State-Level Projections of Cancer-Related Medical Care Costs: 2010 to 2020
Published Online: September 20, 2012
Justin G. Trogdon, PhD; Florence K. L. Tangka, PhD; Donatus U. Ekwueme, PhD; Gery P. Guy Jr, PhD; Isaac Nwaise, PhD; and Diane Orenstein, PhD
Healthcare costs continue to rise nationally and impose greater burdens on state budgets.1 Since cancer-related medical care costs constitute a substantial portion of overall US medical care costs,2-4 accurate projections of future cancer-related care costs are critical. Over the past 20 years, the cost of treating cancer has nearly doubled nationally.2,5 As a result of an aging population and more expensive cancer treatments, the national costs of cancer care are expected to increase significantly in the near future.6 Although previous increases in spending on cancer have occurred despite the decreases in cancer incidence rates and increases in average survival times for patients with many types of cancers,7 researchers have noted many opportunities to further improve cancer detection and treatment while controlling costs.8-10
To take advantage of these opportunities, state-administered insurance providers such as Medicaid and public healthcare providers such as the National Breast and Cervical Cancer Early Detection Program11 need state-level projections of future cancer care costs. Previous projections of cancer prevalence and cancer care costs have focused only on the national level.6 This study produces state-level projections of cancer care costs through 2020. While our goal is not to explain differences across states, our projections do reflect projected changes in the distribution of state residents by age and sex during this period. They provide a useful baseline against which to gauge the impact of current and future cancer policies and could be useful for budget allocations for investments in cancer prevention and early detection.
DATA AND METHODS
First, we generated estimates of the number of adults who had been treated for cancer and the average cost of their treatment by age group (18-44, 45-64, >65 years) and sex (male, female). The small number of children with cancer in our data prevented reliable estimates for children. Second, in our base projections, we assumed that the treatment rate for cancer in each of the 6 age-by-sex groups would remain constant and that the inflationadjusted initial average cancer treatment cost per person would increase at the same rate as Congressional Budget Office (CBO) projections of overall medical spending.12,13 Third, we generated state-level projections of the total number of adults who will be treated for cancer and the costs of their treatment by multiplying treated cancer prevalence and average costs by the Census-projected population of each demographic cell. Therefore, the projections reflect expected changes in the distribution of state residents by sex and age group but assume that there will be no policy changes that could affect cancer treatment costs. For example, the projections do not account for possible changes in national healthcare policies mandated by the Affordable Care Act.
Projections of the Annual Number of US Adults Treated for Cancer
To estimate the number of adults in each state who will be treated for cancer, we used cancer prevalence data from the 2004 to 2008 Medical Expenditure Panel Survey (MEPS)14 and the US Census Bureau’s projections of state population counts for 2010 through 2020. The MEPS, a nationally representative survey of the civilian noninstitutionalized population administered by the Agency for Healthcare Research and Quality, provides data on participants’ use of medical services and on the costs of those services. MEPS provides a single, consistent data source to link disease prevalence and expenditures. Medical conditions are identified in the MEPS medical condition files; we restricted our condition indicators to those for which respondents received care within the interview year. Medical conditions were classified using the
International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes based on self-reported conditions that were transcribed by professional coders. Cancer was defined using clinical classification codes 11 through 43 and 45, which group ICD-9-CM codes into related groups.15 We combined cancers of any site.
We estimated logit models for the probability of cancer treatment that controlled for survey year and survey participants’ age, sex, and region of residence (northeast, south, midwest, and west). We used stepwise regressions to identify significant interactions among these variables to be included in the models. The significant interactions in the stepwise regressions represent age-by-sex-by-region categories with enough sample and power to detect differences in cancer treatment rates. We estimated cancer treatment rates (ie, the proportion of the population treated for cancer) for US adults in each age/sex/region group using coefficients from the logit regressions and adjusted these estimates to account for the nursing home care population using data from the 2004 National Nursing Home Survey.
We used the projected state population counts for 2010 through 2020 generated in 2008 by the US Census Bureau on the basis of data from the 2000 Census.16 For each state, we multiplied the predicted percentage of people treated for cancer in each of the 6 age-by-sex categories by the projected number of state residents in the corresponding category for each year from 2010 through 2020. We then aggregated these projections to project the total number of people who will be treated for cancer in each state in each year.
Projections of Direct Medical Care Costs of Cancer
MEPS measures total annual medical spending, including payments by insurers and out-of-pocket spending by patients (copayments, deductibles, and payments for noncovered services). The costs captured by MEPS represent payments (not charges) from the payer to the provider. MEPS spending data are obtained through a combination of self-reports by the respondents and validation of the self-reports from payers.17
We projected direct cancer-related medical care costs of cancer in 6 steps. First, we estimated per-person medical costs as a function of cancer by using a 2-part regression model (a logit model to predict the probability of any expenditure and a generalized linear model with a gamma distribution and a log link to estimate total annual medical expenditures for people having any such expenditures). To choose among alternative nonlinear estimators, we used an algorithm recommended by Manning and Mullahy18 and found the generalized linear model was the most appropriate for the data. All regressions included the following variables: age; age2; sex; race/ethnicity; education; family income; other sources of health insurance; year indicators; and indicators for cancer, arthritis, asthma, back problems, congestive heart failure, chronic obstructive pulmonary disease, coronary heart disease, depression, diabetes, dyslipidemia, human immunodeficiency virus/acquired immunodeficiency syndrome, hypertension, injuries, other cardiovascular disease, other mental health/substance abuse, pneumonia, pregnancies, renal failure, skin disorders, and stroke.
Second, we calculated expenditures attributable to cancer by comparing predicted expenditures from the 2-part regression model. To ensure no double counting of expenditures for co-occurring diseases, we used the “complete classification” technique described in an earlier study.19 We treated each disease and combination of diseases as a separate entity and, for each unique combination of diseases, compared predicted expenditures with and without the disease(s) holding all else constant. For example, we treated cancer alone and cancer with hypertension as 2 different “diseases.” We then divided the total expenditures attributable to the combinations of diseases back to the constituent diseases using the parameters from the model to construct shares for each constituent disease within a combination (ie, a share of all cancer with hypertension disease costs that are attributable to cancer). The shares attribute a greater share of the joint expenditures to the disease with the larger coefficient in the main effect.19 We estimated per-person costs attributable to cancer for each age/sex/region category on the basis of coefficients from the national model.
Third, we used the 2004 National Nursing Home Survey and National Health Accounts to adjust our per-person cost estimates to account for nursing home spending. We assumed that average per-person, non-nursing home expenditures attributable to cancer were the same for the nursing home population as for the non-institutionalized population. Fourth, we used confidential MEPS data that identified the most populous 30 states and 9 Census divisions to generate state-specific per-person cost estimates. Sample sizes were not large enough for us to replicate the full analysis for each state. We regressed log (positive) medical expenditures on the variables in the model plus state/census division dummies. The coefficients on the dummies provided measures of the differences in average medical care costs across states that we used to scale the national estimates to make them state-specific.
PDF is available on the last page.