Mandated Coverage of Preventive Care and Reduction in Disparities: Evidence From Colorectal Cancer Screening

Abstract

Objectives. We identified correlates of racial/ethnic disparities in colorectal cancer screening and changes in disparities under state-mandated insurance coverage.

Methods. Using Behavioral Risk Factor Surveillance System data, we estimated a Fairlie decomposition in the insured population aged 50 to 64 years and a regression-adjusted difference-in-difference-in-difference model of changes in screening attributable to mandates.

Results. Under mandated coverage, blood stool test (BST) rates increased among Black, Asian, and Native American men, but rates among Whites also increased, so disparities did not change. Endoscopic screening rates increased by 10 percentage points for Hispanic men and 3 percentage points for non-Hispanic men. BST rates fell among Hispanic relative to non-Hispanic men. We found no changes for women. However, endoscopic screening rates improved among lower income individuals across all races and ethnicities.

Conclusions. Mandates were associated with a reduction in endoscopic screening disparities only for Hispanic men but may indirectly reduce racial/ethnic disparities by increasing rates among lower income individuals. Findings imply that systematic differences in insurance coverage, or health plan fragmentation, likely existed without mandates. These findings underscore the need to research disparities within insured populations.

Colorectal cancer (CRC) incidence and mortality rates have declined substantially for the overall population since the late 1990s as the result of a confluence of significant advances in screening, a consensus in screening recommendations for individuals aged 50 to 75 years from the medical community, and evidence of cost-effectiveness of screening.^1–5 Estimates have indicated that increases in screening explain approximately half of the observed decrease in incidence and in mortality.⁶ Yet disparities in CRC screening, incidence, treatment, and mortality persist, even within insured populations.^7,8 The translation of advances in CRC screening procedures and adoption of uniform guidelines into practice occurs in the context of health services delivery settings, including the private health insurance system. Systematic fragmentation in health plan coverage of CRC screening may contribute to systematic disparities. We focused on disparities in screening by race and ethnicity among the insured population, examined the role of socioeconomic status (SES) and community-level factors in explaining these disparities, and estimated changes in these disparities associated with state-mandated insurance coverage of CRC screening procedures. In doing so, we provided an indirect analysis of the potential role of health plan fragmentation.

Health plan fragmentation occurs when differences in the amounts and quality of health care consumed arise because of systematic differences in plan benefits and cost sharing.⁹ Available data have suggested substantial fragmentation existed in coverage of CRC screening in the privately insured population before most state mandates.¹⁰ For example, in a national sample of 180 plans, only 57% covered colonoscopy for average-risk individuals in 1999.¹¹ Between 1999 and 2011, 34 states and the District of Columbia enacted CRC screening mandates. In theory, mandated coverage should lead to more uniform coverage of screening procedures and reduce fragmentation. If fragmentation is correlated with race and ethnicity, then mandates may also reduce racial and ethnic disparities in screening.

Several previous studies have examined the determinants of CRC screening disparities, and the majority of nationally representative studies have suggested that individual SES is a key determinant.^12–19 Indeed, several found that disparities were substantially reduced or even eliminated after adding measures of SES and access to care to regression models.^12–15 Yet, the data used in these studies did not indicate whether individuals’ insurance plans covered CRC screening. If the probability that an individual’s insurance plan covers CRC screening is correlated with observed measures of individual SES, then the apparent relationship between individual SES and screening may be due in part to underlying health plan fragmentation.

Although there is no nationally representative data set that contains information about individuals’ insurance coverage for CRC screening, expansions in insurance coverage introduced through Medicare or state insurance mandates provide natural experiments in which it is possible to study underlying health plan fragmentation indirectly. If these expansions in coverage are associated with reductions in disparities, either directly or indirectly through a shift in screening patterns across the SES distribution, then this would suggest that health plan fragmentation may contribute to observed disparities. Only 1 previous study examined possible changes in screening disparities after state-mandated coverage and found no statistically significant changes in disparities, but this study estimated a 1.4-percentage-point increase in annual endoscopic screening overall.²⁰ A study evaluating the effect of expanded Medicare coverage of CRC screening in 2001, however, found that increased coverage eliminated the gap between non-Hispanic Whites and Blacks but actually increased the gap between non-Hispanic Whites and Hispanics.²¹ Two other studies have estimated the effects of mandates on screening rates in the population as a whole and found small²² or null²³ effects.

In this study, we extend the existing literature in 2 key ways. First, we provide a formal decomposition of disparities in the insured population aged 51 to 64 years that includes a more comprehensive set of state- and county-level measures of the economic environment and local availability of health care than previous work. All previous studies of disparities using national data have included all individuals (insured and uninsured) recommended to receive CRC screening^12–18,24 or have focused only on the Medicare population.^{7,8,21,25–26} The decomposition technique is advantageous because, unlike the sequential regression technique used in most previous studies,^12–19 it is not sensitive to the order of covariates and it allows assessment of the relative contribution of screening determinants. One previous study²⁴ used the decomposition technique, but it included insured and uninsured individuals, had a less expansive set of community-level measures, and used data from 1998.

Second, we estimated changes in observed disparities under state-mandated insurance coverage using a triple-difference (DDD) estimation strategy as in Hamman and Kapinos to account for possible within-state trends in screening behavior.²² Our DDD estimates are based on before- and after-mandate and cross-state variation in screening rates among insured individuals aged 50 to 64 years relative to Medicare age-eligible individuals (aged 65–75 years). This estimation strategy has been used extensively in the policy evaluation literature.^22,27–29 This research is of particular interest since the passage of the Patient Protection and Affordable Care Act (ACA) because the ACA mandated coverage for CRC screening under private insurance plans.³⁰ Thus, our analysis of state policies may foreshadow the likely effects of the ACA.

METHODS

We used individual-level Behavioral Risk Factor Surveillance System (BRFSS) data from 2002 to 2008 matched to state-level mandate enactment dates obtained from the State Cancer Legislative Database Program and the National Conference of State Legislatures. Our sample includes only the years in which the colorectal cancer screening questions were fielded in the core survey: 2002, 2004, 2006, and 2008. We excluded the years before Medicare expanded coverage of CRC screening, which occurred in 2001 (see Hamman and Kapinos for a more detailed discussion of the mandates²²). We also merged in additional sources of state- and county-level data to account for differences in community and institutional environments.

We restricted our sample to insured individuals aged 51 to 75 years at the time of the interview, excluding those who were aged 65 years because we could not ascertain Medicare eligibility in the entire year preceding the interview for those individuals. We excluded individuals with missing values or who refused to answer the questions used to create our main measures. The analysis sample consisted of 193 970 individuals aged 51 to 64 years and 95 863 individuals aged 66 to 75 years (full sample n = 289 833).

Main Measures

Outcomes.

We defined our first outcome variable following previous adherence measures for CRC screening^16,18 as a dichotomous measure of whether an individual was up to date with the US Preventive Services Task Force screening recommendations. During our study period, the guidelines recommended annual blood stool tests (BSTs) and flexible sigmoidoscopy every 5 years or colonoscopy every 10 years.³ The BRFSS does not distinguish between flexible sigmoidoscopy and colonoscopy, so we defined up to date as equal to 1 for any individual who had either a BST within the past year or an endoscopic screening (includes flexible sigmoidoscopy and colonoscopy) within the past 5 years. (The results did not differ if we instead coded those who had an endoscopic screening within the past 10 years as up to date.) Although screening guidelines do not advocate one screening mode over another, BST and endoscopic tests differ substantially in their specificity, sensitivity, cost, and patient preference.³¹ Thus, we defined 2 additional outcome measures to analyze screening models separately. Our BST and endoscopy measures each equaled 1 for individuals who had the test within the recommended time frame (≤ 1 year for BST, ≤ 5 years for endoscopy).

Race and ethnicity measures.

Our race variable included the following Census racial categories: White, Black, Asian, Native Hawaiian or Pacific Islander, and Native American or Alaska Native. (In earlier years, the BRFSS did not distinguish between Asian and Native Hawaiian or Pacific Islander ethnicities. For comparability across years and to ensure adequate statistical power for our analysis, we collapsed these categories into 1 group.) We used a separate variable to measure Hispanic/Latino ethnicity so individuals of any race could also be coded as Hispanic. In our decomposition results and descriptive statistics, however, we constructed a non-Hispanic White category to allow for clearer comparison to the majority group. Although more detailed race and ethnicity categories are available in more recent BRFSS surveys, we adopted this coding for consistency across years and to ensure adequate statistical power to produce estimates.

Socioeconomic status measures.

We used 2 measures of SES in our analysis. First, we used the BRFSS categorical measure of annual household income, which has the following categories: less than $10 000, $10 000 to $14 999, $15 000 to $19 999, $20 000 to $24 999, $25 000 to $34 999, $35 000 to $49 999, $50 000 to $74 999, and $75 000 or more. Second, we included a BRFSS categorical measure of educational attainment with the following 4 categories: less than high school, high school, some college, and 4-year degree or more.

Policy measure.

For our DDD estimation, we defined a policy variable as equal to 1 for individuals aged 50 to 64 years living in states in which a mandate was in effect at least 12 months before the respondent’s interview date. This coding scheme is the fixed-effects specification for a DDD estimator. As explained, we included state, year, and age group fixed effects as well as all 2-way interactions. In the results presented here, we did not interact each racial/ethnic group with these fixed effects because doing so would have substantially increased the number of parameters to estimate. Omitting these interactions imposed the assumption that differences between racial and ethnic groups are time constant and do not vary across states or age groups. Relaxing this assumption to allow for differences in the effect of race across the 2 age groups and over time yielded similar results to those presented. We rounded the dates on which the legal mandate was effective to the next January for mandates enacted in February–December and January of the current year for mandates enacted in January. This measure is standard in the mandate literature and reflects the plan-year renewal cycle.^22,32

Age (treatment) group variable.

We defined a treatment group indicator variable equal to 1 for people younger than the Medicare age cutoff of 65 years. In mandate states, these individuals are expected to experience a change in insurance coverage only when mandates pass. We note that some people in this group may have had coverage for CRC screening before the state mandate. Also, some may have been covered by self-insured plans, which are exempted from state mandates under the Employee Retirement Income Security Act. We were unable to identify these individuals in our data, so they were coded as part of the treatment group. Including them in the treatment group may have biased our estimates toward zero (no effect).

County- and state-level measures.

We included the following county-level measures to capture socioeconomic diversity and differences in access across communities³³: percentage of the county population below the poverty line, percentage of the county population living in rural areas, an indicator equal to 1 if the entire county was designated as a Health Professional Shortage Area (HPSA–Primary Care), and a count of geographical HPSAs within the county. In our decomposition analysis, we included the rate of health maintenance organization (HMO) penetration at the state level by year. (County-level US Census data were obtained from http://www.census.gov/support/USACdataDownloads.html; HPSA data were obtained from the US Department of Health and Human Services Health Resources and Services Administration Data Warehouse, http://datawarehouse.hrsa.gov/topics/shortageAreas.aspx; and HMO penetration data were from the Health Leaders-InterStudy. We omitted HMO penetration from the DDD model because it is perfectly collinear with the state–year fixed effects, which are included in that specification.) We included state fixed effects to account for any unobservable time-constant differences across states, such as climate or regional diet, that may be related to CRC risk. We were unable to match information for BRFSS observations without county Federal Information Processing Standards codes, and thus our sample systematically excluded residents of smaller counties.

Other controls.

To account for differences in CRC risk, we used age-adjusted CRC mortality rates (cases per 100 000) for deaths occurring between 1990 and 1999 by state, gender, and race/ethnicity from Surveillance, Epidemiology, and End Results Program data.³⁴ Missing state values were set equal to the national average for people of the same gender and race or ethnicity. Because Hispanic ethnicity is not mutually exclusive from other racial groups in these data (or in our BRFSS analysis sample), we matched 2 variables—mortality rate for one’s race and for one’s ethnicity—to the BRFSS data by state, gender, and race/ethnicity.

Our specifications also included controls available in the BRFSS that have been shown in previous research to be associated with CRC screening or CRC risk, including age, self-reported overall health, whether the respondent was diabetic, whether the individual was obese, whether the respondent had ever smoked, marital status, and month in which the respondent was interviewed. (We entered age into the analysis using a full set of dummy variables, which allows for arbitrary nonlinear patterns in screening across the age distribution; obesity was based on body mass index calculated from self-reported height and weight and was equal to 1 if body mass index was ≥ 30 kg/m².) Age, obesity, diabetes, and tobacco use have been linked to an increased odds of developing CRC.^35–39 According to the Centers for Disease Control and Prevention, CRC risk is higher for older people, people with a history of inflammatory bowel disease, people with a familial history of colorectal cancer or a genetic syndrome, and people with a lack of regular physical activity, low fruit and vegetable intake, a low-fiber–high-fat diet, obesity, alcohol consumption, and tobacco use.⁴⁰

Statistical Analyses

To analyze correlates of racial and ethnic disparities in CRC screening, we used the Fairlie decomposition method for nonlinear models and survey data.⁴¹ These estimates were produced using the Fairlie command and Stata version 12 (StataCorp LP, College Station, TX). Like the Oaxaca-Blinder decomposition used for linear models, the Fairlie decomposition estimates the proportion of observed disparities attributable to differences in observed characteristics (known as differences in the coefficients) as opposed to differences in the effects of those characteristics on screening behavior (differences in the effects).

To examine the effects of mandated insurance coverage on CRC screening disparities, we produced the DDD estimate on the basis of adjusted screening rates for those in mandate and nonmandate states, those in the treatment group (aged 51–64 years), and those in the comparison group (Medicare age eligible), over time:

where Screen is 1 of the 3 outcome measures defined earlier (up to date, BST in the past year, or endoscopy in the past year) for each group. Like a differences-in-differences model, the DDD is based on before- and after-policy variation across states that introduce mandates and those that do not. The DDD adds an additional within-state control group, Medicare-eligible individuals aged 65 to 75 years. These individuals are also recommended to receive CRC screenings, but they should not be affected by private health insurance state mandates because they already had coverage through Medicare. Although Medicare-eligible individuals likely have higher screening rates to begin with, the DDD model compares changes in relative screening behavior across the younger and older groups over the study period and explicitly controls for differences in initial screening levels and separate trends over time.

The DDD estimates were produced using 6 separate logistic regression models, 1 for each of our 3 dependent variables (up to date, BST, and endoscopy) by gender, adjusting for individual and county-level controls as described earlier but omitting state HMO penetration, which is perfectly collinear with the state–year fixed effects. The regressions included interactions between the policy variable and the race and ethnicity measures, which provided the estimated change in screening rates in the minority group relative to the majority group under mandated coverage. The regressions included year, state, and age group fixed effects and all 2-way interactions between them to control for fixed differences between states and age groups across states, as well as separate nonlinear trends for each state and each age group.

Because previous studies^12–19 have suggested that differences in income explain a substantial portion of screening disparities, we estimated 4 additional logistic regressions—using the BST and endoscopic dependent variables for men and women separately—and replaced the interactions between the policy variable and racial/ethnic categories with interactions between the policy variable and 7 indicator variables representing categories of household income (using incomes of < $10 000 as the reference category). These regressions included controls for race and ethnicity and all the control variables and fixed effects we have described.

We obtained all estimates in Figures 1 and 2 by using the margins and marginsplot commands after logistic regression in version Stata 12.0.⁴² All models were estimated with heteroskedasticity robust standard errors clustered at the state level.⁴³

FIGURE 1—

Triple difference estimates of changes in screening rates under mandated coverage by race/ethnicity for (a) up-to-date men, (b) up-to-date women, (c) endoscopy in men, (d) endoscopy in women, (e) bowel screening test in men, and (f) bowel screening test in women: Behavioral Risk Factor Surveillance System, United States, 2002–2008.

Note. Bars depict changes in screening rates (as decimals) under mandated coverage on the basis of logit estimation of the triple difference model. Estimates were derived using the margins command in Stata version 12.0 for racial and ethnic group. Whiskers indicate 95% confidence intervals.

FIGURE 2—

Triple difference estimates of changes in screening rates by household income for (a) endoscopy in men, (b) endoscopy in women, (c) bowel screening test in men, and (d) bowel screening test in women: Behavioral Risk Factor Surveillance System, United States, 2002–2008.

Note. Points depict changes in screening rates (as decimals) under mandated coverage on the basis of logit estimation of the triple difference model. Estimates were derived using the margins command in Stata version 12.0 at each value of the categorical household income variable. Whiskers indicate 95% confidence intervals.

RESULTS

Table 1 contains summary statistics describing the characteristics of non-Hispanic White, Black, Asian, Native American, and Hispanic people aged 50 to 64 years with insurance coverage. Disparities were evident for all groups except Blacks. The proportions of Blacks who were up to date or who had a BST in the past year were both 3 percentage points higher than the proportions for non-Hispanic Whites. This may be because Blacks perceive a greater risk of CRC because of higher population mortality rates. All other patterns were as expected. Non-Hispanic Whites reported being in better health, had lower rates of diabetes and obesity, had higher household income, and lived in more affluent counties than Blacks, Native Americans, and Hispanics. Asians were more similar to non-Hispanic Whites.

TABLE 1—

Screening Rates and Characteristics of Insured People Aged 50–64 Years and Their Communities by Race/Ethnicity: Behavioral Risk Factor Surveillance System, United States, 2002–2008

Characteristic	Non-Hispanic White	Black	Asian	Native American	Hispanic
Proportion up to date	0.53	0.56	0.45	0.45	0.43
Proportion BST past year	0.16	0.19	0.13	0.15	0.12
Proportion endoscopy past 5 y	0.46	0.47	0.38	0.38	0.37
Proportion female	0.50	0.54	0.44	0.46	0.49
Age, y, mean	56.79	56.76	56.43	56.47	56.51
Median self-reported health status	Very good	Good	Good	Good	Good
Proportion previous smokers	0.53	0.52	0.32	0.62	0.49
Diabetes rate	0.10	0.23	0.14	0.21	0.20
Obesity rate	0.28	0.41	0.09	0.36	0.33
Mortality rate per 100 000 in 1990–1999, mean	22.94	29.86	14.05	16.07	21.31
Median household income category,a $	50 000–75 000	25 000–35 000	50 000–75 000	25 000–35 000	35 000–50 000
Median educational attainment	High school	High school	Some college	High school	High school
Proportion married	0.65	0.38	0.71	0.50	0.58
Regular doctor	0.93	0.93	0.94	0.86	0.87
Proportion living in HPSA county	0.02	0.02	0.00	0.05	0.01
No. of geographical HPSAs in county, mean	20.26	44.46	52.10	30.77	73.89
Average state HMO penetration, %	24.76	23.74	35.16	28.61	35.23
Average of county in poverty, %	11.94	14.91	11.74	13.57	14.47
Average county rural, %	19.43	11.28	5.79	20.19	7.43
No.	170 583	14 630	2 952	2 234	4 442

Note. BST = bowel screening test; HMO = health maintenance organization; HPSA = Health Professional Shortage Area. Averages produced using Behavioral Risk Factor Surveillance System sampling weights. Medians are reported to conserve space; analyses included categorical measures corresponding to 5 categories of self-reported health status, 8 categories of income, and 4 categories of educational attainment. Hispanic ethnicity and race are not mutually exclusive designations.

^aExcluding people who responded “don't know” and who refused to provide income information.

Decomposition of Racial and Ethnic Disparities

Table 2 illustrates how the differences in observable characteristics displayed in Table 1 contribute to screening disparities. For example the observed difference between non-Hispanic Whites’ and Hispanics’ screening rates was 10 percentage points, 8 of which were explained by observable characteristics. Observable characteristics explained a substantial fraction of the disparities between Whites and Hispanics and Native Americans, but little of those between Whites and Blacks and Asians.

TABLE 2—

Decomposition of Differences in Up-to-Date Colorectal Cancer Screening Rates: Behavioral Risk Factor Surveillance System, United States, 2002–2008

Decomposition	Black/Non-Hispanic Whitea	Asian/Non-Hispanic Whitea	Native American/ Non-Hispanic Whitea	Hispanic/Non-Hispanic Whitea
Overall
Total estimated difference	−0.028	0.076	0.080	0.099
Total explained by differences in observed characteristics	0.018	0.004	0.077	0.082
Differences attributable to
Female	0.000	0.000	0.000	0.000
Age	0.000	0.006**	0.005**	0.004**
Self-reported health status	−0.006**	0.002**	−0.008**	−0.008**
Previous smoker	−0.000*	−0.002*	0.001*	−0.000*
Diabetes	−0.003**	−0.001*	−0.002**	−0.002*
Obesity	−0.000	−0.000	0.000	0.000
CRC mortality rateb	−0.016	0.031	0.016	0.024
Income	0.023**	−0.001*	0.021**	0.024**
Education	0.013**	−0.012**	0.016**	0.024**
Marital status	0.006**	−0.002**	0.002**	0.002*
Regular doctor	−0.001**	−0.002**	0.015**	0.014**
Living in HPSA county	0.000	−0.000	0.000	−0.000
No. of geographical HPSAs in county	0.001	0.003	0.001	0.005
State HMO penetration	−0.001	0.012	0.004	0.011
% of county in poverty	0.004*	−0.000*	0.002*	0.003*
% of county rural	−0.005**	−0.008*	0.000**	−0.006**
State of residence	0.000	−0.011	−0.000	−0.012
Survey year	0.001**	0.001**	0.001**	−0.001**
Survey month	−0.000	−0.000	0.000	0.000
Total, no.	185 213	173 535	172 817	175 025
Minority, no.	14 603	2 952	2 234	4 442

Note. CRC = colorectal cancer; HMO = health maintenance organization; HPSA = Health Professional Shortage Area. Results based on Fairlie decomposition technique for nonlinear models with random ordering of covariates. H₀: Difference attributable to covariate is equal to 0. Hispanic ethnicity and race are not mutually exclusive designations. A full set of results including a description of the Fairlie method is available as a supplement to the online version of this article at http://www.ajph.org.

^aExcluding Hispanic Whites.

^bCRC mortality rates come from Surveillance, Epidemiology, and End Results data from deaths between 1990 and 1999 and vary by race, gender, and state.

*P < .05; **P < .01.

Negative estimates indicated differences in that characteristic that led to higher screening rates for the minority group relative to non-Hispanic Whites. For example, the difference in mortality rates between Blacks and Whites accounted for a 1.6-percentage-point higher screening rate among Blacks, although this estimate was not statistically significant. Positive estimates indicated that the factor contributed to the disparity. Individual income, educational attainment, health status, past smoking behavior, diabetes, marital status, having a regular doctor, county poverty rate, and percentage of the county that is rural were also statistically significant predictors of disparities. As in previous work,^12–19 individual income and education explained most of the observed disparity for all groups in this population except Asians. For example, lower income and educational attainment together accounted for 59% of the explained disparity.

State Mandate Triple-Difference Results

Figure 1 displays the results of the DDD analysis. There were several statistically significant estimates for men. Estimates implied that up-to-date screening rates among White and Hispanic men increased by 4 and 8 percentage points, respectively, but the difference between these estimates was not statistically significant. Estimates indicated that rates of endoscopic screening increased by 10 percentage points among Hispanic men under mandated coverage, and this gain was statistically significantly larger than the 3-percentage-point gain for non-Hispanic men (P < .01), which implied a reduction in the disparity. We found a statistically significant increase in men’s BST screening rates for all groups except Hispanics, but none of the increases in the minority groups were statistically significantly larger than the increase among White men.

For women, estimated changes in up-to-date and endoscopic screening were positive for all groups except Hispanics, but no changes were statistically significant from zero. However, we did find statistically significant increases in the up-to-date and endoscopic screening disparities between Hispanic and non-Hispanic women because rates for non-Hispanic women fell while those for Hispanic women rose. We found no statistically significant changes in BST screening rates among women.

Figure 2 displays the estimated changes in screening rates under mandated coverage at each level of household income by gender and by screening mode. Here we found an interesting pattern for both men and women. The largest, and in some cases the only statistically significant, increases in endoscopic screening rates were among households with incomes in the $10 000 to $15 000 range. Although we did see some positive and statistically significant changes in BST screening among lower income men as well, we did not find any among women.

DISCUSSION

We examined the determinants of disparities in CRC screening. We found that the distribution of SES characteristics across racial and ethnic groups explained a substantial proportion of observed disparities for all groups except Asians. This could be because average income and education are very similar among Whites and Asians or, as previous studies have suggested, there may be substantial heterogeneity within the Asian population and important cultural and language barriers for some subgroups.⁴⁴ We found several increases in screening associated with mandates for minority group men, but they were generally similar in size to increases in White men, and we could not conclude that disparities decreased. The sole exception was endoscopic screening among Hispanic men, for which we estimated a 7-percentage-point decrease in the disparity. This is a large change considering that the estimated disparity between Hispanics and non-Hispanic Whites was 9.9 percentage points. However, we found the up-to-date and endoscopic screening disparities increased by 4 and 5 percentage points, respectively. This finding is consistent with those of an earlier study that found that the Medicare expansion of coverage of CRC screening actually exacerbated disparities between non-Hispanic Whites and Hispanics.²¹ The differences in our findings by gender are consistent with findings of previous studies, but there is no clear explanation for them.³⁸ It is especially puzzling that we found differences by gender for the same ethnic group.

Mandates were, however, associated with statistically significant improvements in endoscopic screening rates among lower income individuals only. This result is in keeping with the possibility that health plan fragmentation may drive some of the observed relationship between income and screening. This finding also implies that mandates may indirectly improve minority screening rates because minority populations (except Asians) are overrepresented in poor households.

Although our results suggest that insurance mandates may reduce screening disparities associated with household income, we note that mandates do not eliminate all health plan fragmentation. State mandates do not set limits on cost sharing. The federal mandate in the ACA does, but coverage may remain fragmented. For example, if an endoscopic test includes a polypectomy, it may be billed as a diagnostic procedure and no longer subject to coverage under the ACA. The Obama administration issued guidance stating that the preventive care mandates included in the ACA should apply to endoscopic screenings that include a polypectomy; however, variance in coding of procedures and confusion over coverage of colonoscopies that follow positive BST tests may still lead to fragmentation in coverage.^45,46

Although the BRFSS had several advantages for this study, it did have some key limitations. First, we were unable to observe which individuals actually experienced a change in coverage when mandates went into effect. Thus, many in our treatment group may not actually have been treated. This could have biased our estimates toward zero, which may mean that the response to the policy among those who were actually affected is larger than our estimates. Also, we added county-level information to the data set using the county Federal Information Processing Standards codes available in the data files, but codes are not provided for the very smallest counties to protect anonymity, and thus, these results systematically exclude smaller communities. Finally, we were unable to examine more detailed race or ethnicity categorizations because of a lack of statistical power, and heterogeneity within these groups may also have contributed to some of the insignificant findings.

This discussion and our findings underscore the need to understand the problems that arise in the translation of evidence-based medical procedures to utilization. Universal insurance coverage may not achieve substantial reductions in disparities if fragmentation in coverage persists. Disparities research must move beyond a focus on insurance coverage as a binary construct and adopt methods that reflect the institutional complexity of health care transactions. In the absence of more detailed plan data, many natural experiments are arising through ACA implementation that offer opportunities for future research similar to the policy analysis in this study.