The Breast and Cervical Cancer Prevention and Treatment Act in Georgia

Effects on time to Medicaid enrollment

Authors

  • E. Kathleen Adams PhD,

    Corresponding author
    1. Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, Georgia
    • Rollins School of Public Health, Emory University, 1518 Clifton Road, Room 654, Atlanta, GA 30322
    Search for more papers by this author
    • Fax: (404) 727-9198

  • Li-Nien Chien MPH, MS,

    1. Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, Georgia
    Search for more papers by this author
  • Curtis S. Florence PhD,

    1. Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, Georgia
    Search for more papers by this author
  • Cheryl Raskind-Hood MS, MPH

    1. Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, Georgia
    Search for more papers by this author

Abstract

BACKGROUND:

Implementation of the Breast and Cervical Cancer Prevention and Treatment Act of 2000 (BCCPTA) allowed states to extend Medicaid to any woman aged <65 without insurance screened and found to need treatment either for breast or cervical cancer or for a precancerous cervical condition through the National Breast and Cervical Cancer Early Detection Program (NBCCEDP) or in Georgia, other provider sites.

METHODS:

The authors used linked Georgia Comprehensive Cancer Registry (GCCR) and Medicaid data to test the: 1) likelihood of Medicaid enrollment in a given month and 2) time-to-enrollment (months) for those eventually enrolling. The authors used difference-in-differences analysis to estimate the effects of BCCPTA for breast or cervical cancer cases relative to a control group of women with other cancers. The authors controlled for sociodemographics, stage at diagnosis, year of diagnosis, and county level factors related to insurance levels in the area.

RESULTS:

Compared with the control cancer group, the hazard ratio of Medicaid enrollment for women with breast and cervical cancers increased post- vs pre-BCCPTA implementation. The estimated effect of this increase was that out of every 1000 women with breast cancer, BCCPTA led to 1.7 more (from 2.8 to 4.5 per month) enrolling in Medicaid. The results for women with local or later stages of cervical cancer indicated that of 1000 women with these cancers, the number enrolling in a given month increased by 3.4 due to BCCPTA. Results on time-to-enrollment indicated that the time between cancer diagnosis and enrollment was shortened by 7 to 8 months.

CONCLUSIONS:

The Georgia Medicaid program, in response to national legislation, increased the probability of women enrolling in Medicaid earlier and in turn, likely increased their cancer treatment options. Cancer 2009. © 2009 American Cancer Society.

The outlook for early/appropriate treatment among poor women diagnosed with breast and/or cervical cancer changed significantly on October 24, 2000 with the signing of the Breast and Cervical Cancer Prevention and Treatment Act of 2000 (BCCPTA). The BCCPTA allows states to extend Medicaid to any woman aged <65 years without insurance who is screened through the National Breast and Cervical Cancer Early Detection Program (NBCCEDP) and needs treatment for breast or cervical cancer or for a precancerous cervical condition. The NBCCEDP, which is funded through the Centers for Disease Control and Prevention, has provided screening and some diagnostic follow-up for uninsured women with incomes <250% of the Federal Poverty Level since 1990. It served almost 400,000 women in 2006.1 Research has indicated that the aging of the NBCCEDP over 5 years was associated with a 3 percentage point increase in mammography rates in the late 1900s.2 The BCCPTA was the first effort to use a population-based public health screening program (NBCCEDP) to establish a pathway to a publicly funded health insurance program.

The BCCPTA was adopted quickly by all 50 states, and 12 states expanded beyond NBCCEDP providers to create greater access.3 Georgia began enrollment in the Women's Health Medicaid Program on July 1, 2001. It was 1 of the 12 states that extended BCCPTA eligibility to women screened by non-NBCCEDP providers, thereby widening the network of screening providers beyond public health departments. State data indicated that nearly 75% of Georgia's BCCPTA enrollees entered through non-NBCCEDP–funded providers and that the BCCPTA expanded Georgia's coverage of women with breast and/or cervical cancer by >33% in 2003 alone.4 BCCPTA enrollees were more likely to have breast cancer rather than cervical cancer, to be older than other Medicaid enrollees with these cancers, and to look most like the disabled eligibility group in terms of costs.

The BCCPTA has the potential to help reduce the decades-old ‘treatment gap’ for low-income women.5 Earlier studies reported that Medicaid enrollees were more likely to be diagnosed at a late disease stage6-8; because, before the BCCPTA, cancer patients often became eligible for Medicaid only after a serious illness/disability or catastrophic expense. The BCCPTA not only may increase the number of women enrolling in Medicaid but also may decrease the time to enrollment. This is particularly important because time delays from initial symptoms of 3 to 6 months have been associated with lower survival, and this effect is mediated through the correlation between delay and advanced disease stage.9

To evaluate the BCCPTA, we linked Georgia Comprehensive Cancer Registry (GCCR) and Medicaid administrative data to distinguish women who enrolled in Medicaid before their diagnosis through traditional eligibility categories from women who enrolled once they were diagnosed. Only the latter group could be affected by the BCCPTA. Although the ultimate outcomes of BCCPTA are decreased time to treatment and, in turn, reduced morbidity and mortality, we studied 2 intermediate outcomes—Medicaid enrollment and time to enrollment. To our knowledge, this is the first study to evaluate the BCCPTA with detailed data on the timing of diagnosis and enrollment. If earlier enrollment means that treatment occurs at an earlier stage of disease, then the current study is an important first step in analyzing this legislation.

MATERIALS AND METHODS

Data

We linked the GCCR and Georgia Medicaid enrollment files using encrypted Social Security numbers for incident cases of breast cancer, cervical cancer, or 1 of 5 control cancers (bladder cancers, colorectal cancers, melanomas of the skin, non-Hodgkin lymphomas, and thyroid cancers) that were identified in the Georgia registry between January 1, 1999 and December 31, 2004. The GCCR is a statewide (since January 1999), population-based cancer registry of cancer cases diagnosed in any Georgia county. The Georgia administrative files provide monthly enrollment information back as far as January 1996 and through December 2005. We used the Federal Information Processing Standard code from the GCCR to link data from the Area Resource File and the Georgia County Business File. With the linked data, we tested the effects of health system capacity and economic conditions (eg, employment) that could affect the probability that a woman otherwise would be uninsured and, hence, in need of Medicaid coverage. The Area Resource File, which is assembled by the Health Resources and Services Administration, provides a variety of county-level data for the 159 counties in Georgia as well as across the nation. The Federal Information Processing Standard is a 5-digit code that uniquely identifies counties and county equivalents in the United States. The Georgia County Business File provides data on employment and economic activity by size of firm.

We used a quasiexperimental design to evaluate the assumed causal relations between our outcomes and the BCCPTA by identifying women who had the treatment and control cancers in the periods before and after the BCCPTA was passed. The total study sample consisted of all women who were diagnosed with breast cancer, cervical cancer, or 1 of the control cancers at ages 14 to 64 years before and after the BCCPTA. Individuals aged ≥65 years were not newly eligible under the BCCPTA. The final sample from the GCCR was 34,023 women who had a first primary diagnosis of breast cancer (n = 22,199), cervical cancer (n = 1846), or a control cancer (n = 9978). We excluded 2198 women who already were eligible and enrolled in Medicaid before and in their month of their diagnosis, because they would not be affected by the new eligibility rules under the BCCPTA. This resulted in a sample of 31,825 women who, after they were diagnosed with cancer, faced the ‘hazard’ of enrolling in Medicaid.

For the enrollment analysis, we used the 31,825 sample, although those with missing stage data (around 3% of the sample) were omitted from regression analyses. Patients who died without enrolling in Medicaid were treated as censored observations (n = 3544). The maximum time observed for the hazard function was 84 months. Women who enrolled in their month of diagnosis (but did not enroll in the previous month) were counted as new enrollees. The time to enrollment sample included 4260 women ages 14 to 64 years with breast cancer, cervical cancer (local or later stages), or control cancers who enrolled in Medicaid during or after the month of their diagnosis and before December 2005.

Variables

For dependent variables, we used 1) the odds of enrolling in Medicaid in a given time and 2) the time to enrollment in months. The key independent variable for both outcomes was the pre/post-BCCPTA period, which was interacted with cancer site (breast or cervical). We also used a set of covariates that could affect each outcome independent of the BCCPTA.

For individual covariates we used 1) age at diagnosis (ages 14-24 years, 25-34 years, 35-44 years, or 45-64 years), 2) race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, or other/unknown), 3) marital status (single, married, separated, divorced, widowed, or unknown), 4) the number of cancer diagnoses in a lifetime (1 primary, 2 primaries, or >2 primaries), and 5) summary stage according to the National Cancer Institute Surveillance, Epidemiology, and End Results Program (in situ, localized, regional, or distant). For county control covariates, we used 1) geographic area of the resident county (central city large metropolitan, fringe county large metropolitan, small metropolitan, or rural), 2) teaching hospital in county to measure the ability to diagnose/refer and treat, 3) the percentage of total employment in the service sector, 4) the unemployment rate, and 5) the percentage of small firms (<10 employees). Because the registry does not include insurance information, we included many of these covariates as proxies for being uninsured and in need of Medicaid coverage. We used unemployment rate, employment in the service industry, and employment in small firms to proxy insurance coverage, because the bulk of private insurance is through the place of employment and small-firm workers. Furthermore, service sector employees have higher uninsurance rates than other types of workers.10 We included dummies for the variable year of diagnosis to control for secular changes independent of the BCCPTA. The median household income and the percentage poverty in the county were examined in preliminary models but were identified as insignificant and were omitted.

Models and Statistical Analysis

Enrollment model

When the time to event (here, enrollment) varies for each observation and is not observed for some, the hazard function, h(t), provides the likelihood of an event at a given time for a woman who did not enroll before that time.11 We estimated a likelihood function of enrollment by time (t), as shown in Equation 1 below:

equation image

where the hazard (h) is enrollment in Medicaid in month t after the cancer diagnosis, Cancer is a dummy variable denoting the primary cancer site, BCCPTA is a dummy indicating prepolicy/postpolicy implementation, Cancer*BCCPTA is a dummy variable denoting the interaction term of pre/post-BCCPTA for primary cancer site, Xi is a vector of individual covariates, Tt is a dummy for year; and Ct is a vector of county covariates.

The effect of BCCPTA was derived by using difference-in-difference analysis. That is, we estimated the effect of the BCCPTA by using our control cancers to ‘difference’ out the effects of other factors that changed before and after the BCCPTA that could affect the probability that any woman with cancer, not just those affected by the BCCPTA, was enrolled in Medicaid. Statistically, we interacted the dummy variable indicating the pre- and post-BCCPTA period with cancer site (Eq. 1) to test whether the likelihood of enrollment changed pre- and post-BCCPTA for women with, alternatively, breast cancer or cervical cancer differently than for women with control cancers. The footnote to Table 1 illustrates how the estimated equation is used.

Table 1. Derivation of Difference-in-Differences*
Odds of EnrollingControl Cancers (Cancer=0)Breast/Cervical Cancer (Cancer=1)
  • BCCPTA indicates Breast and Cervical Cancer Prevention and Treatment Act.

  • *

    The BCCPTA effect is calculated by taking the difference-in-differences or (α22−α12)−(α21−α11)−exp(β3).

Pre-BCCPTA (BCCPTA=0)α11=exp β0α12=exp(β01)
Post-BCCPTA (BCCPTA=1)α21=exp (β02)α22=exp(β0123)

The widely used Cox proportional hazard rate model (nonparametric model) is more flexible for big datasets like ours, but our data failed to meet the assumption of proportional hazards for several subgroups. Hence, we tested and reported results of the parametric model with Weibull distribution, which had the highest likelihood ratio across alternative functional forms, and the results were robust across functions.

Time-to-enrollment model

To measure the time to enrollment, we had data on the actual date of diagnosis but only had the month of enrollment. Hence, we assigned each date of diagnosis to the appropriate month. Then, we counted the months from the date of diagnosis to the date of either Medicaid entry or death. Women who enrolled in the month of diagnosis were assigned a time of 0.5 months. To estimate our models on the time to enrollment, we used ordinary least-square models with log transformation, as shown in Equation 2 below:

equation image

Again, the difference-in-differences was measured by using the interaction term from the estimated equation.

Marginal effects

Given the difficulty in interpreting the interaction term in nonlinear models,12 we also estimated the marginal effect of BCCPTA on the probability of enrollment. We did this by predicting the pre- and post-BCCPTA probabilities for the breast/cervical cancers and the control cancers and calculating the difference. Bootstrapping was used to estimate the standard deviation of this measure. We derived a marginal effect of BCCPTA on the time to enrollment using a smearing transformation because of the log form of the dependent variable. We used bootstrapping to estimate the standard error of the marginal effect.

RESULTS

The means for all independent variables are presented in Table 2. They indicated a fairly balanced panel pre-BCCPTA (n = 12,356) and post-BCCPTA (n = 19,469). There were more breast cancers by far (around 65%) than cervical cancers (around 5%) among the total of treatment plus control cancers (30%). In part, this was caused by the omission of in situ cervical cancer cases from the cancer registry. Sample women predominantly were white, largely were in the group ages 45 to 64 years, were married, and had either in situ first cancers (15%-17%) or localized first cancers (approximately 48%). Around 12% of the regression sample enrolled in Medicaid sometime during the study period.

Table 2. Characteristics of Registry Women Before and After the Breast and Cervical Cancer Prevention and Treatment Act
CharacteristicPre-BCCPTA (January 1999 to June 2001) N=12,356Post-BCCPTA (July 2001 to December 2005) N=19,469P
No.%No.%
  • BCCPTA indicates Breast and Cervical Cancer Prevention and Treatment Act; SD, standard deviation.

  • *

    Chi-square test (the proportion of registry women pre- and post-BCCPTA among the groups).

  • Student t test (the mean difference of time to enrollment pre- and post-BCCPTA in each group).

Cancer site    <.001*
 Breast cancers830267.212,65965 
 Cervical cancers6705.49134.7 
 Control cancers338427.4589730.3 
Age, y     
 <18-241160.91590.8 
 25-34737611435.9 
 35-44245319.9383419.7 
 45-64905073.214,33373.6 
Race/ethnicity    <.05*
 White928875.214,34273.7 
 Black260021431322.2 
 Hispanic2311.94142.1 
 Others1691.43071.6 
 Unknown680.6930.5 
Marital status    <.001*
 Single157712.8250512.9 
 Married79126412,35163.4 
 Separated, divorced, widowed208916.9308215.8 
 Unknown7786.315317.9 
No. of cancer diagnoses over lifetime    <.001*
 1 Primary in life10,7538717,63690.6 
 First of ≥2 primaries7275.97223.7 
 ≥2 Primaries8767.111115.7 
Stage of cancer at diagnosed    <.001*
 In situ192015.5330817 
 Local597648.4944848.5 
 Regional337427.3495225.4 
 Distant736611275.8 
 Missing3502.86343.3 
Residence    <.001*
 Control city large metro323926.2472924.3 
 Fringe county large metro690855.911,22757.7 
 Small metropolitan area186615.1302415.5 
 Completely rural3432.84892.5 
Teaching hospital in county     
 No854669.213,69670.3 
 Yes381030.8577329.7 
Percentage of service employers, mean±SD12.223±1.6 13.320±2.3 <.001
Percentage small firms (<10 employees), mean±SD73.178±4.0 73.473±3.9 <.001
Unemployment rate, mean±SD3.724±1.3 4.697±1.0 <.001

Enrollment

In Table 3, we provide descriptive data on the incidence of enrollment per 1000 person-months and the mean of time to enrollment observed for our sample of women in the pre- and post-BCCPTA periods. Although the mean months of follow-up were shorter in the post-BCCPTA versus the pre-BCCPTA period, we observed a significant increase in the incidence or rate of enrollment for women with breast and cervical cancers post-BCCPTA. The rate for breast cancers increased from 1.7 to 3.2 per 1000 person-months, and the rate for cervical cancers increased from 4.7 to 10.7 per 1000 person-months. Although there also was an increase in the rate of enrollment for the control cancers, it was smaller in magnitude.

Table 3. Bivariate Comparison of the Incidence Rate of Enrollment Per 1000 Person-Months, Months of Follow-Up, and Time to Enrollment in Medicaid Before and After the Breast and Cervical Cancer Prevention and Treatment Act
Cancer SitePre-BCCPTA (Jan. 1999 to June 2001), N=12,356Post-BCCPTA (July 2001-Dec. 2005), N=19,469
No. EnrollingMean Follow-Up, moIncidence Rate per 1000 Person-Months [95% CI]Mean±SD Time to Enrollment, moNo. EnrollingMean Follow-Up, moIncidence Rate per 1000 Person-Months [95% CI]Mean±SD Time to Enrollment, mo
  • BCCPTA indicates Breast and Cervical Cancer Prevention and Treatment Act; CI, confidence interval; SD, standard deviation.

  • *

    Significantly different from zero at the P = .001 level for log-rank test of the probability of enrolling and Student t test of the mean difference in the time to enrollment pre-BCCPTA and post-BCCPTA.

Breast99169.21.7 [1.6-1.8]*16.7±18.9*161639.63.2 [3.1-3.4]*4.9±10*
Cervical17956.74.7 [4.1-5.5]*13.6±16.7*29029.810.7 [9.5-12]*3.0±9*
Control34854.23.1 [2.7-3.6]*13.8±17.9*44434.74.1 [3.6-4.6]*8.7±6.1*

The hazard ratios for primary cancer site, BCCPTA implementation, and the interaction term are shown in Table 4 after adjusting for all covariates (full results are available upon request). The ratios reflect the relative probabilities of enrolling. A ratio >1 indicates a better outcome (here, enrollment), on average, for women with breast or cervical cancer than for women with control cancers. The results indicated that younger, single, minority women; women experiencing their second cancer; and women in later stages of their disease were more likely to enroll than their counterparts. The greater unemployment in the area, probably associated with higher levels of uninsured, were associated with higher hazard ratios and with the percentage of very small firms in the county.

Table 4. Parametric Regression Model of the Odds of Women Enrolling in Medicaid with Weibull Distribution
Policy CharacteristicBreast vs Control Cancers, N=29,284Cervical vs Control Cancers, N=9165
Adj HRP95% CIAdj HRP95% CI
  • Adj HR indicates adjusted hazard ratio; CI, confidence interval; BCCPTA, Breast and Cervical Cancer Prevention and Treatment Act.

  • *

    The regressions models were adjusted for women's age at diagnosis, race/ethnicity, marital status, number of cancer diagnoses over lifetime, cancer stage at diagnosis, and cancer diagnosis year and county covariates, including urban/rural residence, teaching hospital, percentage employment in the service sector, percentage of employees in small business firms (<10 employees), and percentage unemployment rate.

Cancer site      
 Breast cancers1.40<.0011.23-1.60  
 Cervical cancers  2.63<.0012.16-3.19
 Control cancers1.00  1.00  
BCCPTA implementation      
 Before1.00  1.00  
 After0.69<.010.56-0.850.74<.050.55-1.00
Interaction term      
 Cancer site/BCCPTA implementation*1.60<.0011.35-1.891.66<.0011.30-2.13

Independent of the BCCPTA, women with breast cancer were more likely to enroll in Medicaid relative to women with control cancers (hazard ratio, 1.40). For women with control cancers, the BCCPTA appeared to have lowered their probability of enrollment (hazard ratio, 0.69). Yet, the likelihood of women who had breast cancer enrolling increased significantly with implementation of the BCCPTA compared with women who had the other cancers (hazard ratio, 1.60). Their hazard ratio in the post-BCCPTA period increased to 2.24 (1.60*1.40). The effects of the BCCPTA for cervical cancers also were positive. The hazard ratio of women with the local or later stage cervical cancers enrolling was 2.6 times that of women with control cancers (of local or later stage) pre-BCCPTA but was greater than 4 times (1.66*2.64 = 4.38) their odds post-BCCPTA.

Time to Enrollment

The mean number of months between diagnosis and enrollment for women with breast cancer, cervical cancer, and the control cancers (see Table 3) indicated that, on average, before the BCCPTA, women with breast cancer who eventually enrolled in Medicaid took 16.7 months to do so. This time declined by 11.8 months to approximately 5 months post-BCCPTA, and the time to enrollment for women with cervical cancer declined by 10.6 months. However, there also was reduced time to enrollment (by 5.1 months) among the women with control cancers whose eligibility was not affected by the BCCPTA. The difference between these 2 changes was 6.7 months for women with breast cancer. This perhaps is a rough indication of the effect of the BCCPTA on time to enrollment for these women.

We completed multivariate analyses of the time to treatment (results available upon request) and observed that the time to enrollment increased for women with control cancers after the BCCPTA. The effect of BCCPTA for women with breast and local or later stages of cervical cancer was negative and significant (coefficient, −1.13). This indicated that these women were enrolling in Medicaid more quickly post-BCCPTA versus pre-BCCPTA in Georgia relative to women with control cancers.

Marginal Effects

The estimated marginal effect from both the enrollment and the time-to-enrollment models based on the interaction terms are shown in Table 5. These reflected the difference in the pre- and post-BCCPTA enrollment probabilities for women with breast cancer (and, in turn, cervical cancer). The estimate for women with breast cancer was equal to 1.7 per 1000 person-months and could be interpreted to mean that, of every 1000 women with these cancers, the BCCPTA led to 1.7 more women (range, 2.8-4.5 women per month) enrolling in Medicaid. The results for women with local or later stages of cervical cancer indicated that, of 1000 women with these cancers, the number enrolling in a given month increased by 3.4 before versus after implementation of the BCCPTA in Georgia.

Table 5. Estimated Mean Marginal Effects for the Breast and Cervical Cancer Prevention and Treatment Act*
VariablePre-BCCPTAPost-BCCPTADifference (Post-Pre)
Mean95%CIMean95%CIMean95%CI
  • BCCPTA indicates Breast and Cervical Cancer Prevention and Treatment Act; CI, confidence interval.

  • *

    The estimates were derived from the interaction term in the regression models.

  • These 95% CIs were based on bootstrapping standard error.

Enrollment rate, 1000 person-months      
 Breast cancer2.82.7-3.04.54.3-4.71.71.6-1.8
 Cervical cancer5.04.6-5.48.47.7-9.03.43.1-3.6
Time to enrollment, mo      
 Breast cancer12.912.6-13.14.84.7-4.9−8.1−8.2 to −7.9
 Cervical cancer10.19.8-10.42.92.8-2.9−7.3−7.5 to −7.0

The estimated marginal effect from the time-to-enrollment model indicated that the BCCPTA, after controlling for other factors, shortened the time from diagnosis to enrollment for women with breast cancer by an estimated 8 months. For women with local or later stage cervical cancers, the time between diagnosis and eventual enrollment was shortened by approximately 7 months. Both were significant and sizeable effects.

Limitations

Although the current analyses shed new light on the effects of an important public policy, there are several limitations worth noting. First, the data are for only 1 state and, hence, cannot be generalized. Moreover, Georgia is 1 of the 12 states that allows any provider to refer women into the BCCPTA program. Therefore, the effects observed here may be stronger than those that are observed in more restrictive state policy environments. The linkage of the registry and administrative files for a period pre- and post-BCCPTA, however, has not been done in many states. Hence, the Georgia analysis will be 1 of the first to address these particular effects of the BCCPTA.

There also are several statistical limitations. First, our study faced the common problem of most follow-up studies: right censoring of the data because of the end of the follow-up period. This means that the time to event (enrollment) was not observed for all individuals: We observed that only approximately 12% of women in the registry enrolled in Medicaid. Although the hazard model was designed to analyze such censored data, the marginal effects may have been affected, because a large portion of our sample did not experience enrollment over the time observed. However, we did observe women for an average of at least 30 months; thus, it is likely that there were relatively few enrollees in the tail of the distribution for whom we could not observe enrollment.

A key limitation is that we could not identify the sample of women who probably were eligible for Medicaid either because of low income or, more important, lack of insurance. Rather, we had to estimate our models based on all women who were diagnosed with breast, cervical, or control cancers. This limitation occurred because the registry data did not include income or insurance status at the time of diagnosis. We addressed this limitation by including variables that were correlated with insurance coverage at the individual and county levels; and, as noted, these variables generally had the expected values in the regression analyses.

The major limitation of the study is that we were not able to assess the ultimate goal of this legislation, which is to improve outcomes by providing Medicaid coverage that assures timely treatment. This study only addressed the 2 intermediate outcomes: the numbers of women enrolling in Medicaid and the time between initial diagnoses and eventual enrollment. However, enrollment is a necessary first step. Further studies should examine the association between the BCCPTA and women's health status and explore whether shortening the time from diagnoses to enrollment changes disease stage at the time of treatment and/or treatment patterns.

DISCUSSION

The results presented here indicate that the Georgia Medicaid program, in response to the BCCPTA, through its Women's Health Medicaid Program, increased the probability that women would enroll in Medicaid. In turn, providers and program staff brought them in more quickly. The increase in the number enrolling in a given month and the reduction in the number of months between cancer diagnosis and eventual Medicaid enrollment may be significant in terms of getting women into treatment more quickly and, perhaps, at earlier stages.

Our findings for women with breast and cervical cancers are especially important, because overall progress in the detection and early treatment of breast and other cancers still has left disparities in the stage at diagnosis.13-16 This is true especially among the uninsured and the Medicaid insured.6-8, 17, 18 In contrast to most previous studies, we were able to distinguish between women who were enrolled in Medicaid before versus after diagnosis, like Bradley et al,19 who observed that enrollment after diagnosis but before the BCCPTA also meant a later stage of diagnosis among Medicaid insured individuals and, hence, lower survival.19 Our finding that the BCCPTA reduced the time to enrollment by 7 to 8 months may mean that women enroll at an earlier stage. However, women who newly enter Medicaid still need to find participating specialists and other providers who can provide timely and appropriate care. Future work should examine whether the changes wrought by the BCCPTA are enough to alter differences between Medicaid and other payers with respect to treatment and, in the long run, survival rates.

The current results indicated that the time to enrollment was shorter after the BCCPTA both for women with breast cancer and for women with cervical cancer. However, we could analyze only women with local or later stages of cervical cancer. This cancer is highly preventable, especially with new science on the causal link to the human papillomavirus and the potential use of vaccines. Although our analysis sheds insight on women with cervical cancers that are not in situ, the BCCPTA includes women with all stages of disease and, indeed, those with precancerous cervical conditions. The effects of the BCCPTA on the initial treatment of women with all stages of cervical cancer and the follow-up of those with precancerous conditions will be of great importance for the ability of this public policy to affect the morbidity and mortality of this preventable disease. Future analysis should include all women in the BCCPTA program who have this cancer.

We also note that our finding of a reduction in Medicaid enrollment and an increased time to enrollment associated with the BCCPTA for women with the other (control) cancers is of concern if these results reflect an increased demand from BCCPTA-eligible women for a limited number of participating Medicaid providers. However, it also may mean that, as Medicaid enrolled more women with breast and cervical cancers, this ‘freed up’ some resources within Georgia's Cancer State Aid program for the indigent and uninsured that now could be used for those with other cancers. This question deserves further attention.

Although there are remaining challenges to the use of Medicaid and registry files to analyze treatment patterns (eg, discontinuous enrollment, comorbidities), detailed monthly enrollment data and software that allow for the development of indices of comorbid conditions can help to address these issues. It is also possible that women who are affected by the BCCPTA will exhibit more continuous enrollment patterns, because they are eligible for Medicaid as long as they are considered ‘under treatment’ by their physician. It will be important to examine the timing and type of treatments received by these newly eligible and enrolled women, because the BCCPTA represents a potential model for the expansion of Medicaid to uninsured individuals with other cancers.

Acknowledgements

We thank Kevin Ward and Jonathon Liff for their help in interpreting the Georgia Comprehensive Cancer Registry (GCCR) data and acknowledge the support of the GCCR director, Rana Bayakly.

Conflict of Interest Disclosures

Funding was provided by the American Cancer Society under grant RSGT-05-004-01-CPHPS.

The opinions reflected herein are the authors and do not necessarily reflect those of the funding agency.

The authors made no other disclosures.

Ancillary