This section describes the data sources used to create input for the ARC CLASS Program Model. The main data sources include the 2011 OASDI Trustees Report, the Current Population Survey (CPS), the National Health Interview Survey (NHIS), the National Long Term Care Survey (NLTCS), and the National Nursing Home Survey (NNHS).
The annual Trustees Report presents the current and financial status of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds. The report is put out by the Social Security Administration’s Office of the Chief Actuary.2 Supplemental tables for single years are provided and used as the source for:
- Unemployment rate and labor force participation rate projections by age group, sex, and year
- U.S. population projections by age, sex, and year
- Mortality assumptions projected by age, sex, and year
Labor force data are critical to the calculation of program cash flow. The labor force participation rates and unemployment rates are used to calculate the number of workers that may participate in the program, pay premiums into the system, and ultimately will be eligible for benefits. The default labor force participation and unemployment rates vary by age, sex, and year.
Current Population Survey
The CPS is a monthly survey of households providing data on the labor force, employment, unemployment, hours worked, earnings, and other demographic and labor force characteristics. The CPS is collected each month from a probability sample of approximately 60,000 households through personal and telephone interviews and is designed to be representative of the civilian noninstitutionalpopulation of the US. It is conducted by the Bureau of the Census for the Bureau of Labor Statistics.3 The March 2009 CPS is used in the model for the annual income interval distribution by sex and age group. Below is a table that shows the income distribution of workers for all ages.
The income distribution is used in the model to estimate the population that is eligible to participate in the CLASS program, which requires an individual to earn in one year more than the amount needed to earn one quarter of coverage under the Social Security program. This amount is $1,120 in both 2010 and 2011, but it was $1,090 in 2009, the year of the CPS used in the model. The model uses the distribution of workers by income to estimate the number of eligiblescorresponding to the income requirement selected by the user. The level of the premium depends on the total number of eligiblescompared to the number of eligibles that could immediately qualify for benefits. The cash flow projections also depend on the number of individuals who participate in the program. When the user selects a subsidy for the premiums for low-income individuals, the model calculates the number of individuals receiving the subsidy as the number of workers whose income is above the income requirement and below the poverty line.
National Nursing Home Survey
The NNHS is a nationwide sample survey of nursing homes, residents, discharges, and staff. Nursing homes included in the survey are freestanding or nursing care units of hospitals, retirement centers, or other similar entities, with at least three beds. They must be Medicare or Medicaid certified or state licensed to operate as a nursing home. The survey is conducted by the National Center for Health Statistics, part of the Centers for Disease Control and Prevention. Estimates from the 1999 survey are based on roughly 1,400 responding facilities out of approximately 18,000 nursing homes in the United States.4 The 2004 survey uses 1,174 responding facilities. Prevalence rates and Admission rates are graduated by age using the Whittaker-Henderson graduation algorithm, which maximizes a function that measures both “fit” and “smoothness.” The fit is measured by the sum of the squares of the difference between the graduated prevalence rates and the original prevalence rates and the “smoothness” is measured by the sum of the second differences of the graduation rates by age. The model uses NNHS Survey resident and discharge data by age and sex from the 1985 and 2004 surveys and is the source for:
- Nursing Home Incidence Rates
- Nursing Home Prevalence Rates,
- Average Lengths of Stay, and
- Distributions of lengths of nursing home stays (i.e., continuance tables)
The 1985 survey is used as base information for incidence rates and average length of stay because it analyzed nursing home behavior in more detail than the 1999 or 2004 surveys. There is a difference in how nursing home admissions and lengths of stay are defined in the surveys and how they are used for purposes of insurance. The surveys count every admission to a nursing home, while insurance would concatenate multiple stays into one benefit period whenever there is a transfer between nursing homes or a stay that is briefly interrupted by a hospital stay. The 1985 survey had information that made it possible to concatenate stays, while the more recent surveys do not contain the information to make this possible. In general, it is more difficult to determine accurately admission rates and lengths of stays than it is to determine the number of nursing home residences at a point in time. The ratio of the number of admissions to the population is referred to as the admission rate, while the ratio of the number of residents to the population is referred to as the prevalence rate. There is a loose relationship between prevalence rates (PR), admission rates (AR), and average lengths of stay (ALOS) such that the following equation is approximately true:
PR = AR * ALOS
All of the NNHSs provide relatively accurate counts of the resident population and therefore of prevalence rates. We used the ratio of the prevalence rates obtained from the 2004 survey to those obtained from the 1985 survey to project the admission rates from the 1985 survey up to 2004. We did not project the average lengths of stay.
National Long-Term Care Survey
The NLTCS is a nationally representative longitudinal survey of Medicare beneficiaries designed to study changes in health and functional status of Americans age 65 and over, in both community and institutional settings. The NLTCS defines an institutional setting as one having a full-time medical professional (doctor, nurse, physician assistant, or psychiatrist) on its staff. Data from this survey exist for years 1982, 1984, 1989, 1994, 1999, and 2004. The survey population consists of roughly 36,000 people originally drawn from Medicare enrollment files in 1982 with new people entering each successive survey. With each survey about 5,000 people passing age 65 between each wave are added to replace those that have died. In this way, the NLTCS represents all Medicare beneficiaries age 65 and over both institutionalized and noninstitutionalized(although it excludes individuals in correction facilities). The NLTCS has many components, including disability measures, cognitive test results, medical conditions, education levels, and income. It tracks health expenditures, Medicare service use, and the availability of personal, family, and community resource caregiving. The survey was administered by the U.S. Census Bureau and Duke University.5
Longitudinal data from 1984 to 1989 are used to identify incidence rates according to the methods of Eric Stallard and Robert Yee as presented in their paper Noninsured Home and Community-Based Long-Term Care Incidence and Continuance Tables6 (2000). Data from the 1984 through 2004 NLTC surveys are used to identify prevalence rates of frailty to both community and institutional respondents. These results are graduated using the Whittaker-Henderson formula according to both respondent age and data year. The rates of change for the prevalence rates are used to trend the incidence rates calculated by Stallard and Yee through 2004 by age, sex, and ADL level.
The NLTCS asks each respondent if he has trouble with each of six ADLs, and follows up by asking if he gets help from a person. Those getting help from a person are counted as needing assistance with that ADL. Continence is also assessed by the NLTCS, but separately with the general medical questions. Cognitive impairment is calculated by a mini-mental status exam in each survey except for the 1999 survey, where it is calculated by a more detailed SPMSQ test, which we have calibrated to the mini-mental exams in the other years. The mini-mental status exam has ten simple questions, and answering three or more of those questions incorrectly identifies the respondent as cognitively impaired. Persons interviewed through a proxy are marked as cognitively impaired if they are reported to have Alzheimer’s, dementia, or senility. The community survey for this group is used to obtain the trends in frailty prevalence among those 65 and older. This is used to adjust the Stallard and Yee incidence rates to a more current year.
The model calculates 2004 home care incidence rates using the 1984-1989 Stallard/Yee incidence rates and ratio of home care prevalence in 1989 and 2004 from the National Long Term Care Survey (NLTCS). The 2004 incidence rates are projected to be the 1984-1989 incidence rates multiplied by the ratio of 2004 prevalence to 1989 prevalence. The model does not project average lengths of stay.
National Health Interview Survey
The NHIS is a cross-sectional household interview survey begun in 1957 and conducted annually. It provides data on a broad range of health-related topics for the civilian noninstitutionalized population of the U.S. The NHIS has an expected sample size of approximately 35,000 households each year, and provides information on the amount, distribution, and effects of illness and disability and the services rendered for or because of such conditions. The data include numerous demographic and socioeconomic characteristics. In particular, this study focuses on age, gender, earnings level, cognitive impairments, and ADL limitations that are used to determine eligibility rates for the CLASS model. The survey is conducted by the National Center for Health Statistics.7
In the survey, respondents are asked several times if they have some kind of limitation in their daily life. They are asked separately if they are limited in the work they can do, or if they are limited in daily living, or if they are limited in any way. Those who report some kind of limitation are then asked about specific limitations.
We have set up our CLASS model so that the user can select from among utilization rates based on several possible interpretations of cognitive impairments. Senility (which includes Alzheimer’s and dementia and is denoted by the letter “S” in the model) is always included as a cognitive impairment. A second interpretation also includes mental retardation and developmental disabilities (denoted by “SRD” in the model). The third category also includes ADD, Schizophrenia, bipolar disorder, and other mental disabilities, which are all kept together because they are represented by the same variable in the survey. When all mental impairments are included, the model refers to this option as the “SRDA” option. The survey asks everyone if they have difficulty performing ADLs, including: eating, dressing, transferring, bathing, and using the toilet. However, the survey does not ask about continence as an ADL or in any other fashion.
The 2007-2009 NHIS was the source in the model for frailty prevalence rates for the under 65 population by age and sex. Frailty rates are calculated separately for all three interpretations of cognitive impairment and for one, two, and three ADLs aggregating data from the three years. In addition, frailty rates are calculated for individuals at several income levels. These rates are independently graduated using the Whittaker-Henderson formula by respondent age. Users can select the utilization rates used in the model calculation from among the three interpretations of cognitive impairment and 1+, 2+, or 3+ ADLs. The choice of the interpretation of cognitive impairment has a significant impact on the results because including ADD, Schizophrenia, Mental Retardation, and developmental disabilities more than doubles the frailty rate. Changing between 2 and 3 ADLs has a much smaller effect on the initial frailty rate among those who work, but has a significant impact on frailty for the population as a whole.
The following table shows how the incidence rates used in the ARC Model for 2+ ADLs and the senility plus developmental disabilities (SRD) interpretation of mental impairment compare with the intercompany study from the Society of Actuaries (SOA).
In addition, we take NHIS data from every year from 1997 to 2009 to identify how frailty prevalence rates have changed over time according to age and gender (but not earnings level). Rates are calculated according to the same method and attributes as for the prevalence rates calculated from 2009. These prevalence rates over time are graduated using the Whittaker-Henderson formula by data year and respondent age. The average rate of growth over the 1997-2009 timespan is summarized in the table below.
The increase in frailty rates over time has mostly occurred because of cognitive factors, and the more restrictive the cognitive assumption, the less reported growth there is. For the most generous interpretation of cognitive impairment, frailty grows among all age groups, where if only senility is included, then frailty increases only among older people.