|Home | About | Journals | Submit | Contact Us | Français|
To compare outcomes between women enrolling in Medicaid after being diagnosed with breast cancer and those referred to Medicaid through the Ohio Breast and Cervical Cancer Early Detection Program (BCCEDP).
Using linked data from the 2002 to 2008 Ohio Cancer Incidence Surveillance System, Medicaid, the BCCEDP database, and Ohio death certificates (through 2010), we identified women 40 to 64 years of age diagnosed with incident invasive breast cancer during the study years and enrolled in Medicaid 3 months before or after cancer diagnosis. We compared the following outcomes across BCCEDP one-time and repeat participants and nonparticipants: (1) cancer stage at diagnosis, (2) treatment delays, (3) receipt of standard treatment, and (4) survival. We conducted multivariable logistic regression and survival analysis to examine the association between BCCEDP participation and the outcomes of interest, controlling for potential confounders.
We identified 427 and 654 BCCEDP participants and nonparticipants, respectively; 28.5% of BCCEDP women were repeat participants. Compared with nonparticipants, BCCEDP one-time and repeat participants were significantly less likely to be diagnosed with advanced-stage cancer (one-time: adjusted odds ratio [AOR], 0.64; 95% CI, 0.49 to 0.85; repeat: AOR, 0.34; 95% CI, 0.23 to 0.52), or experience delays in treatment initiation (one-time: adjusted hazard ratio [AHR], 1.29; 95% CI, 1.09 to 1.51; repeat: AHR, 1.38; 95% CI, 1.11 to 1.72). In addition, although we observed no difference in receipt of standard cancer treatment, BCCEDP participants experienced cancer-specific and overall survival benefits.
Compared with nonparticipants, BCCEDP participants experienced earlier breast cancer stage at diagnosis, shorter time to treatment initiation, and survival benefits.
Breast cancer is the most common cancer in women. In 2014, more than 230,000 women were expected to have been diagnosed with incident breast cancer, and 40,000 were expected to die from the disease.1 Although breast cancer is detectable by screening, only 54% of US women 40 years of age or older received mammogram in the past 12 months, according to the 2011 National Health Interview Survey.2 Reasons for the lack of screening include access, knowledge, worry or fear, and cost.3 To primarily address the access and cost barriers, a number of public health programs have been initiated to improve the rates of screening mammography among low-income and uninsured women. One such effort is the Centers for Disease Control and Prevention's Breast and Cervical Cancer Early Detection program (BCCEDP).4
The Ohio BCCEDP provides services to women who are uninsured and live in households with incomes of 200% or less than the Federal Poverty Level. BCCEDP services encompass screening and diagnostic services, including clinical breast examinations for women 40 to 64 years of age, and mammograms for those 50 to 64 years of age. It also provides coverage for diagnostic mammogram in women 40 to 50 years of age. In accordance with the Breast and Cervical Cancer Prevention and Treatment Act enacted in 2000,5 BCCEDP participants diagnosed with cancer are eligible to receive comprehensive health care coverage through Medicaid.
A previous study compared cancer outcomes between Medicaid beneficiaries diagnosed with breast cancer, regardless of when they enrolled in Medicaid, and BCCEDP participants who, by definition, enroll in Medicaid after being diagnosed with cancer.6 Although informative, this study may be biased, given that it included women who had been insured through Medicaid before having been diagnosed with cancer. Because BCCEDP participants enroll in Medicaid only on being diagnosed with cancer, it is imperative that the comparison group in Medicaid also be composed of women who enroll in Medicaid around the time of cancer diagnosis, and who were most likely uninsured or underinsured at the time of diagnosis. Indeed, past studies have documented a great level of heterogeneity within the Medicaid population, by showing significant differences in cancer-related outcomes between Medicaid-insured individuals (ie, those enrolled in Medicaid for some time before cancer diagnosis) and those who use Medicaid as a safety net program (ie, individuals who enroll in Medicaid on being diagnosed with cancer).7,8
The aim of this study is, therefore, to evaluate the benefits of the BCCEDP by evaluating outcomes in patients with breast cancer, comparing BCCEDP participants and non–BCCEDP participants who enrolled in Medicaid on being diagnosed with breast cancer. Furthermore, we distinguish between one-time and repeat BCCEDP participants, as a proxy for the extent to which women value and engage in the cancer prevention efforts of BCCEDP.
We tested the above hypotheses using a unique database, which we created by linking Ohio BCCEDP data with data from the Ohio Cancer Incidence Surveillance System (OCISS), Ohio Medicaid enrollment and claims files, and death certificates.
This study was approved by the institutional review boards of Case Western Reserve University and the Ohio Department of Health, which manages the BCCEDP and OCISS programs. The study was also approved by the Ohio Department of Medicaid.
Patient demographic variables gathered at enrollment are documented in this database as are variables for each patient encounter with the BCCEDP. Thus, it was possible for us to identify one-time versus repeat BCCEDP participants.
The OCISS program is intended to capture data on all incident cases of cancer diagnosed among residents of the state of Ohio; its completeness for female breast cancer was estimated to average at nearly 90% during the study period.
In addition to patient identifiers, OCISS records include sociodemographic variables (age, race, marital status), county of residence, date of cancer diagnosis, and tumor stage. Although OCISS data include tumor size, number of affected lymph nodes, and metastatic status of the disease, we used the SEER summary stage in our analysis, given the high proportion of records with missing data for the aforementioned variables.
The OCISS also includes the patient's address at diagnosis, which we geocoded to characterize the socioeconomic status (SES) in their area of residence, at the census tract level. In addition, we used the patients' census tract to determine whether they resided in medically underserved areas (MUAs), or whether they were part of medically underserved populations (MUP), based on the classification by the Health Resource and Service Administration and the Ohio Department of Health.
The enrollment file includes records for each individual enrolled in the Medicaid program. In addition to patient identifiers and demographic variables, it includes enrollment spans, which we used to construct beneficiaries' enrollment history in Medicaid relative to their date of cancer diagnosis (retrieved from the OCISS). The enrollment file also includes the beneficiaries' eligibility category, including their BCCEDP status, indicating that a woman had enrolled in Medicaid as a result of her referral through the BCCEDP.
The claims files include records for all services received by Medicaid beneficiaries (inpatient, outpatient institutional and noninstitutional, and pharmacy data). Claim records carry dates of service, as well as diagnostic and procedure codes, making it possible to identify cancer treatment patterns, including the timing of treatment initiation.
Ohio death certificate files include a record for almost all decedents who were residents of the state of Ohio, including residents who died out of state. In addition to patient identifiers and demographic variables, death certificates iclude the date and cause of death, which allowed us to examine all-cause and cancer-specific mortality for our study population.
We linked the above data sources by using a multistep, deterministic algorithm based on patient Social Security number, first and last name, and date of birth (month and year), as described in previous studies.9–11
Our study population included women 40 to 64 years of age who were residents of Ohio, diagnosed with incident invasive breast cancer during the years 2002 to 2008, and who first enrolled in Medicaid in the 3 months before or after the date of cancer diagnosis, consistent with previous studies.7,9,12 Women with unstaged or unknown-stage cancer were excluded from the study (n = 40; 12 (2.3%) in the BCCEDP group and 28 (4.4%) in the non-BCCEDP group; P = .058). Furthermore, analysis pertaining to treatment and survival outcomes was limited to women diagnosed with local- or regional-stage cancer.
These included the following:
Our main independent variable was BCCEDP participation status (yes/no) originating from the Medicaid enrollment file, indicating whether a woman was enrolled in Medicaid as a result of having been diagnosed with breast cancer through the BCCEDP. Non–BCCEDP participants are those who first enrolled in Medicaid around the time of breast cancer diagnosis, but not through the BCCEDP program.
As a proxy for women's level of compliance with cancer screening, and for the extent to which they are aware of the benefits of screening mammography, we also accounted for women's patterns of participation in the BCCEDP, thus differentiating between one-time and repeat participants, based on data in the BCCEDP database.
Other independent variables included sociodemographic variables and county of residence: age at the time of cancer diagnosis (40 to 49, 50 to 59, 60 to 64), race (African American v white and all others), and marital status (married v nonmarried), which were retrieved from the OCISS. Comorbid conditions as defined by Elixhauser et al13 were identified from claims for services incurred during the 3 months preceding cancer diagnosis. County of residence, also at the time of diagnosis, was categorized as metropolitan/suburban, Appalachian, and rural. We also identified patients according to whether they resided in an MUA or were part of an MUP, as noted above. We used socioeconomic measures at the census tract level. These included the percentage of the population with incomes below the Federal Poverty Level, percentage of adults with no high school diploma, percentage of female-headed households, and percentage of unemployed. After calculating the z-score for each of these measures (subtracting the mean from the raw score, and dividing by the standard deviation), we summed them to obtain a composite measure for socioeconomic indicators, with a higher score reflecting greater socioeconomic disadvantage. The composite measure was then used in the analysis in quartiles.
In addition to descriptive analysis, we developed multivariable regression models (logistic regression models to analyze binary variables, and survival models for time-to-event analysis) to study the association between BCCEDP status and the outcomes of interest after adjusting for patient covariates. In our regression models, we first compared outcomes separately for one-time and repeat participants versus nonparticipants, and then compared outcomes between one-time and repeat participants. In separate analyses, we also compared outcomes between participants and nonparticipants; the findings from the latter analysis are summarized in Figure 1.
In our regression models, we included all of the variables listed in Table 1, whether they remained statistically significant or not after adjusting for the other variables, given that they were important in how we conceptualized our research questions. Exceptions were variables indicated by dashes in the table, such as receipt of standard treatment, which could not have been used to analyze cancer stage at diagnosis. Given the nonrandomized nature of this study, we tested the propensity score approach to adjust for potential selection bias. The models with and without propensity scores yielded comparable findings. However, given the absence of pertinent variables in the period preceding cancer diagnosis (eg, screening status), our propensity scores were not considered in capturing the differences between the two groups. We used SAS version 9.3 (SAS Institute, Cary, NC) in all of our analyses.
Our study population included 427 BCCEDP participants and 654 nonparticipants. Repeat users constituted 28.5% of participants. Table 1 presents the distribution of the participant and nonparticipants groups by the variables of interest. Compared with nonparticipants, the participant group included fewer African Americans (17.33% v 23.24%), but more married women (33.26% v 25.38%). In addition, a greater percentage of participants than nonparticipants resided in Appalachian counties (24.36% v 18.50%). There were no marked differences between the two groups by socioeconomic measures, or by residence in MUA/MUP areas.
Comparing the percentage of women who presented with comorbid conditions between the study groups, more nonparticipants than participants presented with comorbid conditions (34.40% v 26.93%; P < .05). Similarly, among BCCEDP participants, more one-time than repeat users presented with comorbid conditions (30.49% v 18.03%; P < .01).
With regard to outcomes, we note a significantly higher occurrence of advanced-stage breast cancer in nonparticipants than in participants (63.31% v 48.24%). Furthermore, among participants, a significantly higher percentage of one-time than repeat participants were diagnosed with advanced-stage disease (52.46% v 37.70%). Repeat users had the shortest time to treatment (median, 27 days; 95% CI, 19 to 34 days), followed by one-time participants and nonparticipants (one-time: median, 32 days; 95% CI, 28 to 38 days; nonparticipants: median, 38 days; 95% CI, 33 to 44 days). We observed no significant differences in receipt of standard treatment across the comparison groups.
The Kaplan-Meier curves for overall and cancer-specific survival (Figures 2A and and2B)2B) indicate a survival advantage among BCCEDP participants, with repeat participants having the best survival outcomes, followed by one-time participants and nonparticipants. Table 2 presents the results from the multivariable regression analysis, first comparing outcomes for one-time and repeat BCCEDP participants with nonparticipants, and next comparing outcomes among participants. Figure 1 presents odds ratios and hazard ratios obtained from multivariable models using BCCEDP as a binary variable, with the reference group being nonparticipants. The following summarizes the results for each of the outcomes of interest:
Both one-time and repeat BCCEDP participants were significantly less likely than nonparticipants to present with advanced-stage disease (one-time: adjusted odds ratio [AOR], 0.64; 95% CI, 0.49 to 0.85; repeat: AOR, 0.34; 95% CI, 0.23 to 0.52). In addition, compared with one-time participants, repeat participants were nearly half as likely to be diagnosed with advanced-stage disease (AOR, 0.54; 95% CI, 0.35 to 0.83).
BCCEDP participants experienced fewer delays than their non-BCCEDP counterparts in treatment initiation (one-time: adjusted hazard ratio [AHR], 1.29; 95% CI, 1.09 to 1.51; repeat: AHR, 1.38; 95% CI, 1.11 to 1.72). However, there was no difference between one-time and repeat participants.
Nonparticipants were as likely as BCCEDP one-time or repeat participants to receive standard treatment.
Compared with nonparticipants, both one-time and repeat BCCEDP participants experienced significant overall survival benefits (one-time: AHR, 0.48; 95% CI, 0.32 to 0.73; repeat: AHR, 0.20; 95% CI, 0.08 to 0.49).
Both one-time and repeat BCCEDP participants had significant survival advantage compared with nonparticipants (one-time: AHR, 0.62; 95% CI, 0.39 to 0.98; repeat: AHR, 0.20; 95% CI, 0.06 to 0.64). However, survival was similar between one-time and repeat participants.
This study documents significant differences in cancer-related outcomes between one-time and repeat BCCEDP participants and a comparison group of nonparticipants on Medicaid. Our findings indicate that compared with nonparticipants, and after adjusting for potential confounders, BCCEDP participants experienced favorable outcomes with respect to all of the study variables, with the exception of receipt of adequate cancer treatment. Thus, in addition to treatment delays, there may be other factors, unaccounted for in this study, that may have contributed to poorer survival outcomes in non–BCCEDP participants.
Studies examining outcomes associated with the BCCEDP have generally yielded favorable results with regard to cancer stage at diagnosis,14 timeliness of care,15–18 and treatment patterns.6,19 In contrast, an earlier study reporting unfavorable outcomes in BCCEDP participants indicated that 71% of women who received mammography through the BCCEDP were symptomatic, and more than 50% of the women were diagnosed with late-stage disease.20 The favorable survival outcomes among participants are also in line with those of prior reports showing a reduction in breast cancer mortality associated with participation in the BCCEDP program.21,22
The new contribution of our study is the comparison of outcomes between BCCEDP participants and women who are similarly disadvantaged, thus evaluating the “net effects” of the BCCEDP on cancer-related and survival outcomes. Because BCCEDP participants enroll in Medicaid on being diagnosed with cancer, we defined the comparison group as those women who enrolled in Medicaid on cancer diagnosis, or in the 3-month window before or after the date of cancer diagnosis. Indeed, by having a comparison group of women, who, like BCCEDP participants, enrolled in Medicaid as they were diagnosed with cancer, this study shows what the outcomes would be in the absence of such a program as the BCCEDP, and provides evidence of the added value of the BCCEDP relative to the outcomes of interest.
Several factors may explain the positive findings. First, with regard to earlier stage at diagnosis, despite the noted similarities between BCCEDP participants and nonparticipants in SES or residence in MUAs, there may be important differences between the two groups in knowledge and attitude regarding the benefits of screening mammography, and their ability to identify the screening venue. Second, the case management program of the BCCEDP may have helped women navigate through the system, possibly helping participants to have shorter time to diagnosis resolution after abnormal findings on the mammogram, as well as expedited enrollment in Medicaid, as shown elsewhere.23 Unfortunately, these measures could not be compared between the two groups.
Our findings should be interpreted in light of the following limitations. First, we were not able to distinguish, especially for the Medicaid group, between diagnostic and screening mammography that led to the diagnosis of breast cancer, nor were we able to investigate the underlying factors (eg, differential between the comparison groups on attitude, knowledge, and ability to navigate the system) that would explain these favorable findings in BCCEDP participants. Future, in-depth studies are warranted to gain a better understanding of these factors and their contribution to improved process measures and outcomes. Second, as noted above, given that we were unable to search back in claims data, we were unable to account for screening status or time to diagnosis resolution after abnormal findings on the mammogram. Therefore, we were unable to account for lead time and length time biases. These biases may have a bearing on survival differences between the two groups, especially given that screen-detected tumors tend to grow more slowly than those detected otherwise.24
The main strength of our study lies in our use of an extensive database, which we built by combining data from multiple sources, with each source contributing to the richness of our data. In particular, the BCCEDP database made it possible for us to identify one-time versus repeat participants in the program. Use of census data at the census tract level enabled us to adjust for SES when studying the association between BCCEDP and the outcomes of interest in the multivariable models. Similarly, we were able to account for whether women resided in MUAs, which is not only a proxy for shortage of health care providers, but also a measure of socioeconomic disadvantage in general.
In summary, although the reach of BCCEDP has remained at only approximately 15% nationwide25,26 and approximately 13% in Ohio, clear benefits have been demonstrated with the program.27–29 With the implementation of the Affordable Care Act, important shifts in caseload and locus of care will occur across payers with regard to cancer prevention and control. To guide downstream efforts in improving cancer outcomes in underserved populations, future studies should identify programmatic features of the BCCEDP that may have contributed to these positive outcomes.
Supported by American Cancer Society Research Scholar Grant No. 121913-RSGI-12-093-01-CPHPS (S.M.K.). S.M.K. is also supported by the Clinical and Translational Science Collaborative of Cleveland through Grant No. UL1TR000439 from the National Center for Advancing Translational Sciences component of the National Institutes of Health (NIH) and NIH Roadmap for Medical Research. The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.
Presented in part at the annual meeting of AcademyHealth, San Diego, CA, June 8-10, 2014.
We thank Xiaozhen Han for her assistance in preparing this article and Tsui Chan, MS, for her work on geocoding and retrieving data from the US Census. We also thank Lynn Giljahn, Tina Bickert, Mary Applegate, MD, and Dan Hecht for their careful review of the manuscript and their helpful comments.
Disclosures provided by the authors are available with this article at jop.ascopubs.org.
Conception and design: Siran M. Koroukian, Paul M. Bakaki, Mark Schluchter, Gregory S. Cooper, Susan Flocke
Financial support: Siran M. Koroukian
Administrative support: Siran M. Koroukian
Provision of study materials or patients: Siran M. Koroukian
Collection and assembly of data: Siran M. Koroukian, Paul M. Bakaki
Data analysis and interpretation: All authors
Manuscript writing: All authors
Final approval of manuscript: All authors
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jop.ascopubs.org/site/misc/ifc.xhtml.
Employment: American Renal Associates (I)
Leadership: American Renal Associates (I)
Stock or Other Ownership: American Renal Associates (I)
No relationship to disclose
No relationship to disclose
No relationship to disclose
Research Funding: Medtronic
No relationship to disclose