For some time now, certain analysts, advocates, and policymakers have recommended the creation of a new social insurance program to finance long-term services and supports (LTSS) for the working-age disabled population and, in particular, the elderly disabled population. This demand came close to fruition when the CLASS program was included in the Affordable Care Act (ACA) and used in the estimate of its budgetary effects (scoring) to help fund that legislation over Congress’ 10-year scoring window. CLASS (Community Living Assistance Services and Supports) would have been a national, voluntary insurance program, sponsored and administered by the federal government, for selling insurance to most workers, with no underwriting, to cover a portion of their LTSS needs.
Inherent flaws in the design and financing of CLASS were evident to some even before its passage despite the surprisingly favorable score given by the Congressional Budget Office. These flaws were further demonstrated through several Congressional hearings after its passage; the Obama Administration never implemented the CLASS program, and Congress repealed it. In this post, I make the case that there are also serious flaws in a recent Urban Institute attempt at modeling, employed in Health Affairs, for new public insurance programs for financing LTSS.
The Policy Background
The political compromise that formed the budget and tax legislation that passed at the end of 2012 included the repeal of CLASS; it also set up a Commission on Long-term Care in June 2013 to create a plan for addressing the Nation’s challenges regarding delivering and financing LTSS. The Chair of the Commission was Bruce Chernof, head of the Scan Foundation, and the Vice-Chair was Mark Warshawsky (author).
The Commission coalesced around many recommendations to improve service delivery and the LTSS workforce, and agreed on the need for a comprehensive financing solution emphasizing insurance, with public and private sources of funds and assistance to the poor. However, in its September 2013 report, it split on the appropriate extent of the role of the government. Some wanted to create a full Medicare-like social insurance program for LTSS, with a small optional carve-out for private LTCI; others were in favor of a federal catastrophic insurance program to cover the back-end of the LTSS risk; and still others wanted to rely mainly on private resources and insurance as well as on Medicaid or subsidies for the poor.
While some on the Commission emphasized the lack of retirement savings, others believed that private resources (including housing) are available on a fairly widespread basis or could be forthcoming. The latter, conservative, group wanted to further encourage, through both incentives and penalties and strict enforcement of federal Medicaid eligibility and repayment rules, private insurance solutions and private funding sources for LTSS. They believed that the Medicaid incentive to avoid the purchase of LTCI could be cut back by greater enforcement of the current eligibility and asset-recovery-from-estates-of-deceased-beneficiaries rules, and by making the rules stricter to increase the amount and types of assets (especially retirement accounts) as countable for eligibility purposes. They also believed that, despite somewhat poor current conditions, the market for LTCI and the nature of LTCI products could be improved greatly through more regulatory flexibility and product innovation.
The Urban Institute DYNASIM Model
Despite the split on the Commission (perhaps because of it), there continues to be a strong current of demand in certain political circles for the federal government to provide LTSS insurance directly, even on a social insurance basis. Because concern about cost is an obvious barrier to such efforts, there has also been a demand for models that can score proposals for insurance programs. Urban Institute researchers Melissa M. Favreault, Howard Gleckman, and Richard W. Johnson, funded by the federal government (The Department of Health and Human Services’ Assistant Secretary for Planning and Evaluation), Scan Foundation, AARP, and Leading Age (an organization of non-profit long-term care providers), have developed an elaborate dynamic microsimulation model, DYNASIM3, to analyze policy proposals for financing long-term services and supports (LTSS), with an emphasis on new public programs. The essence of their results was published in a recent article in Health Affairs.
In Health Affairs, Favreault, et al. analyze what they call three new insurance options: a program with a front-end benefit that begins after a 90-day waiting period and covers a maximum of two years of need; a catastrophic-only or back-end program that begins after a waiting period of two years but provides a lifetime benefit thereafter; and a comprehensive program that begins after a 90-day waiting period and provides a lifetime benefit. Each option is modeled as voluntary insurance and as a universal mandatory program for workers. For the voluntary programs, they include subsidized and unsubsidized versions.
The Bipartisan Policy Center (2016) in its project on long-term care financing uses the DYNASIM model, in particular to score the cost to the federal government of a catastrophic insurance program. Apparently, other advocacy groups are also using the Urban Institute model.
Actually, Sometimes, Two Models
The model is quite complex and is based on multiple sources of data and contains hundreds of assumptions. Although the supporting written material is voluminous, in key areas and assumptions it is incomplete and sometimes even opaque; it would be impossible for others to replicate the results found.
According to an Appendix to the Health Affairs article, to evaluate the voluntary front- and back-end programs, the researchers use premium prices based on an actuarial model developed by the consulting firm Milliman for starting values. The researchers say that these premium estimates were based on output from the DYNASIM model and the Milliman model; how the models are combined is not specified. They then “modestly adjust” these premium estimates “to increase consistency” with the DYNASIM projections; the details of, and justification for, this adjustment are not explained.
The researchers say that there are “necessarily” different assumptions about the distribution of risk of needing LTSS, transactions costs, and so on across the two models; the details of, and justification for, these different assumptions are not given. For developing all other premiums, in particular for the mandatory and comprehensive programs, the researchers use DYNASIM alone.
This last modeling choice is particularly perplexing because it is not clear why some financing programs should be assessed using one model alone and others using another, hybrid, model. Even more importantly, the bifurcation is quite concerning because it is distinctly possible that the results generally favoring mandatory and comprehensive programs over voluntary and partial programs may derive entirely from the use of the DYNASIM model, as opposed to the Milliman hybrid model. As we will see below, the DYNASIM model contains many unreasonable assumptions, most of which seem to lower the proportion of the older population that is disabled, the incidence of significant disability, the duration of disability and paid care, and hence the estimated costs of the evaluated programs that finance LTSS.
Projection of Future Paid Care for LTSS is Surprisingly Low
Another supporting document on the HHS website gives some of the basic assumptions and results of the DYNASIM model about the risk of needing care and associated costs now and in the future. The researchers say that the model can predict what percentage of older individuals will develop a disability; have LTSS needs; and use paid LTSS, how much, and for how long. The model also predicts how much care costs and how care needs are currently financed. Both averages and distributions are given.
The focus is on disabilities that meet the criteria set in the Health Insurance Portability and Accountability Act (HIPAA) — that is, a need for assistance with at least two activities of daily living (ADLs, e.g. toileting, bathing, eating, etc.) that is expected to last at least 90 days, or a need for substantial supervision for health and safety threats due to severe cognitive impairment. In other words, the level of need, presumably most of it serviced through paid care, is quite high in HIPAA-eligible conditions because the extent of disability is quite substantial, continual, and on going. The Urban Institute researchers say that the data underlying the DYNASIM model in the areas of health, disability, LTSS use, and long-term care insurance (LTCI) coverage come from the Health and Retirement Study (HRS).
The researchers present projections of the number of persons ages 65 and older, including the number meeting HIPAA disability criteria from 2015 to 2065. The text says that the number with HIPAA-level disability will increase from “6.3 million to almost 15.7 million.” Yet Figure 1 shows that the proportion of the population age 65 and older that is HIPAA-disabled hardly changes — 6.3 million out of about 48 million in 2015 (13.1 percent) to 15.7 million out of about 98 million in 2065 (16.0 percent), a relatively small increase. The small increase is surprising because of the universally accepted projection that the elderly population will age considerably over this long horizon, given declining mortality rates and the movement of the large baby boom generation into retirement. Hence it would have been presumed that the aging elderly population would become proportionately more and more subject to disability. (The probability of being disabled in a year, whatever the definition of disability, increases significantly with age.)
According to the 2015 Social Security Trustees’ Report, 6.2 million Americans were ages 85 and older in 2015, comprising 12.9 percent of the age 65 and older population; in 2065, 19.0 million will be age 85 and older, accounting for 19.7 percent of the 65-plus population. The significant aging of the elderly population over this long horizon suggests that the slow and small increase in the projected proportion of HIPAA-disabled people in the elderly population, as produced by the DYNASIM model, is too low. Although the researchers assume that projected disability rates by age will fall over time as mortality rates decline (by how much is not said), and some decline in disability rates seems reasonable given health improvements among the elderly, their calculations are surprisingly low even so, given that these disability rates must be multiplied by the rapidly increasing number of very aged people.
Moreover, assuming rapidly declining rates of disability over time is not conservative, as would be appropriate for a model used to cost out policy proposals; it would be very good to see the sensitivity of results to this key assumption. In particular, the Urban Institute researchers should also report results based on no morbidity improvement, and on mortality improvements applied to both healthy and non-healthy segments of the population.
But this may not be the most significant questionable assumption in the DYNASIM model. The researchers claim, with no indication of source or empirical basis, that there is a large difference between the need for LTSS and use of paid LTSS. They state the following in the HHS supporting document:
Table 1 displays expected LTSS needs from age 65 to death. It presents life expectancy and then the mean and distribution of the duration of disability for those turning 65 in 2015-2019. The typical person who is alive at age 65 can expected [sic] to live another 20.9 years. Fifty-two percent can anticipate having at least some needs for LTSS; 19 percent are expected to have needs that last less than a year, and about 14 percent are expected to have needs that extend beyond five years. ….
The figures for average LTSS needs mask substantial variation by subpopulation. The average duration of disability is much higher for women than for men – about 2.5 years for all women as compared to 1.5 year for all men. ….
Table 2 examines use of paid LTSS provided for a disability from age 65 onward measured in days. Of the average two years that an older adult will have a disability, they will receive paid care for about one year, and informal caregivers only, such as family and friends, will make up the difference. ….
While on average, individuals will need one year of paid LTSS, 48 percent of individuals will not use paid, formal LTSS at all (….). Among those who need paid LTSS services, about half will need less than a year, and a little more than 10 percent will five years or more.
The researchers are assuming, or perhaps estimating, that fully half of LTSS care at the severe HIPAA-level of disability is covered through informal sources. They need to support and document fully such an important and non-intuitive assertion. As we will see below, such a strong assumption is inconsistent with results in the published literature.
The researchers indicate that there would be an increase in utilization—more spending on new services (presumably mostly replacing uncompensated care-givers)—when new financing programs are introduced. However, this increase, even for the mandatory comprehensive program, is indicated to be only about a third of all program spending, considerably less than the half the need that they claim is not now covered by formal paid care. This does not seem credible: Why assume that once covered by insurance, anyone in a comprehensive program would not simply demand and get what is available to him or her under the rules of the program, given the significant need? Even for incomplete “pool-of-money” private insurance with a somewhat restrictive service reimbursement structure, Milliman assumes that up to 90 percent of benefits are utilized. For a comprehensive public cash benefit program, it is much more reasonable to assume that close to 100 percent of benefits will be utilized. Indeed, this is particularly true when assisted living facilities, which are generally more pleasant than nursing homes, are becoming more prevalent.
Key Model Assumptions are Inconsistent with those found in the Professional Literature, Leading to Lower Estimated Costs
The current life expectancy at age 65 according to DYNASIM is significantly higher than that stated by the Social Security Trustees. DYNASIM has 19.6 more years for men and 22.2 more years for women; the Trustees have 18.3 more years for men and 20.7 more years for women who are age 65 in 2015. So it seems that the DYNASIM model has healthier, longer-lived, individuals in its model than exist in the actual current population.
The estimates of the incidence and duration of LTSS needs by the Urban Institute researchers are on the low side compared to other results in the literature for similar measures of lifetime disability, that is, HIPAA-criteria-to-be-eligible-for-benefits. Based on the 1986 National Mortality Follow-back Survey, Murtaugh, Spillman, and Warshawsky (2001) find that the percent disabled for the population aged 65 and over is 68.5 percent and the expected duration of disability is 2.2 years — 1.6 years for men and 2.6 years for women; thus, both the incidence and duration are higher than stated by the Urban Institute researchers. Brown and Warshawsky (2013), based on the HRS, find a slightly lower disability incidence—42 percent for men and 55 percent for women—than the Urban Institute researchers, but Brown and Warshawsky’s statistics are for a somewhat healthier population: those who would not immediately be rejected by underwriting for purchase of a LTCI policy. The average disability duration found by Brown and Warshawsky, however, is similar to Murtaugh, et al., 1.5 years for men and 2.6 years for women.
Brown and Warshawsky use these statistics, with no haircut for unpaid care, to price various insurance products. Using the Brown and Warshawsky (2013) model, based on the HRS data, for the whole population—men and women combined and including those already disabled—they find (unpublished) that the rate of severe disability incidence is 49.6 percent and the average duration is 2.2 years, somewhat higher than the Urban Institute’s base statistics. (Our thanks to Jason Brown for doing these calculations.)
Based on the National Long-Term Care Survey, Stallard (2011) finds that the average disability duration is 2.24 years, 1.5 years for men and 2.83 years for women. Stallard also finds that, on average, 1.53 years were spent in paid nursing home and home health care, 0.89 years for men and 2.04 years for women. This latter result—that about two-thirds of HIPAA-level disability is paid care, even without considering assisted living—seems to be a more reasonable finding, on its face, than that made by the Urban Institute authors that only half of care for significant disability is paid.
Another, earlier, microsimulation model, based on different data sources, by Kemper, Komisar and Alecxih (2005/6) also finds incidence and duration significantly higher than the Urban Institute model, although the Kemper, et al. definition of disability used was broader than the HIPAA one. In particular, Kemper, et al. report that 69 percent of population turning 65 in 2005 will need LTSS (79 percent of women and 58 percent of men), for, on average, 3 years, women for 3.7 years and men for 2.2 years. About 20 percent of this population will need care for more than five years. If a definition of disability more restrictive than the HIPAA one is instead used, Kemper, et al. find that 61 percent will need LTSS at some time in their lives, for an average 2.2 years. Presumably using the HIPAA definition would produce results somewhere in between these two figures, but significantly larger in any case than used in the DYNASIM model.
In terms of paid care, Kemper, et al. report an average duration of 1.6 years for the somewhat looser definition of disability. According to Murtaugh and Spillman, based on data from the 1993 National Mortality Follow-back Survey, there are 1.7 expected years of HIPAA-level disability and 1.3 years in paid home health and nursing home care. (Murtaugh, Christopher M., and Brenda C. Spillman, “Lifetime Risk of Chronic Disease, Disability and Long Term Care,” Presentation for the 12th Annual Intercompany Long Term Care Insurance Conference, 2012.) Again, all these results, taken together, lead to a judgment that the DYNASIM model is using prevalence and duration statistics for disability and paid care significantly lower and shorter, respectively, than found in the mainstream of the professional literature.
The Urban Institute researchers report that the current sources of payment for LTSS as found in DYNASIM are quite different than found in aggregate statistics reported by the Commission and by the Congressional Budget Office (CBO) (2013). In 2011, according to the CBO, paid LTSS was composed of $68 billion by Medicare, $60 billion by Medicaid, $39 billion out-of-pocket, $12 billion by private insurance, and $12 billion by other sources (such as the federal government through the Veterans Administration or private charitable organizations). CBO also reports that, according to their tabulations of the HRS, about 13 percent of the age 65 and older population has LTCI; the Urban Institute researchers have 8.6 percent of the population covered by LTCI in their baseline.
In terms of spending, the Urban Institute researchers report, according to DYNASIM, that 44.8 percent of LTSS expenditures are from public sources (9.9 percent Medicare, 34.3 percent Medicaid, and 0.6 percent other) and 55.0 percent of spending is from private sources (52.3 percent out-of-pocket and 2.7 percent private insurance). Even after accounting for some conceptual controversy about the exact role of Medicare in paying for LTSS, the proportion covered by out-of-pocket sources is significantly lower for CBO compared to the Urban Institute; this large difference needs a full explanation by the Urban Institute researchers because the CBO figures are not model estimates but actual spending as reported in the national health accounts. These assumptions call into question the reported findings of the Urban Institute researchers on the impact of the various new programs on Medicaid spending.
Policy Analysis Bias
Finally, there seems to be a policy analysis bias inherent in the Urban Institute DYNASIM model. New social insurance schemes are extensively investigated, but recommendations to improve private incentives for insurance and limit Medicaid coverage are not. Although it is hard to tell for sure, it does not seem that the DYNASIM model would be capable of any investigations of the recommendations of the conservative group of the Long-term Care Commission.
The failure of CLASS would have been prevented by a more reasonable score of the proposed program by the CBO, as was produced by the CMS actuary at the time. Similarly, the fundamental source of the growing failures of the health exchanges under the ACA mainly relate to the significant overestimates by many modelers of the take-up of health insurance coverage, which led directly to program design misspecifications. Perhaps if the modeling processes in both instances had been more transparent and open to extensive outside critical comment and advice, and had taken place over a longer time period, the policy outcomes would have been better and the political controversy more limited.
Scoring in the LTSS area needs to be based on widely accepted facts (i.e., a fair reading of the literature and/or settled empirical findings), completely transparent (with all major assumptions disclosed and justified), and robust (to consider all types of policy interventions). Scoring should also be alternatively illustrated by conservative results and assumptions, to give a sense of the possible range of outcomes. Unfortunately, despite the great apparent effort of the Urban Institute, their model, as presently constituted, cannot serve this purpose.