Targeting High-Risk Employees May Reduce Cardiovascular Racial Disparities

September 17, 2014
James F. Burke, MD, MS
James F. Burke, MD, MS

Sandeep Vijan, MD
Sandeep Vijan, MD

Lynette A. Chekan, MBA
Lynette A. Chekan, MBA

Ted M. Makowiec, MBA
Ted M. Makowiec, MBA

Laurita Thomas, MEd
Laurita Thomas, MEd

Lewis B. Morgenstern, MD
Lewis B. Morgenstern, MD

Volume 20, Issue 9

Targeting cardiovascular risk reduction interventions to high-risk patients has the potential to reduce cardiovascular racial disparities, improve health, and reduce costs.


A possible remedy for health disparities is for employers to promote cardiovascular health among minority employees. We sought to quantify the financial return to employers of interventions to improve minority health, and to determine whether a race- or risk-targeted strategy was better.

Study Design

Retrospective claims-based cohort analysis.


Unconditional per-person costs attributable to stroke and myocardial infarction (MI) were estimated for University of Michigan employees from 2006 to 2009 using a 2-part model. The model was then used to predict the costs of cardiovascular disease to the University for 2 subgroups of employees—minorities and high-risk patients—and to calculate cost-savings thresholds: the point at which the costs of hypothetical interventions (eg, workplace fitness programs) would equal the cost savings from stroke/ MI prevention.


Of the 38,314 enrollees, 10% were African American. Estimated unconditional payments for stroke/MI were almost the same in African Americans ($128 per employee per year; 95% CI, $79-$177) and whites ($128 per employee per year; 95% CI, $101- $156), including higher event rates and lower payments per event in African Americans. Targeting the highest risk decile with interventions to reduce stroke/MI would result in a substantially higher cost-savings threshold ($81) compared with targeting African Americans ($13). An unanticipated consequence of risk-based targeting is that African Americans would substantially benefit: an intervention targeted at the top risk decile would prevent 75% of the events in African Americans, just as would an intervention that exclusively targeted African Americans.


Targeting all high-risk employees for cardiovascular risk reduction may be a win-win-win situation for employers: improving health, decreasing costs, and reducing disparities.

Am J Manag Care. 2014;20(9):725-733

Disparities in cardiovascular disease incidence create a potential financial incentive for employers and insurers to improve the health of minorities. While the magnitude of this incentive is small for race-targeted strategies, risk-targeted strategies have the potential of generating a unique win-win-win situation for employers and insurers: improving health, decreasing costs, and reducing disparities.

• 75% of all events that would be prevented by targeting African Americans would be prevented by targeting the highest risk decile of employees.

• Even in the low-risk population studied, the cost-savings threshold is substantial ($198) for even modestly effective interventions targeting the highest risk decile.

Racial and ethnic disparities in the incidence, quality of care, and outcomes of cardiovascular disease and stroke have been widely documented.1-3 These disorders are potentially preventable and appropriate treatment can reduce morbidity and mortality. Thus, there is the hope that targeted interventions can reduce or even eliminate health disparities.4 Government programs to expand healthcare access have reduced access disparities,5,6 but significant outcome disparities persist even in insured populations.7,8

The potentially wider role of organizations outside of government, including employers, in funding programs to eliminate heath disparities deserves consideration. Employers benefit from a diverse, healthy, and productive workforce that minimizes the cost of insurance, employee absenteeism, and turnover. Further, some health promotion activities may be cost saving for employers.9,10 Are there incentives for employers to fund projects to improve minority health and reduce disparities for their employees?

We sought to understand the potential financial incentives for employers to implement interventions to improve minority health. We first estimated the cost differential for cardiovascular disease and stroke in African Americans compared with whites, hypothesizing that average costs would be higher in African Americans, per capita, due to increased disease incidence and severity. We also explored whether the best strategy would be to focus health improvement programs on minority employees, or instead on all employees with risk factors for heart disease and stroke. We then estimated the cost-savings threshold, the break-even point, for interventions to reduce cardiovascular risk targeted at either a) all African American employees, or b) all employees at high risk, defined as employees, white or African American, with an average annual risk of stroke or myocardial infarction (MI) greater than 1.5%. For employers, any intervention that could generate net cost savings represents a rare win-win situation: decreased disease burden and decreased net costs.


Study Population

We analyzed all health insurance claims for University of Michigan employees from 2006 to 2009. The University of Michigan, like many large employers, is self-insured and thus assumes the risk associated with health insurance claims from their employees. The study population included all continuously insured employees and their dependents under the age of 65 years. De-identified demographic and employment information for employees and dependents was obtained through university employment records. Race and ethnicity of employees was obtained from employment records in which it was declared by self-report. The university does not record race and ethnicity of dependents, thus we made a limiting assumption that dependent race and ethnicity was the same as that of the related employee.

The University of Michigan Institutional Review Board reviewed the study protocol and determined it was exempt from formal review as it relied on de-identified data.

Risk Factor and Diagnosis Assignment. An employee was classified as having stroke or MI if that employee had any claim between 2006 and 2009 in which one of the first 3 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes represented stroke or MI. MI was identified through ICD-9-CM codes classified as “Acute MI” (Single-Level Diagnosis Category 100) by the Agency for Healthcare Research and Quality (AHRQ) Clinical Classification Software (CCS).11 Stroke was identified using ICD-9-CM codes 433.x1, 434.x1, and 436.12 Vascular risk factors were classified using CCS categories. Modified Charlson Comorbidity Index (CCI) scores were calculated for each patient.13

Payments. Direct payments for medical care were abstracted from claims. Indirect costs, costs borne by the employer as a consequence of illness that are separate from direct medical expenditures, were estimated by summing 3 sources of employer cost: missed work, long-term disability, and worker replacement. The cost of missed work was estimated by multiplying the days of work lost to outpatient visits or hospitalizations14 by the employee’s daily salary. Long-term disability costs were calculated by multiplying an individual’s total compensation (1.3 times salary to account for benefits) by 0.66 (the fraction of total benefits paid out for long-term disability) for all individuals on long-term disability for the time spent on disability during the study. Costs of worker replacement were estimated for individuals who went onto long-term disability as 0.3 times the individual’s salary.15 All cost data were inflation-adjusted to 2009 dollars using the gross domestic product deflator.16 Because the available mortality data was incomplete (no data was available on employees who died in the first 15 months of the study), we did not account for patients who died. However, this would likely only have a small effect on our estimates because only a very small proportion of the population died; 195 of the entire study population (N = 38,314) died during the last 33 months of the study, including 12 of 502 of the employees with stroke or MI.

Statistical Analysis. Characteristics of the population were summarized with descriptive statistics. Comparisons between racial and ethnic groups were made using Analysis of Variance (ANOVA) for normally distributed continuous variables, χ2 tests for categorical variables, and Kruskal-Wallis tests for non-normally distributed variables.

Modeling Rationale and Approach. To quantify the difference in payments for stroke/MI in minorities, we sought to determine the relationship between race/ethnicity and payments independently of other factors that predict payments (eg, age). Two features of the claims data posed challenges for simple modeling approaches: the large number of patients with zero payments and the heavily right-skewed distribution of payments. To address these issues, we developed a 2-part model. This approach relies on estimating 2 separate quantities for each individual: 1) the probability of having a stroke or MI in a given year, and 2) the estimated cost of stroke/MI, conditional on that particular individual having a stroke or MI.17 By multiplying these 2 estimates, we can estimate the unconditional costs due to stroke or MI in a given year—the predicted costs for stroke and MI for a given employee in a given year.

To implement this approach, we first used logistic regression to estimate an individual employee’s risk of stroke or MI. Specifically, we estimated the probability of an employee having a stroke or MI in a given year (Part 1 of the model; logit [probability of stroke or MI] = β0 + β1 X age + β2 X sex + β3 X race/ethnicity + β4 X hypertension + β5 X diabetes + β6 X hyperlipidemia + β7 X salary decile + β8 X job type + β9 X CCI). We then defined risk deciles based on this model and categorized the number of actual events and race of employees across risk deciles. Second, we estimated payments in employees who had a stroke or MI in a given year. Specifically, ordinary least squares (OLS) regression was used to predict log-transformed annual payments in individuals who were diagnosed with stroke or MI in that year (Part 2 of model; log (annual payments) = β1 X age + β2 X sex + β3 X race/ethnicity + β4 X hypertension + β5 X diabetes + β6 X hyperlipidemia + β7 X salary decile + β8 X job type + β9 X CCI + Ɛ1). By multiplying an individual employee’s risk of stroke/MI by the predicted payments for stroke/MI in patients as if they had an event, we were able to estimate payments for each employee in each year prior to knowing whether they would have an event—unconditional annual payments attributable to stroke or MI for an individual employee. Duan-smear detransformation was used to detransform Part 2 of the model.18 Differences in payments between African Americans and whites were characterized using mean unconditional estimated payments by race. Standard errors were estimated using bootstrapping. Both parts of the model were adjusted for age, sex, vascular risk factors, salary, job category, and CCI.

We explored alternate modeling strategies including both 1- and 2-part generalized linear modeling approaches following the approach of Buntin and Zaslavsky.19 We chose the 2-part logistic-OLS model described above because this model predicted most accurately over the range of payments, had the lowest root mean squared error, and had relatively normally distributed residuals.

Estimating Cost-Savings Thresholds

To determine the cost implications of potential interventions to prevent stroke/MI in targeted groups, we estimated the number of events prevented and racial/ethnic distribution of event prevention for hypothetical interventions of varying efficacy, assuming equal relative treatment effects across strata of risk and race. We then determined the cost-savings threshold of these hypothetical interventions—the point at which the cost of a hypothetical intervention would be equal to the cost savings generated by stroke/MI prevention, estimated by our 2-part model. In contrast to a conventional cost analysis, which specifies the estimated cost of a specific intervention up front, this approach works backwards by first estimating the cost savings associated with event prevention and then dividing by the number of targeted enrollees to estimate the cost per enrollee that would break even, from the perspective of the employer. We generated cost-savings thresholds by solving a linear equation where the cost of a hypothetical intervention was set equal to cost savings attributable to event prevention: intervention cost = cost per event X expected events X (1 — Relative Risk [RR]) (Number of enrollees in target population). Using this equation, we were able to vary the RR of the hypothetical intervention and estimate the intervention costs that would break even. Relative risk was conceptualized as an estimate of intervention effectiveness—ie, efficacy after accounting for imperfect enrollment and compliance with an intervention—rather than pure efficacy. All analyses were performed with Stata, version 12 (Stata Corp, College Station, Texas).


Table 1

Population Characteristics. The study included a total of 38,314 individuals, 16,179 (42%) of whom were male. Eighty-eight percent of the study population was white, 10% was African American, and 3% was Hispanic American. Mean average annual medical payments were $3355 (SD $7825). Both the African American and Hispanic American populations were younger than the white population, although the burden of vascular risk factors was increased in African Americans relative to whites ().

Table 2

Characteristics of Enrollees With Stroke or MI. Vascular events were relatively uncommon in this low-risk cohort; 225 MIs and 286 strokes occurred during 153,256 person-years of observation. So few events (4 MIs and 2 strokes) were recorded in the Hispanic American population that they were excluded from comparative analyses. Mean annual payments among those with vascular events were no different in African Americans with MI or stroke than in whites ($12,130 vs $14,313, P = .09) ().


Figure 1

Predicted Unconditional Payments for Stroke/MI byRace. In the first part of our model (predicting stroke/MI) African American race was non significantly associated with a higher risk of stroke/MI (odds ratio [OR] 1.2; 95% CI, 0.9-1.2) while in the second part of our model (predicting payments per event), African American race was associated with 45% lower [95% CI, 79%-11%] direct payments per event (, available at Estimates from both parts of our model were combined to estimate the unconditional cost of stroke or MI per enrollee (). Unconditional estimated direct payments for stroke/MI in African Americans ($89; 95% CI, $51-$128) per person per year (PPPY) were lower, but not by a significant margin, than in whites ($109; 95% CI, $85-$133), P = .40. When indirect costs were included, estimated unconditional payments were almost the same in African Americans ($128; 95% CI, $80-$176) and whites ($128; 95% CI, $103-$154), P = .98.

We also estimated payments in African Americans as if there were no disparity in payments per event (by estimating payments per event for African Americans using Model Part 2, but excluding the race term) to isolate the cost impact of increased incidence in African Americans. In this case, estimated payments for African Americans increased substantially from $89 (95% CI, $51-$128) to $139 (95% CI, $94-$186) excluding indirect payments and from $128 (95% CI, $81-$175) to $168 (95% CI, $115-$221) including indirect payments.

Table 3

Risk/Race Distribution. As a consequence of the increased risk of stroke/MI associated with African American race and the increased burden of vascular risk factors in African Americans, African Americans were disproportionately represented in the highest risk strata. African Americans comprised 9.9% of the overall population and 14.0% of the highest risk decile. The top risk decile represented significant overall risk; 59% of all patients with stroke/MI and 75% of African Americans with stroke/ MI were in this decile (; decile 10).

Impact of Hypothetical Targeted Interventions. Two key observations emerged from our analysis of hypothetical interventions. First, as expected, considerably more events are prevented when the highest risk groups are targeted; targeting a moderately effective intervention (RR 0.9) at the highest risk decile (number needed to treat [NNT] to prevent stroke/MI: 516) would prevent about 3 times as many events as targeting the second-highest risk decile (NNT 1625). Surprisingly, however, we found that targeting high-risk groups would also prevent a disproportionately large number of events in African Americans and thus potentially reduce disparities. Targeting the highest risk decile would prevent 75% as many events in African Americans (1.2 per year) as an intervention that exclusively targeted African Americans (1.6 per year) and would prevent 59% as many events in whites as targeting the entire population (7.4 vs 12.6 per year).

Figure 2

In this population, we found considerable increases in cost-savings thresholds with increasing risk in the target population and increasing intervention efficacy (). Targeting the highest risk patients (individuals with a predicted annual risk >2% [n = 717]) has a cost-savings threshold of $198 per person. Thus, an intervention (RR = 0.9) that cost less than $198 PPPY to implement is predicted to reduce net costs for an employer because it would reduce expenditures on stroke/MI by $198 PPPY. The cost-savings threshold falls to $81 PPPY when targeting the top risk decile (n = 3819), to $49 when targeting the top 2 risk deciles (n = 7651), and to $13 when targeting African Americans (n = 3714). Increasing intervention efficacy significantly increases the cost-savings threshold. For example, increasing intervention efficacy (decreasing RR from 0.9 to 0.6 in intervention compared with non intervention patients) results in cost-savings threshold increases from $81 to $323 in the top risk decile. The obvious downside to more narrowly targeting interventions is a decrease in the absolute number of events prevented: targeting individuals with a risk >2% would prevent only 21% of the events if the entire population were targeted while targeting the top risk decile would prevent 59% and targeting the top 2 risk deciles would prevent 77%.


In our population, there was no significant per person payment differential attributable to racial disparities in MI and stroke. The primary implication of this finding is that targeting cardiovascular disease reduction strategies to African Americans is unlikely to be cost saving for employers when considering an employee population similar to that of the University of Michigan. Unsurprisingly, we found that across the spectrum of intervention efficacy, risk-targeted strategies will result in lower costs per event prevented than strategies that explicitly target African Americans. Given the higher burden of cardiovascular disease in African Americans, risk-targeted strategies prevent a large number of cardiovascular events in African Americans. However, an unanticipated consequence of risk-targeted strategies, in our population, was the magnitude by which they prevented events in African Americans: 75% of the events prevented in African Americans with race-targeted strategies were also prevented just by targeting the highest risk decile. Consequently, risk-targeted interventions for cardiovascular event prevention have the potential to generate a unique win-win-win for employers: improved health, cost savings, and disparity reduction.

Improving cardiovascular prevention for minorities is a challenge of enormous scope. Reaching this goal will require effort at multiple organizational levels—individual, community, employer, government, and charitable foundations among them. Thus far, little attention has been paid to the potential role of employers in accomplishing this goal. Barriers to more extensive employer involvement include uncertainty about the business case for reducing disparities, the time scale of return on investment for interventions, and taboos surrounding discussions of race in the workplace.20 However, if realistic incentives existed for employers, they could play a role in establishing the business case for improving minority health. Our data suggest that the development of low-cost interventions to minimize cardiovascular risks would create just such an incentive. The magnitude of this incentive may be substantially greater for employers with employees at higher risk of cardiovascular disease; the University of Michigan employee population, by contrast, has a low prevalence of vascular risk factors, high levels of education, and relatively high socioeconomic status.

The current findings illustrate some of the nuances in understanding employer incentives to improve the health of minorities. In spite of its significant disease burden,

the overall cost of cardiovascular disease for employers is relatively modest. This is in part driven by the healthy worker effect21 mediated through the low burden of vascular risk factors in our study population. In our sample, adjusted annual payments were surprisingly lower in African Americans with stroke or MI. The impact of these differences was substantial: estimated annual direct costs of stroke/MI in African Americans were $89, and would increase to $139, if per event payments were held equal across races. The reasons for these differences in the cost of care are unclear. In other contexts, such differences in the costs of care are most credibly explained by a disparity in provided care independent from differences in disease severity,22,23 with a contribution from patient preferences.24,25 In our sample, it is unlikely that event severity was lower in African Americans. If disease severity were lower in African Americans we would have expected lower indirect costs (eg, long-term disability payments), and instead we found the opposite. These data illustrate how racial disparities in the costs of care limit the financial incentive of employers to improve minority health. If, in fact costs of care were higher in African Americans, given the higher incidence of cardiovascular disease in African Americans, healthcare expenditures would be higher in the average African American and employers would have a financial incentive to improve minority health and thus minimize costs.

Our findings establish the rough boundaries of cost and effectiveness that interventions must meet in order to generate cost savings for employers. Unfortunately, little is known about the cost of interventions to improvecardiovascular health.26 One such analysis—the WISEWOMAN project, which effectively reduced vascular risk factor burden through screening in combination with diet and exercise interventions—was cost-effective at the societal level. However, it would only have been cost saving if targeted at the highest risk patients in our population, given an average per person intervention cost of $121 and RR of 0.91.27 However, even if existing interventions are only cost-effective in the highest risk groups, development of more effective or less costly interventions would extend the cost-savings threshold to include more patients. Furthermore, even existing interventions may be cost saving over more widely targeted populations than suggested by our estimates because our estimated cost-savings thresholds do not capture all of the real-world costs to employers of stroke and MI. For example, they do not account for lost productivity at work attributable to cardiovascular disease (presenteeism) or long-term increases in direct medical costs relative to other expenditures.

There are several important limitations to our study. First, lack of data on medication and preventive care utilization leads to the possibility that we have overestimated cardiovascular risk associated with race. If, for example, African Americans used risk-reducing medications less often or underwent fewer risk-reducing procedures than whites, then the higher estimated risk in African Americans may be a reflection of these differences rather than race per se. The potential consequence of such misattribution would be to misestimate the cost-savings associated with targeting African Americans because interventions may have differential effectiveness by race—an assumption we did not account for in our hypothetical interventions. Given the more than 6-fold cost savings differential between risk- and race-targeted strategies, however, it is likely a substantial differential would persist even if interventions were much more effective when targeting African Americans. Second, because race data were not available for dependents, we had to make the assumption that the race of dependents was the same as the race of the enrolled employee. Third, given the small number of Hispanic Americans in our population and the very low incidence of vascular events in this population, our results do not speak to the magnitude of disparities in this important population.28 Fourth, our analysis is limited to the employed population and thus does not include elderly Americans, a relevant cost consideration for employers that provide insurance coverage to retirees.Author Affiliations: Department of Veterans Affairs, VA Center for Clinical Management and Research / VA Healthcare System, Ann Arbor, MI, and Robert Wood Johnson Clinical Scholars Program, University of Michigan, Ann Arbor, University of Michigan, Ann Arbor (JFB, LBM); Department of Internal Medicine, University of Michigan, Ann Arbor (SV); University Human Resources, University of Michigan, Ann Arbor (LAC, TMM, LT).

Funding Source: This work was supported by the Robert Wood Johnson Foundation Clinical Scholars Program, and an associated Veterans Affairs Advanced Fellowship to Dr Burke. Dr Vijan received research support from NIH and the Department of Veterans Affairs. Dr Morgenstern received research support from NIH grant R01NS038916 and AHRQ grant R18HS017690.

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (JFB, SV, LAC, TMM, LT, LBM); acquisition of data (JFB, LAC); analysis and interpretation of data (JFB, SV, LBM); drafting of the manuscript (JFB); critical revision of the manuscript for important intellectual content (SV, LAC, TMM, LT, LBM); statistical analysis (JFB, SV); provision of study materials or patients (LAC, TMM, LT); administrative, technical, or logistic support (SV, LAC, TMM, LT); and supervision (SV, LBM).

Address correspondence to: James F. Burke, MD, MS, Robert Wood Johnson Clinical Scholars Program, University of Michigan Medical School, 6312 Medical Science Building 1, Ann Arbor, MI 48109. E-mail: Institute of Medicine (US) Committee on Understanding and Eliminating Racial and Ethnic Disparities in Health Care; Smedley BD, Stith AY, Nelson AR, eds. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC: National Academies Press; 2003.

2. Mensah GA, Brown DW. An overview of cardiovascular disease burden in the United States. Health Aff (Millwood). 2007;26(1):38-48.

3. Agency for Healthcare Research and Quality. National Healthcare Disparities Report, 2009. Rockville, MD: HHS;2010:1-300.

4. Davis AM, Vinci LM, Okwuosa TM, Chase AR, Huang ES. Cardiovascular health disparities: a systematic review of health care interventions. Med Care Res Rev. 2007;64(5 Suppl):29S-100S.

5. Shone LP, Dick AW, Klein JD, Zwanziger J, Szilagyi PG. Reduction in racial and ethnic disparities after enrollment in the State Children’s Health Insurance Program. Pediatrics. 2005;115(6):e697-e705.

6. Maxwell J, Cortés DE, Schneider KL, Graves A, Rosman B. Massachusetts’ health care reform increased access to care for Hispanics, but disparities remain. Health Aff (Millwood). 2011;30(8):1451-1460.

7. Karter AJ, Ferrara A, Liu JY, Moffet HH, Ackerson LM, Selby JV. Ethnic disparities in diabetic complications in an insured population. JAMA. 2002;287(19):2519-2527.

8. Fiscella K, Franks P, Doescher MP, Saver BG. Disparities in healthcare by race, ethnicity, and language among the insured: findings from a national sample. Med Care. 2002;40(1):52-59.

9. Aldana SG. Financial impact of health promotion programs: a comprehensive review of the literature. Am J Health Promot. 2001;15(5):296-320.

10. Ozminkowski RJ, Ling D, Goetzel RZ, et al. Long-term impact of Johnson & Johnson’s Health & Wellness Program on health care utilization and expenditures. J Occup Environ Med. 2002;44(1):21-29.

11. Elixhauser A, McCarthy EM. Clinical classifications for Health Policy Research, Version 2: Hospital Inpatient Statistics. Rockville, MD: Agency for Health Care Policy and Research; 1996.

12. Ovbiagele B. Nationwide trends in in-hospital mortality among patients with stroke. Stroke. 2010;41(8):1748-1754.

13. Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130-1139.

14. Ivanova JI, Birnbaum HG, Kidolezi Y, Qiu Y, Mallett D, Caleo S. Economic burden of epilepsy among the privately insured in the US. Pharmacoeconomics. 2010;28(8):675-685.

15. Sasha Corporation. Compilation of Turnover Cost Studies. Accessed December 13, 2013.

16. Medical Expenditure Panel Survey / Agency for Healthcare Research and Quality. Using Appropriate Price Indices for Analyses of Health Care Expenditures or Income across Multiple Years. Accessed December 13, 2012.

17. Duan N, Manning WG, Morris CN, Newhouse JP. A comparison of alternative models for the demand for medical care. J Bus Econ Stat. 1983;1(2):115-126.

18. Duan N. Smearing estimate: a nonparametric retransformation method. J Am Stat Assoc. 1983;78(383):605-610.

19. Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? comparing methods of modeling Medicare expenditures. J Health Econ. 2004;23(3):525-542.

20. Lurie N, Somers SA, Fremont A, Angeles J, Murphy EK, Hamblin A. Challenges to using a business case for addressing health disparities. Health Aff (Millwood). 2008;27(2):334-338.

21. Gilbert ES. Some confounding factors in the study of mortality and occupational exposures. Am J Epidemiol. 1982;116(1):177-188.

22. Schulman KA, Berlin JA, Harless W, et al. The effect of race and sex on physicians’ recommendations for cardiac catheterization. N Engl J Med. 1999;340(8):618-626.

23. Kressin NR, Petersen LA. Racial differences in the use of invasive cardiovascular procedures: review of the literature and prescription for future research. Ann Intern Med. 2001;135(5):352-366.

24. Whittle J, Conigliaro J, Good CB, Joswiak M. Do patient preferences contribute to racial differences in cardiovascular procedure use? J Gen Intern Med. 1997;12(5):267-273.

25. Schecter AD, Goldschmidt-Clermont PJ, McKee G, et al. Influence of gender, race, and education on patient preferences and receipt of cardiac catheterizations among coronary care unit patients. Am J Cardiol. 1996;78(9):996-1001.

26. Agency for Healthcare Research and Quality. Strategies for Improving Minority Healthcare Quality: Evidence Report/Technology Assessment, No. 90. Rockville, MD: HHS; 2005:1-198.

27. Finkelstein EA, Khavjou O, Will JC. Cost-effectiveness of WISEWOMAN, a program aimed at reducing heart disease risk among lowincome women. J Womens Health (Larchmt). 2006;15(4):379-389.

28. Burke JF, Brown DL, Lisabeth LD, Sánchez BN, Morgenstern LB. Enrollment of women and minorities in NINDS trials. Neurology. 2011;76(4):354-360.