HEALTH POLICY CENTER RES E A RC H RE P O RT Measuring Medicaid Service Utilization among Dual Medicare- Medicaid Enrollees Using Fee-for- Service and Encounter Claims T-MSIS Analytic Files Data Quality Kyle J. Caswell Timothy A. Waidmann Keqin Wei September 2021 ABOU T THE U RB AN I NS TI T UTE The nonprofit Urban Institute is a leading research organization dedicated to developing evidence-based insights that improve people’s lives and strengthen communities. For 50 years, Urban has been the trusted source for rigorous analysis of complex social and economic issues; strategic advice to policymakers, philanthropists, and practitioners; and new, promising ideas that expand opportunities for all. Our work inspires effective decisions that advance fairness and enhance the well-being of people and places. Copyright © September 2021. Urban Institute. Permission is granted for reproduction of this file, with attribution to the Urban Institute. Cover image by Tim Meko. Contents Acknowledgments iv Executive Summary v Medicaid Service Utilization among Dual Medicare-Medicaid Enrollees, Fee-for-Service and Encounter Claims 1 Data and Methods 2 Medicaid Data 2 Methods 4 Results 6 Part I: All OT and LT Files Claims among Dual Enrollees 6 Part II: Features of Claims among Selected Services 17 Main Findings and Conclusions 29 Notes 32 References 34 About the Authors 35 Statement of Independence 37 Acknowledgments This work was supported by Arnold Ventures. We are grateful to them and to all our funders, who make it possible for Urban to advance its mission. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders. Funders do not determine research findings or the insights and recommendations of Urban experts. Further information on the Urban Institute’s funding principles is available at urban.org/fundingprinciples. iv ACKNOWLEDGMENTS Executive Summary In this data quality report, we investigate the capacity of the 2018 Transformed Medicaid Statistical Information System Analytic Files (TAF) to measure medical services among people dually enrolled in Medicare and Medicaid. We examine services for which Medicaid is the primary payer among dual enrollees: long-term services and supports, including personal care, nonemergency transportation, and other home- and community-based services (HCBS); behavioral health care; and nursing home care. Our analysis produced several key findings: ◼ Five states, New York, South Carolina, Vermont, Utah, and Colorado, have relatively high rates of missing procedure codes for noninstitutional services among Other Services (OT) file claim lines, ranging from 11.5 to 42.1 percent. This raises concerns about the TAF’s ability to accurately measure HCBS, behavioral health services, and nonemergency transportation services in these states. ◼ Five states, Florida, Nebraska, Missouri, Massachusetts, and Hawaii, have relatively high rates of missing or invalid type-of-service codes among Long Term Care (LT) file claim lines, ranging from 21.0 to 98.6 percent. This raises concerns regarding the TAF’s ability to accurately measure nursing home services in these states. ◼ Among services studied on OT and LT file claim lines, beginning and end dates of service are of high quality across all states, except Virginia’s nursing home services in the LT file. ◼ In multiple states, utilization rates for the services of interest are extremely low and in some cases seemingly implausible, even in cases where the analysis of the input data elements needed to measure utilization does not identify quality issues. This underscores the challenges researchers face using the TAF data for specific applications. ◼ The TAF HCBS taxonomy code data elements are unusable. ◼ The type-of-service data element is very imprecise compared with procedure codes for measuring HCBS located on the TAF OT file. Overall, our findings suggest the data fields necessary to identify services commonly used by dual enrollees and paid for by Medicaid are of good quality. However, we identify several states with likely data quality problems based on individual data elements with missing or invalid values or implausibly low levels of implied utilization for specific services. These findings underscore the need to account EXECUTIVE SUMMARY v for both across- and within-state variations in data quality when developing research designs using the TAF. They also suggest TAF data will not support national-level analyses. vi MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Measuring Medicaid Service Utilization among Dual Medicare- Medicaid Enrollees Using Fee-for- Service and Encounter Claims To study those dually enrolled in Medicare and Medicaid (hereafter “dual enrollees”), having administrative data from both programs is vital. On November 8, 2020, the Centers for Medicare & Medicaid Services (CMS) announced the availability of new Medicaid administrative data files spanning calendar years 2014 to 2016 for the research community, namely the Transformed Medicaid Statistical Information System (T-MSIS) Analytic Files (TAF). Importantly, the TAF data are available in CMS’s Chronic Conditions Data Warehouse Virtual Research Data Center (CCW), meaning they can be linked to administrative Medicare data. This linkage opens a range of researchable questions for quantitative analysis, because both Medicare and Medicaid data are necessary to paint a complete picture of health care spending and utilization among dual enrollees. Data quality assessments and a TAF data quality website, DQ ATLAS, offer information by topic and state to users of TAF data.1 Though these resources are valuable, they focus on the Medicaid population more generally. In this work, we contribute to knowledge of the TAF’s data quality by specifically focusing on issues related to dual enrollees. Medicare-Medicaid dual enrollees represent a minority of Medicaid enrollees overall, meaning data quality issues identified in the more general Medicaid population may not apply to the subpopulation of dual enrollees. Also, some data features and data quality questions are specific to dual enrollees. We intend for the information in this report to provide researchers with a basis for understanding the most fundamental quality dimensions of TAF data on dual enrollees by state necessary for designing studies. In doing so, we do not seek to label a state’s data as either “usable” or “unusable”; where we identify issues, researchers may be able to pursue a work-around for their specific research design. This work is intended to identify cases where a work-around or further data quality investigation, at a minimum, is necessary for a specific application. In this data quality report, we investigate the capacity of the TAF Research Identifiable Files (RIF) data to measure the use of the types of services for which Medicaid is the primary payer among dual enrollees. Such services include nursing home care; behavioral health services; and long-term services and supports, such as personal care, nonemergency transportation, and other home and community- based services (HCBS). Given dual enrollees’ growing use of managed care, we investigate both fee- for-service (FFS) and managed-care encounter claims. This report has two main sections. In the first, we investigate the quality of all data elements needed to identify TAF claims representing the medical services of interest. In the second section, we apply the necessary logic to the data elements, create the Medicaid service data for the desired outcomes, and investigate the quality of the resulting utilization measures. Interested readers should refer to the glossary at the end of this report that is intended to facilitate interpretation of the terms commonly used in this report and two related companion reports on the quality of enrollment and Medicaid spending TAF data among dual enrollees (Caswell and Waidmann 2021; Caswell, Waidmann, and Wei 2021). Data and Methods Data on dual enrollment are taken from both the TAF and Medicare Master Beneficiary Summary File (MBSF), as described in our report on the quality of the TAF enrollment data (Caswell and Waidmann 2021). Throughout this work, we report statistics on people identified as dually enrolled in either the TAF or MBSF. We do so because in some states a significant number of dual enrollees identified in the MBSF are included in TAF data as Medicaid-only enrollees. Consequently, identifying dual enrollees using only the TAF would exclude these beneficiaries from analysis. See Caswell and Waidmann (2021) for detailed results on dual enrollment by state, as well as the inputs used from the TAF and MBSF to define dual enrollment. Here we describe the data elements from the TAF claims files used to investigate the utilization measures of interest. Medicaid Data We use the RIF version of the TAF for January 2018, the most recent RIF data available when we conducted this study.2 The following data elements are the basis for identifying HCBS, behavioral health, and nursing home care in the TAF data. We use the following data elements from the TAF’s Other Services (OT) base claim (header) file: 2 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS ◼ encrypted CCW beneficiary identifier (“BENE_ID”) ◼ submitting state ◼ CCW claim identifier ◼ bill-type code ◼ claim-type code ◼ claim beginning date of service ◼ claim end date of service We also use the following data elements from the OT line file: ◼ encrypted CCW beneficiary identifier (“BENE_ID”) ◼ submitting state ◼ CCW claim identifier ◼ type-of-service code ◼ line procedure code ◼ HCBS taxonomy code ◼ claim line beginning date of service ◼ claim line end date of service We use the following data elements from the Long Term Care (LT) base claim (header) file: ◼ encrypted CCW beneficiary identifier (“BENE_ID”) ◼ submitting state ◼ CCW claim identifier ◼ bill-type code ◼ claim-type code ◼ claim beginning date of service ◼ claim end date of service Lastly, we use the following data elements from the LT line file: MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 3 ◼ encrypted CCW beneficiary identifier (“BENE_ID”) ◼ submitting state ◼ CCW claim identifier ◼ type-of-service code ◼ claim line beginning date of service ◼ claim line end date of service Methods In the first section of this report we investigate the quality of the data elements needed to identify TAF claims that represent the services of interest by file. Specifically, we use the OT file data to investigate the completeness of data elements required to measure personal care services,3 nonemergency transportation services,4 other HCBS,5 and behavioral health services.6 And from the LT file we investigate the data elements needed to identify nursing home stays. 7 All service definitions are mutually exclusive. “Other HCBS” is a residual category; it includes all HCBS services except personal care and transportation services and therefore includes a wide range of services (e.g., home- delivered meals, adult day services, caregiver respite support, supported employment). 8 The HCBS category definitions are based on the taxonomy originally developed for CMS by Truven Health Analytics. However, we did not use the corresponding HCBS taxonomy code data elements in the TAF after discovering they were poorly populated and unusable. The OT file claims include procedure codes, located on OT claim lines, which are used to identify services among FFS claims and managed-care encounter claims. Therefore, the procedure code field must be populated to measure these outpatient services. However, we do not expect all OT file claims to have a procedure code. The OT file includes claims from both institutional and professional services (CMS 2020). All professional service claims should have a corresponding procedure code, whereas institutional service claims may not. Therefore, to assess whether the procedure code data field is appropriately populated, one must first limit the OT claims to professional services. The data element “bill-type code” is used to distinguish between these services, and we therefore use it as our first point of inquiry. Specifically, we investigate the following bill-type code features located on OT claim headers by state: 1. Invalid nonmissing values with length less than four and/or inappropriate values 4 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 2. Invalid nonmissing values equal to 099x 3. Unexpected values that should only be populated in LT or inpatient (IP) files (i.e., not on the OT file) 4. Missing values 5. Valid, nonmissing, and expected values (most basic proxy for institutional claims) We then investigate the rate of missing procedure codes on OT claim lines among claims submitted by providers of the services of interest by state. The first type of claim meeting this criterion is professional claims, defined as those with a missing bill-type code (CMS 2019). This definition is based on the fact that professional claims are typically submitted on CMS-1500 (FFS) or 837P (encounter) forms, which do not have a field for bill-type code (CMS 2020). The other claim types have a nonmissing bill-type code, indicating they were submitted using an institutional claim form (UB04 or 837I). The first such type of claim is submitted by home health agencies, a common provider of HCBS, which are not institutional services. These agencies often submit institutional claim forms with a bill-type code with 3 as the second digit (03xx). Finally, several OT claims have nonmissing yet invalid bill-type codes, indicating the provider is not a hospital, nursing facility, clinic, or residential facility. We interpret these cases, where both the second and third position of the bill-type code equal nine (099x), as explicitly reporting a missing value. This suggests that the claim is for professional services (CMS 2020). From the OT file we also investigate the data element “type of service” for valid and nonmissing values among FFS and encounter claims, because some researchers may be interested in defining services by type using this data field. That said, the TAF documentation makes clear that this approach may be problematic, particularly for comparisons across states. Specifically, the documentation states, “Users should be very cautious when using the type of service data element to systematically identify a specific service across different states without including information from other fields, such as procedure or revenue codes” (CMS 2020, 11). The final data element we investigate is “type of service” from LT file claim lines, which we use to define nursing home stays (i.e., codes 009 and 047). LT file claim lines exclude procedure codes that enable researchers to identify services with increased precision, unlike the OT file claims. That said, the services included in the LT file are much less varied than that those in the OT file, so less precision is necessary. We investigate the presence of valid LT file values (i.e., 009, 044-048, 050, 059, 133), nonmissing invalid values, and missing values. Though other “valid” type-of-service codes exist beyond those listed above, only the short list above should present in the LT file. Consequently, the presence MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 5 of “invalid” codes in the LT file could suggest states are submitting claims incorrectly, and that a complete accounting of services expected in IP and OT files may need to scan the LT files for misplaced claims. In the second section, we use procedure codes from OT file claims and type-of-service codes from LT file claims to identify the services of interest. We also investigate features of the resulting claim lines by state. First, we investigate the quality of the beginning and end dates of service, which are fundamental information required for measuring services within a reference period, or the intensity of services delivered. We create four mutually exclusive groupings for invalid, valid but likely inaccurate, and valid and likely accurate dates: 1. Missing beginning or end date (invalid) 2. Beginning date is later than end date (invalid) 3. End date is more than one year after begin date (valid, likely inaccurate) 4. Begin date is both no later than and less than one year from end date (valid, likely accurate) After that, we present utilization statistics for all the services of interest from FFS and encounter claims by state. The metric reported is the percentage of users for a given service, which equals the count of enrollees with one or more claims with a given type of service in the month, divided by all dual enrollees. Lastly, we illustrate two examples (using personal care and nonemergency transportation services) of the degree to which type-of-service codes identify the same OT file services as procedure codes.9 We present percentages from three-by-three cross tabulations, and the columns represent claims where procedure codes (1) define the service, (2) do not define the service (and are nonmissing), or (3) are missing. Results We present our results in two distinct parts. The first focuses on the quality of the data elements needed to identify specific services of interest, whereas the second focuses on the subset of claims that represent specific services. 6 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Part I: All Claims among Dual Enrollees in the OT and LT Files Here we present results on features of the data elements needed to define the services of interest, which are located in the OT and LT files yet represent only a subset of claims in these files. In other words, we investigate the data elements’ abilities to define and extract specific claims from all claims in the respective files. BILL-TYPE CODES IN OT FILE CLAIMS Figure 1 reports state-level results related to the quality of the bill-type code among all FFS and encounter claim headers in the OT file among dual enrollees. A priori, we anticipate some share of claims have valid (nonmissing and expected) values, which are a basic proxy for identifying an institutional provider, and other claims have missing values, which likely identify a professional service claim. Our main findings related to bill-type codes are as follows: ◼ In Illinois, all claim headers have missing values for bill-type code, suggesting no claims from institutional providers exist on the OT file. ◼ In Washington, a significant share of headers, 87.7 percent, have 9 in both the second and third positions, which is invalid. No claim headers have missing values. ◼ In California, the share of claim headers with unexpected codes is 31.6 percent. Unexpected codes identify, for example, inpatient hospital or nursing facility claims, which are valid yet should only present in LT or IP file claims (not OT file claims). ◼ California, New York, Tennessee, and Vermont are outliers, where less than 50 percent of claim headers have missing values, implying a relatively low rate of professional claims. For these states, it is possible bill-type codes are populated inappropriately, which makes distinguishing professional claims from institutional claims in the OT file imprecise. This has implications for our evaluation of the rate of missing procedure codes among professional claims, as discussed in the next section. Washington, however, appears to be a special case. We consider it very likely that the bill-type code data field for professional claims in Washington is populated with “099x” to explicitly communicate a missing value. Indeed, CMS also observed this in Washington in previous work.10 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 7 FIGURE 1 Bill-Type Code in Fee-for-Service and Encounter Claims in the TAF’s OT File, January 2018 Missing Valid (nonmissing and expected) Unexpected (LT and IP codes) Invalid (099x) Invalid values IL UT NV SC MD AZ RI VA PA FL OH TX KS OK NH NC MO MS MI AK AL GA ID DE SD AR OR LA WY KY CT HI MN WV NM IN CO NE NJ ND IA WI DC MA MT ME VT TN NY CA WA 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. 8 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Notes: TAF = T-MSIS Analytic Files. OT = Other Services. LT = Long Term Care. IP = Inpatient. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. Claim headers are limited to fee-for-service and managed-care encounters. MISSING PROCEDURE CODES AMONG NONINSTITUTIONAL SERVICE CLAIMS IN THE OT FILE In this section, we report the rate of missing procedure codes in the OT file among claim lines identified as noninstitutional services. Procedure codes are necessary for identifying services of interest in OT file claims. Figure 2 reports the rate of missing procedure codes among FFS claims in OT file claims. For brevity, we report results only for states with a rate of missing procedure codes at or above 0.1 percent. Though researchers must decide what rate of missing procedure codes is acceptable for their application, we shed light on states that have varying degrees of missing values: ◼ In Vermont and New York, more than 20 percent of noninstitutional FFS claim lines have missing procedure codes, ranging from 20.6 to 26.9 percent. ◼ In South Carolina, Utah, and Colorado, rates of missing procedure codes range from 11.5 to 16.1 percent. ◼ In Texas, Oklahoma, and Iowa, rates of missing procedure codes range from 5.2 to 7.5 percent. ◼ In Wyoming, California, and Kentucky, rates of missing procedure codes range from 1.0 to 2.0 percent. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 9 FIGURE 2 Missing Procedure Codes among Noninstitutional Fee-for-Service Claims in the TAF’s OT File, by State, January 2018 VT NY SC UT CO TX OK IA WY CA KY SD IL AL NC NJ WI LA KS VA PA MD WV AR MO CT 0% 5% 10% 15% 20% 25% 30% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. OT = Other Services. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. Figure 3 reports the rate of missing procedure codes among noninstitutional encounter claims. We report only on states with missing rates at or above of 0.1 percent. Our main findings are as follows: ◼ In New York, the rate of missing procedure codes in noninstitutional encounter claims is 42.1 percent. ◼ In South Carolina, the rate of missing procedure codes is 15.3 percent. 10 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS ◼ In Oregon, Utah, Pennsylvania, Wisconsin, and Illinois, rates of missing procedure codes range from 1.1 to 3.5 percent. FIGURE 3 Missing Procedure Codes among Noninstitutional Encounter Claims in the TAF’s OT File, by State, January 2018 NY SC OR UT PA WI IL MA WV IA OH VA MD TX CO KY WA MS MN KS NE 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. OT = Other Services. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. In the previous section, we identified New York, Vermont, and Illinois as states where bill-type codes may be problematic for identifying noninstitutional claims; namely, they have relatively low rates of missing values, implying a relatively low rate of professional claims. Overall, these findings suggest procedure codes can be used to identify relevant OT services for dual enrollees generally. However, researchers interested in producing national statistics or statistics for the states identified above with relatively high (or nonzero) rates of missing procedure codes should further investigate the sources of this variation. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 11 TYPE-OF-SERVICE CODES IN OT FILE CLAIMS Here we report the proportion of OT claim lines with valid, invalid, and missing type-of-service codes among dual enrollees in January 2018. Some researchers may consider the type-of-service data element a substitute for procedure codes in identifying specific services of interest. For example, depending on the research question, the added specificity of procedure codes relative to type-of- service codes may be unnecessary. Figure 4 shows results for FFS claim lines. For brevity, we only report on states with a rate of valid codes at or below 99.9 percent. The main findings are as follows: ◼ In Missouri, Virginia, Wisconsin, and Pennsylvania, rates of FFS claim lines with missing type- of-service values are relatively high, ranging from 21.1 to 48.0 percent. ◼ In Oklahoma, the rate of missing type-of-service values is 5.3 percent. ◼ In Colorado, the rate of invalid type-of-service values is 5.4 percent. ◼ In Oregon, the rate of invalid type-of-service values is 1 percent. FIGURE 4 Type-of-Service Codes among Fee-for-Service Claims in the TAF’s OT File, January 2018 Valid Missing Invalid MO VA WI PA OK CO OR ND CA LA MN DE FL 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. OT = Other Services. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. 12 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Figure 5 reports results for managed-care encounter claims. We report on states with shares of valid claims at or below 99.9 percent and statistics representing at least 100 enrollees. We find the following: ◼ In Massachusetts, Wisconsin, and Iowa, rates of missing type-of-service values in managed- care encounter claims range from 29.8 to 99.2 percent. ◼ In Pennsylvania, Virginia, Idaho, Missouri, and Nebraska, rates of missing type-of-service values range from 2.7 to 8.9 percent. ◼ In California, Colorado, and Idaho, rates of invalid type-of-service values range from 2.5 to 6.3 percent. FIGURE 5 Type-of-Service Codes among Encounter Claims in the TAF’s OT File, January 2018 Valid Missing Invalid MA WI IA PA ID MO CA VA CO NE OR DE FL 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. OT = Other Services. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. Overall, the type-of-service data element is potentially of poor quality in only a few states, and these concerns may or may not be limited to only FFS or encounter claims in a given state. For MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 13 example, more than 99.0 percent of encounter claims in Massachusetts have missing values for type of service, but no FFS claims had missing or invalid type-of-service codes. TYPE-OF-SERVICE CODES IN LT FILE CLAIMS In this section, we report the proportion of LT claim lines with valid, invalid, and missing type-of- service codes among dual enrollees in January 2018. LT file claim lines do not contain procedure codes, leaving researchers to identify nursing home says among dual enrollees using the type-of- service data element. Figure 6 shows results for FFS claim lines. We report only on states with shares of valid claim lines at or below 99.9 percent and statistics representing at least 100 enrollees. The main findings are as follows: ◼ In Florida, the rate of invalid type-of-service codes in LT file claim lines, which should only present in OT or IP file claims, is 77.5 percent. ◼ In Nebraska and Missouri, rates of missing type-of-service values are high at 22.0 and 21.0 percent. ◼ In Texas, North Dakota, and Louisiana, rates of nonmissing, invalid codes range from 1.3 to 3.7 percent. 14 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS FIGURE 6 Type-of-Service Codes among Fee-for-Service Claims in the TAF’s LT File, January 2018 Valid Missing Invalid FL NE MO TX ND LA NM WV UT WI DC 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. LT = Long Term Care. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. Figure 7 reports results for encounter claims in states with shares of valid type-of-service codes at or below 99.9 percent and statistics representing more than 100 enrollees. The main findings are as follows: ◼ In Massachusetts, the rate of missing type-of-service codes in LT file claim lines is 98.6 percent. ◼ In Florida, the rate of invalid type-of-service codes is 84.2 percent. ◼ In Hawaii, the rate of missing type-of-service codes is 30.1 percent. ◼ In Minnesota, the rate of invalid type-of-service codes is 2.2 percent. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 15 FIGURE 7 Type-of-Service Codes among Encounter Claims in the TAF’s LT File, January 2018 Valid Missing Invalid MA FL HI MN NM AZ 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. LT = Long Term Care. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. These findings suggest researchers aiming to study nursing home stays among dual enrollees using type-of-service codes from the LT file in the states identified above or nationally should consider the implications of missing and/or invalid values. For example, Florida has an extremely low rate of valid codes among both FFS and encounter claims, and we find that the majority of claim lines in the file include codes that should only present in the OT and IP files. Therefore, further investigation would be necessary to determine whether these claims were simply categorized in the incorrect files, or if the claims themselves are faulty or unusable. High rates of missing type-of-service codes present a different problem than invalid codes: researchers hoping to use type-of-service codes would need to investigate other features of these claims to determine, for example, whether the encounter data are usable in Massachusetts and Hawaii, or whether the FFS data in Nebraska and Missouri are usable. 16 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Part II: Features of Claims among Selected Services Whereas the first part of the results section above analyzed all claims among dual enrollees in the OT and LT files, this section focuses on the subset of claims that represent the specific medical services of interest among dual enrollees and the data elements necessary to identify such claims. DATES OF SERVICE FOR HCBS AND BEHAVIORAL HEALTH SERVICES IN THE OT FILE Conditional on the HCBS (including personal care and nonemergency transportation) and behavioral health services of interest in the OT file, all claim lines have valid and likely accurate service beginning and end dates in all states (rounding to the nearest tenth of a percentage point). For brevity, we do not report these results. DATES OF SERVICE FOR NURSING HOME STAYS IN THE LT FILE Among all nursing home claim lines, we found only two states where the rate of valid and likely accurate beginning and end dates of service is below 100 percent (data not shown). ◼ In Virginia, the share of claim lines with valid and likely accurate service dates is 79.7 percent, and the remaining 20.3 percent of claim lines have missing beginning or end dates of service. ◼ In Kansas, the share of claim lines with valid and likely accurate service dates is 99.5 percent, which is very high for most research purposes. UTILIZATION ESTIMATES Here we present the rate of any service use among all dual enrollees by state for each of the five service types of interest. We combine FFS and encounter claims. States have varying service utilization rates for several reasons: Utilization rates may relate to data quality, possibly reflecting the data issues identified above. However, such rates may also relate to the wide variation in state policies on the provision of these services. Thus, varied utilization rates may not reflect differences in data quality. This is especially true for optional services under the Medicaid statute and services provided under waivers. However, it is beyond the scope of this report to identify and discuss each state’s policies. Rather, the subsequent tables intend to shed light on how the results presented above may affect utilization estimates. The tables also provide illustrative examples of how the data elements studied above are combined to measure utilization by state. Figure 8 reports personal care utilization by state. We observe that in California, Maryland, Nebraska, Pennsylvania, Utah, and Vermont, less than 0.5 percent of dual enrollees have one or more claims for personal care services, as measured here. These findings are consistent with data reported MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 17 elsewhere in certain states where personal care services are not offered or rarely used. In California, however, other data sources show higher rates of personal care utilization. 11 This suggests TAF data for some states may be missing services, and further investigation may identify these claims in other ways. 18 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS FIGURE 8 Personal Care Service Use, January 2018 WA TX MO AK WY NM KS DC NH OK MN MS IN NJ NC WI MT AR RI ID NV MA CO GA NY LA ME SC IA MI WV AZ TN OH IL DE ND HI VA SD KY FL OR AL CT UT VT CA MD PA NE 0% 5% 10% 15% 20% 25% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 19 Figure 9 reports nonemergency transportation use by state. We observe that in Utah, Pennsylvania, and South Carolina, 1 percent or less of dual enrollees have one or more nonemergency transportation service claims. One potential reason for such low rates of utilization is that many states, such as South Carolina,12 use private brokers to arrange transportation and delivery. If states do not require brokers to submit event- or beneficiary-level claims, the TAF data may exclude the data necessary to measure nonemergency transportation. 20 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS FIGURE 9 Nonemergency Transportation Use, January 2018 MN MA WA AZ RI ME NY IN AK WI ID CO NE NJ MO TN OK AR OH OR LA IL KS SD DE HI IA KY MS CT NV NC WY WV GA MT ND FL NH TX MD VT AL CA NM DC MI VA SC PA UT 0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 21 Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. Figure 10 reports other HCBS use by state. We observe that in Vermont, 0.4 percent of dual enrollees have one or more other HCBS claims. This low rate likely results from incomplete data. Vermont has a substantial HCBS waiver program, Choices for Care, that provides a range of services.13 Figure 11 reports behavioral health use by state. Unlike the other services discussed above, no state has extremely low utilization rates for behavioral health services. 22 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS FIGURE 10 Other HCBS Use, January 2018 MN WY ID IA KS MO IL CO WI NH SD AK NJ DE MA OK IN SC RI WV AZ MS HI OH CT FL ME GA MI WA UT NV MT AR TX KY TN AL DC MD ND VA NM NC CA OR NE LA PA NY VT 0% 5% 10% 15% 20% 25% 30% 35% 40% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: HCBS = home- and community-based services. TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 23 FIGURE 11 Behavioral Health Service Use, January 2018 VT IA MA ID NH DC MN MI RI WY ME UT NE NJ WV MD SD IN OH MT IL MO PA OK AZ AR KS HI WA AK CT SC CA WI NC OR NV ND NM AL CO VA KY LA TN DE MS TX NY FL GA 0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. 24 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Figure 12 reports nursing home care utilization by state. We observe that in North Carolina, Hawaii, Nevada, and New Hampshire, 0.8 percent or less of dual enrollees have one or more nursing home claims. Though nursing home care use is known to vary considerably across states, the low rates seen in these states likely result from incomplete data. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 25 FIGURE 12 Nursing Home Care Use, January 2018 ND NE WY KS IA MO SD IN OH RI PA OK AR MD MN MT NJ CT LA WV DE IL NY TX KY CO UT MS GA MI AL MA TN DC SC WA VT FL CA WI NM OR AZ VA AK ME ID NH NV HI NC 0% 5% 10% 15% 20% 25% 30% URBAN INSTITUTE Source: Authors' calculations using administrative TAF RIF data. 26 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. EXAMPLES OF USING TYPE-OF-SERVICE CODES TO IDENTIFY OT FILE SERVICES As discussed above, researchers may consider using type-of-service codes to identify medical services in the OT file in lieu of procedure codes. In this final section, we provide two examples, for personal care and nonemergency transportation, that illustrate the overall accuracy of this approach at the national level. Table 1 shows the personal care services example and reports results separately for FFS and encounter claims. Using procedure codes, the columns show whether claims are identified as personal care services, not identified as personal care services, or have missing procedure codes. Using type-of- service codes, the rows similarly identify whether the same claims are identified as personal care services, not personal care services, or have a missing type-of-service code. The table cells where rows and columns intersect show column percentages; the denominator equals the number of claims identified using the procedure code definition in the column, and the numerator equals the number of claims identified using type-of-service codes in the rows. We report FFS and encounter claim results separately within each table. Among FFS claims identified as personal care services using procedure codes, 61.1 percent are also identified as personal care services using type-of-service codes, 37.1 were not identified as personal care services, and the remaining 1.8 percent had a missing type-of-service code (table 1, top panel). Said differently, among claims we believe to be personal care services, less than two-thirds are also considered personal care services using type-of-service codes. Among claims explicitly not identified as personal care services using procedure codes, 4.6 percent are (incorrectly) identified as personal care services using type-of-service codes, 91.3 percent are (correctly) not identified as personal care services, and the remaining 4.1 percent have missing type-of-service codes. The results for encounter claims (table 1, bottom panel) are less encouraging, however: less than one-quarter of personal care encounter claims identified using procedure codes are also identified as such using type- of-service codes. Table 2 reports similar results for nonemergency transportation. Among FFS claims, 64.6 percent of nonemergency transportation services identified using procedure codes were also identified using type-of-service codes (top panel), whereas a slightly greater share, 73.6 percent, were correctly identified using encounter claims (bottom panel). Overall, the results from both tables suggest that using type-of-service codes alone is an imprecise way to identify claims of interest. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 27 TABLE 1 Identification of Personal Care Services Using Procedure Codes versus Type-of-Service Codes, January 2018 Percent Procedure Code Personal care Not personal care Missing Fee-for-service claims Type of service Personal care 61.1 4.6 2.5 Not personal care 37.1 91.3 96.7 Missing 1.8 4.1 0.9 Encounter claims Type of service Personal care 22.7 1.4 0.0 Not personal care 74.3 92.8 98.6 Missing 3.0 5.7 1.4 Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. TABLE 2 Identification of Nonemergency Transportation Services Using Procedure Codes versus Type-of- Service Codes, January 2018 Percent Procedure Code Transportation Not transportation Missing Fee-for-service claims Type of service Transportation 64.6 1.8 0.0 Not transportation 34.5 94.3 99.1 Missing 1.0 3.9 0.9 Encounter claims Type of service Transportation 73.6 5.2 0.0 Not transportation 17.9 89.8 98.6 Missing 8.6 5.0 1.4 Source: Authors' calculations using administrative TAF RIF data. Notes: TAF = T-MSIS Analytic Files. RIF = Research Identifiable Files. All dual enrollees identified in either the TAF or Medicare Master Beneficiary Summary File are reported. 28 MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS Main Findings and Conclusions The results of this this report highlight several implications for using the TAF to conduct research among those dually enrolled in Medicare and Medicaid. One overarching implication is that for most states, the data fields necessary to identify services commonly used by dual enrollees and paid for by Medicaid are of good quality. However, we identify several states with likely data quality problems based on either missing or invalid individual data elements. We also find unexpectedly low rates of utilization, even when the quality of individual data elements was acceptable, which suggest data on some services are incomplete. Examining these types of problems, however, is not within the scope of this study, and we did not conduct a systematic search for data incompleteness relative to an external benchmark. Thus, researchers contemplating using TAF data in studies of utilization would be wise to seek out external validation before drawing conclusions solely from the TAF. MEASURING DUAL ENROLLEES’ SERVICE USE USING FFS AND ENCOUNTER CLAIMS 29 Appendix A. Glossary Chronic Conditions Data Warehouse Virtual Research Data Center (CCW). Centers for Medicare & Medicaid Services research database and secured virtual technology available to approved Medicare and Medicaid researchers.14 Dual Eligible Special Needs Plan (D-SNP). Medicare special needs plans for people enrolled in both Medicare and Medicaid. Fully Integrated Dual Eligible Special Needs Plan (FIDE SNP). A type of D-SNP that requires plans to assume the risk for all Medicare and Medicaid services (Archibald et al. 2019). Home- and community-based services (HCBS). Services provided in a person’s home or community instead of an institutional setting. Medicaid HCBS services are optional, vary significantly by state, and include services such as personal care and nonemergency transportation. 15 HCBS is a subset of LTSS. Long-term services and supports (LTSS). A broad term that spans institutional and community-based services, including a “variety of health, health-related, and social services that assist individuals with functional limitations due to physical, cognitive, or mental conditions or disabilities” (Thach and Wiener 2018). It largely addresses needs related to activities of daily living and instrumental activities of daily living. Master Beneficiary Summary File (MBSF). Administrative Medicare enrollment and medical claims data. Medicaid Statistical Information System (MSIS). Administrative Medicaid data system with enrollment, medical care utilization, and spending information. Medicare-Medicaid Plan (MMP). A specific managed-care plan for those dually enrolled in Medicare and Medicaid that assumes the risk for benefits in both programs, has a high degree of integration, and is available in select states through the Centers for Medicare & Medicaid Services Financial Alignment Initiative.16 Medicare Savings Program (MSP). Four Medicaid-administered programs for eligible Medicare enrollees with limited resources that pay for select Medicare expenditures, including premiums and cost sharing, depending on the program. 30 APPENDIX Program for All-Inclusive Care for the Elderly (PACE). A program for dual enrollees eligible for nursing home care that allows enrollees to remain safely in the community rather than enter an institutional environment.17 Qualified Disabled and Working Individuals (QWDI). One of four MSPs for working people with disabilities under age 65.18 Qualified Individual (QI). One of four MSPs that offers Medicare Part B premium assistance only to eligible enrollees.19 Qualified Medicare Beneficiary (QMB). The most generous of the four MSPs, QMB offers Medicare Part B and cost-sharing assistance to eligible enrollees. “QMB Plus” is distinct from “QMB Only”; the former also includes full Medicaid benefits (beyond just Medicare), whereas the latter excludes Medicaid benefits. Research Identifiable Files (RIF). A specific version of data available to researchers with appropriate permissions via the Chronic Conditions Data Warehouse Virtual Research Data Center. Special Needs Plan (SNP). A specific type of Medicare Advantage plan with limited eligibility for people with specific needs targeted by the plan (e.g., chronic conditions, institutionalization, dual enrollment).20 Specified Low-Income Medicare Beneficiary (SLMB). An MSP that offers Medicare Part B payment support. “SLMB Plus” is distinct from “SLMB Only”; the former also includes full Medicaid benefits (beyond just Medicare), whereas the latter excludes Medicaid benefits. T-MSIS Analytic Files (TAF). A version of the T-MSIS data intended to be more user friendly. Transformed Medicaid Statistical Information System (T-MSIS). Administrative Medicaid data system that superseded the MSIS circa 2014. The transition date from MSIS to T-MSIS varies by state. APPENDIX 31 Notes 1 See the following resources for more information: “TAF Data Quality Resources,” Research Data Assistance Center, accessed August 9, 2021, https://www.resdac.org/taf-data-quality-resources; and “DQ Atlas,” Medicaid and CHIP Business Information Systems, accessed August 9, 2021, https://www.medicaid.gov/dq- atlas/welcome. 2 The Research Data Assistance Center established data use agreements with CMS whereby approved team members access the data through the secure CCW. 3 Personal care service procedure codes include S5126, 99456, 99509, S5125, T1019, T1020, T1023, and T2022. 4 Nonemergency transportation codes include T2003, A0170, K0812, K0807, T2004, A0180, K0808, K0806, T2049, A0434, T2007, S0215, S0209, T2001, A0130, A0425, T2005, and T2002. 5 Other HCBS procedure codes include the following: A0210, S5131, S5111, S5175, T1021, G0177, H0017, S5135, S5102, T2019, S5100, T2031, 97537, T2038, H0045, H0018, 98960, H0008, T2018, T1022, H0024, S5150, 97535, H0041, S9124, A0190, T2048, H1003, S5108, H2023, T2026, S5130, T1027, H2024, T2017, T2014, H2030, S5115, S9123, S9436, H0019, T2012, T1002, S9444, H2025, T2015, T1030, 97761, S5116, S5145, S9446, S5101, T2021, S5136, T1031, S9445, T2041, H0034, H0012, T2013, H0013, G0154, 90993, T1004, S5110, 98961, T2044, S5151, G0156, S5109, S5120, S9442, S5140, 98962, H2026, H2014, G0109, 97760, H2031, S5105, G8402, H0010, S5141, T1003, T2016, 90989, T2033, T2020, T2040, T2027, S5170, S5146, H1010, T1000, H2001, S9125, T1005, G0108, S9453, T1001, G0128, S5121, H0009, S9455, T2032, T2030, and S9122. 6 Behavioral health procedure codes include the following: 90826, 90855, 90814, 96150, H2015, G0411, 90853, 90839, H2027, 90822, 90863, H2034, H0007, H2019, 99510, H0021, S9482, H0043, 90813, H0022, H0044, H2021, H2013, 90816, H0016, 96110, H2033, H0020, 90899, G0396, 90857, H0036, H1011, H0030, 90805, 90792, 90802, 96153, 90819, 96116, H0038, 96118, 90875, H0025, H0003, 90834, H0039, 96154, 90845, H0048, H2011, 90837, 90811, 90833, H0004, 90801, 90847, 90869, 90840, 90843, 90823, S0257, 90809, 96152, 90824, H0023, 96120, 90838, 90880, S9484, 90887, H0035, 90821, 90844, S9127, 90865, 90817, H2018, 90812, G0409, H0011, 96155, 90876, 90806, H2017, 90829, 90836, 96111, H0047, H0005, 90808, 90868, G0410, H2020, S9475, 90882, H0001, 90810, 90815, 90832, H0049, H0046, T1012, 90818, 90791, H0014, T1015, 90785, G0155, G8431, H0037, S9485, 90828, T1006, 90827, 90804, H2012, S9480, H2036, 90846, 90870, 96101, 96119, H0002, H2016, 90849, 90867, H0050, 90807, H0031, 96151, H0015, T1007, 96102, 90842, H2035, T2034, H2022, H0040, H2028, H2029, and 90862. 7 Nursing home stays are identified using type-of-service codes equal to 009 or 047. 8 HCBS are categorized based on the taxonomy originally developed for CMS by Truven Health Analytics (Truven Health Analytics 2012). The crosswalk of taxonomy categories to procedure codes was developed by researchers at Mathematica Policy Research, the contractor responsible for production of MAX (Medicaid Analytic eXtract) and TAF data (Wenzlow, Peebles, and Kuncaitis 2011). 9 Type-of-service codes used to define personal care services include 051, 065, and 080. The type-of-service code for nonemergency transportation is 056. 10 “Type of Bill – OT,” DQ Atlas, accessed August 9, 2021, https://www.medicaid.gov/dq- atlas/landing/topics/single/map?topic=g12m25&tafVersionId=16. 11 “Medicaid Personal Care Services Participants,” Kaiser Family Foundation, accessed August 9, 2021, https://www.kff.org/health-reform/state-indicator/personal-care- 32 NOTES participants/?currentTimeframe=0&sortModel=%7B%22colId%22:%22Location%22,%22sort%22:%22asc%22 %7D. 12 “Transportation Beneficiary Information,” South Carolina Department of Health and Human Services, accessed August 9, 2021, https://www.scdhhs.gov/site-page/transportation-beneficiary-information. 13 For an early description of this program, see Crowley and O’Malley (2008). 14 “Introduction to CCW,” Chronic Conditions Data Warehouse, accessed August 9, 2021, https://www2.ccwdata.org/web/guest/about-ccw/introduction-to-ccw-video. 15 “Home- and Community-Based Services,” Medicaid and CHIP Payment and Access Commission, accessed August 9, 2021, https://www.macpac.gov/subtopic/home-and-community-based-services/. 16 “Medicare-Medicaid Plan (MMP) Enrollment,” Centers for Medicare & Medicaid Services, last modified December 22, 2020, https://www.cms.gov/Medicare-Medicaid-Coordination/Medicare-and-Medicaid- Coordination/Medicare-Medicaid-Coordination- Office/FinancialAlignmentInitiative/MMPInformationandGuidance/MMPEnrollment. 17 “Program of All-Inclusive Care for the Elderly,” Centers for Medicare & Medicaid Services, accessed August 9, 2021, https://www.medicaid.gov/medicaid/long-term-services-supports/program-all-inclusive-care- elderly/index.html. 18 “Qualified Disabled and Working Individuals (QDWI) Program,” Benefits.gov, accessed August 9, 2021, https://www.benefits.gov/benefit/6180. 19 “Qualifying Individual Program,” Benefits.gov, accessed August 9, 2021, https://www.benefits.gov/benefit/6176. 20 “Special Needs Plans (SNP),” Medicare.gov, accessed August 9, 2021, https://www.medicare.gov/sign-up- change-plans/types-of-medicare-health-plans/special-needs-plans-snp. NOTES 33 References Archibald, Nancy, Michelle Soper, Leah Smith, Alexandra Kruse, and Joshua Wiener. 2019. Integrating Care through Dual Eligible Special Needs Plans (D-SNPs): Opportunities and Challenges. Washington, DC: US Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation. Caswell, Kyle J., and Timothy A. Waidmann. 2021. Dual Medicare-Medicaid Enrollment and Integrated Plan Identification: T-MSIS Analytic Files Data Quality. Washington, DC: Urban Institute. Caswell, Kyle J., Timothy A. Waidmann, and Keqin Wei. 2021. Medicaid Spending on Managed-Care Capitation Claims and Fee-for-Service Claims among Medicare-Medicaid Dual Enrollees: T-MSIS Analytic Files Data Quality. Washington, DC: Urban Institute. CMS (Centers for Medicare & Medicaid Services). 2019. “Using the Type of Bill to Classify Institutional Claims in 2017.” Baltimore: Centers for Medicare & Medicaid Services. ———. 2020. TAF Technical Guidance: Claims Files. Baltimore: Centers for Medicare & Medicaid Services. Crowley, Jeffrey S., and Molly O’Malley. 2008. “Vermont’s Choices for Care Medicaid Long-Term Services Waiver: Progress and Challenges as the Program Concluded Its Third Year.” Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Thach, Nga T., and Joshua M. Wiener. 2018. An Overview of Long-Term Services and Supports and Medicaid: Final Report. Washington, DC: US Department of Health and Human Services, Office of the Assistant Secretary of Planning and Evaluation. Truven Health Analytics. 2012. “Medicaid Home and Community-Based Services (HCBS) Taxonomy.” Ann Arbor, MI: Truven Health Analytics. Wenzlow, Audra, Victoria Peebles, and Stephen Kuncaitis. 2012. “The Application of the Taxonomy in Claims Data: A First Look at Expenditures for Medicaid HCBS.” Ann Arbor, MI: Mathematica Policy Research. 34 REFERENCES About the Authors Kyle J. Caswell is a senior research associate in the Health Policy Center at the Urban Institute. His research covers multiple areas related to health and economic well-being, with a focus on vulnerable populations. He is currently working with colleagues to evaluate a demonstration to coordinate health care for dually eligible Medicare-Medicaid beneficiaries, and on a study to evaluate how disability status affects Medicare spending among the elderly. Previous projects include an evaluation of economic well-being among elderly individuals with mental health impairments and disability insurance, the financial burden of medical spending, the impact of managed care among Medicaid beneficiaries, uncompensated health care, and inequalities in health outcomes. Before joining Urban, Caswell was an economist in the US Census Bureau’s Health and Disability Statistics Branch, where he contributed to the medical out-of-pocket spending component of the Supplemental Poverty Measure. During his previous tenure at Urban, he worked with colleagues to develop estimates of potential savings in medical spending attributable to preventive health services. Caswell holds a PhD in economics. Timothy A. Waidmann is a senior fellow in the Health Policy Center. He has over 20 years of experience designing and conducting studies on varied health policy topics, including disability and health among the elderly; Medicare and Medicaid policy; disability and employment; public health and prevention; health status and access to health care in vulnerable populations; health care utilization among high-cost, high-risk populations; geographic variation in health care needs and utilization; and the relationships between health and a wide variety of economic and social factors. Waidmann’s publications based on these studies have appeared in high-profile academic and policy journals. He has also been involved in several large-scale federal evaluation studies of health system reforms, assuming a central role in the design and execution of the quantitative analyses for those evaluations. Before joining Urban in 1996, Waidmann was assistant professor in the School of Public Health and postdoctoral fellow in the Survey Research Center at the University of Michigan. He received his PhD in economics from the University of Michigan in 1991. Keqin Wei is a senior research programmer in the Office of Technology and Data Science at the Urban Institute. She supports health care policy researchers with statistical methods, data visualization, and big data analytics. Wei has been working with Medicare Parts A, B, and D and Medicaid claims data for seven years. Through her work on the Financial Alignment Initiative demonstration evaluation, she has ABOUT THE AUTHORS 35 extensive experience working with the interim versions of the T-MSIS Analytic Files data in the Virtual Research Data Center environment. In that work, she has developed extensive code scripts to analyze data quality and present findings both graphically and in table form to facilitate interpretation. 36 ABOUT THE AUTHORS STATEMENT OF INDEPENDENCE The Urban Institute strives to meet the highest standards of integrity and quality in its research and analyses and in the evidence-based policy recommendations offered by its researchers and experts. We believe that operating consistent with the values of independence, rigor, and transparency is essential to maintaining those standards. As an organization, the Urban Institute does not take positions on issues, but it does empower and support its experts in sharing their own evidence-based views and policy recommendations that have been shaped by scholarship. Funders do not determine our research findings or the insights and recommendations of our experts. Urban scholars and experts are expected to be objective and follow the evidence wherever it may lead. 500 L’Enfant Plaza SW Washington, DC 20024 www.urban.org