State variations in hospital expenditures: an ecological approach.(ORIGINAL PAPER)

Abstract Hospital expenditures vary across states both in terms of the levels and growth rates. Economic status, insurance coverage (or lack thereof), health risk factors, and demographic factors are used to explain these differences. Interestingly, the prevalence of poverty rates across states does not seem to be a good predictor of differences in hospital expenditures but the percent without health insurance does relate to higher hospital expenditures, when the factors listed above are all considered. Policy discussions about universal health insurance may be missing a point if better health care coverage resulted in lower hospital costs.

Keywords Hospital costs * The uninsured * Poverty rates

Introduction

Hospital costs continue to be the largest share of health care expenditures in the United States. In 2006 they represented 31% of total health care expenditures and actually were more substantial than that percentage because a portion of physician services (21%) and of prescription drug expenditures (10%) are hospital derived. (1) From 1999 to 2006, hospital costs grew at a 7.5% annual rate, while overall inflation in the United States measured by the CPI only increased by 2.5% per year. (2) An aging population is partly responsible for these increases, but other obvious and not-so obvious factors are also at work.

The purpose of this study is to investigate hospital expenditures at the state level in order to identify factors that differentiate states having higher than average hospital expenditure increases from those having lower than average hospital expenditure increases. Explaining why states differ with respect to hospital cost increases may shed light on why hospital expenditures are increasing at almost triple the rate of inflation. Two states, Alabama and New Mexico, experienced only 5% increases over this period, and 16 other had increases in the 6% range while six states (Oregon, Idaho, Vermont, New Hampshire, Nevada and Alaska) had increases in excess of 10% per year, and another nine had increases in the 9% range. (3) This kind of variation offers an opportunity to identify causes for differences in the rate of increase, and perhaps unexpected causes of hospital expenditure increases in general. Such discoveries may lead to helpful policies in controlling hospital expenditures.

Hospital expenditures by state are linked to factors that logically may drive hospital costs. For example, admissions and admissions squared (to allow for economies or diseconomies of scale), of course, drive total expenditures and control for population differences. Per capita income captures both a demand for hospital services impact and a cost of nursing services factor since states with higher state incomes pay higher nursing wages. Poverty levels in a state can depress low skill wages and can affect costs in an a priori ambiguous way. It may be that the poor receive fewer hospital benefits because of the potential inability to pay or it may be that more expenditure must be spent on the poor because their admittances are for more serious diagnoses. The type of insurance coverage (covered by Medicare, Medicaid, privately insured, or uninsured) also can affect hospital expenditures. It has been noted that the uninsured do not use emergency room services at a significantly different rate than the insured but that does not answer the effect of not having insurance on total hospital expenditures. (4) Two major health risks that are measurable at state levels, prevalence of smoking and diabetes, can also be expected to affect hospital costs. Finally, demographic factors, such as gender and ethnicity and age distribution, can potentially affect hospital expenditures. What follows is an attempt to link these factors to state hospital expenditures in order to discern the causes of the difference in hospital cost growth rates.

Methodology

Sample Data for this analysis come from a variety of sources. Hospital expenditures and hospital admissions are taken from the American Hospital Association's Statistics volume from 2005, which includes data from 1999 through 2003. (5) Per capita income, poverty rates, and the CPI are taken from Bureau of Labor Statistic sources. (6) Age variables, ethnicity rates, and gender proportions are taken from Census Bureau publications. (7) The prevalence of smoking at the state level is found in CDC Morbidity and Mortality Weekly Reports, (8) whereas the prevalence of diabetes is found in CDC National Diabetes Surveillance System reports. (9) Insurance coverage data is published in HCFA reports. Much of the data is available for 2004 and 2005 but at the time of estimation Medicaid data was only available through 2003. (10)

Variables Examined Real total hospital expenditures for inpatient and outpatient services but excluding nursing home care by state is the dependent variable. Using state data allows for a comparison of differences at the state level but also subjects the interpretation of results to the ecological fallacy--state differences do not imply individual differences. Some problems are avoided by using state data, for example individual hospital costs are generally skewed but the distribution of state costs is not. Also endogeneity problems that exist with individual hospital cost estimations are much less present when state data is employed. Total hospital expenditures reported by the AHA are deflated by the CPI.

This variable is related to the number of admissions in a given state and to the number of admissions squared, which term permits testing for economies or diseconomies of scale at the state level. Real hospital expenditures (RHEs) are also related to real per capita income per state, which allows for wealthier states to both spend more on hospital services and spend more on nursing care which itself raises hospital expenditures. RHEs are also related to prevalence of poverty in a state. The percent of those with incomes below the poverty line, through a variety of mechanisms, may lead to less money being spent per admission due to ability-to-pay issues but may also lead to more money ultimately being spent per admissions if the threshold for admitting the poor is higher for a given diagnosis than for the non-poor. RHEs are also related to Medicare and Medicaid insurance coverage and to the rate of the uninsured by state (the variable for those with private insurance is omitted to prevent exact multicollinearity among these four variables). With any estimation involving health care costs, there almost always is some kind of simultaneous bias found, and here it may exist in the relation between Medicaid coverage (at the states' discretion) and state hospital costs. Here it is claimed that Medicaid coverage affects hospital costs but causality may run the other way as well. Two other estimations were attempted using the Federal Medicaid Assistance Percentage (FMAP) and Medicaid coverage lagged 1 year as instruments for the current level of Medicaid coverage and results were not substantially different than those presented here. With 51 observations (the 50 states plus the District of Columbia) over 5 years, the resulting 200 and 55 observations present ample variation for parameter estimation and help reduce small sample biases possibly resulting from simultaneity problems.

In addition, RHEs are related to two major risk factors, the incidence of smoking and of diabetes, both of which can affect hospital costs. Three age variables are employed, but one serves a dual purpose. The percent of population over 85, the percent of a population over 65 (and thus covered by Medicare) and the percent of a population between 25 and 34 are used. Finally RHEs are related to the percent of a population that is female, the percent that are of African-American recent descent and the percent that are of recent Hispanic origin. No attempts, except for admissions, were made to allow for non-linearities of either the quadratic or cross-effect types. Possibly RHEs are affected if someone is poor and uninsured more so than the combined effects of being either poor or being uninsured, but with the number of variables used here, such an analysis would add so much serendipity that results would be correctly suspect since, with so many extra variables, eventually some variable would show significance.

Data Analysis Real hospital costs (hospital costs deflated by the CPI) per state from 1999 to 2003 are linearly related to the following variables: (A) Economic: real per capita income per state, percentage of each state whose income falls below the poverty line; (B) Insurance coverage: percentage of each state's population covered by Medicaid, percentage of each state's population covered by Medicare (measured by the percentage of each state over 65), and percentage of each state's population without health insurance; (C) Health risk factors: percentage of each state who smoke and percentage of each state who have type I or type II diabetes; (D) Demographic Factors: percentage of each state who are women, percentage of each state who are of recent African-American heritage, percentage of each state who are of Hispanic origin, percentage of each state whose age is 85 or greater and the percentage of each state's population who are between the ages of 25 and 34, and (E) Admission Rates: the number of hospital admissions per state and the number of those admissions squared. A pooled time-series cross-sectional estimation approach is taken. When dealing with panel data, an important choice is whether to employ a fixed effects or random effects model. With the fixed effects model, the effect of all omitted variables (some of which may not be measureable, such as cultural, legal, or political differences in the states that may affect hospital costs) are lumped into a single coefficient, the fixed constant of the regression. Here, the random effects model is chosen so that idiosyncrasies of the types listed above or other types are estimated as individual constants for each state.

Print Get the Mag Weekly Updates