Medical Underwriting In Long-Term Care Insurance: Market Conditions Limit Options For Higher-risk Consumers

Abstract

A key feature of private long-term care insurance is that medical underwriters screen out would-be buyers who have health conditions that portend near-term physical or cognitive disability. We applied common underwriting criteria based on data from two long-term care insurers to a nationally representative sample of individuals in the target age range for long-term care insurance (50–71 years of age). The screening criteria put upper bounds on the current proportion of Americans who could gain coverage in the individual market without changes to medical underwriting practice. Specifically, our simulations show that, for the target age range, approximately 30% of individuals whose wealth meets minimum industry standards for the suitability of long-term care insurance would have their long-term care insurance application rejected for medical reasons. Among the general population–without considering restrictions on wealth–we estimate that 40% would be disqualified. In evaluating long-term care financing reforms and their potential to increase private insurance rates, as well as to reduce financial pressure on public safety-net programs, policymakers need to consider the role of underwriting in the market for long-term care insurance.

INTRODUCTION

Most Americans are unprepared for the financial risk posed by the potential need for care as they age. When faced with significant disability, older adults tend to rely on family caregivers, exhaust modest personal savings, and rely on the Medicaid program as a payer of last resort. As the older population grows and the cost of long-term care continues to increase, one potential remedy might be to expand private insurance coverage for long-term care.

Long-term care comprises a wide range of services, including the services of paid homemakers and aides, adult day care, assisted living, and nursing home care. In 2015, the median national rate for a year of nursing home care in a semiprivate room was $80,300; single occupancy in an assisted living facility was $43,200; and the services of a home-health aide for four hours per day were $29,200 (1). Such costs are growing faster than the rate of inflation (2), and the number of Americans expected to need paid long-term care is projected to increase from 12 million in 2012 to 27 million people in 2050–an increase from 3.7% to 6.8% of the total population (3,4).

The majority of households have insufficient personal savings to maintain their living standards during a healthy retirement, much less a cushion to pay for potential long-term care needs (5). A typical long-term care insurance plan might provide coverage for up to three years of nursing home care and even greater amounts of home care (usually with a delay after the onset of disability) to beneficiaries who need assistance with at least two activities of daily living or who need supervision because of cognitive impairment. The average age of individual long-term care insurance buyers in 2010 was 59 (6). Among current 65-year-olds, most will never reside in a nursing home; between four and ten percent, however, will live in a nursing home for five years or more (7), a prospect that would exhaust the savings of most households. Insurance provides value by protecting people from catastrophic financial risk without needing to set aside funds to cover the maximum possible expense (i.e., paying regular premiums instead). Yet few people avail themselves of long-term care insurance: about 10% of Americans ages 60 – 65 had a long-term care insurance policy in 2010 (8). A variety of market failures have restricted the reach of long-term care insurance to a narrow slice of relatively healthy and affluent buyers.

Proposals for subsidies, outreach, and tax preferences for long-term care insurance need to be understood in the economic context that shapes the individual (nongroup) market, which in 2014 comprised roughly two-thirds of policies in force and 86% of new policies (9). Adverse selection can create instability in any market for insurance where consumers buy products voluntarily: Discrepancies between individuals’ and actuaries’ knowledge about need and preference for long-term care can dissuade healthier and lower-risk consumers from buying the product, resulting in premium prices that reflect higher-than-average risk, which in turn appeal to an even narrower market of consumers.

One strategy for broadening the risk pool is to offer guaranteed- or simplified-issue plans to a defined group, such as through an employer. In guaranteed issue, coverage does not require applicants to undergo an exam or answer any questions about medical history questions; in simplified issue, no exam is required, but applicants typically must answer some questions about their medical history. Few employers, however, offer long-term care benefits, and the group market has contracted in recent years (10). Instead, in the context of the individual market, underwriters attempt to correct for information asymmetry by screening would-be buyers for health conditions that portend current, or near-term (within 5–7 years) physical or cognitive disability. Underwriting accuracy confers a competitive advantage, and companies protect their protocols as confidential assets (see Appendix, Section 1 and Exhibit 1, for a description and visual representation of the underwriting process).

Exhibit 1.

Prevalence of underwriting characteristics and their effects on probability of approval

	Insurance Applicantsa N=15659		U.S. Population Sampleb N=13770
	Prevalence	Effect on probability of approvalc	Prevalence
Approved for LTCI	0.758
Age
18 – 49	0.122	d	e
50 – 59	0.341	−0.007	0.525
60 – 69	0.484	−0.037****	0.417
70 – 71	0.053	−0.081****	0.058
Female	0.449	0.016**	0.524
Socioeconomic characteristics
Attained college degree	0.476	0.020***	0.313
Employed	0.625	0.030****	0.585
Cognitive and functional ability
Delayed word-recall score < 7/10	0.300	−0.035****	0.830
Experiences memory loss	0.236	−0.020**	0.217
Difficulty taking medication	0.016	−0.094***	0.025
Difficulty with activities of daily living	0.004	−0.522****	0.122
Diagnosed health conditions
High blood pressure	0.502	−0.078****	0.495
Back pain	0.409	−0.101****	0.389
Arthritis	0.245	−0.111****	0.469
Diabetes	0.201	−0.415****	0.183
Heart problems	0.199	−0.130****	0.161
Psychiatric illness	0.184	−0.123****	0.196
Lung problems	0.102	−0.086****	0.080
Cancer	0.057	−0.111****	0.100
Stroke	0.016	−0.528****	0.046
Health care use
Hospitalization, previous 2 years	0.533	−0.065****	0.213
Long-term care, previous 2 years	0.016	−0.083**	0.055
Health behaviors
Drinks alcohol	0.890	0.024**	0.655
Ever been a smoker	0.379	−0.013*	0.570
Currently a smoker	0.085	−0.114****	0.187
Body mass index (BMI)
Underweight (BMI<18)	0.007	−0.174****	0.009
Normal or overweight (BMI 18 – 30)	0.595	d	0.614
Obese (BMI 30 – 40)	0.374	−0.046****	0.322
Extremely obese (BMI 40+)	0.024	−0.268****	0.055
Constant		0.920****

Source/Notes: SOURCES

^aAuthors’ analysis of insurance sample, which comprises applicants for long-term care insurance for two US firms in 2008 – 2012.

^bAuthors’ analysis of population sample, which is the Health and Retirement Study (HRS) 2010–2011, respondents age 50–71.

NOTES Prevalence estimates from the HRS are weighted to correspond to the American Community Survey, a sample of noninstitutionalized U.S. adults.

^cWe modeled probability of approval in a multivariate regression. These estimates represent the difference in approval probability as compared to the reference group. Standard errors are reported in Exhibit A.5 of the Appendix.

^dReference category.

^eHRS respondents under age 50 are excluded from analysis.

^*p<0.1,

^**p<0.05,

^***p<0.01,

^****p<0.001

Factors specific to long-term care exacerbate uncertainty for insurers. People purchase long-term care insurance as protection against financial risk that may be decades in the future. This long time horizon exposes firms not only to the risk of changing longevity and volatile interest rates, but also fluctuations in health costs and disability trends (11). In contrast to life insurance, where mortality is more stable and predictable over the long term, long-term care insurance underwriters have to consider how changes in population health and functional status, health technology, and consumer preferences (e.g., the shift away from nursing home care and toward home- and community-based care) will change expected claims. The added uncertainty makes long-term care exceptionally challenging to underwrite and insure at a stable price relative to other types of insurance offered in voluntary markets. Moreover, in an extended period of low interest rates and rising long-term care costs, firms selling long-term care insurance have experienced steady losses. They have responded by increasing prices and tightening underwriting requirements or, in many cases, exiting the market altogether (9,12).

In considering long-term care insurance reforms, policymakers would benefit from a deeper understanding of how medical underwriting can affect who has access to insurance. Existing analyses simulate underwriting practices by applying heuristics from industry experts or from field-underwriting guides for insurance agents (13,14,15). There is a need for empirical evidence, however, which can inform policy. To meet that need, we used the actual coverage decisions of insurance firms to model insurance eligibility of Americans who are at prime ages for purchasing long-term care insurance. Our analysis is based on data collected from those firms, on which they based underwriting decisions.

METHODS

Data

We developed coverage-approval models from a dataset composed of application decisions for roughly 1500 individuals who applied for long-term care insurance policies at two carriers in 2008–2012. Underwriting variables used in our analysis are listed in Exhibit 1 and include demographics, socioeconomic factors, cognitive and functional abilities, health conditions, use of health care, health behaviors, and body mass index (BMI). (For additional detail on data collection, see Appendix, Section 2.1 (16).)

Data on the U.S. population were taken from the public use files of the Health and Retirement Study (HRS), a nationally representative survey of older U.S. (17). We use a database from the RAND Center for the Study of Aging (18). The HRS includes information on the health, living arrangements, employment, income and assets, and insurance status of respondents. The detailed information from the HRS allows us to align the underwriting items to the HRS survey questions. To compose a snapshot of the current long-term care insurance market, in the analysis we used survey responses in the HRS collected in 2010 and 2011, which roughly overlapped with the timing of insurance applications.

Analysis

We estimated an empirical model of underwriters’ coverage decisions to identify factors that determined whether a firm offered coverage to an applicant. We give the full model specification in the Appendix, Exhibit A.5 (19). We used the model to impute underwriting probabilities for each respondent in the HRS between the ages of 50 – 71 (the HRS sample is representative of the population over 50, and the maximum age in the insurance data is 71). We designated those whose predicted probability of approval was 0.5 or less to be “likely disqualified” and those whose probability of approval was greater than 0.5 to be “likely approved.” We used the assets, income, ethnicity, and race information in the HRS to refine the average predicted rates of approval (e.g., by differentiating among socioeconomic groups).

Industry guidelines, which are encoded in most state regulations and are almost universally followed by insurance sellers, compel agents to confirm that applicants meet minimum financial benchmarks before proceeding (19,20). To reflect this practice, we created a subsample of individuals for whom long-term care insurance would be considered financially suitable. The sample was composed of individuals whose income and assets exceeded $20,000 and $30,000, respectively (for a couple, $30,000 and $50,000). Within the subsample, we also report results for those whose assets, excluding housing, exceeded $250,000–a benchmark that we chose because it represents the approximate cost of three years of care in a nursing home.

Limitations

Our study provides the first detailed analysis of how underwriting policies of companies that sell long-term care insurance may limit market size. It is important, however, to note the study’s limitations.

First, when we extrapolate our underwriting results to the general population, we generate predictions for a population-representative sample from a model estimated on an applicant pool that differs from the general population in both observed (better health, less health-care use, and more education) and unobserved aspects (generally better off and more financially savvy, but perhaps with higher demand for paid care). In generalizing from the applicants to the HRS respondents, we describe a hypothetical scenario, based on the best available empirical evidence.

Second, our applicant sample represents approximately 5% of the total market over the period studied and caution should be taken in extrapolating our findings to the industry as a whole. The disqualification rates that our model predicts are similar to industry-wide rates among people of ages similar to those in our sample. We discuss the generalizability of our model in the Appendix, Section 4.2 and Exhibit A.9 (19).

Third, some health characteristics that are important components of the underwriting model are not available in the HRS. For instance, although the underwriters use a proprietary, multi-item instrument to predict future cognitive decline, the HRS cognitive function questions do not correspond exactly. We match the model on self-reported memory and ability to recall a 10-word list. Several “knockout conditions,” such as multiple sclerosis and other degenerative chronic diseases, which would usually disqualify an applicant at the field underwriting stage, were not available in the HRS. These discrepancies make the underwriters’ criteria more sensitive than the HRS in identifying the risk of needing long-term care. In missing these knockout conditions, we potentially overestimate the general population’s probability of obtaining underwriting approval. In addition, some results were counterintuitive, which may potentially reflect limitations in our analytic approach. For instance, a history of long-term care use would normally be an automatic disqualifier. We were unable to obtain a good estimate of this effect, though, because few applicants in our sample reported a history of long-term care use, perhaps because such individuals had already been disqualified in field underwriting. The positive estimate of the influence of alcohol intake on approval most likely does not signal that drinking improves the likelihood of approval. Instead, people who abstain may be more likely to have other disqualifying health conditions (21,22).

In addition, although we extrapolate our findings to compare predicted approval rates for white, Black, and other nonwhite minorities, these findings are based solely on the health and demographic profile of those populations in aggregate. Our underwriting data do not contain information on applicants’ race or ethnicity.

RESULTS

Underwriting and Approval for Long-Term Care Insurance Purchase

The sample of 15,659 applicants for long-term care insurance consisted of 3,782 individuals (24%) who were not approved in the underwriting process and 11,877 (76%) who were qualified to purchase a policy (Exhibit 1). Older applicants had lower approval rates: when we controlled for health and socioeconomic characteristics, each 10-year increase in age significantly decreased approval probability, and the average applicant over age 70 had a 8.1%-point lower probability than an applicant under 50 in similar health (p=<0.01). Applicants with a college degree were 2%-points more likely to qualify than those with no college education, and those who were employed were 3%-points more likely to be approved than those who were unemployed. Having any chronic health diagnosis significantly lowered the probability of approval, in comparison with not having the condition (p<0.01). Among the most influential conditions were diabetes (41.5%-point lower) and history of stroke: applicants with history of those conditions had 41.5%-point and 52.8%-point lower approval rates respectively. Back pain, arthritis, heart problems, psychiatric illness, and cancer were each associated with at least a 10%-point decreases in approval rates. In the realm of health behaviors, current smokers were 11.4%-points less likely to be approved than nonsmokers, and those who drank any alcohol were 2.4%-points more likely to be approved than nondrinkers (p<0.01). Applicants who had previously needed long-term care services were only 8.4%-points less likely to qualify (p=0.02) (although it should be noted that very few insurance applicants had a history of long-term care use), while applicants who had hospital use in the prior two years had a 6.5%-point decrease in approval (p<0.01). Applicants who were dangerously underweight (body mass index <18) or extremely obese (BMI>40) were much more likely to be disqualified compared to someone whose weight was within the normal to overweight range (BMI 18 – 27). Being underweight or extremely obese decreased approval by 17.4%-point and 26.8%-points, respectively; for a 5′8″ tall person, those BMI thresholds correspond to 118 and 263 pounds.

Underwriting applied to the representative U.S. population

In the 2010 – 2011 wave of the HRS, 13,770 individuals fell within our targeted age range of 50 – 71. About half (53%) met the minimum recommended financial-suitability standards, and of those, about half (48%) had non-housing assets exceeding $250,000. In column 1 of Exhibit 2, we report the mean of the predicted probabilities in each financial category. For those whose wealth met financial-suitability guidelines, the mean was 0.611 (95% confidence interval: 0.60 – 0.62). Column 2 shows the percent who were likely approved (i.e., percent of the sample whose predicted probability exceeded 0.5). Seventy percent of the population in the financially suitable category was in the likely approved category for long-term care insurance (95% CI: 69 – 72%). Among this group, 75% of the individuals in the households with nonhousing assets over $250,000 were likely approved (95% CI: 73 – 77%). In the target age group of the general population–regardless of income and savings–60% (95% CI: 58 – 61%) were in the likely approved category.

Exhibit 2.

Summary of imputed estimates of underwriting approval for long-term care insurance.

Category of suitability of long-term care insurance for respondent:a	Mean approval probabilityb	Likely to be approvedb,c	Sample size	Population, millionsb
Suitable	0.611	70.1%	6166	37.98
Assets $30,000–250,000d	0.580	65.6%	3349	19.50
Assets $250,000 and overd	0.644	74.9%	2817	18.48
Not suitable	0.441	47.4%	7604	33.08
Full sample	0.532	59.6%	13770	71.07

SOURCE Authors’ analysis of the Health and Retirement Study (HRS), 2010–2011, respondents ages 50 – 71.

NOTES Probability estimates for HRS respondents were predicted from a multivariate logistic regression model, summarized in Exhibit 2, that the authors estimated from approval information on the insurance applicants. The full model with odds ratios is reported in Exhibit A.4 of the Appendix.

^aFinancial suitability of long-term care insurance for the respondent is defined as household yearly income>$20,000 and nonhousing assets>$30,000 for a single person, and respectively $30,000 and $50,000 for a couple.

^bPopulation estimates were weighted to correspond to the American Community Survey, a sample of noninstitutionalized U.S. adults. Estimates with standard errors are available the Appendix, Exhibit A.6.

^cRespondents were designated as “likely to be approved” with imputed approval probability > 0.5. Thus the percentage reported in this column represents the estimated population proportion with predicted approval > 0.5.

^dAssets include the net total of all non-housing assets: property, business assets, other real estate, and financial wealth (including retirement accounts), less non-mortgage debt.

Exhibit 3 shows that in both the applicant pool and HRS sample, approval rates declined steeply with age, with the largest decrease occurring at age 60 and above in the financially suitable HRS sample.

EXHIBIT 3. Comparison of approval probabilities of long-term care insurance applicants to the US population, mean predicted probabilities by age.

Source/Notes: SOURCE Authors’ analysis of the Health and Retirement Study (HRS), 2010 – 2011 (N=13770).

NOTES Among insurance applicants, estimates are the proportion approved for long-term care insurance. Probabilities for HRS sample are predicted from a multivariate logistic regression model estimated from approval information on the insurance applicants. HRS statistics were weighted to correspond to the American Community Survey, a sample of noninstitutionalized, adults in the U.S. Financial suitability was defined as yearly income>$20,000 and nonhousing assets>$30,000 for a single person; respectively $30,000 and $50,000 for a couple.

Exhibit 4 shows that approval probabilities increased steadily with household wealth, even above the 40th percentile (net assets > $100,000), and that mean approval probabilities ranged from approximately 0.6 to 0.75 among those in their 50’s, and 0.45 to 0.6 among those in their 60’s.

EXHIBIT 4. Mean approval probabilities by wealth percentile, US population age 50 – 59 and 60 – 69.

Source/Notes: SOURCES Authors’ analysis of Health and Retirement Study (HRS) 2010 – 2011 (N=13770).

NOTES Approval probabilities are predicted from a multivariate logistic regression model estimated from approval information on the insurance applicants. HRS statistics are weighted to correspond to the noninstitutionalized US population from the American Community Survey. Total wealth includes all housing, property, and financial wealth, less debt.

When we applied underwriting parameters to different racial and ethnic populations in the U.S., we found differential approval rates resulting from underlying differences in health status across groups. For instance, using information from the HRS, 59.6% of whites were likely to be approved (95% CI: 57.7 – 61.2), compared to only 45.0% (42.1 – 47.9%) of blacks and 52.0% (46.7 – 57.3%) of those from other nonwhite minority groups.

As expected, approved applicants had health profiles that were substantially different from applicants who were disqualified. The differences translated to an approximately threefold higher probability of disability within 5 years among those applicants whom we identified as likely-disqualified relative to those who were likely approved (see Section 4.3 and Exhibit A.10 in the Appendix (19)).

DISCUSSION

Current medical underwriting would exclude a large proportion of Americans from being able to buy long-term care insurance in the voluntary, private market. Due to self-selection and field underwriting, long-term care insurance applicants are considerably healthier and wealthier than the general population. We estimate that disqualification rates in the general population would be higher than the rate observed among applicants: 40% of the general population of middle-aged Americans aged 50 – 71 would most likely be rejected if they applied for long-term care insurance and experienced underwriting standards comparable to those experienced by individuals in our sample. Underwriters would likely disqualify a little more than one-quarter of people with assets between $30,000 and $250,000, even if subsidies, information campaigns, or other inducements were able to encourage them to apply for coverage. Individuals with assets in that range are presumably the ones who can afford premiums and are motivated to protect their assets, rather than spend down their savings until eligible for Medicaid, but who are not necessarily wealthy enough to self-insure.

Medical underwriting limits access to insurance regardless of the affordability of the policies. Thus, the ability of approaches that make premiums more affordable to spur insurance purchase is further limited by medical underwriting, although the limits will diminish somewhat as greater numbers of individuals apply for insurance. In fact, the modest impact to date of strategies such as long-term care insurance tax credits or enhanced Medicaid asset protection for individuals who purchase policies should be interpreted in the context of an insurance market where underwriting plays an important role (23,24).

In particular, underwriting limits access of people living with chronic conditions that are relatively common among middle-aged adults. Diabetes, a history of stroke, extreme obesity, and having any difficulty with at least one activity of daily living (ADL) were the most important factors in the underwriting decision. Diabetes and ADL difficulty are widespread, affecting an estimated 18% and 12% respectively of people in this age group.

Because of underlying health differences, racial and ethnic differences exist in access to long-term care insurance. These differences are a function of disparities in health status for racial and ethnic minorities in aggregate–something that we observe in the HRS data and that has been well-documented elsewhere (25). Nonwhite Americans are at higher lifetime risk than whites of entering a nursing home, and if they do, and more likely to be in homes of lower quality (26). African Americans are less likely to purchase long-term care insurance than whites and are less price-responsive (32). Our analysis suggests that underwriting may explain part of that discrepancy in insurance purchase, though consumer preference may also play a role.

Implications for policy

Although adverse selection creates market conditions that necessitate medical underwriting for voluntary insurance, these dynamics are not static. For instance, reforms could mitigate adverse selection by expanding the risk pool and encouraging purchase at younger ages, by offering reinsurance, or by mandating insurance. Underwriters could relax their criteria in response to a broader, younger applicant pool if they believed that adverse selection had lessened and this would potentially lower rejection rates. Firms might also respond by reducing premiums and loosening underwriting criteria. But the task of convincing households to forego present consumption to protect against risks that could be decades in the future still would be tremendously difficult, even if the economic advantages were substantial. Incentives targeted toward the group- and employer-based markets, where insurers can offer guaranteed-issue policies or relaxed underwriting standards, could potentially encourage households to plan for long-term care needs at younger ages, when they are less likely to have disqualifying conditions. Recent evidence indicates, however, that the group market has stagnated and that many long-term care insurers have discontinued sales and marketing efforts to new groups (11). The Community Living Assistance and Support (CLASS) Act–a provision of the Affordable Care Act (ACA) that would have created a voluntary, public long-term care insurance program–was an effort to enroll younger populations. Yet, projections showed low enrollment and spiraling costs due to adverse selection. The Obama administration declared the law actuarially unsustainable and Congress quietly repealed it in 2013 (27). It is difficult to imagine a voluntary program that could surmount the problem of adverse selection, without incurring the same limitations that currently affect the private market.

One recent proposal put forward to address long-term care financing needs includes a catastrophic social insurance program that would cover nursing care after a 2-year waiting period and offer a lifetime benefit thereafter, covering 95% of the population while substantially offsetting Medicaid spending and out-of-pocket costs (28). If such a program existed, the role of voluntary long-term care insurance would shift to become a more limited product, covering the interim period after the onset of disability and before the catastrophic insurance took effect. Lower premium prices for this limited product could attract a healthier pool of consumers and alter underwriting considerations, which could result in lower underwriting rejection rates. A catastrophic social-insurance program would represent a large new program, however, and securing the necessary revenue would be challenging in the current fiscal and political environments.

For individual health insurance, the ACA addressed adverse selection by mandating that individuals either buy private insurance or face penalties. Although a mandate for public or private insurance could address adverse-selection in long-term care insurance, public support for such a mandate is weak, and likely to remain so in the near term.

The public-policy goals of bringing more private dollars into the long-term care system are to provide financial protection to older people and their families, reduce the growth in public spending for long-term care, and support the service infrastructure to meet growing demand. Making progress toward these goals will grow more urgent as the U.S. population ages. Our findings do not preclude a role for the private market in long-term care reform. In fact, politically viable solutions are likely to consist of some combination of public safety-net programs together with incentives for increased personal savings and long-term care insurance purchase (29). Unless policymakers can find ways to broaden the risk pool significantly, however, current underwriting practices are likely to persist. Thus, simply subsidizing the voluntary, nongroup market, without addressing the market conditions that necessitate underwriting, will provide protection only to those with the lowest risk. It will not achieve the goal of expanding insurance protections to as many Americans as possible and assuring them access to a secure financing system for long-term care.