CRIW Measuring and Modeling Health Care Costs
October 18-19, 2013
Hitoshi Shigeoka, Simon Fraser University
Shigeoka exploits a sharp reduction in patient cost sharing at age 70 in Japan, using a regression discontinuity design to examine its effect on utilization, health, and financial risk arising from out-of-pocket expenditures. Because of the national policy, cost sharing is 60 to 80 percent lower at age 70 than at age 69. The author finds that both outpatient and inpatient care are price sensitive among the elderly. While he finds little impact on mortality and other health outcomes, the results show that reduced cost sharing is associated with lower out-of-pocket expenditures, especially at the right tail of the distribution.
Ralph Bradley, Bureau of Labor Statistics, and Colin Baker
While previous studies on obesity's effects on healthcare costs conclude that obesity increases costs, they do not control for the endogeneity of insurance and estimate a Tobit model for the corner solution when individuals have no medical expenditure. Baker and Bradley recognize that there are unobserved heterogeneous factors that guide choices on health insurance, body mass index (BMI) and visiting a provider. Therefore, neither health insurance nor BMI can be treated as exogenous in estimating a cost function and a Tobit model must be used to account for corner solutions when the individual does not visit a provider and incurs no medical costs. The authors find that obesity raises medical costs by $430.33, and that a 10 percent reduction in the BMI of each obese person would only lower costs by $45.28. The obesity elasticity with respect to cost is only .0115 percent.
Frank Lichtenberg, Columbia University and NBER
Lichtenberg examines the relationship across diseases between the long-run growth in the number of publications about a disease and the change in the mortality rate from the disease. The diseases analyzed are almost all the different forms of cancer, that is, cancer at different sites in the body (lung, colon, breast, etc.) The National Cancer Institute publishes annual data on cancer incidence as well as on cancer mortality, by cancer site. Failure to control for the growth in incidence (which is not feasible to do for non-cancer diseases) may bias estimates of the effect of publication growth toward zero, because growth in the number of publications is positively correlated across diseases with growth in incidence. Time-series data on the number of publications pertaining to each cancer site were obtained from PubMed. For articles published since 1975, it is possible to distinguish between publications indicating and not indicating any research funding support. The author's estimates indicate that mortality rates: 1) are unrelated to the (current or lagged) stock of publications that had not received research funding; 2) are only weakly inversely related to the contemporaneous stock of published articles that received research funding; and 3) are strongly inversely related to the stock of articles that had received research funding and had been published five and ten years earlier. The effect after ten years is 66 percent larger than the contemporaneous effect. The strong inverse correlation between mortality growth and growth in the lagged number of publications that were supported by research funding is not driven by a small number of outliers.
Brian Chansky, Corby Garner, and Ronjoy Raichoudhary, Bureau of Labor Statistics
The hospital industry has experienced significant operational and technological advances over the past two decades, but the Bureau of Labor Statistics (BLS) currently does not produce an industry labor productivity series that measures these gains. As with most service-providing industries, the difficulty in measuring hospital productivity lies in defining output because of the complex nature of the services provided as well as the atypical relationship between consumer and provider. Chansky, Garner, and Raichoudhary present initial research by the BLS on measuring productivity growth in private hospitals from 1993 to 2010. Three measures of hospital productivity are developed based on different output concepts. Two measures are based on volume of services provided, while the third is based on industry revenues adjusted for price change. Output of private hospitals includes both outpatient and inpatient care. Inpatient stays are more difficult to measure. These stays can be counted as single units of output where output is defined as the entire course of treatment, or they can be disaggregated into more detailed services, where each medical procedure is counted separately. The two alternative output measures based on volume of output - a course of treatment-based measure and a procedures-based measure - correspond to these two concepts of inpatient care. Additional factors such as the quality of care and the outcome of the treatment are also taken into consideration. Trends in output and labor productivity derived from each of the three models of hospital output are examined. The models show differing rates of positive long-term growth in hospital output and hospital productivity over the period of 1993 to 2010. For a model to be broadly accepted the data must be highly accurate and the definition of output must be compelling. To that end, the authors examine the accuracy and robustness of the various data sources used in each model. Although the output of the hospital industry can be measured in a number of ways, the authors argue that the most natural way to define the output of an industry is to answer the question: what services are the consumers buying? For hospitals, the authors conclude that the consumer is purchasing the overall course of treatment for a specific health problem, and therefore, counting overall courses of treatment is the most effective method of measuring output for private hospitals.
Jacob Glazer, Boston University; Thomas McGuire, Harvard University and NBER; and Julie Shi, Harvard University
Glazer, McGuire, and Shi develop and implement a statistical methodology to account for the equilibrium effects (adverse selection) in the design of risk adjustment formulas in health insurance markets. The authors' setting is modeled on the situation in Medicare and the new state exchanges where individuals are sorted into discrete sets of plan types (here, two). Their "Silver" and "Gold" plans have fixed characteristics, as in the well-known research on selection and efficiency by Einav and Finkelstein (EF). The authors build on the EF model in several respects, including by showing that risk adjustment can be used to achieve the premiums that will lead to efficient sorting. The target risk adjustment weights can be found by use of constrained regressions, where the constraints in the estimation are conditions on premiums that should be satisfied in equilibrium. The authors illustrate implementation of the method with data from seven years of the Medical Expenditure Panel Survey.
Paul Schreyer, OECD, and Matilde Mas,IVIE and University of Valencia
Health expenditure accounts for between 4 and 15 percent of GDP in OECD countries and constitutes one of the most important expenditure items. Schreyer and Mas review the concepts that the System of National Accounts foresees for the measurement of nominal and real output of health service providers and compare practices in OECD countries. There is significant variation in countries' methods although the degree of non-comparability is difficult to establish. The authors also present new estimates of cross-country comparisons of hospital service prices that are currently developed by the OECD. These draw on well-defined, representative case types that are costed across countries. This is made possible by an increasing number of countries that use diagnosis-related group (DRG) type approaches in health care administration. The increasing uptake of DRGs will also be instrumental in improving national accounts estimates of the volume and prices of health services.
Murray Aitken, IMS Institute for Healthcare Informatics; Ernst Berndt; Barry Bosworth, the Brookings Institution; Iin Cockburn, Boston University and Institute for Healthcare Informatics; and Bradley Shapiro, MIT,
Aitken, Berndt, Bosworth, Cockburn, and Frank examine six molecules facing initial loss of U.S. exclusivity (loss of exclusivity, LOE, from patent expiration or challenges) between June 2009 and May 2013 that were among the 50 most prescribed molecules in May 2013. They examine prices per day of therapy (from the perspective of average revenue received by retail pharmacy per day of therapy) and utilization separately for four payer types (cash, Medicare Part D, Medicaid, and other third party payer, TPP) and age under versus 65 and older. The authors find that quantity substitutions away from the brand are much larger proportionately and more rapid than average price reductions during the first six months following initial LOE. Brands continue to raise prices after generics enter. Expansion of total molecule sales (brand plus generic) following LOE is an increasingly common phenomenon compared with earlier eras. The number of days of therapy in a prescription has generally increased over time. Generic penetration rates are typically highest and most rapid for TPPs, and lowest and slowest for Medicaid. Cash customers and seniors generally pay the highest prices for brands and generics, TPPs and those under 65 pay the lowest prices, with Medicaid and Medicare Part D in between. The presence of an authorized generic during the 180-day exclusivity period has a significant impact on prices and volumes of prescriptions, but this varies across molecules.
Pinar Karaca-Mandic, University of Minnesota and NBER; Jean Abraham and Roger Feldman, University of Minnesota; and Kosali Simon, Indiana University and NBER
The Affordable Care Act (ACA) will dramatically alter health insurance markets and the sources through which individuals obtain coverage. In addition to expanding the size and importance of the individual market, establishment of Small Business Health Options Program exchanges in 2014 will simplify the health insurance shopping experience for small employers (50 or fewer full-time equivalent employees). The ACA also increases regulation of health insurers and health insurance markets, for example, by controlling premium increases through rate review regulation and by regulating insurers medical loss ratios (MLRs), which broadly represent the proportion of health insurance premium revenues that is paid out in medical claims. Monitoring the ACA as it is implemented will help uncover intended and unintended consequences of these regulations. To evaluate the changes in health insurance markets linked to the ACA, it is important to consistently measure the size and structure of health insurance markets, as well as the performance of participating health insurers, prior to and post-ACA. Karaca-Mandic, Abraham, Simon, and Feldman discuss challenges of describing the size, structure, and performance of the individual and small group markets. Next, they discuss improvements in data availability starting in 2010 to address some of these concerns. Finally, using data from the National Association of Insurance Commissioners (NAIC), they evaluate insurance market structure and performance during 2010 to 2012, focusing on enrollment, the number of participating insurers, premiums, claims spending, MLR, and administrative expenses. They provide a synthesis of the research available to measure and evaluate the size, structure and performance of the individual and small group markets. They discuss the availability and use of different data sets in measuring these concepts and highlight important measurement problems and possible solutions to consider when assessing the performance of health insurance markets as the ACA is fully implemented. Finally, they present new estimates from 2012 using the NAIC data.
The authors find that Federal household surveys give widely differing estimates of how many individuals were covered in the individual market prior to the ACA. This finding suggests it may be difficult to track changes in enrollment and to conduct studies based on a pre/post-ACA design using the Federal household surveys because of the limitations in properly estimating the size of the individual market at the baseline. Unlike in the individual market, there are better estimates of the small group market enrollment from the Medical Expenditure Panel Survey-Insurance Component (MEPS-IC).
The NAIC was the only source available to identify insurers operating in the individual and group markets until 2011. However, the NAIC data were quite limited until 2010, when major improvements occurred. The new NAIC exhibits allow for estimating participation of non-health insurers (for example, life insurers) in health insurance markets and provide a breakdown of the group market into small and large groups. Although the authors only have one "pre-ACA" year (2010) for early implemented ACA provisions such as the MLR regulation, they can make some assessments of ACA effects. Despite the fact that MLR measurement from the NAIC does not exactly match CMS's measurement of MLR for rebates, the NAIC data seem to perform well in predicting rebates.
Armando Franco, University of California at Berkeley; Dana Goldman, University of Southern California and NBER; Adam Leive, University of Pennsylvania; and Daniel McFadden, University of California at Berkeley and NBER
Administrative data such as insurance claims offer a potentially powerful data source to examine the relative benefits and costs of competing drug treatments. Motivated by a 2011 Food and Drug Administration (FDA) warning about possible side effects of angiotensin-II receptor blockers (ARBs), Franco, Goldman, Leive, and McFadden analyze the benefits and risks of ARBs compared to other classes of hypertension drugs using Medicare Parts A, B, and D data between 2006 and 2009. The authors study treatment adherence and crossover as well as non-random treatment assignment in detail and illustrate how different approaches to handling these issues impact their results. They find little evidence that ARBs increase cancer rates and weak evidence that they increase stroke rates, but falsification tests raise doubts that any associations are causal. Overall, their results suggest comprehensive, robust analyses are needed in using observational data for comparative effectiveness analysis.
Rena Conti, University of Chicago, and Ernst Berndt
Chris Stomberg, Bates White Economic Consulting
Anne Hall and Tina Highfill, Bureau of Economic Analysis
Disease-based medical care expenditure indexes are currently of interest to measurement economists and have been the subject of several recent papers. These papers, however, produced widely different results for medical care inflation and also varied in the datasets and methods used, making it difficult to compare them. In this paper, using two data sources and two different methods for calculating expenditure indexes for the Medicare population, Hall and Highfill compare the indexes produced and establish some results that may be helpful in choosing indexes for this population. First, they find that the two methods examined (primary diagnosis and a regression-based method) produce the same results for the aggregate index and have a moderate level of agreement in which diseases contribute the most to growth in per capita health care spending. Since the primary diagnosis method is more transparent, this implies that the regression-based method may be used when the data is not suitable for the primary diagnosis method without too great a loss of accuracy. Second, the authors find that the two data sources, the Medicare Current Beneficiary Survey (MCBS) and the Medical Expenditure Panel Survey (MEPS), produce very similar results in the aggregate but there is some evidence that they treat chronic illnesses differently. As the MCBS has a larger sample and more comprehensive coverage of Medicare beneficiaries than the MEPS,it seems that a regression-based expenditure index based on the MCBS may be more useful for fee-for-service Medicare beneficiaries. The authors discuss further avenues for research, such as comparing their results with indexes created with commercial groupers, and what data to use for Medicare private plan enrollees.
Partitioning medical spending into conditions is essential for understanding the cost burden of medical care and for developing strategies to improve the efficiency of the health care system. In this paper, Cutler develops a cost attribution method to attribute spending to conditions. His model uses propensity score methods to compare spending by people with a condition with otherwise similar people without a condition. He compares this propensity score stratification method to other proposed methods of cost estimation, including assigning each claim to a condition and using regression analysis in the entire sample of patients. The estimates show that the three methods have important differences in spending allocation, making the choice between them consequential
Abe Dunn and Eli Liebman, Bureau of Economic Analysis, and Adam Shapiro, Federal Reserve Bank of San Francisco
Laurence Baker and Kate Bundorf, Stanford University and NBER, and Anne Royalty, Indiana University
Didem Bernard and Thomas Selden, Agency for Healthcare Research and Quality, and Yuriy Pylypchuk, Georgetown Public Policy Institute