Background
Female breast cancer (BC) constitutes the highest proportion of Medicare cancer spending and is projected to increase by an additional 32% in 2020, suggesting a higher economic burden on Medicare.1 The initial phase of care (12 months) after BC diagnosis constituted 37% of BC Medicare spending, which is substantially higher due to surgery and adjuvant therapy.2–6 Additionally, elderly women aged ≥65 years have higher BC incidence compared with their younger counterparts,7 which may result in higher use of Medicare resources.
Several studies on Medicare expenditures during the initial phase of BC care used cancer case data only to 2003, from fewer cancer registries, or did not include all BC stages.1,2,6,8–10 A study that identified significant determinants of expenditures during the initial phase of BC care was conducted using Medicare beneficiaries in Virginia but is outdated.11 Furthermore, there is a dearth of studies on expenditures associated with BC care among rural women and from non-SEER states.
Numerous rural areas in the United States are economically underdeveloped and medically underserved12—Appalachia is one such geographic region whose residents are designated as a special population of interest by the NCI.13 West Virginia (WV), the only state lying entirely in Appalachia, has all of the characteristics of rural areas in the United States, including high poverty, low levels of education, aging population, high rates of chronic diseases, poor health behaviors, and poor healthcare infrastructure.14–19 Although WV has lower BC incidence among its older women (372.8 per 100,000 vs 410.6 per 100,000), it has higher mortality (110.4 per 100,000 vs 98.6 per 100,000) compared with national estimates,20 and higher rates of advanced BC.21–24 Hence, it is highly likely that Medicare spending during the initial phase of care among WV women with BC may be higher than national averages. However, the estimates of Medicare expenditures during the initial phase of care among elderly women with BC from a rural and medically underserved state (WV) and how these vary from national averages, are not available.
Therefore, the objectives of the study were to (1) determine Medicare expenditures during the initial phase of care among elderly WV Medicare fee-for-service (FFS) female beneficiaries with BC and compare them with national estimates derived from the SEER-Medicare data; and (2) determine the factors associated with expenditures. The post hoc objective of the study was to understand the extent to which differences in observed characteristics between WV Medicare FFS female beneficiaries with BC and national estimates explain the differences in expenditures between WV and the remainder of the United States.
Methods
Study Design
A retrospective observational cohort study was conducted from the Medicare perspective to assess direct medical expenditures paid by Medicare during the initial phase of care among WV Medicare FFS female beneficiaries with incident BC.
Data Source
WV Cancer Registry-Medicare Linked Data Set: This study used data from the WV Cancer Registry (WVCR) to identify WV women with BC. Established by the WV Department of Health and Human Resources in 1993, the WVCR is a state-level cancer registry that provides information on cancer incidence and mortality in WV.25 The WVCR was linked with Medicare claims files to derive Medicare utilization and expenditures. The details of the data set development and linkage processes with the WV Medicare database are described elsewhere.26
SEER-Medicare Linked Data Set: This study used data from the SEER program that gathers data on clinical and enrollment status about cancer cases from 17 population-based tumor registries that in turn collect data from hospitals, outpatient visits, laboratories, private practitioners, hospices, autopsy reports, and death certificates; it represents 26% of the US population.27 Approximately 94% of SEER cancer cases were matched to their Medicare claims from their eligibility through to death based on their name, social security number, sex, and date of birth.28 The Medicare claims files were linked to the SEER cancer cases to identify a nationally representative cohort of elderly women with BC. Details of the SEER-Medicare data set and its linkage are described elsewhere.27
The Area Resource File was linked to the WVCR-Medicare and SEER-Medicare data sets using the state and county Federal Information Processing Standards codes for each beneficiary to identify the county level information on the availability of healthcare facilities, healthcare providers, and socio-economic characteristics of the region's population.
Study Cohorts
The study cohorts consisted of women aged ≥66 years at the first primary incident BC (ICD-9 codes 174.xx, 233.0x, 238.3x, and 239.3x) during 2003 through 2006. Women who were continuously enrolled in Medicare Parts A and B at least 12 months before and up to the follow-up period of 12 months after BC diagnosis, who were not enrolled in health maintenance organizations at any time during this period, and had a reported stage at diagnosis were included in the study. Women diagnosed via death certificates/autopsy, who died within 12 months after diagnosis, and had zero total Medicare expenditures were excluded.
Of 2,814 women with BC from the WVCR-Medicare data set, 1,387 WV women were included based on the inclusion and exclusion criteria. In addition, of 112,719 women with BC from the SEER-Medicare data set, 39,525 were included to form a nationally representative population for comparison.
Measures
Initial Phase of Care: Because the prime objective was to evaluate Medicare expenditures during the clinically relevant treatment phase, and to be consistent with the previous studies, the initial phase of care was identified as the year after the date of BC diagnosis.1,2,4–6,11,29
Medicare Expenditures: All Medicare files, including inpatient, hospital outpatient, carrier, hospice, skilled nursing facility, home health agency, and durable medical equipment (DME), were used to derive direct Medicare expenditures. Expenditures were defined as the amount reimbursed by Medicare. Reimbursements are calculated using reimbursement formulas that exhibit average healthcare use for the particular healthcare service.30 Average Medicare expenditures and expenditures for specific services (inpatient services, outpatient services, physician services, and other services, such as DME, hospice care, and home health agency) were also calculated separately for both of the study cohorts. Even though expenditures associated with chemotherapy covered under Medicare Part B physician services and DME were captured and included in the analyses, expenditures associated with chemotherapy and prescription drugs (Medicare Part D) were not included in the analyses because these were not covered by Medicare before 2007. Additionally, because prescription drug benefits were not available to Medicare beneficiaries during the study period, prescription drug expenditures were not included. All of the expenditures were adjusted with the previously used method6 and were reported in 2015 $USD to control for variation over time.31
Independent Variables: The Andersen behavioral model of healthcare services utilization was used to determine the significant predictors of expenditures.32,33 This conceptual model has been used for decades to understand health services use in population-level studies. According to this model, healthcare utilization and expenditures are the function of the predisposition of individuals to use services, factors that enable or impede use, the need for care, healthcare use, and external healthcare environmental factors.
Predisposing factors consisted of age at BC diagnosis and race, and enabling factors included census tract median household income and percentage of people aged ≥25 years with ≥4 years of college education. Because person-level information on household income and education level are not available with both SEER-Medicare and WV Medicare cohorts, census tract information for these variables were used in the study. Need-related factors consisted of stage at diagnosis,34 tumor grade, estrogen receptor status, and comorbidity35,36 and mental conditions (depression and/or anxiety) derived from co-occurring chronic conditions in the 12 months before BC diagnosis. Healthcare use comprised the number primary care physician (PCP) visits in the year following BC diagnosis,37 categorized as high or low based on the median value; type of initial treatment in the year following BC diagnosis (definitive surgery only; nonsurgical treatment, such as chemotherapy or radiation therapy or both; definitive surgery + nonsurgical therapy; and no treatment); and inpatient use. External healthcare environmental factors comprised location of residence and number of hospitals offering oncology-related services in the area of residence, categorized as high or low based on the median value.
Statistical Analyses
Chi-square statistics were used to describe the significant differences between the study cohorts. Average Medicare expenditures and expenditures for specific services were statistically compared using t tests and ratio of means. Breusch-Pagan/Cook-Weisberg test and simplified White test were used to check for heteroscedasticity and kurtosis of log-scale residuals in the expenditures data.38 Park tests indicated that generalized linear model (GLM) with log link function and gamma distribution should be used for these data.39
Multivariable GLM regressions were separately conducted to model expenditures for both study cohorts. Also, GLM regression was conducted on average expenditures after controlling for setting (WV Medicare vs SEER-Medicare) and other independent variables, to determine significant differences in expenditures between the cohorts. Findings with P<.05 are discussed. All analyses were conducted within statistical analysis system software SAS 9.4 (SAS Institute Inc., Cary, NC) and Stata 13 (Stata-Corp 2013, College Station, TX).
Blinder-Oaxaca Decomposition Technique: A postregression Blinder-Oaxaca linear decomposition technique was used to estimate how much of the differences in average expenditures between the cohorts can be explained by differences in characteristics.40,41 This method has been widely used for decades to understand the disparities in healthcare utilization, and thereby expenditures, and access to care among vulnerable groups. For instance, a study by White-Means42 used this technique to evaluate the racial differences in the use of medical services among the disabled elderly US population. The Blinder-Oaxaca linear decomposition method uses the differences in means of independent variables and parameter estimates (betas) from the regressions to generate 2 components: the “explained” component, which provides differences in expenditures due to differences in observed characteristics between 2 cohorts, and the “unexplained” component, which provides differences in expenditures that could not be explained, either because of differences in the regression parameter estimates between the 2 cohorts or differences in unobservable/unmeasured independent variables (eg, provider level, organizational level variables). The explained portion was calculated by multiplying the differences in the average characteristics between the cohorts using pooled regression weights. The pooled regression weights are the coefficients of the characteristics from the regression model, which included WV Medicare versus SEER-Medicare as one of the independent variables.
Results
Descriptive Characteristics
Most of the WV Medicare cohort (97.6%) was white compared with 89% from the SEER-Medicare cohort (Table 1). A higher proportion of the WV Medicare cohort had a household income <$35,000 (91%) and resided in areas where <15% of the population had some college education (51%) and with a lower number of hospitals with oncology services (60%). However, most of the SEER-Medicare cohort resided in metropolitan areas (84%), had household income >$35,000 (75%), and resided in areas where >15% of the population had some college education (70%), and with a lower number of hospitals with oncology services (55%).
Average Medicare Expenditures and Expenditures by Types of Services: Average expenditures for the WV Medicare cohort were $25,626 compared with $29,502 for the SEER-Medicare cohort; a difference of $3,876 (Table 2). Average inpatient expenditures for the WV Medicare cohort ($6,070) were also significantly lower than those for the SEER-Medicare cohort ($6,775), with a ratio of means of 0.90. Average expenditures for physician services were significantly lower for the WV Medicare cohort ($11,197) compared with the SEER-Medicare cohort ($13,925), with a ratio of means of 0.80.
Factors Associated With Medicare Expenditures: In the WV Medicare cohort, women who were diagnosed with regional and distant BC stage, had surgery with adjuvant therapy or adjuvant therapy only, had inpatient visits, had ≥2 comorbidities, and mental health conditions had significantly higher expenditures during the initial phase of BC care (Table 3). In the SEER-Medicare cohort, women who resided in areas with higher education level, were diagnosed at advanced BC stages, did not have a well-differentiated tumor grade, had negative estrogen receptor tumor status, had any kind of treatment in form of surgery or adjuvant therapy, had inpatient
Patient Characteristics
Average Medicare Expenditures and Expenditures by Specific Services
Differences in Average Expenditures Between Study Cohorts: Compared with the SEER-Medicare cohort, average expenditures during the initial phase of BC care were significantly lower for the WV Medicare cohort by $708 in a multivariable regression (Table 4).
Factors Explaining Lower Average Expenditures in the WV Medicare Cohort: Using the pooled weights and logarithmic transformation of expenditures, there was a difference of −0.1678 in logarithmic terms. Of this difference, only −0.0266 or 15.85% of the difference in average Medicare expenditures between the cohorts was explained by the included beneficiary characteristics (Table 5). Enabling resources contributed 6.86%, healthcare use 7.55%, and external healthcare environmental factors 3.33% to the total explained portion. Detailed examination of the decomposition results revealed that type of initial treatment explained 12.25% of the differences in the average expenditures between the cohorts. This can be interpreted as follows: keeping all the other characteristics same, if the WV Medicare cohort had the same course of initial treatment as the SEER-Medicare cohort, then the WV Medicare cohort would result in lower expenditures. A total of 84.15% of the differences in average expenditures between the cohorts remained unexplained.
Discussion
To our knowledge, this study is first of its kind to determine Medicare spending during the initial phase of care among elderly women (aged ≥66 years) with BC residing in a rural and medically underserved region of WV. As expected, the populations studied were statistically different in terms of most of their characteristics; contrary to the hypothesis, average Medicare expenditures and expenditures for inpatient services, physician services, and other services in elderly WV women with BC were significantly lower than those for the SEER-Medicare cohort.
These findings are surprising given the fact that elderly WV women with BC have a higher comorbidity burden and possibly greater proportion of negative health outcomes, which should increase the Medicare expenditures. Indeed, a previous study reported lower Medicare expenditures among rural elderly women with BC, although it focused on end-of-life care only.43 In addition, some of the lower expenditures in the rural WV region compared with the SEER-Medicare regions may be due to the differences in Medicare spending across various geographic US regions.44 However, no study examined the extent to which independent variables explain the differences in expenditures in rural regions versus nationally. The unique contribution of this study is that it provides an assessment of the extent to which the constructs included in the Andersen behavioral model explained the expenditure differences using a WVCR-Medicare database and a nationally representative population from the SEER-Medicare database using an advanced Blinder-Oaxaca linear decomposition technique.
Average Medicare expenditures during the initial phase of BC care among elderly WV women were significantly lower by $3,876 compared with that for the SEER-Medicare cohort, which reduced
Factors Associated With Average Medicare Expenditures Using Generalized Linear Model Regressions
Differences in Average Medicare Expenditures Using Generalized Linear Model Regression
The Blinder-Oaxaca linear decomposition technique showed that 16% of the expenditure difference between cohorts was explained by the variables adjusted in the analyses. The enabling resources (household income and education) and healthcare use (PCP visits, type of initial treatment, uses of inpatient services) primarily explained 14% of these differences. These findings are consistent with those of previous studies reporting that the differences in healthcare spending are largely explained by different treatment patterns owing to the supply of physicians and hospital resources in the area, and also to household income and education.46–49 A noteworthy finding of this study is that 84% of the difference in Medicare expenditures between the cohorts remained unexplained. It is plausible that some portion of this difference may be due to variables such as health status, severity of comorbid conditions, body mass index, patient preferences, and propensity to seek care that were not captured in the database.
Another portion of the “unexplained portion” was assumed to be due to external factors, which may include differences in the proportion of women covered by Medicare managed care, Medicaid, and Medicare supplemental insurance in WV and the rest of the United States. Due to the unavailability of individual level data, state-level data of Medicare managed care penetration, dual enrollment in Medicare/Medicaid, and Medicare supplemental coverage were identified for WV and compared with the US rates. In 2006, Medicare managed care penetration rate was lower in WV than the national average (9%50 vs 16%51) and the percentage of Medicare beneficiaries with supplemental coverage was lower
Factors Explaining Lower Expenditures Using Blinder-Oaxaca Linear Decomposition Analysis
In elderly WV women with BC, factors associated with higher expenditures during the initial phase of BC care were advanced stages of disease, surgery with adjuvant therapy or adjuvant therapy only, inpatient use, presence of ≥2 physical chronic conditions, and mental conditions. But older age was associated with lower expenditures in these women compared with those aged 66 to 69 years. These findings were consistent with those reported in the previous study on the Virginia population,11 and highlight the importance of encouraging women to use preventive screenings to diagnose BC at earlier stages and curb higher Medicare spending. Additionally, the findings emphasize the importance of better comanagement of physical and mental chronic conditions in elderly women with BC in order to lower overall expenditures to Medicare in WV.
Findings from this study raise important policy issues and have implications for payers (Medicare), providers, and patients. There is a geographic variation in Medicare spending among elderly women that indicates the Centers for Medicare & Medicaid Services to evaluate if lower expenditures among patients with BC in rural WV translates to similar, better, or worse health outcomes (eg, survival, quality of life) compared with national averages, and accordingly develop and/or modify reimbursement policies that can help bend the ever-increasing cancer spending curve.
Several study limitations are worth noting. Healthcare services obtained outside of Medicare settings may not be captured, thereby underestimating the expenditures. The Andersen model has some limitations when applied to claims data. For instance, data on enabling factors (supplemental coverage), need-related factors (severity of comorbidities, health status), personal health practice (body mass index, physical activity, smoking), and patient preferences, which may impact selection of treatment and hence Medicare expenditures, were not captured. Additionally, the study findings are generalizable to elderly patients with BC only covered by Medicare within WV.
Conclusions
The difference in the average Medicare expenditures during the initial phase of BC care between elderly WV women and their national counterparts remained significantly lower after multivariate adjustment. The explained portion of this difference was mainly driven by enabling and healthcare use factors, whereas 84% of it remained unexplained.
The authors have disclosed that they have no financial interests, arrangements, affiliations, or commercial interests with the manufacturers of any products discussed in this article or their competitors.
This study was part of Ami Vyas' doctoral dissertation at West Virginia University. It was supported by AHRQ Grant (R24HS018622-03) and some additional salary support was received by Drs. Madhavan and Sambamoorthi from a National Institute of General Medicine Sciences IDeA Clinical and Translational Research Grant (U54GM104942) obtained by West Virginia University. The content is solely the responsibility of the authors and does not necessarily represent the official views of AHRQ and NIGMS.
This project was presented at the 20th Annual International Meeting of the International Society for Pharmacoeconomics and Outcomes Research; May 16–20, 2015; Philadelphia, PA.
References
- 1.↑
Mariotto AB, Yabroff KR, Shao Y et al.. Projections of the cost of cancer care in the United States: 2010-2020. J Natl Cancer Inst 2011;103:117–128.
- 2.↑
Brown ML, Riley GF, Schussler N, Etzioni R. Estimating health care costs related to cancer treatment from SEER-Medicare data. Med Care 2002;40(8 Suppl):IV-104–117.
- 3.
Campbell JD, Ramsey SD. The costs of treating breast cancer in the US: a synthesis of published evidence. Pharmacoeconomics 2009;27:199–209.
- 4.↑
Riley GF, Potosky AL, Lubitz JD, Kessler LG. Medicare payments from diagnosis to death for elderly cancer patients by stage at diagnosis. Med Care 1995;33:828–841.
- 5.
Warren JL, Brown ML, Fay MP et al.. Costs of treatment for elderly women with early-stage breast cancer in fee-for-service settings. J Clin Oncol 2002;20:307–316.
- 6.↑
Yabroff KR, Lamont EB, Mariotto AB et al.. Cost of care for elderly cancer patients in the United States. J Natl Cancer Inst 2008;100:630–641.
- 7.↑
Howlader N, Noone AM, Krapcho M et al., eds. SEER Cancer Statistics Review, 1975-2013, National Cancer Institute. Bethesda, MD, based on November 2011 SEER data submission, April 2012. Available at: http://seer.cancer.gov/csr/1975_2013/. Accessed April 21, 2016.
- 8.↑
Hanchate AD, Clough-Gorr KM, Ash AS et al.. Longitudinal patterns in survival, comorbidity, healthcare utilization and quality of care among older women following breast cancer diagnosis. J Gen Intern Med 2010;25:1045–1050.
- 9.
Xu X, Herrin J, Soulos PR et al.. The role of patient factors, cancer characteristics, and treatment patterns in the costs of care for Medicare beneficiaries with breast cancer. Health Serv Res 2016;51:167–186.
- 10.↑
Warren JL, Yabroff KR, Meekins A et al.. Evaluation of trends in the cost of initial cancer treatment. J Natl Cancer Inst 2008;100:888–897.
- 11.↑
Penberthy L, Retchin SM, McDonald MK et al.. Predictors of Medicare costs in elderly beneficiaries with breast, colorectal, lung, or prostate cancer. Health Care Manag Sci 1999;2:149–160.
- 12.↑
Behringer B. Health care services in Appalachia. In: Couto R, Simpson NK, Harris G, eds. Sowing Seeds in the Mountains: Community-Based Coalitions for Cancer Prevention and Control. NIH Publication No. 94-3779. Bethesda, MD: National Institutes of Health; 1994:62–80.
- 13.↑
Appalachia Community Cancer Network Web site. Available at: http://www.accnweb.com/. Accessed March 22, 2017.
- 14.↑
Kaiser State Health Facts. Overweight and obesity rates for adults by sex, 2009. Menlo Park, CA: Kaiser Family Foundation; 2009. Available at: http://www.statehealthfacts.org. Accessed April 21, 2016.
- 15.
Kaiser State Health Facts. Percent of adults who smoke by sex, 2008. Menlo Park, CA: Kaiser Family Foundation; 2008. Available at: http://www.statehealthfacts.org. Accessed April 21, 2016.
- 16.
United States Census Bureau. Persons 25 years old and over with a bachelor's degree or more, 2008. Washington, DC: US Census Bureau; 2010. Available at: http://www.census.gov/compendia/statab/2010/ranks/rank19.html. Accessed April 21, 2016.
- 17.
United States Census Bureau. Persons below the poverty level, 2008. Washington, DC: US Census Bureau; 2010. Available at: http://www.census.gov/compendia/statab/2010/ranks/rank34.html. Accessed April 21, 2016.
- 18.↑
Arcury TA, Preisser JS, Gesler WM, Powers JM. Access to transportation and healthcare utilization in a rural region. J Rural Health 2005;21:316–338.
- 19.↑
West Virginia Healthcare Authority. Broadband infrastructure in West Virginia healthcare. Available at: http://www.hca.wv.gov/policyandplanning/br/Documents/Broadband_Report_09_10.pdf. Accessed November 15, 2016.
- 20.↑
State Cancer Profiles. Incidence and mortality rate reports for breast cancer for West Virginia by county. Available at: http://statecancerprofiles.cancer.gov/. Accessed April 21, 2016.
- 21.↑
Centers for Disease Control and Prevention. United States Cancer Statistics. Rankings by state: 2006, male and female, all cancer sites combined. Available at: https://nccd.cdc.gov/USCS/cancersrankedbystate.aspx. Accessed April 21, 2016.
- 22.
Centers for Disease Control and Prevention. United States Cancer Statistics. State vs. national rates: 2006, female, West Virginia. Available at: https://nccd.cdc.gov/USCS/statevsnational.aspx. Accessed April 21, 2016.
- 23.
Lengerich EJ, Chase GA, Beiler J, Darnell M. Increased risk of unknown stage cancer from residence in a rural area: health disparities with poverty and minority status. Available at: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.160.1128. Accessed April 21, 2016.
- 24.↑
Wingo PA, Tucker TC, Jamison PM et al.. Cancer in Appalachia, 2001-2003. Cancer 2008;112:181–192.
- 25.↑
West Virginia Department of Health & Human Resources. Cancer incidence in West Virginia, 1993-2009. West Virginia Cancer Registry: 2012 Annual Report 2012. Available at: http://www.dhhr.wv.gov/oeps/cancer/Documents/WVCR%202012%20Annual%20Report_1_8_2013.pdf. Accessed April 21, 2016.
- 26.↑
Nadpara PA, Madhavan SS. Linking Medicare, Medicaid, and cancer registry data to study the burden of cancers in West Virginia. Medicare Medicaid Res Rev 2012;2:pii: 10.5600/mmrr.002.04.a01.
- 27.↑
Warren JL, Klabunde CN, Schrag D et al.. Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 2002;40(8 Suppl):IV-3–18.
- 28.↑
Engels EA, Pfeiffer RM, Ricker W et al.. Use of Surveillance, Epidemiology, and End Results-Medicare data to conduct case-control studies of cancer among the US elderly. Am J Epidemiol 2011;174:860–870.
- 29.↑
Taplin SH, Barlow W, Urban N et al.. Stage, age, comorbidity, and direct costs of colon, prostate, and breast cancer care. J Natl Cancer Inst 1995;87:417–426.
- 31.↑
United States Department of Labor. Consumer Price Index, medical care services, 1982-84. Available at: https://data.bls.gov/pdq/SurveyOutputServlet. Accessed April 21, 2016.
- 32.↑
Andersen R, Newman JF. Societal and individual determinants of medical care utilization in the United States. Milbank Mem Fund Q Health Soc 1973;51:95–124.
- 33.↑
Andersen RM. Revisiting the behavioral model and access to medical care: does it matter? J Health Soc Behav 1995;36:1–10.
- 34.↑
Young JL Jr, Roffers SD, Ries LA et al., eds. SEER summary staging manual-2000: codes and coding instructions, NIH Pub. No. 01-4969. Bethesda, MD: National Cancer Institute; 2001.
- 35.↑
Klabunde CN, Potosky AL, Legler JM, Warren JL. Development of a comorbidity index using physician claims data. J Clin Epidemiol 2000;53:1258–1267.
- 36.↑
National Cancer Institute. SEER-Medicare: Calculation of comorbidity weights. Available at: https://healthcaredelivery.cancer.gov/seermedicare/considerations/calculation.html. Accessed April 21, 2016.
- 37.↑
Yu X, McBean AM, Virnig BA. Physician visits, patient comorbidities, and mammography use among elderly colorectal cancer survivors. J Cancer Surviv 2007;1:275–282.
- 38.↑
Manning WG. The logged dependent variable, heteroscedasticity, and the retransformation problem. J Health Econ 1998;17:283–295.
- 39.↑
Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J Health Econ 2001;20:461–494.
- 40.↑
Blinder AS. Wage discrimination: reduced form and structural estimates. J Hum Resour 1973;8:436–455.
- 41.↑
Oaxaca R. Male-female wage differentials in urban labor markets. Int Econ Rev (Philadelphia) 1973;14:693–709.
- 42.↑
White-Means S. Racial patterns in disabled elderly persons' use of medical services. J Gerontol B Psychol Sci Soc Sci 2000;55:S76–89.
- 43.↑
Wang H, Qui F, Boilesen E et al.. Rural-urban differences in costs of end-of-life care for elderly cancer patients in the United States. J Rural Health 2016;32:353–362.
- 44.↑
Super N. The geography of Medicare: explaining differences in payment and costs. NHPF Issue Brief 2003:1–19.
- 45.↑
United States Department of Health and Human Services. Administration for Community Living. Administration on Aging. A profile of older Americans: 2014. Available at: http://aoa.acl.gov/Aging_Statistics/Profile/2014/8.aspx. Accessed November 20, 2016.
- 46.↑
Brooks GA, Li L, Sharma DB et al.. Regional variation in spending and survival for older adults with advanced cancer. J Natl Cancer Inst 2013;105:634–642.
- 47.
Fisher ES, Wennberg DE, Stukel TA et al.. The implications of regional variations in Medicare spending. Part 1: the content, quality, and accessibility of care. Ann Intern Med 2003;138:273–287.
- 48.
Fisher ES, Wennberg DE, Stukel TA et al.. The implications of regional variations in Medicare spending. Part 2: Health outcomes and satisfaction with care. Ann Intern Med 2003:138:288–298.
- 49.↑
Zuckerman S, Waidmann T, Berenson R, Hadley J. Clarifying sources of geographic differences in Medicare spending. N Engl J Med 2010;363:54–62.
- 50.↑
Table 12.8. Medicare Advantage and Other Private Health Plan Penetration, (Percent of Medicare Beneficiaries Enrolled), by Geographic Area: December 2006. Available at: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MedicareMedicaidStatSupp/Downloads/2007_Section12.pdf. Accessed November 22, 2016.
- 51.↑
The Henry J. Kaiser Family Foundation. Medicare Advantage. May 2016. Fact Sheet. Available at: http://files.kff.org/attachment/Fact-Sheet-Medicare-Advantage. Accessed November 22, 2016.
- 52.↑
Kaiser Family Foundation. Medicare Policy. Medigap: Spotlight on Enrollment, Premiums, and Recent Trends. April 2013. Available at: https://kaiserfamilyfoundation.files.wordpress.com/2013/04/8412-2.pdf. Accessed November 22, 2016.
- 53.↑
Centers for Medicare & Medicaid Services. Medicare-Medicaid Enrollee State Profile: West Virginia. Available at: http://www.integratedcareresourcecenter.com/PDFs/StateProfileWV.pdf. Accessed November 22, 2016.
