Medicaid Expansion Associated with Reduction in Geospatial Breast Cancer Stage at Diagnosis Disparities

Abstract

Context:

Prior studies demonstrate that Medicaid expansion has been associated with earlier stage breast cancer diagnosis among women with low-income, likely through increased access to cancer screening services. However, how this policy change has impacted geospatial disparities in breast cancer stage at diagnosis is unclear.

Objective:

To examine whether there were reductions in geospatial disparities in advanced stage breast cancer at diagnosis in Ohio after Medicaid expansion.

Design:

The study included 33 537 women aged 40-64 years diagnosed with invasive breast cancer from the Ohio Cancer Incidence Surveillance System between 2010 and 2017. The space-time scan statistic was used to detect clusters of advanced stage at diagnosis before and after Medicaid expansion. Block group variables from the Census were used to describe the contextual characteristics of detected clusters.

Results:

The percentage of local stage diagnosis among women with breast cancer increased from 60.2% in the pre-expansion period (2010-2013) to 62.6% in the post-expansion period (2014-2017), while the uninsured rate among those women decreased from 13.7% to 7.5% during the same period. Two statistically significant (p < .05) and six non-significant spatial clusters (p > .05) of advanced stage breast cancer cases were found in the pre-expansion period, while none were found in the post-expansion period. These clusters were in the four largest metropolitan areas in Ohio and individuals inside the clusters were more likely to be disadvantaged along numerous socioeconomic factors.

Conclusions:

Medicaid expansion has played an important role in reducing geospatial disparities in breast cancer stage at diagnosis, likely through the reduction of advanced stage disease among women living in socioeconomically disadvantaged communities.

Introduction

Breast cancer remains the second leading cause of cancer death in US women.¹ The Affordable Care Act’s Medicaid expansion was a highly anticipated policy development for breast cancer prevention and control because it provided insurance coverage to many uninsured and low-income women who have faced large barriers to routine medical care and cancer screening.² Initial studies have demonstrated that Medicaid expansion was associated with increased rates of early-stage breast cancer diagnosis among low-income women, likely through increased uptake of breast cancer screening.^3-6

Residential location appears to be a long-standing surrogate for access to care related social determinants of health, which in turn is associated with breast cancer stage at diagnosis. Prior studies have used geospatial methods to detect clusters of advanced-stage breast cancer at diagnosis^7-13 and linked these clusters to higher levels of neighborhood deprivation.^14,15 However, these studies did not account for health policy interventions, such as the Affordable Care Act, on advanced-stage diagnosis. Hence, revisiting these location-based disparities is important, especially in the setting of an improved health insurance landscape. In the Medicaid post-expansion period, such a landscape would translate into greater access to cancer screening services, and therefore earlier stage at diagnosis.

Thus, we examined whether there were reductions in geospatial disparities in advanced-stage breast cancer at diagnosis in Ohio before and after Medicaid expansion. Ohio expanded their Medicaid program in January 2014 to include nonelderly adults with incomes at or below 138% of the federal poverty level. We hypothesized that Medicaid expansion would be associated with reduced geospatial disparities in advanced stage at diagnosis. We also hypothesized that these reductions would be concentrated in the most socioeconomically deprived communities, since these communities traditionally had the highest uninsured rates, and thus the highest barriers to accessing care and cancer screenings. If present, these improvements could translate into improved survival among the most vulnerable populations since cancer stage at diagnosis is one of the most important prognostic factors.

Methods

Overview

To determine whether gains in insurance coverage have translated into geospatial improvements in breast cancer stage at diagnosis, we compared the presence, size, and significance of advanced-stage clusters before and after Medicaid expansion. Cases diagnosed during 2010-2013 were classified into the pre-expansion period, while those from 2014-2017 were classified into the post-expansion period. Specifically, we explored whether pre-expansion space-time clusters of advanced-stage diagnosis were reduced in the post-expansion period. To examine possible spatiotemporal trends in cancer stage at diagnosis independent of Medicaid expansion, we conducted a sensitivity analysis on women who had private or other insurance at the time of diagnosis. These women were selected because they were unlikely to be affected by Medicaid expansion.

This study was approved by the Institutional Review Board of each of Case Western Reserve University (IRB# 2016-1752) and the Ohio Department of Health (IRB# 2017-50).

Data sources

We identified our breast cancer cases from the Ohio Cancer Incidence Surveillance System (OCISS), a population-based incidence registry of all cases of cancer diagnosed in Ohio residents.¹⁶ All providers or hospitals, by law, must report incident cases within 6 months of date of diagnosis and/or first contact with the facility. The OCISS data fields utilized in the study included patient’s cancer site, stage, year of diagnosis, age, race/ethnicity, as well as insurance status and geocoded address of residence at the time of diagnosis. We obtained contextual characteristics at the block group level from the US Census - American Community Survey 5-year Estimates, 2009-2013 for pre-expansion and 2014-2018 for post-expansion periods.^17,18

Study population

Our study population included women aged 40-64 years who were first diagnosed with invasive breast cancer in Ohio during 2010-2017. Patients with unstaged/unknown-stage cancers or without accurate residential location were excluded. We did not include women younger than 40 years of age because most guidelines do not recommend regular screening for this population assuming they have an average risk for breast cancer,^19-21 and 40 is also the age when Ohio Medicaid begins covering regular mammograms.²² Those 65 years of age and older were also excluded because they were likely enrolled in Medicare and thus unlikely to have been impacted by Medicaid expansion.

Variables of interest

The outcome of interest was stage at the time of diagnosis based on the SEER (Surveillance, Epidemiology, and End Results) Summary Stage variable from the OCISS, which classifies tumors into three ordinal categories: local stage, regional stage, and distant stage. Specifically, the three stages correspond to an order of 1, 2, and 3, where increasing numbers represent more advanced disease. This arrangement of the outcome variable was designed to accommodate the ordinal-based space-time scan statistic discussed in the next section.

Individual-level characteristics included age at diagnosis (40-49 and 50-64 years), race/ethnicity (Non-Hispanic Black and All Other), marital status (married/partnered and single), and insurance status (i.e., the primary payer at diagnosis variable classified as uninsured, Medicaid, and private/other insurance). We note that 97% of our study population in the “All Other” category was Non-Hispanic White, and the remaining 3% was Hispanic, American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific Islander.

Contextual characteristics at the block group level from 5-year ACS data included insurance status (percent with no insurance, Medicaid, and private/other insurance), median household income (measured in inflation-adjusted dollars), educational attainment (percent of women over 25 years of age with high school degree or higher), household type by tenure (percent of renter occupied housing), and household vehicle availability (percent of households with no vehicle). We accounted for insurance status variable at the contextual level as a complement to the individual-level variables because the number of individuals with no insurance or with Medicaid in some clusters may be too small to show. In addition, as a marker of resource deprivation, we included the Area Deprivation Index (ADI) with the fourth quartile of ADI representing the most deprived areas.²³ The ADI is a block group-level indicator that includes factors such as income, education, employment, and housing quality.

The space-time scan statistic

The analyses were performed using the SaTScan statistical software²⁴ to identify geographic areas with high proportions of women diagnosed with advanced-stage breast cancer before and after Medicaid expansion. We used individuals’ coordinates geocoded from their residential addresses at the time of diagnosis. The space-time scan statistic in SaTScan was used to test whether individuals with advanced-stage diseases were randomly distributed or clustered in space and time.²⁵ To preserve more information on stage at diagnosis, we used an ordinal-based space-time scan statistic with all three stage categories included in the algorithm.²⁶ The algorithm first compares the distribution of cases among all three stages in each location against the expected distribution in the overall study area. If no cluster was detected, the algorithm would create a new outcome variable by combining two adjacent stages together (i.e., [local + regional] vs distant, or local vs [regional + distant]). Then it would repeat the comparison with a binary outcome against the expected distribution. The maximum number of individuals found in a detected cluster, regardless of their stage at diagnosis, was set to 50 percent of the overall study population,²⁷ which means that the size of the scanning window for cluster detection was variable up to a maximum of 50 percent of all cases in the study area. To optimize clustering information, clusters were allowed to overlap with the limitation that the centroid of any cluster cannot be contained in another reported cluster.

The space-time statistic was used because it makes no assumptions about whether and where a cluster exists, identifies a cluster at any location at any time of any size up to a maximum size, and minimizes the problem of multiple statistical tests.²⁵ The space-time scan statistic is defined by a cylindrical window with a circular geographic base and with height corresponding to time. By gradually moving the window across space and time, the algorithm calculates the number of observed and expected observations inside and outside the window for each location for the space-time model. Based on these numbers, the statistical likelihood for each window is calculated.²⁸ The most likely cluster is then the area and time with the maximum likelihood and with more than its expected number of cases. The p-value indicating the significance of the test statistic is evaluated using Monte Carlo hypothesis testing, with the number of Monte Carlo replications set at 999.

SaTScan can evaluate different cluster locations and sizes without restrictions imposed by administrative geographical boundaries, minimizing assumptions about the geographic cluster location and size²⁷ and reducing the effect of the modifiable areal unit problem (MAUP), a source of bias in many geographic studies.²⁹ To protect patient confidentiality, any cluster with a number smaller than 11 in any of the stage categories (i.e., local, regional, or distant stages) was either removed or the number was combined with its neighboring stage category of the cluster if the number of cases in the combined stage category (i.e., [local + regional] or [regional + distant]) was greater or equal to 11.

Results

Overview

There were 35 478 newly diagnosed invasive breast cancer cases among women aged 40-64 years in Ohio during the years 2010-2017. We excluded patients with unstaged/unknown-stage cancers (n=553) and those without an accurate residential location (n=1388), resulting in a final study population of 33 537 women. We note that the percentage of unstaged/unknown-stage cases decreased from 1.8% before the expansion to 1.5% after the expansion. More information about the trend and the exclusion of unstaged/unknown-stage cases is included in Supplemental Digital Content Table 1.

Among the study population, 20 608 (61.4%), 10 916 (32.5%), and 2 013 (6.0%) were diagnosed with local, regional, and distant stage disease, respectively. In the pre-expansion period (2010-2013), the percentage of cases diagnosed with local stage disease was 60.2% (95% Confidence Interval: 59.5 – 61.0%) compared to 62.6% (95% Confidence Interval: 61.9 – 63.4%) in the post-expansion period (2014-2017), while the uninsured rate among those women decreased from 13.7% (95% Confidence Interval: 13.2 – 14.2%) to 7.5% (95% Confidence Interval: 7.1 – 7.9%) during the same period. Starting in 2014, the year in which Medicaid expansion was implemented, we observed a steady and significant increase in the percentage of local stage cases, and a decrease in regional stage cases (P < 0.001), while the percentage of cases diagnosed with distant-stage disease remained relatively stable (Figure 1). We also included breast cancer incidence by stage-at-diagnosis for Ohio women aged 40-64 years during 2010-2017 in Supplemental Digital Content Figure 1.

Figure 1.

Temporal trends of percentage of breast cancer cases by stage at diagnosis in Ohio

Space-time clusters

There were eight space-time clusters detected during the years 2010-2017, all of which were in the pre-expansion period (none in the post-expansion period). These clusters were located in the five largest metropolitan areas in Ohio (Figure 2). As described above, the ordinal-based algorithm compares the distribution of all three cancer stages in each location to the distribution in the overall study area. If no cluster is detected, the algorithm then combines two adjacent stages and repeats the comparison. As a result, clusters 1, 2, 5, 6, and 8 were based on an ordinal outcome of local vs regional vs distant, cluster 3 was based on a binary outcome of [local + regional] vs distant, and clusters 4 and 7 were based on a binary outcome of local vs [regional + distant]. Because presenting small cell counts can constitute a risk to patient privacy, we combined regional and distant stages in the tables regardless of which outcome the clusters were based on (Table 1 and Supplemental Digital Content Table 2).

Figure 2.

Space-time clusters of more-advanced-stage breast cancer in the pre-expansion period in Ohio (no cluster found in the post-expansion period)

Table 1.

Individual and block group level characteristics of clusters 1 and 2 of more-advanced-stage breast cancer in Ohio.

	Cluster 1		Cluster 2		Ohio
P-value	0.009		0.03
Period	Pre-E	Post-E	Pre-E	Post-E	Pre-E	Post-E
Number of patients	143	108	56	62	16,433	17,104
Individual characteristics of study subjects
Stage at diagnosis (%)
Local	57 (39.9)	65 (60.2)	14 (25.0)	33 (53.2)	9,896 (60.2)	10,712 (62.6)
Regional or Distanta	86 (60.1)	43 (39.8)	42 (75.0)	29 (46.8)	6,537 (39.8)	6,392 (37.4)
Median age at diagnosis	55	58	53.5	55.5	55	56
Race/ethnicity (%)
Non-Hispanic Black	62 (43.4)	44 (40.7)	29 (51.8)	36 (58.1)	1,954 (11.9)	2,010 (11.8)
All Other	81 (56.6)	64 (59.3)	27 (48.2)	26 (41.9)	14,479 (88.1)	15,094 (88.2)
Marital status (%)
Married/Partnered	45 (31.5)	43 (39.8)	26 (46.4)	23 (37.1)	10,210 (62.1)	10,577 (61.8)
Single	98 (68.5)	65 (60.2)	30 (53.6)	39 (62.9)	6,223 (37.9)	6,527 (38.2)
Medicaid enrollment (%)	35 (24.5)	32 (29.6)	<11 (<19.6)	14 (22.6)	1484 (9.0)	2104 (12.3)
Block group characteristics of study subjects b
Percent of population by type of insurance (%)
Not insured	26.1	11.7	15.5	9.5	13.7	7.5
Medicaid	22.6	37.3	8.3	17.2	7.5	13.2
Privately/other insured	51.4	51.0	76.2	73.3	78.8	79.3
Median household income (US dollar)	24.0K	26.5K	42.2K	43.3K	45.8K	51.7K
Percent women 25 years and older with high school degree	78.5	79.9	88.2	89.9	88.9	90.6
Percent renter-occupied housing	52.5	55.6	42.4	41.6	32.5	34.0
Percent households with no vehicle	23.3	22.9	13.9	11.9	8.3	8.2
Percent people living in c
Most deprived area (4th quartile)	84.5	87.4	22.5	27.5	25.0	25.0
Less deprived area (1-3rd quartile)	15.5	12.6	77.5	72.5	75.0	75.0

Abbreviations: Pre-E, Pre-expansion; Post-E, Post-expansion.

^aRegional and distant stages were combined due to a small number of distant-stage cases in some clusters.

^bBlock groups overlapped with the corresponding cluster and contained at least one patient from the cluster.

^cMeasured by area deprivation index.

Notes: All space-time clusters were detected in the pre-expansion period. Patients under the Post-E columns were within the geographic boundary of the corresponding pre-expansion cluster.

All clusters had higher percentages of regional and distant stage cases combined compared to that of Ohio statewide, pre- or post-expansion. Specifically, clusters 1, 2, 4, and 5 had more than 20% fewer local stage cases (39.9%, 25.0%, 39.5%, and 37.3%) than overall Ohio pre-expansion (60.2%) or post-expansion (62.6%). Meanwhile, the percentages of distant stage cases in all clusters (from 10.4% to 17.3%) were more than one and half times that of overall Ohio either pre- or post-expansion (6.0%). Also, the percentages of regional stage cases for all clusters (from 38.1% to 46.8%) were higher than overall Ohio pre-expansion (33.8%) or post-expansion (31.4%).

Among all clusters, clusters 1 and 2 were statistically significant (P=0.009 and 0.03, respectively) based on Monte Carlo simulations. Cluster 1 (n=143) covered the entire downtown Toledo and several suburbs, and cluster 2 (n=56) covered a northern suburb of Cincinnati. Notably, despite its small size, the percentage of advanced stage cases in cluster 2 prior to expansion represented the highest level among all clusters – almost double the statewide percentage: 75.0% vs 39.8%. With expansion, we observed a sharp decline in the proportion of advanced stage cases in this cluster: from 75% to 46.8%.

The individual-level characteristics of women with breast cancer located within both clusters showed that they were more likely to be Non-Hispanic Black and not married/partnered compared to Ohio pre- or post-expansion. Cluster 1 had more women with no insurance coverage or on Medicaid compared to Ohio. The distribution of insurance status for cluster 2 was suppressed due to insufficient number of patients in the cluster. We also observed similar results for clusters 3-8 (see Supplemental Digital Content Table 2).

Contextual characteristics of clusters

We also explored the characteristics of the population living in cluster 1 (Toledo) and cluster 2 (a suburb of Cincinnati) in both pre- and post-expansion periods (Table 1). For people living in cluster 1, the majority also lived in the most deprived areas, i.e., with an ADI falling in the highest quartile (84.5% and 87.4% for pre- and post-expansion). The median household income measured as US dollars for people living in cluster 1 was about half of that for Ohio. Compared to the general population of Ohio, fewer women in cluster 1 had at least a high school degree. The percentages of the population for renter-occupied housing (versus owner-occupied) and household vehicle availability in cluster 1 were also less favorable compared to Ohio overall. Cluster 2 had a similar profile along the dimensions of area deprivation, median household income, and education attainment, while the percentages of renter-occupied housing, and household vehicle availability were significantly higher than those for Ohio overall. The biggest change among all contextual characteristics between pre-expansion and post-expansion was in insurance status. We observed great reductions of people with no insurance for cluster 1 (from 26.1% to 11.7%), cluster 2 (from 15.5% to 9.5%), and Ohio (from 13.7% to 7.5%). These changes in the uninsured rate were likely due to Medicaid expansion since the percentage of people with private/other insurance stayed almost the same through the study period. Those with Medicaid increased by almost 15 percentage points in cluster 1 while the percentage of people enrolled in Medicaid more than doubled in cluster 2 between the pre- and post-expansion periods. We also observed similar results for clusters 3-8 (see Supplemental Digital Content Table 2).

Sensitivity analysis

The sensitivity analysis was designed to rule out the possibility that the spatiotemporal clusters we observed in the main analysis was a result of an existing spatiotemporal trend rather than the effect of Medicaid expansion. Thus, the sensitivity analysis focused on women with private or other insurance (including Medicare, Military, and other government-sponsored programs) at the time of diagnosis, a subgroup that was relatively unaffected by Medicaid expansion. We used the same ordinal-based space-time scan statistic to detect clusters of advanced stage at diagnosis.

The results showed that among the five detected clusters, none was statistically significant at P < 0.05 (see Supplemental Digital Content Figure 2). All five clusters were located in the areas of Columbus, Cleveland, and Toledo. Four of them were detected pre-expansion and one was detected post-expansion. The one detected post-expansion was the largest among the five and was in the southwest of the city of Cleveland, partially overlapping with cluster 3 in the main analysis. There was no cluster found in or around Cincinnati where cluster 2 in the main analysis was identified.

Discussion

Our study identified space-time clusters with significantly higher percentages of more-advanced-stage disease pre-expansion that were no longer identified as clusters post-expansion. These findings suggest that in the absence of Medicaid expansion, women living in these pre-expansion clusters would have likely experienced higher rates of advanced stage breast cancer diagnosis into the post-expansion period. The contextual characteristics of the clusters suggest that these women were more likely to be disadvantaged along numerous dimensions of social determinants of health. Additionally, these clusters experienced the largest improvements in the uninsured rate in the post-expansion period. Collectively, the evidence suggests that Medicaid expansion is associated with a greater improvement in reducing breast cancer advanced stage at diagnosis in low-SES urban and suburban areas of Toledo and Cincinnati, and potentially in Columbus and Cleveland. Our study adds new evidence that Medicaid expansion has likely played an important role in reducing long-standing place-based disparities in breast cancer stage at diagnosis. These improvements were mostly due to the reduced regional-stage cases and the increased local-stage cases (Figure 1). The fact that the proportion of distant stage cases remained essentially unchanged may suggest that there are a subset of the most aggressive breast cancer cases which are less amenable to screening. Impacting this subset of cases through improved healthcare access may prove difficult. Observation across broader populations and longer timeframes will provide further insight into this possible phenomenon.

To our knowledge, this is the first study that accounts for both the geographic variation of advanced stage breast cancer and the effect of Medicaid expansion. While these findings indicate space-time differences in advanced stage at diagnosis for breast cancer, it’s unclear if the same relationships will hold for breast cancer mortality. An important topic for future research is the extent to which stage-specific survival is improved through better access to treatment and survivorship care. These are benefits which may ultimately accrue to even patients diagnosed prior to Medicaid expansion.

Our study includes several methodological strengths as well. First, in contrast to previous studies that aggregated individual point locations to area units for cluster detection, ^7,9-11,15 we used individual residential locations directly for cluster detection, which improved the accuracy of the results and potentially reduced the bias caused by MAUP. Furthermore, previous studies have used the Bernoulli-based model which adopted a binary classification of the stage variable grouping regional stage and distant stage together into advanced stage^7,11 or the Poisson model which regarded distant-stage cases as incidence and all-stage cases as population.^9,10 Both models resulted in a loss of granularity of the outcome variable by accounting for two stage categories only. To leverage the data on cancer stage, we used an ordinal-based model incorporating all three cancer stages, allowing for greater granularity in our study. In addition, we conducted a sensitivity analysis focusing on those with private/other insurance, which was designed to examine existing spatiotemporal trends that could explain the observed results. Unlike in in the main analysis, there were no significant changes in the presence of cluster before and after Medicaid expansion. This is an expected result since those with private/other insurance were unlikely to be affected by Medicaid expansion. Thus, the association between Medicaid expansion and the improvements in the geospatial disparities of breast cancer stage-at-diagnosis from the main study was likely to be independent of an existing spatiotemporal trend.

A limitation of our study is that the space-time scan statistic may have limited power to detect irregularly shaped clusters using a circular scanning window. This problem is particularly acute in rural areas. On the one hand, a larger circular window may cover an irregularly shaped true cluster in a rural area but also cover its surrounding urban areas with substantially more cases. On the other hand, a smaller circular window may only cover part of the true cluster, which may reduce the likelihood of a cluster being reported within the current window. Both situations could result in the loss of statistical power in cluster detection. Researchers have been developing methods in detecting irregularly shaped clusters,^30-34 but these methods have not been extended to consider the temporal dimension of cluster detection. We have partially overcome this issue by allowing overlapping clusters to present in the condition that they do not contain the center of one another. However, by not being able to fully detect these irregularly shaped clusters, our study may underestimate both the magnitude of prevailing place-based disparities in breast cancer stage-at-diagnosis along with the impact that Medicaid expansion has had on reducing them.

In conclusion, our study suggests that Medicaid expansion has played an important role in reducing place-based disparities in breast cancer stage at diagnosis, especially in socioeconomically disadvantaged communities. As new policy interventions are implemented, spatiotemporal methods, as the ones utilized in this study, can offer a nuanced evaluation on how effectively these interventions reduce place-based disparities.

Implications for Policy & Practice.

• Spatial clusters of advanced stage breast cancer diagnosis detected prior to Medicaid expansion in Ohio were not seen after the expansion.

• Health policy interventions that improve access to routine health care and cancer screenings can reduce geospatial disparities in cancer outcome.

• The ACA’s Medicaid expansion is associated with improvements in geographic disparities in breast cancer stage-at-diagnosis, which tended to impact women living in socioeconomically disadvantaged neighborhoods.

• Disadvantage tends to concentrate geographically. According to the current study, so too did the occurrence of advanced stage breast cancer prior to Medicaid expansion.

• In the case of breast cancer stage at diagnosis, our analysis suggests that the benefits of Medicaid expansion accrue effectively to disadvantaged communities.