Healthcare Spending Index for Employer-Sponsored Insurance: Methodology and Baseline Results

Healthcare Spending Index for Employer-Sponsored Insurance: Methodology and Baseline Results Gary Pickens, PhD Elisse Moldwin, MA William D. Marder, PhD November 2010 White Paper

Table of Contents Healthcare Spending for Employer-Sponsored Insurance.................... 1 Background.............................................. 2 Data Sources and Methodology................................... 3 How representative is MarketScan data of the comparable U.S. population?......... 4 How important are the industry distribution differences between CPS and MarketScan data?..................................... 7 How representative is MarketScan data of healthcare spending in the comparable U.S. population?.................................... 9 Baseline Results........................................... 14 How does the HSI-ESI compare to other measures of healthcare spending and to growth in prices and the economy overall?......................... 16 References.............................................. 17

The Truven Health Analytics Healthcare Spending Index for Employer- Sponsored Insurance (HSI-ESI) measures historical and current levels of per capita spending on healthcare for individuals with private health insurance. The HSI-ESI estimates are based on the MarketScan Databases, which contain paid claims for inpatient and outpatient services, as well as outpatient prescription drugs. The current release of the HSI-ESI is confined to self-insured employers offering healthcare coverage to employees and dependents. Only non-elderly individuals and those with non-capitated health insurance coverage are included in the dataset used to compute the HSI-ESI. We estimate that in 2009 this segment of the privately insured population represented roughly 25 percent of the U.S. total population (300+ million). Healthcare Spending For Employer-Sponsored Insurance In 2002, an average of 4.9 million employees and dependents in the MarketScan Database were used to construct the HSI-ESI. By 2009, the number of employees and dependents used for the HSI-ESI had grown to more than 12 million. The HSI-ESI will be released quarterly, approximately 90 days after the end of the quarter in which healthcare services were used. The most recent four quarters of estimates are preliminary and are subject to revision in subsequent quarterly releases of the indexes. Methods of constructing final and provisional estimates are evaluated in this paper. The HSI-ESI will be reported at several levels. Per capita physician (professional), hospital (facility), drug, and total spending levels are reported. Out-of-pocket spending by consumers also will be tracked. The HSI-ESI measures levels of annual per capita spending on a rolling four-quarter basis. The baseline of the index (value of 100) is set at 2002 Q4 for all reported components. The initial index release, covering expenditures through 2010 Q2, indicates that of the reported components, hospital care has had the highest per capita spending inflation since 2002. Annual spending rates as of 2010 Q2 are highest for hospital care and lowest for prescription drugs. All of these important changes are compelling those involved to look at how they do business and make decisions through a new lens. The ability to mitigate risk and subsequently manage costs becomes paramount, and understanding a patient population is the first step in better managing its risk. Although there are several avenues in which organizations will need to maneuver to make the most of this developing landscape, the use of predictive analytics is becoming key by offering insights to identify and engage in actionable events. Healthcare Spending Index for Private Insurance 1

Background The pace of healthcare spending increases was a topic mentioned frequently during the national debate leading up to the passage of the Patient Protection and Affordable Care Act (PPACA) and the accompanying legislation in the spring of 2010. During this debate, predictions were made about the future growth of healthcare costs. Given the current state of the healthcare delivery system, the Centers for Medicare & Medicaid Services (CMS) Office of the Actuary estimated that national spending on healthcare would grow to $4.6 trillion by 2019, nearly 20 percent of the projected Gross Domestic Product (GDP). 1 A Truven Health white paper, written by Bob Kelley, senior vice president, analytics, and Ray Fabius, MD, concluded that maintaining the healthcare share of the GDP at 2009 levels (about 17 percent) would require a 5 percent absolute reduction in expenses (focused on waste) every year for the next decade. 2 Many other studies have also emphasized the importance of managing healthcare costs from a public policy viewpoint. Bending the healthcare cost curve is a key policy initiative from almost every angle. Studies on the growth in healthcare spending often highlight the importance of tracking inflation rates, especially as the PPACA legislation is implemented. The annual Milliman Medical Index tracks medical expenses for a family of four. 3 The Bureau of Labor Statistics (BLS) measures increases in consumer medical prices and various healthcare producer prices on a monthly basis. 4 The BLS measurements are extremely important for tracking increases in price inflation for a fixed market basket of goods and services but do not provide insight into the total impact of healthcare cost inflation, which is a product of unit price and service utilization. In addition, the BLS producer price indexes are based on samples of healthcare providers and services and may not provide a complete picture of healthcare cost inflation due to the limitations of the survey method of data collection. The Truven Health Analytics Healthcare Spending Indexes are a family of estimates that were created to provide timely and comprehensive information about various levels of healthcare expenditures in the United States. The indexes are based on prior work done by William D. Marder, Ph.D., and others. 5 The first of these indexes, described in this paper, focuses on the private health insurance market.

Data Sources and Methodology MarketScan Data Source The primary data source supporting the Truven Health Analytics Healthcare Spending Index for Employer-Sponsored Insurance is the MarketScan Commercial Claims Database. The dataset contains information on more than 130 million covered lives (more than 23 million in 2009 alone) and includes fully integrated medical and drug claims data at the patient level. The data is derived from the products and services Truven Health provides to self-insured employer and health plan customers. With 125 employers and 13 health plans contributing data during the past three years, the MarketScan Databases are one of the largest of this kind. The dataset includes information on all plan designs, as well as payment information on carve-outs, mail-order drugs, injectables, and patient copayments. Complete MarketScan Commercial Claims data are available annually through 2009. Current preliminary MarketScan claims are maintained in a series of quarterly Early View data releases. These datasets represent claims paid based on a 60-day lag between the service date and the date of the Early View data release. For example, the Early View release in September 2010 contains claims for services incurred and paid between January 2010 and July 2010. Early View contains all contributors and data available up to the creation date. However, it does not contain all claims for the service dates covered; claims with payments that have longer lag times are omitted. Adjusting the Early View data to ensure completeness is discussed in the following section on spending estimates. Accompanying the MarketScan Commercial Claims Database are the MarketScan Commercial Insurance Weights. These weights were constructed using the Household Component of the Medical Expenditures Panel Survey. 6 This survey, conducted by the Department of Health and Human Services (HHS) Agency for Healthcare Research and Quality (AHRQ), provides national estimates of the number of people with employersponsored insurance (ESI). These estimates were used to weight individuals in the MarketScan Databases to reflect the national ESI distribution. The following strata were used to construct the weights: Age (five groups: 0-17, 18-44, 45-64, 65-74, 75+) Gender (male, female) Region (northeast, north central, south, west) Metropolitan Statistical Area (MSA) classification (urban, rural) Relationship to the insurance policy holder (policy holder, spouse/dependent) Healthcare Spending Index for Private Insurance 3

MarketScan Representation The current release of the Truven Health Analytics Healthcare Spending Index for Employer-Sponsored Insurance (HSI-ESI) is limited to self-insured employers offering healthcare coverage to employees and dependents. Only nonelderly individuals and those with non-capitated health insurance coverage are included in the dataset used to compute the HSI-ESI. Based on information from the 2009 Medical Expenditure Panel Survey (MEPS) and 2009 Truven Health Insurance Coverage Estimates, we estimate that this population segment represents approximately 25 percent of the 2009 total U.S. population. Spending index sample sizes (quarterly averages) from the MarketScan Databases have grown substantially over time (Figure 1). We estimate that the 2009 sample size of just over 12 million lives represents more than 15 percent of the employer-sponsored, privately insured, non-elderly, non-capitated U.S. population. Figure 1: Covered Lives in the MarketScan Spending Index Sample: 2002 2009 15,000,000 10,000,000 5,000,000 0 2002 2003 2004 2005 2006 2007 2008 2009 How representative is MarketScan data of the comparable U.S. population? To determine the answer, MarketScan enrollee totals for 2009 were compared to weighted 2009 estimates from the March supplement of the Current Population Survey (CPS). 7 MarketScan enrollees were selected based on participation in employer-sponsored private insurance, age less than 65, and non-capitated coverage. CPS respondents were selected based on participation in employer-sponsored private insurance and age less than 65. Weighted and unweighted totals were computed from MarketScan data and compared to the weighted CPS totals for several key demographics: age, gender, industry, hourly/ salary status, and metropolitan residence. MarketScan weights were based on the MEPS. The weighted MarketScan data agreed closely with the CPS estimates for gender, hourly/ salary status, and metropolitan residence (Figures 3-5). There is reasonable agreement between the sources on age and census region (Figures 2, 6), as well. 4 Healthcare Spending Index for Private Insurance

Figure 2: Comparison of MarketScan and Current Population Survey Data, 2009: Age Groups 30% 25% 20% 15% 10% 5% 0% 00-17 18-24 25-34 35-44 45-54 55-64 CPS MarketScan Unweighted MarketScan Weighted Figure 3: Comparison of MarketScan and Current Population Survey Data, 2009: Gender 60% 50% 40% 30% 20% 10% 0% Female CPS MarketScan Unweighted MarketScan Weighted Male Figure 4: Comparison of MarketScan and Current Population Survey Data, 2009: Hourly/Salary 53% 52% 51% 50% 49% 48% 47% 46% 45% Hourly CPS MarketScan Unweighted MarketScan Weighted Salary Healthcare Spending Index for Private Insurance 5

Figure 5: Comparison of MarketScan and Current Population Survey Data, 2009: Metropolitan Residence 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% CBSA Residence CPS MarketScan Unweighted MarketScan Weighted Non-CBSA Residence Figure 6: Comparison of MarketScan and Current Population Survey Data, 2009: Census Region 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% 00-17 18-24 25-34 35-44 CPS MarketScan Unweighted MarketScan Weighted 6 Healthcare Spending Index for Private Insurance

Figure 7: Comparison of MarketScan and Current Population Survey Data, 2009: Industry Construction Oil/Gas/Mining Manufacturing Transportation/ Communication/ Utilities Retail/ Wholesale Trade Finance, Insurance, Real Estate Services 0% 10% 20% 30% 40% 50% 60% CPS MarketScan Unweighted MarketScan Weighted There are some substantial differences between the sources on industry of employment (Figure 7). In particular, the MarketScan Databases have a higher representation than CPS in the manufacturing and transportation/communications/utilities industries and a lower representation in service industries. How important are the industry distribution differences between CPS and MarketScan data? To assess the impact of the differences in the industry of employment category, we constructed a regression model using MarketScan 2009 total per capita spending and eligibility data based on age, gender, urban/ rural residence, and census region. We also included information on the industry and hourly/salary status of the employee policyholder. We used the same selection criteria discussed above for inclusion in the spending index sample. After restricting the sample based on data completeness and quality criteria, more than 7 million enrollees were included in the analysis. Age and gender are overwhelmingly the most important factors driving per capita spending in our sample. Industry and other factors play a much smaller role in explaining per capita spending variation. For example, in our analysis, the average spending per member per month (PMPM) was approximately $375. Figure 8 contains estimated spending by age group (controlling for gender, census region, metropolitan residence, industry, and hourly/salary worker status). Figure 9 contains estimated spending by industry group (controlling for all other factors). Age has a large impact on spending; variation across the age groups is over $500 PMPM. In contrast, the spending variation across industries is much smaller. The difference between the services and manufacturing industries, the largest discrepancy in the CPS and MarketScan data comparisons, is approximately $42 PMPM. It is likely that the industry distribution has a minor impact on the per capita spending estimates. Healthcare Spending Index for Private Insurance 7

Figure 8: Comparison of MarketScan and Current Population Survey Data, 2009: Industry Per Member Per Month Spending $800 $700 $600 $500 $400 $300 $200 $100 $0 <18 18-24 25-34 35-44 45-54 55-64 Figure 9: Comparison of MarketScan and Current Population Survey Data, 2009: Industry Retail Trade Construction Manufacturing, Nondurable Goods Manufacturing, Durable Goods Transportation/ Communication/ Utilities Finance, Insurance, Real Estate Oil and Gas Extraction, Mining Services $0 $50 $100 $150 $200 $250 $300 $350 $400 Per Member Per Month Spending 8 Healthcare Spending Index for Private Insurance

How representative is MarketScan data of healthcare spending in the comparable U.S. population? We compared per capita spending estimates from the MEPS 6 with corresponding estimates from the MarketScan Databases from 2002-2007. Only individuals less than 65 years of age, with employer sponsored insurance and non-capitated insurance coverage, were selected from both datasets. MarketScan and MEPS per capita spending estimates exhibit similar time trends for similar covered populations (Figure 10). Generally, MarketScan per capita spending estimates are higher than MEPS estimates, even when spending outliers are removed. This finding holds when spending estimates are decomposed into inpatient, outpatient, and ambulatory drug components (Figure 11). Aside from the differences in spending levels, MEPS and MarketScan data exhibit similar time trends and cross-sectional spending composition patterns. Figure 10: Comparison of MarketScan and Current Medical Expenditure Panel Survey Data, 2002 2007: Total Spending 4,500 4,000 3,500 3,000 2,500 2,000 1,500 1,000 500 0 2002 2003 2004 2005 2006 2007 MarketScan: Total MEPS: Total Figure 11: Comparison of MarketScan and Current Medical Expenditure Panel Survey Data, 2007: Spending Components 2,500 2,000 1,500 1,000 500 0 Inpatient Outpatient Drug MarketScan MEPS Healthcare Spending Index for Private Insurance 9

There are several reasons that MEPS produces lower estimates than MarketScan data. Some of these reasons have been documented by third parties: 8,9 MEPS staff has documented that it misses very expensive ($25,000+) events. Conversely, the size of the MarketScan sample allows us to capture rare, expensive events. MEPS staff has also documented under-reporting in separately billed laboratory services and durable medical equipment expenditures. Others have noted as much as a 17 percent shortfall in pharmaceutical spending among Medicare patients that may be indicative of wider under-reporting for prescription drug purchases. Individuals with extremely high medical and drug expenses have a large impact on per capita healthcare spending. The HSI-ESI uses all reported spending for enrollees meeting our spending index sample selection criteria. We measured the impact of spending outliers by creating a sample of enrollees with the highest 0.1 percent of total PMPM spending from 2009 MarketScan data. This group of enrollees contributes approximately 10.5 percent of total healthcare spending on an annualized basis. Excluding these outliers decreases total per capita spending by nearly 12 percent. Spending Estimates The HSI-ESI describes levels and trends of per capita healthcare spending and consequently depends on accurate per capita spending estimates. Conceptually, estimation of per capita spending is straightforward (the following discussion assumes that MarketScan enrollees selected on the basis of participation in employer-sponsored private insurance, age less than 65, and non-capitated coverage): Determine categories of spending for reporting Compute weighted total spending by category Compute weighted total enrollment Estimate per capita spending as the quotient of these two sums This technique is acceptable for historical MarketScan data (in practice, 2009 and earlier) where the lag between the incurred service date and the claims payment date (claims runoff) is not an issue. Data collected for more recent time periods requires some correction for lack of completeness. Fortunately, we can use historical MarketScan data to simulate the claims completion process and to evaluate alternate methods of estimating complete per capita spending. Claims completion distributions vary, depending on the type of service for the claim. Ambulatory prescription drug claims, for example, have very short incurred-payment lag times, due in part to electronic real-time connections between many retail pharmacies and insurance carriers (see Figure 12). Some medical outpatient claims complete very quickly (e.g., office visits for established patients), while other medical outpatient claims (e.g., repetitive medical therapies that are billed periodically) have longer mean completion times. Generally, hospital inpatient claims have the longest completion times. This is due to the time required to abstract the medical record for billing purposes and possibly more adjudication cycles for higher-priced hospital services. Regardless of the type of service, all claims are virtually complete after nine months have elapsed from the date of service. 10 Healthcare Spending Index for Private Insurance

Figure 12: Claims Completion Distributions by Type of Service: MarketScan Data 2008 Cumulative Percent of Total Payments 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0 1 2 3 4 5 6 7 8 9 Months Elapsed Between Service and Payment 2008-IP Hospital 2008-IP Physician 2008-OP Hospital 2008-OP Physician 2008-IP Drug Several methods of estimating complete per capita spending from incomplete MarketScan claims data were evaluated. All of these rely on complete historical MarketScan payment data and the ability to analytically create current payment data samples based on incurred service dates and payment dates. Incurred Date Method: Estimate complete incurred per capita expenditures based on incomplete incurred per capita expenditures in an incurred service quarter through regression. Paid Date Method: Estimate complete incurred per capita expenditure in a quarter based on payments in that quarter (regardless of service date) through regression. Completion Factor Method: Estimate complete incurred per capita expenditure by use of completion factors. A completion factor is the reciprocal of the claims runoff distribution (see Figure 13) for a given lag between date of service and date of reporting. Many completion factor variants were tested, and two are reported here: Employer-specific completion factors for each type of service Cohort-level completion factors for each type of service (no differentiation among employers) Time Series Method: Estimate complete incurred per capita expenditure by time series models fit to historical quarters of per capita spending, by type. The SAS Time Series Forecasting System was used to identify optimal models for each spending type (Holt-Winters Additive Smoothing models were selected for all spending types). 10 The models were used to forecast per capita spending for the future quarters using the historical series lagged four quarters. For example, to estimate spending for 2006 Q1, historical data from 2000 Q1 to 2004 Q4 were utilized.

Results of applying the various methods are summarized in Table 1. Mean absolute percentage errors (MAPEs) and other fit statistics were computed between the MarketScan complete per capita spending estimates and estimates generated by each method. The methods were tested on six quarters of historical data (2006 Q1, 2006 Q2, 2007 Q1, 2007 Q2, 2008 Q1, and 2008 Q2). Table 1: Mean Absolute Percentage Errors by Estimation Method and Spending Type Metric Spending Type Incurred Date Paid Date Time Series Comp. Factor (Employer Specific) Comp. Factor (Cohort) MAPE Total 3.3% 3.1% 4.9% 2.6% 0.8% MAPE Drug 7.8% 7.3% 7.5% 9.0% 3.5% MAPE Inpatient 2.6% 2.9% 2.3% 3.6% 2.9% MAPE Outpatient 3.9% 3.9% 6.1% 2.8% 1.6% None of the methods tested was a hands-down winner in predicting per capita spending. Regarding the completion factor method, there is no advantage in using employerspecific completion factors. Aggregate cohort completion factors by type of spend work just as well as those computed on an employer-specific basis. The paid date method performs fairly well but would require changes in MarketScan database operations to implement in a timely fashion. Of the three remaining methods (cohort completion factors, incurred date regression, and time series models) the time series models perform better in terms of Mean Squared Error, R-squared, and MAPE. The completion factor technique is attractive because it is used widely in practice and because it lacks the black-box attributes of the regression methods, especially the time series techniques. Also, the time series methods, since they are extrapolative, can provide misleading estimates when there is a shift in historical spending patterns. The completion factor method was tested in greater depth using datasets that provide a sequence of estimates for time periods 2008 Q4 and 2009 Q1, based on sequentially more mature MarketScan data. Completion factors were computed separately for inpatient facility, inpatient professional, outpatient facility, outpatient professional, and drug claims and were quarter-specific. 2007 MarketScan data was used to compute the completion factors. Figure 13 illustrates the sequence of completion factor estimates for services incurred in 2009 Q1 (results for 2008 Q4 are similar). Displayed in the chart are estimates starting with MarketScan data available in 2009 Q2 and moving forward in time. Note that there is some bias in the leading quarter (2009 Q2), especially for outpatient physician and hospital expense, which diminishes as MarketScan data release dates advance. 12 Healthcare Spending Index for Private Insurance

Figure 13: Completion Factor Estimate Sequence for 2009 Q1 Services 450 400 350 300 250 200 150 100 50 0 IP Hospital: Estimate OP Hospital: Estimate 2009 Q2 2009 Q3 2009 Q4 2010 Q1 MarketScan Data Release Date IP Hospital: Actual OP Hospital: Actual IP Physician: Estimate OP Physician: Estimate IP Physician: Actual OP Physician: Actual Based on these findings, we elected to construct quarterly per capita spending estimates based on a blend of methodologies. We found that the time series estimates are very accurate and are probably safe to use with relatively short forecast period length. We also found biases in the completion factor methods in the early quarters following the incurred service date. After much analysis, we were unable to identify a systematic source for this bias. Consequently, we elected to average the completion factor and time series estimates using the weighting system defined in Table 2. This weighting system gives 80 percent weight to the time series estimate in the MarketScan release quarter immediately following the incurred service. The time series weight decreases 20 percent with each subsequent quarter until five quarters have elapsed. At that point in time, claims runoff is virtually complete and the estimated per capita spending is based entirely on the weighted per capita estimate for the incurred service quarter. Table 2: Methodology Blending Weights MarketScan Release Quarter Time Series Weight Completion Factor Weight Incurred Quarter+5 0% 100% Incurred Quarter+4 20% 80% Incurred Quarter+3 40% 60% Incurred Quarter+2 60% 40% Incurred Quarter+1 80% 20% Healthcare Spending Index Computation There are many approaches that could be taken when converting the per capita spending estimates to index values. We elected to work with four quarter moving sums of per capita expense (moving annual totals or MATs). This has the advantage of handling seasonality, which is significant for some medical expenses, and also stating per capita spending on an annual basis. While other spending breakdowns are possible, we chose to publish index results for hospital, physician, drug, and total spending. Out-of-pocket expenses will be tracked as a separate breakdown of total expense. Note that premium expense is not a separate HSI-ESI component. We reserve the right to introduce other components or modify those published in future releases of the HSI-ESI. Healthcare Spending Index for Private Insurance 13

We elected to base our index on annual per capita spending as of 2002 Q4. We chose this date for several reasons. First, our ability to accurately identify facility versus professional claims using the MarketScan Databases improved substantially in 2002. Second, the period of time around 2000, which we also considered as a possible baseline, exhibited some historically aberrant rates of inflation in expenses, especially for prescription drugs. Basing the index components in 2002 Q4 provides a starting point more in line with recent historical spending inflation. Baseline Results Historically, hospital (facility) per capita spending has increased most rapidly in the period 2002 to 2010 Q2, with an index value of 174 in 2010 Q2 (an annualized rate of increase of 7.7 percent, see Figure 14 and Table 3). In 2004, increases in prescription drug spending matched those for hospital care but have slowed steadily since then, resulting in a value of 147 in 2010 Q2, the smallest of the components being tracked with an annualized rate of increase of 5.3 percent. The long-term increase in per capita spending on physician (professional) care was near that for total spending, with an index value of 157 in 2010 Q2, or an annualized increase of 6.2 percent. Table 3: Healthcare Spending Index for Employer-Sponsored Insurance: 2010 Q2 Spending Category 2009 Q2 Index 2010 Q1 Index 2010 Q2 Index Annual % Change Total 153 160 162 6.3% 1.3% Hospital 161 172 174 8.2% 1.5% Physician 149 155 157 5.5% 1.2% Drug 142 146 147 3.4% 0.8% Quarterly % Change Figure 14: Healthcare Spending Index for Employer-Sponsored Insurance 180 170 160 150 140 130 120 110 100 90 80 Total Index Hospital Index Physician Index Drug Index Preliminary 10/1/2002 1/1/2003 4/1/2003 7/1/2003 10/1/2003 1/1/2004 4/1/2004 7/1/2004 10/1/2004 1/1/2005 4/1/2005 7/1/2005 10/1/2005 1/1/2006 4/1/2006 7/1/2006 10/1/2006 1/1/2007 4/1/2007 7/1/2007 10/1/2007 1/1/2008 4/1/2008 7/1/2008 10/1/2008 1/1/2009 4/1/2009 7/1/2009 10/1/2009 1/1/2010 4/1/2010 Annual changes in the HSI-ESI can be used to estimate per capita healthcare spending inflation. There have been marked fluctuations in year-over-year changes in spending, as illustrated in Figure 15. Per capita spending inflation for prescription drugs increased dramatically prior to 2002 and was in fact the most rapidly increasing index in our series around 2004. Since then, the rate of increase has fallen, with the result that prescription drug annual spending inflation is the lowest of the components we are tracking in 2010 Q2 (3.4 percent, see Figure 15). In contrast, hospital per capita spending has increased 14 Healthcare Spending Index for Private Insurance

at the most rapid rate since mid-2007, with an annual inflation rate of 8.2 percent in 2010 Q2. Overall, we estimate the per capita healthcare spending for those covered by Employer-Sponsored Insurance is increasing at a rate of 6.3 percent annually in 2010 Q2. Figure 15: Annual Rate of Change in the Healthcare Spending Index for Employer-Sponsored Insurance 14% 12% 10% 8% 6% 4% 2% 0% Total Index Annual % Change Hospital Index Annual % Change Physician Index Annual % Change Drug Index Annual % Change Preliminary 10/1/2003 1/1/2004 4/1/2004 7/1/2004 10/1/2004 1/1/2005 4/1/2005 7/1/2005 10/1/2005 1/1/2006 4/1/2006 7/1/2006 10/1/2006 1/1/2007 4/1/2007 7/1/2007 10/1/2007 1/1/2008 4/1/2008 7/1/2008 10/1/2008 1/1/2009 4/1/2009 7/1/2009 10/1/2009 1/1/2010 4/1/2010 Quarter Beginning Per capita out-of-pocket spending includes payments made by healthcare consumers for copayments, deductibles, and coinsurance. Out-of-pocket estimates are available starting in 2002 and are presented here on an annual basis. The base of the index (value of 100) is set at 2002 Q4. Figure 16 compares total and out-of-pocket spending indexes by year from 2002-2009 and suggests that the consumer portion of total payments increased between 2002 and 2009. Figure 16: Healthcare Spending Index for Employer-Sponsored Insurance: Out-of-Pocket and Total Spending Spending Index Level (2002=100) 170 160 150 140 130 120 110 100 90 80 Total Index Out-of-Pocket Index 2002 2003 2004 2005 2006 2007 2008 2009 Healthcare Spending Index for Private Insurance 15

How does the HSI-ESI compare to other measures of healthcare spending and to growth in prices and the economy overall? Figure 17 relates annual rates of change in the HSI-ESI to the overall and medical consumer price indexes (Bureau of Labor Statistics, U.S. city average 4 ), as well as estimated percent change in U.S. GDP (Bureau of Economic Analysis 11 ). Generally, the HSI-ESI rate of inflation has been greater than annual change in either the CPI measure or GDP growth. (Note that the HSI-ESI captures changes in unit prices and services utilized, while the price indexes measure price changes for a fixed set of services and goods). In particular, the gap between healthcare spending inflation, as measured by the HSI-ESI and economic growth/inflation, is most noticeable in 2009, when the full effect of the recession was felt. Figure 17: Healthcare Spending Index for Employer-Sponsored Insurance: Comparison with Other Indexes and Economic Metrics 10% 8% 6% 4% 2% 0% -2% -4% HSI-PI: Total Index Annual % Change GDP: Annual % Change CPI: Annual % Change Medical CPI: Annual % Change 2003 2004 2005 2006 2007 2008 2009 16 Healthcare Spending Index for Private Insurance

References 1 Andrea M. Sisko, Christopher J. Truffer, Sean P. Keehan, John A. Poisal, M. Kent Clemens, and Andrew J. Madison. National Health Spending Projections: The Estimated Impact of Reform Through 2019. Health Affairs, October 2010; 29(10): 1933-1941. 2 http://factsforhealthcare.com/reduce/. 3 http://www.milliman.com/expertise/healthcare/products-tools/mmi/. 4 http://www.bls.gov/data/http://www.bls.gov/data/. 5 William D. Marder, Ernst R. Berndt, Larry Levitt and Joseph P. Newhouse. A New Approach to Measuring Increases in Private Health Expenditures, unpublished manuscript, September 2003. 6 Medical Expenditures Panel Survey, Agency for Healthcare Research and Quality, http://www.ahrq.gov/data/mepsix.htm. 7 http://www.census.gov/cps/. 8 Samuel Zuvekas and Gary Olin, Validating Household Reports of Health Care Use in the Medical Expenditure Panel Survey, Health Services Research 44 (5), 2009. 9 Ana Aizcorbe, Eli Liebman, Sara Pack, David M. Cutler, Michael E. Chernew, and Allison B. Rosen. Measuring Health Care Costs of Individuals with Employer-Sponsored Health Insurance in the U.S.: A Comparison of Survey and Claims Data, http://www. bea.gov/papers/pdf/measuring_health_care_costs_of_individuals_aizcorbe.pdf. 10 http://support.sas.com/rnd/app/ets/cap/ets_forecasting.html. 11 http://www.bea.gov/national/xls/gdpchg.xls. Healthcare Spending Index for Private Insurance 17

Get Connected Send us an email at info@truvenhealth.com or visit truvenhealth.com ABOUT TRUVEN HEALTH ANALYTICS Truven Health Analytics delivers the answers that clients need to improve healthcare quality and access while reducing costs. We provide market-leading performance improvement built on data integrity and empirical truth. For more than 30 years, our insights and solutions have been providing hospitals and clinicians, employers and health plans, state and federal government agencies, life sciences companies, and policymakers the facts they need to make confident decisions that directly affect the health and well-being of people and organizations in the U.S. and around the world. Truven Health Analytics owns some of the most trusted brands in healthcare, such as MarketScan, 100 Top Hospitals, Advantage Suite, Micromedex, Simpler, ActionOI, Heartbeat Experts, and JWA. Truven Health has its principal offices in Ann Arbor, Mich.; Chicago; and Denver. For more information, please visit truvenhealth.com. truvenhealth.com 1.734.913.3000 2015 Truven Health Analytics Inc. All rights reserved. All other product names used herein are trademarks of their respective owners. TRU 15667 0515