National Long-Term Care Survey

Overview of NLTCS Research

Changes in Chronic Disability in the U.S. Elderly Population - A number of papers (e.g., Manton et al., 1997; Freedman and Soldo, 1994) described changes in the prevalence of chronic disability in the U.S. elderly population to 1994. While disability declines were evident from 1982 to 1994, Manton and Gu (2001) extended that analysis to include the 1999 NLTCS. Findings include a) declines in chronic disability accelerated to 2.6% per annum 1994 to 1999 (1.7% 1982 to 1999), compared with 1.6% 1989 to 1994, (a result robust to a number of sample design factors); b) the institutional population declined rapidly with a drop in absolute size; and c) starting in 1989, the African-American population had a faster rate of decline in chronic disability than U.S. whites. This confirmed and elaborated findings of declines from 1989 to 1994 (Manton et al., 1997). These declines have been replicated in other U.S. surveys; e.g., SIPP (Freedman and Martin, 1998), CMBS (Waidmann, 1998) and the NHIS (Crimmins et al., 1989). Declines in the institutional population agree with declines in institutionalization rates in MEPS (Bishop, 1999) and the 1985 to 1995 NNHS (Strahan, 1997). Most recently, a comparison of 1991 to 1994 with 1994 to 1999 CMBS data found a modest acceleration in the rate of decline of disability (Corder et al., 2002) consistent with Manton and Gu (2001).

Severe Cognitive Impairment, a concern because of its high prevalence at late ages, its heavy impact on the individual and his/her family, and the projected growth of elderly U.S. populations to 2070 (Herdman and Behney, 1992) has been addressed in a number of papers. In the 1982 to 1994 NLTCS, evidence was found of significant (33%) declines in the prevalence of 'Severe Cognitive Impairment' (Manton et al., 1998). This continued in the 1999 NLTCS when, after age standardization, prevalence declined from 5.4% in 1982 to 2.8% in 1999; or 3.8% per year - a decline from 1.8 million forecast (1982 rates) to 1.0 million observed in 1999. This is consistent with declines in the U.S. nursing home population in Manton and Gu (2001) and the reduced rate of institutionalization in the MEPS (Bishop, 1999) and NNHS (Strahan, 1997). This finding was replicated by Freedman et al. (2001), using supplements to the 1993 AHEAD and the 1998 HRS surveys.

Changes in Medicare and Social Security Expenditures - Singer and Manton (1998) examined effects of declines in disability prevalence on a modified 'disability' adjusted dependency ratio to evaluate the health (as opposed to demographically) constrained dependency ratio by U.S. population aging - especially Post WW II baby boom cohorts. They found the altered dependency ratio could be maintained at 1994 levels if disability declined, to 2070, at 1.5% per annum. Declines observed 1982 to 1999 were 1.7%; from 1994 to 1999, 2.6%. Thus, the 1.5% rate could be surpassed (Manton and Corder, 1995). Manton and Singer (2002) analyzed Medicare costs and changes, 1982 to 1999, in defined disability categories. They concluded that disability declines have a salutary effect on Medicare expenditures.

Age and gender specific effects of nutrition on functioning of U.S. elderly and oldest-old are now possible to be addressed in the NLTCS using questions on nutrition added to the 1994 and 1999 NLTCS. These studies were conducted in collaboration with Professor Rush, then head of the Epidemiology program at the Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University. Nutritional factors, and their relation to physical activity, are important for the elderly in terms of longevity and in maintaining function.

Methodological investigation of the evolution of disability and mortality processes is also possible using this dataset. The longitudinal design of the NLTCS offers analytic opportunities and complexities. An advantage is that detailed assessments are made at multiple time points for a large elderly sample. In the 1982 to 1999 NLTCS there are 42,000 persons followed with ~22,000 deaths 1982 to December 31, 1999. Another advantage is that each person is linked to a continuous record of service use and mortality (i.e., dates of death, as well as dates and types of service use, are recorded) through Medicare files (the NLTCS sample is drawn from computerized Medicare lists). Additionally, records will now be linked to NCHS Multiple Cause of Death files for 1982 to 2004.

An important analytic issue is that NLTCS disability processes are multidimensional and only 'partly observed.' This 'time' series is different from those usually encountered in that, while series usually have many time points (a dense sampling over time) and a modest sampling of variables (a sparse sampling of the theoretical universe of state variables), the NLTCS data have a sparse time sampling (six times, five years apart) but a dense state variable sampling (e.g., 27 variables assessed on 42,000 persons). That is, the data present a multi-dimensional picture of processes in individuals at survey dates. The sparse time sampling is justified because of the temporal stability of chronic disability compared to economic indices. Additionally, there is a complex temporal pattern of additions (new sample; persons passing age 65) and deletions (mortality) to the case space of the process.

Yashin and Manton developed likelihood methods to use information on the joint distribution of continuously-observed Medicare mortality (time to death; now more highly dimensional with multiple-cause data) and Medicare service use variables (e.g., hospitalization, time, duration, outcome, diagnosis, which are broadly indicative of health changes between surveys) and discretely (in time) observed disability measures (high dimensional space measured with a high degree of 'granularity' and some error related to case position relative to state variable partition boundaries). Use of the joint distribution improved estimates of disability process and mortality parameters (Yashin and Manton, 1997; Manton and Yashin, 2000) and ALE by better describing the complex processes of accretion and deletion (partial sample paths whose endpoints are randomly generated) within survey intervals in estimating disability dynamic parameters (Manton and Akushevich, 2002). Modeling is facilitated by the statistical implementation of a high dimensional fuzzy state space model in GoM (Manton et al., 1994).

Professors Manton and Tolley examined use of stochastic process models of disability and mortality to analyze the 1982, 1984, and 1989 NLTCS data to assess Social Security and Medicare cost implications of changes in disability over time. These studies (Tolley and Manton, 1996a; Tolley and Manton, 1996b) assess the benefit to Trust Fund stability if the normal retirement age for Social Security (or, alternatively, Medicare benefit age) were changed from current levels, given trends in age and gender-specific disability distributions and age, gender, and disability-specific mortality.

Professor Stallard has used the NLTCS to study LTC insurance both for research (e.g., Stallard and Yee, 1994) and for U.S. professional actuarial associations. The NLTCS has been used by health insurers and reinsurers in the U.S., Germany, and France because it contains a wide range of disability measures that can be used to form actuarial indices as well as data to evaluate cost and health implications. The Health Insurance Portability and Accountability Act (HIPAA) of 1996, designed to clarify tax treatment of LTC services and insurance contracts, was based on NLTCS measures (e.g., 90+ days chronicity; 6 ADLs, with one being incontinence). Research showed the necessity of having data to extreme ages to evaluate actuarial options because of the strong correlation of disability prevalence, by type and intensity, with age. With serum chemistry data, if funded as a supplement to the 2004 survey, will allow researchers to generate physiological age measures or indices to see how well they explain chronological age, health state variation and covariation. While Manton and Yashin (2000) developed improved mathematical formulations for stochastic process models of disability, Manton et al., (1994b) generalized GoM to better estimate parameters to describe longitudinal changes in health and function using a three-way tensor (Levi-Civita, 1947) decomposition of a) groups of similar persons, and b) types of disability trajectories (Woodbury et al., 1992).

APOE and WRN and Relations to Alzheimer's Disease and Cardiovascular Disease is another potential area of research questions. In the 1999 NLTCS, it is possible to analyze the relation of APOE alleles, and with AD and CVD population risks. Data were collected (639 bloods and 2,083 buccal wash samples) and are now undergoing genetic (PCR) analyses. WRN is a gene coding for a novel helicase of 1,432 amino acids that may be associated with premature aging. Homozygous mutations, all resulting in truncation of the protein, lead to Werner's syndrome. Little is known about the role of WRN in 'normal' aging. Professor Martin identified WRN polymorphisms that may modulate common age-related disorders and perhaps aging (Castro et al., 2000).

Pilot and Methodological Studies Conducted for the 1999 NLTCS are ongoing, and more could be funded through NIA's R03 grants for secondary analyses. Several studies of the quality of the 1999 NLTCS data are ongoing and provide important insights for the 2004 NLTCS.

This very brief review of topics shows the wealth of data that is present. With risk factor and nutrition data in the 1994 and 1999 NLTCS, researchers can examine how those factors contribute to chronic disability and mortality declines. Researchers can isolate disability, mortality, and disability specific mortality (by cause), changes by a) cohort, b) gender, c) education, d) economics, and e) family and housing status. Social and health factors are likely to be important in determining current and future changes in life expectancy and ALE. Changing patterns of use of personal assistance, special equipment, and housing services by disabled elderly persons can be studied, as well as institutional risk and the level of family and social resources necessary to maintain persons with different levels and types of disability in the community - especially as Medicare HHA benefits and SNF use have, and will, change (Manton et al., 1997). Because of large NLTCS samples at late ages, researchers can examine changes in not only the oldest-old population (i.e., 85+), but even in populations aged 95+ and 100+. The 2000 Census counted 337,238 persons 95+; in 20 years (2015) that population is expected, under middle variant mortality assumptions, to grow to an adjusted (normed to 2000 Census counts vs. the 377,000 ten-year estimate used in 1999 sample weighing) 722,000 (1.13 million under adjusted high life expectancy assumptions); and by 2035 to 1,287,000 (adjusted middle variant) i.e., the size of the 85+ population in 1982. The 85+ population is projected to be 10.9 million (middle variant) in 2035. The 1994 and 1999 NLTCS have been augmented with over-samples of persons aged 95+ (540 in 1994; 600 in 1999; 800 proposed in 2004) - a group for which we have little nationally-representative health data (e.g., numbers of 95+ persons in the NHANES and NHIS samples are small; the NHIS sample does not cover institutions), though that population is expected to grow rapidly, both relatively and absolutely, in the U.S. With the NLTCS' elderly sample and augmentation samples of persons 95+, there is the potential for researchers to determine how increasing disability risks and mortality selection of disabled persons operate to affect the population prevalence of chronic disability trajectories by level and type (specific to age and gender). Analyses (Manton, 2000) suggest gender-specific age and disability-specific mortality will reach such high levels by 95 that they constrain disability prevalence and mortality rates.