SEARCH

SEARCH BY CITATION

Keywords:

  • medically uninsured;
  • insurance coverage;
  • mammography;
  • state health plans;
  • health services accessibility;
  • mass screening;
  • breast neoplasms

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES

BACKGROUND:

The National Breast and Cervical Cancer Early Detection Program (NBCCEDP) provides free or low-cost breast and cervical cancer screening to low-income, uninsured or underinsured women. The authors analyzed the impact of the NBCCEDP on breast cancer mortality rates.

METHODS:

The data consisted of observations for each state and year for the period from 1990 through 2004. The outcome variable was the breast cancer mortality rate for women ages 40 to 64 years. Independent variables included the proportion of women ages 40 to 64 years screened under NBCCEDP. The impact of screening intensity was estimated using least-squares regression with state and year fixed effects.

RESULTS:

In 2004, 1.2% of women ages 40 to 64 years were screened under NBCCEDP. The NBCCEDP screening rate was related significantly and negatively to breast cancer mortality in the same year. Results indicate that, for every 1000 women screened, there were 0.6 fewer deaths because of breast cancer among women ages 40 to 64 years. Changes in screening rates were unrelated to breast cancer mortality ≥2 years in the future.

CONCLUSIONS:

In the current study, there was some evidence suggesting that the NBCCEDP led to a reduction in breast cancer mortality rates. However, the failure to detect an impact of screening on mortality rates in subsequent years suggests that caution is needed in interpreting these results as strong evidence in favor of the effectiveness of the NBCCEDP in reducing breast cancer mortality. Cancer 2010. © 2010 American Cancer Society.

Screening mammography is effective in reducing mortality rates associated with breast cancer,1, 2 and regular screening is recommended by the US Preventive Services Task Force beginning at age 40 years.3 Despite the long history of evidence regarding effectiveness and broad use, screening rates are substantially lower among uninsured women,4, 5 and these women are more likely to present with advanced-stage cancer compared with privately insured patients.5, 6

To reduce disparities in breast cancer mortality, the US Congress passed the Breast and Cervical Cancer Mortality Prevention Act of 1990 (Public Law 101-354, 1990). The law gave the US Centers for Disease Control and Prevention (CDC) the authority to establish and administer the National Breast and Cervical Cancer Early Detection Program (NBCCEDP). The program was established formally in 1991 to provide free or low-cost breast cancer screening to uninsured or underinsured women aged 40 to 64 years with an annual income <250% of the Federal Poverty Level. The breast cancer screening services provided by the NBCCEDP include clinical breast examinations, mammograms, diagnostic testing if results are abnormal, and referrals to treatment.

Currently, the program is administered through cooperative agreements with health agencies in the 50 states, the District of Columbia, 5 US territories, and 12 American Indian/Alaska Native tribal organizations. Since 1991, the NBCCEDP has provided >8.2 million breast and cervical screening examinations to >3.3 million women and has diagnosed >37,000 breast cancers. The NBCCEDP screens approximately 15% of the eligible population for breast cancer.7

In this article, we estimate the impact of the NBCCEDP on breast cancer mortality rates. There are 2 pathways by which the NBCCEDP could affect mortality. First, NBCCEDP facilitates access to treatment among women with breast cancer. Before the year 2000, there was no formal treatment program for these women, but almost every woman who was diagnosed through the program was able to receive free care. After the year 2000, following passage of the Breast and Cervical Cancer Prevention and Treatment Act, state Medicaid programs covered US citizens who were diagnosed with cancer through the program. Second, the NBCCEDP may increase mammography rates. Women who have early stage breast tumors diagnosed by mammography experience lower breast cancer mortality rates because early stage tumors are more amenable to treatment.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES

Overview

We performed a state-level analysis of the impact of the NBCCEDP on breast cancer mortality rates. We observed mortality rates over 14 years (from 1990 to 2004) for each state. We estimated the impact of the NBCCEDP based on changes in breast cancer mortality within states.

Data

We obtained mortality rates for breast cancer and other causes from Multiple Cause of Death files.8 We used the 72 cause-of-death recode variable, which records the “underlying cause of death,” to ascertain causes of death between 1990 and 1998, and we used the 113 recode variable to ascertain causes of death between 1999 and 2004 (the 72 cause-of-death variable was discontinued in 1998, and the 113 cause-of-death variable was not available before 1999). Data on the number of women screened under the NBCCEDP by state and by year were obtained from the NBCCEDP surveillance database. Data on uninsurance rates and Medicaid coverage rates were obtained from the historic health insurance tables produced by the US Census Bureau. Data on unemployment rates were obtained from the local area unemployment statistics produced by the Bureau of Labor Statistics. Data on income levels were obtained from the regional economic accounts produced by the Bureau of Economic Analysis.

Variables

The main outcome of interest was breast cancer mortality per 100,000 women ages 40 to 64 years, which is the target age group of the program. Medicare covers mammography and treatment for women aged ≥65 years. We also measured mortality for heart disease, colon cancer, and homicides, suicides, and accidents for use in falsification tests.

The idea behind the falsification tests is to estimate the impact of the NBCCEDP on deaths from causes not targeted by the program (for example, heart disease). A finding that the NBCCEDP is related negatively to mortality for 1 of these causes could indicate that our study design lacks specificity or that some omitted variable explains the results. Of course, we expect some of the coefficients to be significant based on chance alone.

The Multiple Cause of Death files do not include income or insurance variables; therefore, we were unable to limit the sample to women in the NBCCEDP target population. The Multiple Cause of Death files do include a variable that measures educational attainment, but it was not coded consistently across the period covered by the current study.

We separately tested the impact of 2 measures of the NBCCEDP. The first measure was the percentage of women ages 40 to 64 years screened under the NBCCEDP, which was 0 in the years before program adoption. The measure varies based on 1) when the state implemented an NBCCEDP program and 2) the NBCCEDP screening rate as it varies within states over time. The second measure is an indicator variable for whether or not the state has an NBCCEDP program. This measure varies based only on whether or not a state has adopted an NBCCEDP program.

When estimating the impact of the NBCCEDP on mortality, we controlled for the uninsurance rate for adults aged ≤64 years, the proportion of adults aged ≤64 years already covered by Medicaid, the unemployment rate, and the per capita income level in the state. Note that these variables, including mortality rates, measure rates and averages in the entire nonelderly population statewide and not in the low-income population targeted by the NBCCEDP.

Statistical Methods

We estimated 2 models. In Model 1, the measure of program effectiveness is the NBCCEDP screening rate. We estimated the impact on breast cancer mortality rates using least-squares regression with fixed effects for state and year (ie, state-specific and year-specific intercepts). We also estimated a version of the model in which we replaced the year fixed effects with a time trend variable.

In Model 2, the measure of program effectiveness was the indicator variable for whether or not the state had an NBCCEDP program. We estimated the impact on breast cancer mortality rates using ordinary least-squares regression with fixed effects for state and year. The coefficient on the indicator variable can be interpreted as a “difference-in-difference” estimate.

Under each model, the impact of the NBCCEDP is identified based on changes in mortality within states, and each state serves as a control for the other states. It is easiest to explain this concept in the context of Model 2. For example, consider the case of Virginia, which began its program in 1998, and West Virginia, which began its program in 1991. If the program has an impact on breast cancer mortality, then we would expect a steeper decline in breast cancer mortality rates in West Virginia compared with Virginia in 1991. Likewise, we would expect a steeper decline in breast cancer mortality rates in Virginia compared with West Virginia in 1998. This approach does not assume or require that breast cancer incidence or mortality rates are similar across the states. Rather, it assumes that, in the absence of the NBCCEDP, trends in rates would have been similar across states during the study period.

Both models were estimated using the procedure described by Stock and Watson,9 which relaxes the assumption inherent in ordinary least-squares that the error terms in the model have equal variance across observations. All regressions were estimated using MATLAB version 6.1 (The MathWorks, Inc., Natick, Mass). We also estimated models using lagged measures of the NBCCEDP program (ie, screening rates and adoption) and estimated the impact of NBCCEDP measures on mortality rates from other causes as a falsification test.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES

Table 1 describes the history of NBCCEDP-funded screening and breast cancer diagnosis and US breast cancer mortality rates. The first NBCCEDP programs began screening women in 1991. By 1999, all 50 states and the District of Columbia had programs in place. In 2004, the NBCCEDP paid for >560,000 mammograms, screening approximately 1.2% of all women ages 40 to 64 years. There were 2000 women diagnosed by the NBCCEDP annually at a rate of 4 per 100,000 women ages 40 to 64 years. Breast cancer mortality rates among women ages 40 to 64 years have declined steeply from 51 per 100,000 in 1990 to 34 per 100,000 in 2004.

Table 1. Screening, Diagnosis, and Mortality Rates by Year Among Women Ages 40 to 64 Years
YearNo. of States With ProgramsaNBCCEDP Program Details 
MammographiesCancer DiagnosesDeaths
No.Per 100KNo.Per 100KNo.Per 100K
  • NBCCEDP indicates National Breast and Cervical Cancer Early Detection Program;

  • a

    Washington DC was included here as a state, because a program was implemented there in 1997.

19900000016,78551
19912170150016,75950
1992936,49910576016,16747
199314114,806323248116,14345
199417166,530458397116,07844
199525223,578596627216,34144
199635277,058720810215,97942
199747325,651820948215,87540
199850325,3417941111315,65338
199951329,1217801087315,22736
200051353,6518151230315,70536
200151381,1608551343316,00036
200251404,2488851547316,01635
200351524,07111201974416,33235
200451565,19711822111416,31434

Table 2 displays summary statistics for the main variables of interest. The average proportion of women ages 40 to 64 years who were screened across the 50 states and the District of Columbia in 2004 was 1.5% (standard deviation, 1.1%). The standard deviation is large relative to the mean, indicating that screening rates (measured as the percentage of women in the target age group, and not eligible women) vary substantially across NBCCEDP programs.

Table 2. Summary Statistics
Variable1991, N=512004, N=51
MeanSDMinMaxMeanSDMinMax
  1. SD indicates standard deviation; Min, minimum; Max, maximum; NBCCEDP, National Breast and Cervical Cancer Early Detection Program.

Breast cancer mortality per 100,000 women ages 40-64 y, %4911271083452157
Proportion of women ages 40-64 y screened by NBCCEDP, %<.001<.001<.001.0064.015.011.003.057
Uninsured, %1558291641028
On Medicaid, %103518134622
Unemployment rate, %612105148
Per capita income ($1000s)19314283252450
Proportion of women ages 40-64 y aged ≥50 y, %56920764462158

Results from the baseline regressions are displayed in Table 3. In Model 1, the main variable of interest was the proportion of women ages 40 to 64 years screened under each state's NBCCEDP program. In Model 2, the variable of interest was an indicator variable equal to 1 if the state had an NBCCEDP program and 0 otherwise. Parameter estimates for state and year fixed effects are not shown. The R2 statistic was 0.48 in each case.

Table 3. Estimated Impact of the National Breast and Cervical Cancer Early Detection Program on Breast Cancer Mortality per 100,000 Women Ages 45 to 64 Years
VariableModel 1Model 2
Coeff95% CICoeff95% CI
  • Coeff indicates coefficient; CI, confidence interval.

  • a

    P < .05.

  • b

    P < .10.

Proportion of women screened−62.7−118.2 to −7.2a  
State has a screening program  −1.08−2.38 to 0.22
Percentage uninsured0.41−0.04 to 0.87b0.40−0.05 to 0.86b
Percentage on Medicaid0.30−0.11 to 0.710.30−0.12 to 0.71
Unemployment rate0.13−0.47 to 0.730.13−0.47 to 0.73
Per capita income0.22−1.00 to 1.440.20−1.03 to 1.42
Percentage of women aged ≥50 y0.470.20 to 0.74a0.460.19 to 0.74a

The estimated impact of the NBCCEDP on mortality rates in Model 1 is statistically significant at the 5% level. The point estimate is −62.7, indicating that, for every 1000 women screened (1% of 100,000), there are 0.6 fewer deaths (=0.01 × −62.7) because of breast cancer among women ages 40 to 64 years. We caution readers against using this estimate to project the impact of increases in screening outside the range of the data, because the coefficient, 62.7, is greater than the number of breast cancer deaths per 100,000 among women ages 40 to 64 years (ie, increasing the screening rate to 100% would not result in a negative mortality rate). When we replaced the year fixed effects with a year time trend (model not shown), the coefficient the NBCCEDP screening rate was −96 (95% CI, −149 to −43; P < .001). In Model 2, the parameter indicates that implementing an NBCCEDP program reduces the breast cancer deaths by 1.0 per 100,000 (P = .103).

The coefficients on the variables that measure the uninsurance rate, the percentage of residents insured by Medicaid, the unemployment rate, and the proportion of women aged ≥50 years have the expected values. The coefficient on per capita income is positive, but the confidence interval is very wide.

Table 4 explores the impact of lagging the measures of the NBCCEDP program. For example, a model with a 1-year lag estimates the impact of screening rates in year t on breast cancer mortality rates in year t + 1. The justification for estimating these models is that NBCCEDP-funded screens and treatment may not reduce breast cancer mortality rates instantaneously. Instead, it may take some time before screens and treatments lead to avoided deaths. Note that separate models were estimated for each lag.

Table 4. Estimated Impact of Lagged National Breast and Cervical Cancer Early Detection Program Measures on Breast Cancer Mortality per 100,000 Women Ages 40 to 65 Years
Program MeasureLag Time
None (Base Model)1 Year2 Years
Coeff95% CICoeff95% CICoeff95% CI
  • Coeff indicates coefficient; CI, confidence interval.

  • a

    P < .05.

  • b

    P < .10.

Proportion of women screened−62.7−118.2 to −7.2a−55.5−114.0 to 3.1b−19.1−64.5 to 26.3
State has a screening program−1.08−2.38 to 0.22−0.22−1.49 to 1.05−0.75−2.01 to 0.51

For Model 1, in which the impact of the NBCCEDP is measured based on the proportion of women screened, a 1-year lag slightly reduced the coefficient on the screening rate, from −62.7 to −55.5. This estimate differs significantly from zero at the 10% level but does not differ significantly from the coefficient in the baseline model (P = .81). The coefficient in a model with a 2-year lag is negative (−19.1) but is not significantly different from zero. For Model 2, in which the impact of the program is measured by a dichotomous variable, neither of the lagged models yielded coefficients that differed significantly from zero. Regression models with longer lags (ie, 3 years and 4 years) yielded insignificant coefficient estimates for both Model 1 and Model 2.

We present the results of falsification tests in Table 5. We estimated models separately for women ages 40 to 64 years and women aged ≥65 years. The top portion of Table 5 shows results for models in which the impact of the NBCCEDP was measured according to the proportion screened among women ages 40 to 64 years. The NBCCEDP screening rate was not a significant predictor of mortality among women ages 40 to 64 years for any cause other than breast cancer.

Table 5. Estimated Impact of National Breast and Cervical Cancer Early Detection Program Measures on Mortality per 100,000 Women (Falsification Tests)
Program Variable/Cause of DeathAge Group
40-64 Years≥65 Years
Coeff95% CICoeff95% CI
  • Coeff indicates coefficient; CI, confidence interval.

  • a

    P < .05.

  • b

    P < .10.

Proportion of women screened    
 Breast cancer (base model)−62.7−118.2 to −7.2a−46.−194.5 to 102.3
 Colon cancer−28.1−67.6 to 11.5−52.1−228.7 to 124.4
 Heart disease4.9−36.7 to 46.624.2−293.5 to 341.9
 Homicides/suicides/accidents15.0−48.5 to 78.6−78.8−114.9 to −42.7a
State has a screening program    
 Breast cancer (base model)−1.08−2.38 to 0.220.94−1.91 to 3.79
 Colon cancer−0.82−1.77 to 0.13b2.90−1.35 to 7.16
 Heart disease−0.96−2.12 to 0.2011.603.21 to 19.99a
 Homicides/suicides/accidents0.18−0.96 to 1.32−2.57−5.61 to 0.47b

NBCCEDP screens were related negatively to breast cancer mortality among women aged ≥65 years, but the coefficient was not significant. The NBCCEDP screening rate was related significantly and negatively to mortality from homicides, suicides, and accidents among women aged ≥65 years at the 10% level.

The bottom portion of Table 5 shows results from models in which the impact of the NBCCEDP was measured by an indicator variable for whether or not a state had a program. The indicator variable was related negatively to colon cancer mortality among women ages 40 to 64 years at the 10% level.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES

According to our baseline estimate, there are 0.6 fewer deaths annually from breast cancer among women ages 40 to 64 years for every 1000 women screened under the NBCCEDP. However, we were unable to detect an effect of screening on mortality rates ≥2 years in the future. To the best of our knowledge, this is the first study estimating the direct impact of the NBCCEDP on patient outcomes. Previously, Adams et al10 reported that implementation of the Breast and Cervical Cancer Prevention and Treatment Act reduced the time from diagnosis to enrollment in Medicaid by approximately 7 months.

Given the low rate of breast cancer mortality among nonelderly women and the size of the NBCCED—both in relation to the entire population of women in the target age group and in relation to the number of eligible women7—detecting an effect (assuming there is one) is difficult. The variation in within-state trends in breast cancer mortality rates explained by the NBCCEDP is small in relation to the total variation. We also note that our estimates reflect the mix of never-screened and previously screened women among NBCCEDP participants. According to the National Health Interview Survey, 42% of women with no health insurance and family incomes <250% of the Federal Poverty Level reported having been screened within the previous 2 years.7 In general, screening would be expected to have a greater impact on mortality in a never-screened population.

We must address several questions before determining whether these results reflect a causal relation. First, is there a plausible mechanism behind the effect? There are 2 pathways by which the NBCCEDP might reduce breast cancer mortality rates. First, screening permits the detection of early stage tumors that, in the absence of early intervention, would have developed into metastatic disease. Second, the NBCCEDP facilitates access to treatment among women who are diagnosed with breast cancer. Even in patients for whom treatment does not lead ultimately to a cure, it may delay tumor progression.

Second, are there omitted (ie, confounding) variables? Our models for estimating the impact of the NBCCEDP implicitly control for time-invariant factors that differ across states and years. In this context, a variable would confound the observed relation if it was changing over time in a way that is correlated with both year-to-year changes in NBCCEDP screening rates and breast cancer mortality rates within states. It is difficult to think of a variable that fits this profile. We cannot test directly the assumption of unconfoundedness (ie, the absence of variables that we do not control for but are related to both changes in screening rates and changes in breast cancer mortality rates within states). Instead, we performed falsification tests. The results from models in which we measured the impact of the NBCCEDP based on the proportion of women screened were consistent with the assumption of unconfoundedness. This measure is not unrelated to mortality from colon cancer, heart disease, or homicides/suicides and accidents or to breast cancer mortality outside the target population. The results from models in which we measured the impact of the NBCCEDP using an indicator variable were mixed. We observed that implementing the NBCCEDP program was associated with a reduction in colon and heart disease mortality, but the confidence interval was wide, and these estimates were not statistically stable. These may be false-positive results (ie, Type I errors), or the results may indicate that estimates of the impact of implementing an NBCCEDP program on breast cancer mortality are biased by omitted variables. Conversely, it is conceivable that the NBCCEDP, a multicomponent program, has a positive spillover effect to other forms of preventive care. For example, the Well Integrated Screening and Evaluation for Women Across the Nation (WISEWOMAN) program is a component of the NBCCEDP that was developed for cardiovascular disease prevention and other lifestyle screening interventions.11 Studies evaluating the WISEWOMAN program have indicated that participants significantly improved their systolic and diastolic blood pressure, high-density lipoprotein, and cholesterol and had a significant improvement (8.7%) in their 10-year risk of coronary heart disease.11, 12

Third, is the magnitude of the effect plausible? Assume for the time being that the mechanism of action is screening. Studies based on data from randomized controlled trials13, 14 and an observational study similar to our current investigation15 indicated that regular screening prevents between 1.8 and 3 deaths from breast cancer per 1000 women screened. Our estimate (0.6 deaths per 1000 screened) is on the low side but is certainly a plausible estimate. Next, assume that the treatment of late-stage tumors accounts for the mechanism of action. Our finding that implementing a program reduces deaths by 1.2 per 100,000 women is not out of proportion to the number of cancer tumors detected under the program (in 2004, 4 per 100,000 women) and deaths in the target age group (in 2004, 34 per 100,000 women). In general, our results are consistent with findings from an in-progress modeling study estimating that the NBCCEDP has saved >100,000 life-years over the last 15 years (unpublished observations).

Fourth, is the temporal nature of the relation consistent with what we know about the epidemiology of breast cancer tumors? We observed that the impact of the NBCCEDP dissipated quickly with lagged measures (Table 3). This result seems to be inconsistent with the hypothesis that the NBCCEDP reduced mortality by early detection. Tumors grow slowly; and, in general, we would not expect an increase in screening rates to lead to an immediate reduction in mortality. It took at least 2 years for a difference to emerge in mortality rates between treatment and control groups in the Swedish randomized controlled trials of mammography.16 Of course, detecting lagged effects may be more difficult; the sample size is equal to the number of states multiplied by the number of years of observation. Introducing a 2-year lag reduces the effective sample size, because we can no longer use the variation in mortality rates in 1991 and 1992 to estimate the program effect.

The finding that the NBCCEDP program measures have an instantaneous (ie, same-year) impact on mortality rates is consistent with the hypothesis that the NBCCEDP reduces breast cancer mortality by facilitating access to treatment for women with late-stage disease. In 2001, the Committee on the Early Detection of Breast Cancer at the Institute of Medicine recommended that the program should be expanded to reach ≥70% of eligible women.17 Screening this many women would require a substantial increase in the NBCCEDP budget; therefore, these extra costs must be weighted against the benefits of screening versus the benefits that could be obtained from funding other health-related programs. We observed some evidence suggesting that the NBCCEDP led to a reduction in breast cancer mortality rates. However, the failure to replicate these results in models with lagged program measures suggests that caution should be used in interpreting these results as strong evidence in favor of the effectiveness of the NBCCEDP in reducing breast cancer mortality.

CONFLICT OF INTEREST DISCLOSURES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES

Dr. Howard is supported in part by the Centers for Disease Control and Prevention under an Interagency Personnel Agreement (05IPA47710). The authors made no other financial disclosures.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. CONFLICT OF INTEREST DISCLOSURES
  7. REFERENCES
  • 1
    Humphrey LL, Helfand M, Chan BK, Woolf SH. Breast cancer screening: a summary of the evidence for the U.S. Preventive Services Task Force. Ann Intern Med. 2002; 137: 347-360.
  • 2
    Berry DA, Cronin KA, Plevritis SK, et al. Effects of screening and adjuvant therapy on mortality from breast cancer. N Engl J Med. 2005; 353: 1784-1792.
  • 3
    U.S. Preventive Services Task Force. Screening for breast cancer: recommendations and rationale. Ann Intern Med. 2002; 137: 344-346.
  • 4
    Centers for Disease Control and Prevention. Self-reported use of mammography and insurance status among women aged 40 years—United States, 1991-1992 and 1996-1997. MMWR. 1998; 47: 825-830.
  • 5
    Ward E, Halpern M, Schrag N, et al. Association of insurance with cancer care utilization and outcomes. CA Cancer J Clin. 2008; 58: 9-31.
  • 6
    Halpern MT, Ward EM, Pavluck AL, Schrag NM, Bian J, Chen AY. Association of insurance status and ethnicity with cancer stage at diagnosis for 12 cancer sites: a retrospective analysis. Lancet Oncol. 2008; 9: 222-231.
  • 7
    Tangka FK, Dalaker J, Chattopadhyay SK, et al. Meeting the mammography screening needs of underserved women: the performance of the National Breast and Cervical Cancer Early Detection Program in 2002-2003 (United States). Cancer Causes Control. 2006; 17: 1145-1154.
  • 8
    National Center for Health Statistics. Compressed Mortality File, 1999-2004 [machine-readable data file and documentation, CD-ROM Series 20, No. 2J]. Hyattsville, Md: National Center for Health Statistics; 2006.
  • 9
    Stock JH, Watson MW. Heteroskedasticity—robust standard errors for fixed effect panel data regression. Econometrica. 2008; 76: 155-174.
  • 10
    Adams EK, Chien LN, Florence CS, Raskind-Hood C. The Breast and Cervical Cancer Prevention and Treatment Act in Georgia: effects on time to Medicaid enrollment. Cancer. 2009; 115: 1300-1309.
  • 11
    Will JC, Farris RP, Sanders CG, Stockmyer CK, Finkelstein EA. Health promotion interventions for disadvantaged women: overview of the WISE-WOMAN projects. J Womens Health. 2004; 13: 484-502.
  • 12
    Finkelstein EA, Khavjou O, Will JC. Cost-effectiveness of WISEWOMAN, a program aimed at reducing heart disease risk among low-income women. J Womens Health. 2006; 15: 379-389.
  • 13
    Keen JD, Keen JE. What is the point: will screening mammography save my life [serial online]? BMC Med Inform Decis Mak. 2009; 9: 18.
  • 14
    Tabar L, Vitak B, Yen MFA, Chen HHT, Smith RA, Duffy SW. Number needed to screen—lives saved over 20 years of follow-up in mammographic screening. J Med Screen. 2004; 11: 126-129.
  • 15
    Swedish Organised Service Screening Evaluation Group. Reduction in breast cancer mortality from organized service screening with mammography: 1. Further confirmation with extended data. Cancer Epidemiol Biomarkers Prev. 2006; 15: 45-51.
  • 16
    Nystrom L, Rutqvist LE, Wall S, et al. Breast cancer screening with mammography: overview of the Swedish randomised trials. Lancet. 1993; 341: 973-978.
  • 17
    Institute of Medicine. Mammography and Beyond. Washington, DC: National Academy Press; 2001.