status, obesity status, alcohol consumption, smoking status presence of a personal physician and health insurance coverage. All of these were determined and categorized a priori, based on a PubMed literature search.21, 22, 23, 24, 25 Age was grouped into 50–59, 60–69, 70–79 and 80 or above years. Education was categorized as not graduated from high school, high-school graduate and some college or more. Employment status was classified as currently employed, currently unemployed, student/homemaker/unable to work and retired. Marital status was categorized as currently married, divorced/widowed/separated and never married. Annual income was grouped into less than $25 000,$25 000–$49 999, 50 000–$75 000 and more than \$75 000. Race/ethnicity was categorized into Whites, African–American, Hispanics and others. Self-reported health status was dichotomized by combining excellent/very good/good and fair/poor. Obesity status was divided into the following based on the BMI: neither obese nor overweight (BMI<25), overweight (BMI: 25–30) and obese (BMI>30). Alcohol consumption was defined as nondrinkers (no drink within last 1 month), light drinkers (one drink per day during last 1 month) and heavy drinkers (two or more drinks per day during last 1 month). Smoking status was defined as nonsmokers (<100 cigarettes during lifetime and not regularly smoking during last year), previous smokers (>100 cigarettes during lifetime and not regularly smoking during last year) and current smokers (>100 cigarettes during lifetime with regular smoking during last year). Presence of a personal physician and health insurance coverage were treated as dichotomous variables (yes or no).

#### Aggregate level variables (geographic, physicians’ availability, metropolitan area)

Fifty states of the United States, the District of Columbia and the three territories of Puerto Rico, the US Virgin Islands and Guam were treated as level-2 units. Predictors examined at level-2 are numbers of doctors per 100 000 persons in each state, number of urologists per 100 000 persons in each state, Census Bureau regions (that is, Northeast, Midwest, South and West)26 and metropolitan statistical area (MSA) codes. Data for the prevalence of doctors were obtained from the 2007 Census Bureau estimates.27 Data for number of urologists per 100 000 persons in each state was obtained from the area resource files of the Health Resources and Services and Administration, Department of Health and Human Services.28 Data for Census Bureau Regions and MSA codes were obtained from the BRFSS 2010 data set. Based on the distribution (percentiles) of the doctors per 100 000 persons in each state, the prevalence of doctors per state was divided it into the following categories: less than 220.6 doctors per 100 000 population (0–25%; first quartile), between 220.6 and 274.9 doctors per 100 000 population (26–75%; second and third quartile), and >274.9 doctors per 100 000 population (76–100%; fourth quartile). Similarly, based on the distribution (percentiles) of the urologists per 100 000 persons in each state, the prevalence of urologists per state was divided into: less than 27.4 urologists per 100 000 population (0–25%; first quartile), between 27.4 and 35.3 urologists per 100 000 population (26–75%; second and third quartile), and >35.3 urologists per 100 000 population (26–75%; second and third quartile). MSA codes were included as recorded in the BRFSS, that is, a categorical variable with five classes: in the center city of an MSA, outside the center city of an MSA but inside the county containing the center city, inside a suburban county of the MSA, in an MSA that has no center city and not in an MSA. These are categorized as per the Centers for Disease Control and Prevention classification of the MSA codes.

### Statistical analysis

Multilevel models are appropriate when predictor variables are measured simultaneously at different levels.15 We used two-level (multilevel) logistic regression models to analyze the influence of aggregate data and individual data on receiving any type of PCS in the last year, dependent variable, with individuals (level-1) residing within the states (level-2). The build-up approach was used to build the final models, starting with the intercept-only models.15 Two separate models were built, differing in the dependent variables: (1) PCS consisting of PSA, DRE or both (that is, PCS) and (2) screening that included PSAT. Dependent variables were treated as binary variables (yes/no screening) in both of the above models.

In the first step, an intercept-only model was used to test the influence of state on receiving PCS. For this purpose, the null hypothesis of variance of the random intercept being zero was tested using the likelihood ratio test, or deviance difference, of the model without random effects and without parameters and the random intercept multilevel model with one parameter and the random intercept. Under the null hypothesis, the distribution of this statistic of the deviance difference can be approximated by the χ2-distribution with 1 degree of freedom.15 In the second step, individual data significantly associated (P<0.05) with the dependent variable were added. Next, significantly associated (P<0.05) aggregate variables were added, that is, number of doctors per 100 000 persons in each state, Census Bureau regions and MSA code. All of the level-1 individual variables were tested for random slopes. Cross-level interactions were also examined.

Intraclass correlation coefficients (ICCs) and proportional change in variance were estimated for both PCS and PSAT models. ICC is the measure of proportion of total variance in the outcome (that is, screening for prostate cancer) that is attributable to level-2 (that is, the state-level) in a two-level model.29 ICCs were calculated using the linear threshold model method.30, 31, 32, 33 Proportional change in variance is the estimate of the proportion of total variance that is attributable to the added predictors in the model. In other words, proportional change in variance may be interpreted as the amount of variance ‘explained’ by addition of predictors to the model.29, 33 All statistical analyses were conducted in SAS 9.2 (Cary, NC, USA).

## Results

### Sociodemographic characteristics

The median response rate for the 2010 BRFSS survey was 54.60% (range=39.05–68.78%), Appendix 1. Nonetheless, selection bias is not a substantial problem in the 2010 BRFSS data set based on the differences between BRFSS and US population data with respect to sex, age and race/ethnicity by state.34 Table 1 displays the individual (level-1) descriptive statistics of the study population. Geographic distribution of men by states is displayed in Appendix 2a and 2b. A total of 108 245 individuals were included in the study of which 92% received PCS during the last year. Further, 81.2% of the men aged 50 or above received PSA test during the last year. In comparison with men that did not receive screening, men who received screening were of older age (mean age=60.45±10 vs 65.5±10, P<0.01), whites (85.2%), attended some college or more (63.6%), currently married (66.7%), covered by at least some health insurance (94.2%), had at least one personal physician (92.1%), and had self-reported health status of good or above (77.1%). Table 2 displays the aggregate (level-2) variables of the study population. The majority of men that received PCS lived in MSA with or without center city (65.1%) and in states with the number of doctors ranging from 220.6 to 274.9 doctors per 100 000 persons. In the bivariate analysis, all of the examined level-1 and level-2 sociodemographics were significantly different (P<0.01) between men who did and did not receive PCS during the last year.

### Multilevel logistic regression analysis

The convergence criteria were satisfied in the both of the final models.30 We found significant state-level geographic variability (P<0.0001) in utilization of PCS (variance=0.129, s.e.=0.028), as well as PSAT (variance=0.033, s.e.=0.008) in the United States. ICCs were 3.80% and 1.00% for PCS and PSAT models, respectively. Proportional change in variances was 48.06% and 41.07% for PCS and PSAT models, respectively. For PCS, all individual variables were independent predictors (P<0.05), except self-reported health status (Table 3). When aggregate variables were added (level-2), MSA code was found to be an independent predictor of PCS. We found that in comparison with men not living in an MSA, men living in the city center of an MSA or outside the city center of an MSA but inside the county containing the center have lower odds to screen for prostate cancer, that is, OR=0.82, 95% CI=0.76–0.90 and OR=0.86, 95% CI=0.78–0.95, respectively. No significant cross-level interactions were found in this model.

For PSAT, all individual variables were independent predictors (P<0.05), except self-reported health status (Table 4). When aggregate variables were added (level-2), we found the number of doctors per 100 000 persons by state to be a significant level-2 predictor of utilization of PSAT. More specifically, we found that in comparison with men living in states with fourth quartile (76–100%) of availability of doctors (>274.9 doctors per 100 000 population), men living in states with first quartile (0–25%) of availability of doctors (<200.6 doctors per 100 000 persons) have lower odds to PSAT (OR=0.78, 95% CI=0.63–0.94). No significant cross-level interactions were found in this model.

Table 5 lists the types of association between PCS and its predictors as found in the multivariate analysis.

## Discussion

The main findings of this study are that after adjusting for individual characteristics there are health disparities at the state-level in PCS and PSAT in the United States, P<0.0001. We also found MSA code and the number of doctors per 100 000 persons per state being independent predictors of variations in PCS and PSAT, respectively. In total, the added predictors (individual and aggregate) explained 48.06% and 41.07% of the total variances in PCS and PSAT, respectively. Although we acknowledge the limitations associated with the use and interpretation of ICC in multilevel logistic regression models, we found ICCs for PCS and PSAT to be 3.8% and 1.0%, respectively.30 Further, our findings for the individual-level predictors are in congruence with previous literature reports on sociodemographic predictors of PCS.21, 22, 23, 24, 25

The current study is also important from the methodological context as it accounts for (1) clustering of data between individuals residing in the same geographic locations and (2) violation of logistic regression assumption of independence of observations due to the presence of within group (that is, state) variability because of nested data structure.

### Andersen model of health services utilization

PCS services in the United States can be explained based on the Andersen model of health services utilization (Figure 1).35 The Andersen model argues that the use of health services, such as PCS, is a function predisposing, enabling and need characteristics of individuals. The predisposing component is based on the idea that some individuals have greater tendencies to use health care services than others and these tendencies can be predicted through demographics, social structures and health beliefs. The enabling factors center on the idea that although people may be predisposed to use health services, means of obtaining these services are also required. The need factors constitute the most immediate cause of health services use and have two dimensions: (1) amount of illness that an individual perceives to exist, and (2) professionally evaluated need to use health services. In our analyses, each individual-level variable was an independent predictor of screening for prostate cancer, as per the theoretical framework of the Andersen model. One exception was self-reported health status, which was not significantly associated with screening for prostate cancer in the multivariate analysis.

We note that the Andersen model does not directly capture the influence of physician’s perceptions on PCS. However, such effects can be reflected indirectly in utilization of PCS through the existing structure of the model, that is, predisposing, enabling and need characteristics. For example, physicians may impact the utilization of PCS by influencing the need of individuals through patient-provider communication.

### Urban men have lower odds to screen for prostate cancer

In PCS, we found that people who live in urban areas have significantly (P<0.05) lower odds of screening for prostate cancer (Table 3). Higher probabilities of screening for prostate cancer among the rural population indicate its potential overuse in the rural population. However, current evidence on this is limited and further work is required to test this hypothesis. Future studies should examine drivers of these regional disparities in screening for prostate cancer. One such driver can be the differences in value clarification and risk communication in patient-provider relationships between the rural and urban populations. Future studies should also focus on exploring the health disparities in receiving PCS at other levels, such as cities and counties.

### State-level prevalence of physicians is a predictor of utilization of PCS

State-level availability of physicians was found to be an independent aggregate level predictor of PSA-based screening, after adjusting for individual-level predictors. In particular, we found men living in states with the lowest quartile of doctors (<220.6 per 100 000 persons) have lower odds (OR=0.78, 95% CI=0.63–0.94) of PSAT in comparison with men living in states with the highest quartile of doctors (>274.9 per 100 000 persons). Notably, this was in addition to the presence of a personal physician being an individual-level independent predictor of receiving PSA-based screening (OR=3.26, 95% CI=3.26–3.70). These results demonstrate health care system barriers in receiving PSA-based screening. Examples of health care system barriers are organizational factors such as the length of consultation and treatment, consultancy appointment and waiting time, and acquisition of a referral system.36, 37

### Requirement of risk-benefit assessment of PSA-based screening

The best practice evidence available on PSA-based screening for prostate cancer suggest that no substantial benefit is associated with using this procedure to screen for prostate cancer. Based on a review undertaken by the Agency for Healthcare Research and Quality, the US Preventive Services Task Force concluded that for every 1000 men screened for prostate cancer, less than one man will benefit while 43 men face serious harm during treatment. Harm includes incontinence, erectile dysfunction, cardiovascular events and life-threatening blood clots.38 Therefore, in a recent report (July 2012), the US Preventive Services Task Force recommended against the use of PSA-based screening in men, regardless of age, that is, a grade D recommendation.38 The American Cancer Society also recommends that patients make informed decisions about whether to undergo PCS after discussion with their doctors and by considering their own views on the benefits and side effects of the screening.39 We found that the majority of men aged 50 or more (82.2%) received PSA-based screening in the United States during the last year. In particular, men with high socioeconomic status were found to have higher odds of screening for prostate cancer, as suggested by indicator variables such as education level, annual income and employment status. However, we found substantially higher rates of PSA-based screening in comparison with the previous literature, which may be due to the exclusion of those men that did not respond yes or no when asked about having had a PSA test or a DRE during the last year. Nonetheless, future research is required for risk-benefit assessment of PSA-based screening on such a large scale.

### Comparison with previous literature

Previous studies have found evidence of significant health disparities at the level-2 in clinical practice patterns in many different areas of prostate cancer related health care delivery in the United States.40, 41, 42 Although none of these studies focused on screening for prostate cancer, the findings from these studies add to the validity of our results. Li et al.40 found that the higher level of neighborhood deprivation is associated with increased prostate cancer mortality, after adjusting for individual-level factors. Xiao et al.41 investigated racial differences in prostate cancer incidence, stage and grade in Florida using individual, community and environmental data. They found a significant association between late cancer stage, low median income and low percentage of people with some college education at the community level. Nambudiri et al.42 used the Veterans Affairs population to analyze the variations in prostate cancer treatment across different facilities (an aggregate variable) and concluded that the prostate cancer treatment varies substantially across different facilities.

### Study limitations

Several important limitations should be considered in our study. First, the median response rate for the 2010 BRFSS survey was 54.60% (range=39.05–68.78%). Therefore, the results of this study are prone to nonresponse bias. Second, the BRFSS respondents are limited to only those households that have landline telephones, thus our findings are not generalizable to certain populations. Third, as the BRFSS survey interviews are conducted in English or Spanish, people that do not speak either of these languages could be under-represented in the data. Fourth, like any observational study, our findings are susceptible to the residual confounding. Fifth, as discussed earlier, we found substantially higher rates of PCS (that is, PSAT and DRE) in our study. Therefore, our results may overestimate screening rates and should be interpreted with caution.

In conclusion, we found significant health disparities, at the state-level and individual-level, in clinical practice patterns of PCS in the United States. Men living in urban areas were found to have lower odds of overall PCS (with a PSA test and/or a DRE) in comparison with people living in rural areas of the United States (PCS). Men living in states with lower state-level availability of physicians were found to have lower odds of PSA-based screening. Future studies are required to quantify participation-levels of patients in decision-making for PSA-based screening and to examine potential health disparities at other levels such as cities and counties in the United States, as well as treatment outcomes associated with positive PCS.

## References

1. 1

Siegel R, Naishadham D, Jemal A . Cancer statistics, 2012. CA: Cancer J Clin 2012; 62: 10–29.

2. 2

Howlader N, Noone A, Krapcho M, Neyman N, Aminou R, Waldron W et al(eds) SEER Cancer Statistics Review, 1975–2008. National Cancer Institute: Bethesda, MD, http://seer.cancer.gov/csr/1975_2008/ 2011.

3. 3

American Cancer Society. What are the key statistics about prostate cancer? Atlanta, GA, 2012.

4. 4

Roehrborn CG, Black LK . The economic burden of prostate cancer. BJU Int 2011; 108: 806–813.

5. 5

American Urological Association. AUA Disputes Panel’s Recommendations on Prostate Cancer Screening, In: Lacy SS (ed). http://www.auanet.org/content/media/USPSTF_AUA_Response.pdf Linthicum, MD, 2012.

6. 6

Chou R, Croswell JM, Dana T, Bougatsos C, Blazina I, Fu R et al. Screening for prostate cancer: a review of the evidence for the US Preventive Services Task Force. Ann Intern Med 2011; 155: 762–771.

7. 7

Moyer VA . Screening for prostate cancer: U.S. preventive services task force recommendation statement. Ann Intern Med 2012; 157: 120–134.

8. 8

Andriole GL, Crawford ED, Grubb RL, Buys SS, Chia D, Church TR et al. Prostate cancer screening in the randomized prostate, lung, colorectal, and ovarian cancer screening trial: mortality results after 13 years of follow-up. J Natl Cancer Inst 2012; 104: 125–132.

9. 9

Schröder FH, Hugosson J, Roobol MJ, Tammela TLJ, Ciatto S, Nelen V et al. Prostate-cancer mortality at 11 years of follow-up. N Engl J Med 2012; 366: 981–990.

10. 10

Sirovich BE, Schwartz LM, Woloshin S . Screening men for prostate and colorectal cancer in the United States. JAMA 2003; 289: 1414–1420.

11. 11

Chan ECY, Barry MJ, Vernon SW, Ahn C . Brief report: physicians and their personal prostate cancer-screening practices with prostate-specific antigen. J Gen Intern Med 2006; 21: 257–259.

12. 12

Schootman M, Jeffe DB, Baker EA, Walker MS . Effect of area poverty rate on cancer screening across US communities. J Epidemiol Community Health 2006; 60: 202–207.

13. 13

Lian M, Schootman M, Yun S . Geographic variation and effect of area-level poverty rate on colorectal cancer screening. BMC Public Health 2008; 8: 358.

14. 14

Semrad TJ, Tancredi DJ, Baldwin LM, Green P, Fenton JJ . Geographic variation of racial/ethnic disparities in colorectal cancer testing among medicare enrollees. Cancer 2011; 117: 1755–1763.

15. 15

Hox JJ . Multilevel analysis: Techniques and applications. Taylor & Francis: Mahwah, NJ, 2010.

16. 16

Centers for Disease C. Behavioral Risk Factors Surveillance System Website. Atlanta, GA 2010.

17. 17

Nelson DE, Powell-Griner E, Town M, Kovar MG . A comparison of national estimates from the National Health Interview Survey and the Behavioral Risk Factor Surveillance System. Am J Public Health 2003; 93: 1335–1341.

18. 18

Nelson DE, Holtzman D, Bolen J, Stanwyck CA, Mack KA . Reliability and validity of measures from the Behavioral Risk Factor Surveillance System (BRFSS). Soz Praventivmed 2001; 46 (Suppl 1): S3–42.

19. 19

Stein AD, Lederman RI, Shea S . The Behavioral Risk Factor Surveillance System questionnaire: its reliability in a statewide sample. Am J Public Health 1993; 83: 1768–1772.

20. 20

Wolf A, Wender RC, Etzioni RB, Thompson IM, D'Amico AV, Volk RJ et al. American Cancer Society guideline for the early detection of prostate cancer: update 2010. CA Cancer J Clin 2010; 60: 70–98.

21. 21

Vadaparampil ST, Jacobsen PB, Kash K, Watson IS, Saloup R, Pow-Sang J . Factors predicting prostate specific antigen testing among first-degree relatives of prostate cancer patients. Cancer Epidem Biomarker Prev 2004; 13: 753–758.

22. 22

Seo HS, Lee NK . Predictors of PSA screening among men over 40 years of age who had ever heard about PSA. Kor J Urol 2010; 51: 391–397.

23. 23

Chiu BCH, Anderson JR, Corbin D . Predictors of prostate cancer screening among health fair participants. Public Health 2005; 119: 686–693.

24. 24

Austin OJ, Valente S, Hasse LA, Kues JR . Determinants of prostate-specific antigen test use in prostate cancer screening by primary care physicians. Arch Fam Med 1997; 6: 453–458.

25. 25

Nelson TF, Naimi TS, Brewer RD, Wechsler H . The state sets the rate: the relationship among state-specific college binge drinking, state binge drinking rates, and selected state alcohol control policies. Am J Public Health 2005; 95: 441–446.

26. 26

U.S. Census Bureau . 2007 Economic Census: Regions and Divisons 2007.

27. 27

US Department of Commerce. State Rankings -- Statistical Abstract of the United States: Doctors per 100,000 resident population 2007.

28. 28

US Department of Health and Human Services. Area Resource Files 2012.

29. 29

Merlo J, Yang M, Chaix B, Lynch J, Råstam L . A brief conceptual tutorial on multilevel analysis in social epidemiology: investigating contextual phenomena in different groups of people. J Epidemiol Community Health 2005; 59: 729–736.

30. 30

Goldstein H . Multilevel statistical models vol. 847. John Wiley & Sons Inc: London, 2010.

31. 31

Rasbash J Modelling UoLCfM. A user's guide to MLwiN. University of London, Institute of Education, Centre for Multilevel Modelling: London, 2000.

32. 32

Snijders TAB, Bosker RJ . Multilevel Analysis: An Introduction To Basic And Advanced Multilevel Modeling. SAGE publications Ltd: Thousand Oaks, CA, 1999.

33. 33

Merlo J, Chaix B, Ohlsson H, Beckman A, Johnell K, Hjerpe P et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health 2006; 60: 290–297.

34. 34

Office of Surveillance Epidemiology and Laboratory Services. Behavioral Risk Factor Surveillance System 2010. Summary Data Quality Report Centers for Disease Control and Prevention: Atlanta, GA 2010.

35. 35

Andersen RM . Revisiting the behavioral model and access to medical care: does it matter? J Health Soc Behav 1995; 36: 1–10.

36. 36

Scheppers E, Van Dongen E, Dekker J, Geertzen J . Potential barriers to the use of health services among ethnic minorities: a review. Fam Pract 2006; 23: 325–348.

37. 37

Osterberg L, Blaschke T . Adherence to medication. N Engl J Med 2005; 353: 487–497.

38. 38

Chou R, Croswell JM . Screening for prostate cancer. Ann Intern Med 2012; 156: 540.

39. 39

American Cancer S. Can prostate cancer be found early? Atlanta, GA 2010.

40. 40

Li X, Sundquist K, Sundquist J . Neighborhood deprivation and prostate cancer mortality: a multilevel analysis from Sweden. Prostate Cancer Prostatic Dis 2012; 15: 128–134.

41. 41

Xiao H, Gwede CK, Kiros G, Milla K . Analysis of prostate cancer incidence using geographic information system and multilevel modeling. J Natl Med Assoc 2007; 99: 218.

42. 42

Nambudiri VE, Landrum MB, Lamont EB, McNeil BJ, Bozeman SR, Freedland SJ et al. Understanding variation in primary prostate cancer treatment within the veterans health administration. Urology 2012; 79: 537–545.

## Author information

Authors

### Corresponding author

Correspondence to D W Raisch.

## Ethics declarations

### Competing interests

The authors declare no conflict of interest.

## Rights and permissions

Reprints and Permissions

Garg, V., Raisch, D., Selig, J. et al. Health disparities in clinical practice patterns for prostate cancer screening by geographic regions in the United States: a multilevel modeling analysis. Prostate Cancer Prostatic Dis 16, 193–203 (2013). https://doi.org/10.1038/pcan.2013.3

• Revised:

• Accepted:

• Published:

• Issue Date:

### Keywords

• cancer screening
• prostate-specific antigen test
• digital rectal examination
• geographic variation
• multilevel modeling

• ### Racial and Ethnic Variation in PSA Testing and Prostate Cancer Incidence Following the 2012 USPSTF Recommendation

• Kevin H Kensler
• , Claire H Pernar
• , Brandon A Mahal
• , Paul L Nguyen
• , Quoc-Dien Trinh
•  & Timothy R Rebbeck

JNCI: Journal of the National Cancer Institute (2020)

• ### Prostate examination among adult and elderly subjects in southern Brazil: a cross-sectional population-based study

• Kevin Francisco Durigon Meneghini
• , Hsu Yuan Ting
•  & Samuel Carvalho Dumith

Sao Paulo Medical Journal (2020)

• ### Geographical Variations in Prostate Cancer Outcomes: A Systematic Review of International Evidence

• Paramita Dasgupta
• , Joanne F. Aitken
• , Nicholas Ralph
• , Suzanne Kathleen Chambers
•  & Jeff Dunn

Frontiers in Oncology (2019)

• ### Association Between Antidiabetic Medications and Prostate-Specific Antigen Levels and Biopsy Results

• Kerri Beckmann
• , Danielle Crawley
• , Tobias Nordström
• , Markus Aly
• , Henrik Olsson
• , Anna Lantz
• , Noor Binti Abd Jalal
• , Hans Garmo
• , Martin Eklund
•  & Mieke Van Hemelrijck

JAMA Network Open (2019)

• ### Epidemiological Determinants of Advanced Prostate Cancer in Elderly Men in the United States

• Jinani Jayasekera
• , Eberechukwu Onukwugha