User Support

Table of Contents

Overview

Population Data Sources

AARP State Profiles

American Community Survey (ACS)

Area Health Resource File (AHRF)

Behavioral Risk Factors Surveillance System (BRFSS)

Bureau of Economic Analysis (BEA)

Bureau of Labor Statistics (BLS)

Census

Centers for Medicare & Medicaid Services (CMS)

Current Population Survey (CPS)

Economic Research Service (ERS)

Health, United States

Health Resources and Services Administration (HRSA)

Kaiser Family Foundation (KFF) State Health Facts

Medical Expenditure Panel Survey (MEPS)

National Asthma Survey (NAS)

National Comorbidity Survey Replication (NCS-R)

National Health and Nutrition Examination Survey (NHANES)

National Health Interview Survey (NHIS)

National Survey on Drug Use and Health (NSDUH)

National Vital Statistics System (NVSS)

National Women’s Health Information Center (NWHIC)

Small Area Health Insurance Estimates (SAHIE)

State and Local Interview Integrated Telephone Survey (SLAITS)

Statistical Abstract of the United States (US Stat)

Surveillance, Epidemiology, and End Results (SEER)

Survey of Income and Program Participation (SIPP)

Vendor

Washington, Wyoming, Alaska, Montana, and Idaho Rural Health Research Center (WWAMI RHRC)

APPENDICES

Appendix A. National Population Data

Appendix B. Using Payer Population Estimates From the CPS with HCUP Data

Payer vs. Insurance Conceptual Differences

Other Payer vs. Uninsured

Total vs. Non-Institutionalized Populations

Time Dimension Differences

Uniform Payer Tangent

Other Considerations

Appendix C. Calculating Standard Errors for Population-Based rates From HCUP Data

Step-By-Step Example

OVERVIEW

The Healthcare Cost and Utilization Project (HCUP) is a family of health care databases and related software tools and products developed through a Federal-State-Industry partnership and sponsored by the Agency for Healthcare Research and Quality (AHRQ). HCUP includes the largest collection of longitudinal hospital care data in the United States, with all-payer, encounter-level information beginning in 1988. These databases enable research on a broad range of health policy issues, including cost and quality of health services, medical practice patterns, access to health care programs, and outcomes of treatment at the national, state, and local market levels. The HCUP databases include the following:

• The National (Nationwide) Inpatient Sample (NIS) – the largest all-payer inpatient care database in the United States, containing data on nearly eight million hospital stays per year.
• The Kids' Inpatient Database (KID) – the only all-payer inpatient care database for children in the United States.
• The Nationwide Ambulatory Surgery Sample (NASS) – the largest all-payer ambulatory surgery database in the United States, yielding national estimates of major ambulatory surgery encounters performed in hospital-owned facilities.
• The Nationwide Emergency Department Sample (NEDS) – the largest all-payer emergency department database publicly available in the United States, containing information from over 30 million records for ED visits.
• The Nationwide Readmissions Database (NRD) – a unique and powerful database designed to support various types of analyses of national readmission rates for all payers and the uninsured.
• The State Inpatient Databases (SID) – inpatient discharges from a census of hospitals in participating states.
• The State Ambulatory Surgery and Services Databases (SASD) – encounter-level data for ambulatory surgery and other outpatient services from hospital-owned facilities.
• The State Emergency Department Databases (SEDD) – hospital-owned emergency department visits that do not result in hospitalizations.

The objective of this report is to identify relevant sources of population data that can be used with the HCUP databases to calculate rates of hospital care events per population. This compilation includes data sources that provide nationwide population counts based on people, and not hospitals, physicians, or local resources. Not included are data sources collected by associations (e.g., the March of Dimes) or data collected by individual states. State surveys can be a valuable resource for information on specific subpopulations.

When possible, collections with a mixture of information (i.e., demographic, health status, resource information) or other helpful resources are also mentioned.

Table 1 lists the array of population data sources referenced in this report.

Table 1. Population Denominator Data Consistent with Information in the HCUP Databases
This table lists possible sources of population denominator data. Please consult the description of the data source for more details. A list of abbreviations appears at the conclusion of this table.

POPULATION DATA SOURCES

The descriptions on the following pages provide details on the data sources listed in Table 1 in alphabetic order. For each data source, the following information is provided:

• Sponsor organization
• Data collection method
• Population targeted
• Types of information available
  
  • Demographic - age, gender, race/ethnicity, income level, payer type, etc.
  • Geographic aggregation - national, regional, state, county, and ZIP Code
  • Diseases and health statistics - health status, diabetes, asthma, tobacco use, etc.
• Update cycle for data
• Web references (when available) for the data, published statistics, and online query tools.

NATIONAL POPULATION ESTIMATES

Appendix A provides commonly used estimates from the Bureau of the Census on the resident population for the U.S., for regions (Northeast, Midwest, South, and West), and for the States from 1990 to the most recent year for which data are available. In addition, estimates are provided by gender, five-year age groups, and race-ethnicity at the national level. Total resident population by community income quartile and urban-rural location are based on data from Nielsen, a vendor that compiles and adds value to the U.S. Bureau of Census data. Nielsen uses intra-census methods to estimate household and demographic statistics for geographic areas. Population estimates are available in Excel format on the HCUP User Support Web site under the Method Series (www.hcup-us.ahrq.gov/reports/methods.jsp).

APPENDICES

APPENDIX A. National Population Data

Appendix A includes a collection of national population tables to facilitate the use of this type of information. The information is provided in a separate Excel file to facilitate use.

The following population data from the U.S. Census Bureau are available:

• Table A.1 Annual estimates of the resident population for the U.S., census regions, and states
  • Table A.1.1 Annual Estimates for July 1, 2010 to July 1, 2020
  • Table A.1.2 Annual Estimates for July 1, 2000 to July 1, 2010
  • Table A.1.3 Annual Estimates for July 1, 1990 to July 1, 1999
• Table A.2 Annual estimates of the resident U.S. population by gender and selected age groups
  • Table A.2.1 Annual Estimates for July 1, 2010 to July 1, 2020
  • Table A.2.2 Annual Estimates for July 1, 2000 to July 1, 2010
  • Table A.2.3 Annual Estimates for July 1, 1990 to July 1, 1999
• Table A.3 Annual estimates of the resident U.S. population by gender and age in years
  • Table A.3.1 Annual Estimates for July 1, 2010 to July 1, 2020
  • Table A.3.2 Annual Estimates for July 1, 2000 to July 1, 2010
  • Table A.3.3 Annual Estimates for July 1, 1990 to July 1, 1999
• Table A.4 Annual estimates of the resident U.S. population by gender, race, and Hispanic origin
  • Table A.4.1 Annual Estimates for July 1, 2010 to July 1, 2020
  • Table A.4.2 Annual Estimates for July 1, 2000 to July 1, 2010
  • Table A.4.3 Annual Estimates for July 1, 1990 to July 1, 1999

National population counts from Nielsen that are consistent with the HCUP data elements are also available:

• Table A.5 Annual estimates of the resident U.S. population by national income quartiles, 2000 to 2020.
  • In the HCUP databases, the data element ZIPINC_QRTL contains the national quartile based on the median household income of the ZIP Code of the patient’s residence.
• Table A.6 Annual estimates of the resident U.S. population by a four-category urban-rural designation based on Urban Influence Codes, 2000 to 2020.
  • In the HCUP databases, the data element PL_UR_CAT4 is a four-category urban-rural designation for the patient's county of residence. The categorization is a simplified adaptation of the 2003 version of the Urban Influence Codes (UIC). The 12 categories of the UIC are combined into four broader categories that differentiate between large and small metropolitan, micropolitan, and a non-urban residual.
• Table A.7 Annual estimates of the resident U.S. population by a six-category urban-rural classification scheme developed by the National Center for Health Statistics (NCHS), 2000 to 2020.
  • In the HCUP databases, the data element PL_NCHS2006 is a six-category urban-rural classification scheme for U.S. counties developed by the National Center for Health Statistics (NCHS) especially for use in health care research. The classification emphasizes urban distinctions and is unique in differentiating between central and fringe counties of large metropolitan areas. Smaller metropolitan counties are subdivided by population. Non-metropolitan counties are divided simply into micropolitan and non-core categories.

APPENDIX B. USING PAYER POPULATION ESTIMATES FROM THE CPS WITH HCUP DATA

Rosanna M. Coffey, Katharine Levit, and Marguerite Barrett

During the development of AHRQ Quality Indicator (QI) estimates from HCUP data for the first National Healthcare Quality and Disparities Report (NHQDR), the possibility of using the Current Population Survey (CPS) for population denominators related to QIs stratified by hospital bill payer was considered and abandoned. The difficulty was that the NHQR measures were categorized by the primary expected payer of the hospital bill (a mutually exclusive concept in HCUP), while the CPS captures the health insurance coverage of the population, where one respondent can have multiple types of coverage. Our assessment was that there was not a ready translation from CPS counts by health insurance coverage (including uninsured) to HCUP primary payer categories (including no payment, government subsidy programs, and liability insurance) and that a quick attempt at such a translation was not defensible. When we attempted to create mutually exclusive payer categories from CPS data, the payer categories underestimated some payer populations.

Below we raise some of the differences between HCUP and CPS that we are aware of. We present them here, in case HCUP data users are considering using the CPS as population denominators. Solutions to these issues would require an investment of resources and more exploration.

To understand the following discussion, it is important to be familiar with how HCUP retains information on the expected primary payer. The coding schemes utilized by state-specific data sources are retained as provided in the HCUP data element PAY1_X. During the processing of the data into HCUP uniform files, the state-specific coding in PAY1_X is mapped into a uniform coding scheme in the HCUP data element PAY1. For example, any state-specific values in PAY1_X that refer to either fee-for-service and managed care Medicare patients are mapped to the value one (1) for PAY1. The uniform coding scheme of PAY1 simplifies the analysis of the expected primary payer across states, but often obscures the additional detail available in PAY1_X.

Payer vs. Insurance Conceptual Differences

Problem: Because the concepts of HCUP “payer” and CPS “insurance” are inherently different, there will never be an exact match between the two. CPS represents health insurance coverage (including multiple coverages) plus an estimate of the uninsured. HCUP retains the hospital primary bill payers (including a category for when there is no payment). The devil is in the details.

Potential Solution: To address the issue of conceptual differences in payers, one could impute ‘primary’ insurance coverage for individuals in CPS who have multiple coverages. This would be done by employing a set of rules as to which insurance takes precedence over other insurance coverages when it comes to paying bills. Such an order of precedence would most likely be Medicare, other government insurance (CHAMPVA, CHAMPUS), private insurance, and Medicaid. After each individual with multiple coverages in CPS is assigned a primary insurer, then the CPS data can be tabulated to create “primary payer” denominators for Medicare, CHAMPVA/CHAMPUS, private insurance, and Medicaid.

Tabulations could also be made of the uninsured, which for CPS would be any person not covered by one of these previously mentioned insurance programs. Similar counts of the uninsured in HCUP include “no charge,” “self-pay,” and government program payments sometimes identified in the state-specific payer field (PAY1_X). In addition, CPS tabulations could be made of workers, which would be an appropriate denominator for analysis of certain states where the data element PAY1_X indicates workers’ compensation as the primary payer. (Ninety-six percent of all workers are covered by workers’ compensation, so this is a reasonable denominator for this payer type.)

Note that CPS would not be useful in creating population denominators similar to state-specific payer categories (PAY1_X) for programs that pay for services but are not insurance (e.g., maternal and child health programs, state-county run mental health and substance abuse programs, black lung program, corrections system, etc. ) or for homeowners or automobile liability insurance payments. CPS respondents who only have coverage through these programs would be considered uninsured.

Other Payer vs. Uninsured

Problem: HCUP uniform coding of payer (PAY1) identifies the following government subsidy programs (other than Medicaid) in an "other" payer group: CHAMPUS, CHAMPVA, Indian Health Service, child health insurance programs, maternal and child health programs, state-county run mental health and substance abuse programs, black lung program, corrections system, and other general assistance state and county programs. Also included in the "other" payer group are non-government payers such as workers' compensation¹ accident insurance, etc. CPS, which aims to measure health insurance coverage of the population, counts people in government assistance programs (other than Medicaid) as part of the uninsured population if they have no other health insurance coverage. HCUP uniform coding (PAY1) includes a category for “self-pay" and "no charge" (uninsured from the hospital perspective).

Potential Solution: For analyses of specific states, it would be possible to regroup the state-specific payer categories (PAY1_X) involving government subsidy programs (other than Medicare, Medicaid, and CHAMPUS/CHAMPVA) as uninsured for comparability with the CPS. This may or may not be helpful, depending on the purpose of the analysis.

Total vs. Non-Institutionalized Populations

Problem: The CPS universe covers only the non-institutionalized population, whereas HCUP covers hospital stays for the entire population. The most obvious segment of the population missing from the CPS universe but included in HCUP is the aged population residing in nursing homes.

Potential Solution: This mismatch is difficult to solve because the data provided to HCUP does not contain accurate information on which admissions are from or to nursing homes. Because many patients from nursing homes are admitted to the hospital through the emergency department, identifying admissions from and discharges to nursing homes from the HCUP data elements for the patient admission route and discharge disposition underestimates the actual number.

Time Dimension Differences

Problem: Another issue to consider is the time dimension of the two measures – a point in time for a HCUP discharge versus recall for a year in the CPS. Even if the two sources could be joined, the estimates will undoubtedly differ by a factor related to time dimension because some people change insurance categories during the year (e.g. uninsured to Medicaid).

Potential Solution: No immediate solution.

Uniform Payer Tangent

Problem: A tangential issue that is worth considering is the inconsistency in the state-specific reporting of payer categories which limits the creation of truly consistent uniform HCUP categories for PAY1. For example, in scrolling through state-specific HCUP documentation for the uniformly coded expected payer PAY1, Hill-Burton charity care is often classified as “no-charge,” but sometimes included under “other payer” when the state-specific data source combines Hill-Burton and other government programs into one “other government” category. In addition, state-specific data sources sometimes do not distinguish other government programs from other non-government programs. This distinction would be beneficial for HCUP and its consistency with CPS.

Solution: No immediate solution.

Other Considerations

There are two other considerations for better aligning HCUP to CPS and for using other, more definitive sources for some denominators:

Using HCUP Secondary Payer. There may be other ways to use HCUP payer data, depending on the purpose of the analysis. For example, analysts might use both the primary and secondary payer categories (available for some states) in HCUP to get a better idea of the number and types of discharges that are covered by various insurance payers and programs. All of the combinations of payers in the CPS could be compared with the relevant combinations in HCUP. This would allow for more careful construction of payer denominators for studies of specific populations. Rules (such as who is likely to pay the bill, as noted above) also could still be used to collapse to more manageable groups for analysis. Again, the other payers would have to be grouped into no insurance, and people covered in the hospital by workers’ compensation would have to be paired with CPS “counts-of-workers” denominators or left out.

More Definitive Sources. Alternatively, beneficiary counts for major payers (Medicare, private insurance, Medicaid, no charge, and self-pay) should be able to be tied to population estimates nationally and at the state level from definitive programmatic and other survey sources. It may be best to use CMS Medicare and CMS or state Medicaid enrollment as the definitive source for those programs and to use CPS for private insurance and uninsured. For other payers, depending on the analytic need, it might be best to obtain counts (person-years) for many of these programs directly from the agencies involved. For example, the National Academy for Social Insurance develops counts of people covered by workers’ compensation in recent years (www.nasi.org/publications2763/publications_show.htm?doc_id=385937), while the Indian Health Services counts people eligible for IHS services. Counts for other populations might be explored (e.g., the nursing home population).

Some of the issues presented above need to be assessed. The more important issues that would have a major impact on population estimates would need to be dealt with in some effective way for HCUP numbers to be related to population insurance estimates in some reasonable manner.

APPENDIX C. CALCULATING STANDARD ERRORS FOR POPULATION-BASED RATES FROM HCUP DATA

When calculating population-based rates using HCUP data as the numerator and Census population data as the denominator, standard errors for each component must be carefully calculated. For estimates based on the HCUP nationwide databases – Nationwide Inpatient Sample (NIS), Kids’ Inpatient Sample, and Nationwide Emergency Department Sample (NEDS) – the standard errors should be calculated as described in the HCUP report entitled Calculating Nationwide Inpatient Sample (NIS) Variances (Houchens, et al., 2005). This report will simply be referred to as the NIS Variance Report throughout this appendix. The method for calculating standard errors takes into account the cluster and stratification aspects of the NIS, KID, and NEDS sample design when calculating these statistics using the SAS procedure PROC SURVEYMEANS. For estimates based on the HCUP state databases – State Inpatient Databases (SID), State Emergency Department Databases (SEDD), and State Ambulatory Surgery Databases (SASD) – the same procedure omitting the cluster and stratification features should be used. For population counts based on Census data, there is no sampling error.

Step-By-Step Example

Consider the following example of calculating a population-based rate and standard error using the NIS as the numerator and Census data as the denominator. The rate for the adult admissions for diabetes in the US per 100,000 U.S. population, age 18 and above is defined as follows:

Number of weighted NIS discharges age 18 and above with a diagnosis of diabetes * 100,000
U.S. adult population count from the Census for the same data year

Step1: Define the population of interest in the NIS. Create a 0/1 variable named DIABETES in which the value 1 indicates adult discharges with a diagnosis of diabetes and the value 0 is used for all other discharges.

Step 2: Calculate the numerator and standard deviation of the numerator using the NIS. The example below for calculating the numerator and standard deviation uses the SAS procedure PROC SURVEYMEANS. It is important to include statements for cluster, strata, and weight because the NIS is a stratified sample. The NIS Variance report provides more detail on the purpose of these SAS statements and also provides example code for other statistical software packages.

ods listing close;
proc surveymeans data=NISdataset SUM STD ;
  cluster hospid ;
  strata  nis_stratum;
  var     DIABETES;
  weight  discwt;
  ods output statistics=NIS_COUNTS; 
  title2 "NIS weighted counts for 0/1 variable DIABETES";
ods listing;

In the PROC SURVEYMEANS output data set, the sum is the numerator of the rate (call this “NIS_TOP” for this example) and std is the standard deviation of numerator (call this “NISTOP_SD” for this example). The actual configuration of the output data set depends on the version of SAS.

Step 3: Determine the appropriate population denominator from Census data. There is no error associated with the Census population count. For this example, call the population count "ADULT_US_POP"

Step 4: Calculate rate and standard error of the rate. Calculate the population-based rate per 100,000 population and the standard error of the rate as follows:

Rate =      NIS_TOP
--------------------- * 100,000
Adult_US_POP
SE of Rate = 	    NISTOP_SD 
		------------------------ * 100,000
 			Adult_US_POP

¹Workers' Compensation is state mandated, but is fully funded by employers. It becomes the sole payer of services once liability is established.