Out-of-pocket costs attributable to dementia: A longitudinal analysis

Abstract

Background:

Alzheimer’s disease and related dementias (ADRD) affects 5.7 million Americans, and is expensive despite the lack of a cure or even treatments effective in managing the disease. The literature thus far has tended to focus on the costs to Medicare, despite the fact that one of the main characteristics of ADRD (the loss of independence and ability to care for oneself) incurs costs not covered by Medicare.

Methods:

In this paper, we use survey data for 2002–2016 from the Health and Retirement Study to estimate the out-of-pocket costs of ADRD for the patient and their family through the first 8 years after onset of symptoms, as defined by a standardized 27-point scale of cognitive ability. A two-part model developed by Basu and Manning (2010) allows us to separate the costs attributable to ADRD into two components, one driven by differences in longevity and one driven by differences in utilization.

Results:

We identified a cohort of 3,619 incident dementia cases, 38.9% were male, and 66.9% were non-Hispanic White. Dementia onset was 77.7 years of age, on average. OOP costs attributable to dementia are $8,751 over the first 8 years after onset. These incremental costs are driven by nursing home expenditures, which are largely uninsured in the US. OOP spending is highest for whites and women.

Conclusion:

The financial burden of ADRD is significant, and largely attributable to the lack of wide-spread long-term care insurance.

1. INTRODUCTION

Alzheimer’s disease and related dementias (ADRD) affects 5.7 million Americans, and forecasts project that number to grow to 13 million by 2050.¹ Costs associated with ADRD are increasing simultaneously – estimates project annual per person costs doubling from 2010 to 2050.² Assistance with personal activities and supervision (i.e., long-term care) are essential components of care for people living with ADRD. However, there is a gap in insurance coverage for these services, making long-term care (LTC) one of the largest financial risks facing older Americans.³ We measure the incremental out-of-pocket (OOP) costs attributable to ADRD on a dementia-incident cohort to quantify this financial risk. Second, we examine differences in OOP spending by race/ethnicity and sex to quantify who is bearing the financial burden.

1.1. The insurance gap for long-term care

Medicare, health insurance for older and disabled Americans, does not cover LTC. Medicare covers skilled therapy during short nursing home stays (up to 100 days following hospital stays) and at home for homebound individuals if the services must remain medically necessary and are reordered every 60 days.

Medicaid, health insurance for individuals in and near poverty, covers custodial care in nursing homes and home and community-based settings for eligible beneficiaries. However, eligibility rules and generosity of benefits vary by state. Dual eligibility (enrollment in both Medicare and Medicaid) can range from full coverage to Medicaid covering some Medicare premiums (i.e., Medicare Savings Programs). Limited state funding for home and community-based waiver programs, which cover LTC received in the home, means applicants have typically been subject to long wait periods.⁴

Veterans can receive LTC through the Veteran’s Health Administration’s health care system (VA), provided they meet eligibility requirements based on service-connected (disability) status, level of disability, and income.⁵

Private long-term care insurance (LTCI) must be purchased when the beneficiary is in good health and is typically costly, as reflected in the characteristics of LTCI purchasers, who are more likely to be college-educated, working, and wealthier.⁶ Only 7.5 million Americans have private LTCI as of Jan 1, 2020—and the market is shrinking.

Most Americans, and 72.9% of our baseline sample, are without any LTCI coverage. They must choose between paying OOP for their LTC needs or forgoing some or all formal care, relying on informal/unpaid care to fill the gaps.

1.2. Literature on ADRD-Related Spending

Most research on the costs of ADRD focuses on estimating the costs to the Medicare and Medicaid programs.^7,8 A small but growing literature evaluates OOP costs associated with ADRD. Dwibedi et al. use spending measures from the Medicare Current Beneficiary Survey, finding that individuals with ADRD are paying an additional $1,390 on average, per year.⁹ Three studies use measures from the HRS, the same data source as we use,^10–12 finding average annual incremental costs ranging from $5,491–6,194. These studies use a cross-sectional design, which relies on the prevalence of dementia in the population during a given period, and estimates average costs for people at various stages of the disease. Because costs are not constant over the course of the disease to Medicare^7,8 or Medicaid,⁸ and prevalence is changing over time,² estimates produced using these methods may not generate accurate forecasts.¹¹ Kelley et al. address this by focusing their analysis around the years leading up to death, however this approach may exclude the time of diagnosis and symptom onset, which tends to be characterized by higher costs to Medicare.^7,8 Hudomiet et al. use simulation models to estimate OOP spending from age 65 to death, finding that dementia increases spending by $38,540 on average, largely driven by nursing home use.¹³

While the literature estimating spending due to dementia by race/ethnicity and gender is scant, there is considerable literature suggesting that racial and ethnic minorities tend to use less formal/paid care.¹⁵ Due to higher life expectancy and lower socioeconomic status relative to men, we also know that women are more likely to use formal LTC.¹⁶ To account for heterogeneity in spending, we also stratify the sample, estimating period-specific total and incremental costs for men, women, non-Hispanic White, and Black/Hispanic cases.

2. METHODS

2.1. Study design and data sources

We use 2000–2016 data from the Health and Retirement Study (HRS), a longitudinal survey of a nationally representative sample of adults over the age of 50. The HRS conducts interviews every two years, collecting information on a variety of topics including health status, health care utilization, and expenditures. The HRS also collects information from a proxy respondent following the respondent’s death. We add data from these exit interviews to the HRS core data to a complete picture of OOP spending. We link this data with standardized covariate data from the RAND HRS Longitudinal File (2016, Version 2).¹⁷

2.2. Study population

We identify ADRD cases using a survey-based method of measuring cognitive function, designed and validated by Langa-Weir.¹⁸ At each wave, the HRS administers cognitive tests using an adapted version of the Telephone Interview for Cognitive Status (TICS), with missing measures then imputed by the HRS. Respondents are classified as having ADRD after scoring in the range of 0–6 out of 27. If a respondent is unable to complete the assessment, a proxy (typically a spouse or other family member) is asked about the respondent’s impairment and difficulties with activities of daily living. When combined with the interviewer’s assessment of the respondent’s level of impairment, respondents are classified as having ADRD after scoring 6 or more out of 11. We assume respondents have ADRD in all subsequent waves, and require at least one wave of data prior to identify ADRD-onset.

36,176 respondents completed at least one wave of the HRS during our study period. ADRD-diagnosed respondents were required to be present at least one wave prior to diagnosis, leaving 33,701 respondents. The majority of those dropped had a dementia diagnosis at the start of our study period. After dropping respondents who report race as “Other”, 30,486 respondents (3,780 cases) have complete information on all matching criteria. To select a control group, we match cases with potential controls based on HRS entry wave, birth year, gender, race/ethnicity (non-Hispanic White and Black/Hispanic), and education. Controls are assigned the ADRD-qualifying date of their matched cases to ensure a comparison of equivalent time-periods.

We allow cases to act as controls prior to ADRD onset to ensure comparable longevity, provided that the assigned qualifying date was at least 2 waves (4 years) before the actual qualifying date. Matching cases to controls based on the assigned group and requiring that all pairs were present in the survey during the ADRD-qualifying wave, up to five controls were randomly assigned to each case, resulting in 21,030 respondents (3,619 cases). Our cases and controls are in the sample for an average of 2.8 waves. There are 5,751 unique individuals in the analysis, acting as cases only (3,303), controls only (2,132), or both (316).

2.3. Measures

OOP spending are medical costs paid by the household. We use RAND-imputed values.¹⁹ The HRS asks respondents to estimate OOP medical costs paid over the previous two years (since the last interview). If the respondent does not provide an exact amount, the HRS uses an unfolding bracket method to improve response rates.

Respondents report OOP spending in several categories; we sum over all categories (hospital, prescription drug, doctor, nursing home, home health, special health facility, outpatient surgery, and dental visits) to create a total OOP spending variable. We estimate total spending and spending on nursing homes in the primary analysis, but also provide estimates for spending on hospitals, prescription drugs (where costs are shown as monthly averages), and doctor visits (Supplementary Table S1). All spending is adjusted for inflation using the Personal Consumption Expenditures price index for health care (CPI-U) and reported in 2017 dollars.

2.4. Statistical analysis

We calculate the marginal effect of ADRD on OOP spending using the multi-step Basu and Manning¹⁴ method, which estimates costs under censoring. (Detail can be found in the Supplementary text S1.) Costs are estimated with a two-part model, first estimating the probability of incurring any OOP cost in each wave using a logit model, then using a generalized linear model to estimate the magnitude of costs, conditional on a cost incurred. OOP costs are estimated separately for waves prior to death or censoring, and the wave of death, due to the concentration of spending at the end of life. To estimate survival functions, while accounting for censoring, we use an accelerated failure time model based on the lognormal distribution.

Once the costs and survival functions are estimated, we use recycled prediction methods to produce averages over all marginal effects and obtain standard errors via bootstrapping with 1,000 iterations. We use our cost and survival estimates to estimate OOP spending for the ADRD cohort. We then produce the counterfactual case of this same cohort not having ADRD, which eliminates any baseline differences in spending between cases and controls. The difference in these cost estimates is the incremental cost due to ADRD.

We use this method to estimate costs for the full sample as well as by gender and race/ethnicity. All models control for covariates measured at the time of the first ADRD- qualifying score: gender, age, race/ethnicity, income, and whether the respondent reported ever having high blood pressure, diabetes, cancer, lung disease, heart problems, a stroke, psychiatric problems, or arthritis. We also include time since diagnosis (in 2-year waves), an interaction term for ADRD status and time since diagnosis, and indicators for each wave since diagnosis. All analyses are unweighted.

3. RESULTS

Summary statistics comparing ADRD cases with first-matched controls in the diagnosis wave are in Supplementary Table S1. Because we match on these characteristics, there are no statistically significant difference in gender, age, race/ethnicity, and education. Average income is higher among the controls ($42,166 vs $34,468). Overall, our controls appear healthier at baseline; they are less likely to have ever had diabetes, cancer, lung disease, heart problems, a stroke, or psychiatric problems. There are differences in insurance status, even at baseline, where 72.6% of our cases are without LTCI coverage compared to 73.2% of our controls, with higher Medicaid enrollment (14.8% vs 10.9%). Total OOP spending was higher at baseline for cases than for controls ($5,155 vs $4,974, respectively). This pattern holds in all sub-categories of spending except prescription drugs, although not always statistically significant.

Figure 1 presents the unadjusted OOP spending over time for both the cases and controls. Prior to diagnosis, OOP spending is quite similar in these two cohorts, both in levels and in time trends. Upon diagnosis, the ADRD cohort’s OOP spending increases, while the control cohort’s OOP spending remains relatively stable. The difference in OOP spending remains high between these two cohorts until 10 years after diagnosis, when the spending on the ADRD cohort drops, but remains higher than the control cohort.

Figure 1:

Unadjusted Out-of-Pocket Spending around the time of ADRD diagnosis, for cases and controls.

Blue: Cases

Orange: Controls

X-Axis: Years Pre- and Post-ADRD diagnosis

Y-Axis: Annual Out-of-Pocket Expenditures

The figure represents the unadjusted spending of all individuals observed in each period.

Table 1 presents the estimated total and incremental costs for the ADRD cohort. Column 1 presents the spending estimates for those with ADRD, which start at $7,506 in the first two years after onset of symptoms and decrease to $3,630 by years 7 and 8. Respondents with ADRD are spending $22,795 OOP for all categories. Column 2 presents the counterfactual estimates – what respondents with ADRD would have paid OOP over the same period if they had not developed ADRD. Over the first eight years this sums to $14,044 OOP. Column 3 presents the incremental costs of ADRD, or the difference between columns 1 and 2; ADRD costs an additional $8,751 OOP per person in an incident dementia cohort over the first 8 years after symptom onset.

Table 1.

Period-specific absolute and incremental, Total Out-of-Pocket Spending

	Total Costs		Incremental costs
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Participants with dementia diagnosis	Predicted costs without dementia	Incremental costs	Incremental costs due to changed survival	Incremental costs if survival held constant	Incremental costs conditional on being alive	Difference in cumulative survival probability
Full sample
1 – 23 months	$7,506 (6,963; 8,137)	$4,472 (4,253; 4,697)	$3,034 (2,484; 3,594)	−$30 (−37; −24)	$3,064 (2,518; 3,621)	$3,104 (2,508; 3,663)	−0.29%
24 – 47 months	$6,832 (6,378; 7,364)	$4,135 (3,952; 4,326)	$2,697 (2,263; 3,159)	−$310 (−372; −253)	$3,007 (2,577; 3,452)	$3,730 (3,150; 4,294)	−4.22%
48 – 71 months	$4,827 (4,389; 5,301)	$3,041 (2,882; 3,208)	$1,787 (1,389; 2,233)	−$457 (−539; −378)	$2,244 (1,861; 2,668)	$3,762 (3,045; 4,436)	−9.30%
72 – 95 months	$3,630 (3,217; 4,059)	$2,397 (2,260; 2,527)	$1,233 (849; 1,663)	−$505 (−595; −415)	$1,738 (1,373; 2,123)	$3,934 (2,998; 4,982)	−11.64%
Total	$22,795 (21,236; 24,398)	$14,044 (13,544; 14,597)	$8,751 (7,354; 10,217)	−$1,303 (−1,531; −1,077)	$10,053 (8,634; 11,473)	$14,530 (12,255; 16,842)	−5.63%

Notes: Month 1 is the month in which the respondent received a dementia-qualifying score. 95 percent confidence intervals reported in parentheses.

Column 1 presents the estimated out-of-pocket costs for the dementia cohort as a whole. Column 2 presents the estimated out-of-pocket costs for the counter-factual case where the dementia cases did not have dementia, but all other characteristics were the same. Column 3 is the difference (column 1-column 2).

The incremental costs (column 3) are then decomposed into to differences in survival (column 4) and differences in utilization (column 5). To calculate column 4, the expected costs for each time interval are held constant between the dementia cohort and the counter-factual case, and are multiplied by the estimated cumulative survival probabilities for the two groups. To calculate column 5, the cumulative survival probability is held constant between the dementia cohort and the counter-factual case and is applied to expected costs for each time interval for the two groups. Column 6 presents the estimates of the incremental costs of ADRD conditional on being alive during the entire 2-year window. Column 7 presents the difference in the cumulative survival probability between having ADRD and not having ADRD.

Columns 4 and 5 break down the incremental costs into two components – differences due to changes in survival (column 4) vs. differences due to health care service utilization (column 5). Lower survival rates among individuals with ADRD lowers the predicted OOP spending by $1,303 over 8 years per individual. Utilization differences increase the OOP spending ($10,053), and shorter survival offsets the incremental OOP spending.

Column 3 presents the incremental OOP spending over the first 8 years for the incident cohort, which is relevant for population-level decisions and policies. For individuals who die during this window, their $0 OOP spending is included in the average. Column 6 presents results conditional on respondents being alive in each period, which may be more relevant at the individual-level, or for forecasting incremental costs of ADRD starting at a later point in the disease course. Column 7 presents the differences in cumulative mortality between our cases and controls. Our findings show that among survivors, average OOP spending is relatively stable over time, totaling to $14,530 over the first eight years from symptom onset. Supplementary Table S2 present the incremental out-of-pocket spending by category of spending.

Table 2 illustrates that among those with ADRD, women and non-Hispanic White respondents have the highest incremental OOP spending in total ($13,706 and $16,766, respectively). Nursing home spending is the primary driver of the increased OOP spending for all groups. Non-Hispanic White respondents with ADRD are paying less than they would have on other OOP spending categories (i.e., hospital and doctor visits) such that OOP spending on nursing homes exceeds total average incremental spending ($37,902 and $16,766, respectively). The Black/Hispanic cohort have the smallest OOP spending attributable to ADRD ($853).

Table 2.

Period-specific incremental costs by Group, Total and Nursing home

	(1) Total	(2) White, Non-Hispanic	(3) Non-White	(4) Men	(5) Women
Total OOP
1 – 23 months	$3,034 (2,484; 3,594)	$5,870 (4,934; 6,875)	$120 (−336; 586)	$2,355 (1,626; 3,131)	$4,323 (3,348; 5,312)
24 – 47 months	$2,697 (2,263; 3,159)	$5,055 (4,313; 5,811)	$265 (−106; 662)	$1,837 (1,291; 2,443)	$4,110 (3,413; 4,878)
48 – 71 months	$1,787 (1,389; 2,233)	$3,410 (2,755; 4,106)	$245 (−82; 585)	$1,030 (536; 1,577)	$2,990 (2,333; 3,675)
72 – 95 months	$1,233 (849; 1,663)	$2,430 (1,781; 3,190)	$223 (−170; 644)	$522 (4; 1,055)	$2,283 (1,643; 2,982)
Total	$8,751 (7,354; 10,217)	$16,766 (14,305; 19,380)	$853 (−441; 2,209)	$5,744 (3,815; 7,801)	$13,706 (11,393; 16,322)
Nursing homes
1 – 23 months	$2,324 (1,892; 2,806)	$5,235 (4,324; 6,216)	$45 (−57; 174)	$1,624 (1,049; 2,299)	$3,386 (2,623; 4,162)
24 – 47 months	$2,539 (2,125; 3,103)	$6,976 (4,334; 23,166)	$194 (63; 407)	$1,554 (936; 3,339)	$3,939 (3,216; 4,753)
48 – 71 months	$1,710 (1,349; 2,355)	$9,498 (2,869; 59,061)	$215 (98; 417)	$1,299 (608; 4,721)	$2,662 (2,089; 3,308)
72 – 95 months	$1,371 (929; 2,868)	$16,193 (1,794; 13,1390)	$262 (111; 481)	$1,752 (448; 9,020)	$1,780 (1,323; 2,266)
Total	$7,945 (6,546; 10,409)	$37,902 (14,064; 21,9697)	$716 (281; 1,387)	$6,229 (3,293; 18,669)	$11,767 (9,813; 13,894)
Nursing homes / Total OOP Spending	90.8	226.1	83.9	108.4	85.9

Notes: Month 1 is the month in which the respondent received a dementia-qualifying score. 95 percent confidence intervals reported in parentheses. Each column is a separate sub-sample.

4. DISCUSSION

We use survey data for 2000–2016 from the HRS to estimate the incremental OOP costs of ADRD over the first 8 years after symptom onset. We consider total OOP spending, as well as spending by category (i.e. hospital, nursing home, doctor, and prescription drug) in order to identify the drivers of OOP spending. We estimate that the total incremental cost of ADRD over the first 8 years is $8,751. We find that OOP spending attributable to dementia on nursing homes ($7,945) accounts for over 90 percent of total OOP spending. OOP costs are highest around the time of symptom onset, declining thereafter. This is consistent with the pattern for the costs of ADRD to the Medicare program.^7,8 This pattern, while partially a function of mortality, may be Reflective of spending down assets²⁰ to the point in which Medicaid coverage is available, and is consistent with the increasing time trend in incremental Medicaid spending on ADRD found in recent work.⁸

Our findings showing that women with ADRD are spending nearly $12,000 more OOP on nursing homes than women without ADRD are perhaps the most striking results of this study. Furthermore, incremental OOP spending by women is more than double that of men in this category. Because women tend to live longer and are more likely to live alone in older age,²¹ they are more likely to require formal/paid care, however they are also more likely to have fewer assets with which to pay for this care.²¹ These findings show that women are disproportionately affected by the gap in insurance coverage for LTC.

Finally, we find that incremental OOP spending due to ADRD is much lower among the Black/Hispanic subsample compared to the White, Non-Hispanic cohort. This could be due to a variety of factors. Access to health care differs among racial and ethnic groups, perhaps limiting OOP spending on health care.²² Diagnosis rates among different racial and ethnic groups differ,²³ and thus the timing between an ADRD-qualifying score and triggering a health care response could be different.²⁴ Differences in income and asset levels in retirement are stark by race and ethnicity, which could lead to lower OOP spending due to a lack of resources to spend as well as a higher likelihood of Medicaid coverage.²⁵ Finally, access to care, expense of care, or preferences lead to different LTC utilization patterns, with lower utilization of formal/paid long-term care among racial and ethnic minorities.¹⁵ This difference in utilization would also be consistent with our findings. Future work could identify the exact causes and their relative roles in this disparity.

4.1. Limitations and Strengths

Survey-based methods of identifying dementia have several drawbacks, including heavy reliance on language and memory, sensitivity to education, and a limited ability to gauge the level of cognitive impairment.²³ However, claims-based methods have been shown to underestimate dementia rates among minorities due to delayed diagnosis.²³ Because onset of symptoms may induce ADRD-related OOP spending prior to the formal receipt of a diagnosis, the survey-based approach is more appropriate for this context.

We rely on survey self-response to measure OOP spending, which may be subject to recall bias. However, the HRS’s OOP spending measures have been shown to provide quality data when compared with other sources of data on OOP spending, including the Medical Expenditures Panel Survey (MEPS) and the Medicare Current Beneficiary Survey (MCBS).²⁶

We match cases and controls on a few select attributes. We do not match on income since it is likely endogenous, so we instead rely on matching by education, which is more likely to be pre-determined. We use the methods of recycled predictions, which predicts OOP spending for the same cohort of individuals both with and without ADRD, in order to mitigate the effect of the differences between our cases and controls. Because our controls are higher income, healthier, and have lower Medicaid eligibility at baseline, our OOP spending estimates are likely conservative.

The gap in insurance coverage for LTC services likely contributes to the prevalence of informal care received.²⁷ However, this paper only accounts for OOP spending, not the substantial costs of informal care,²⁸ or any interaction between informal care receipt and formal care payments. Informal care could offset some of the OOP spending by decreasing use of nursing homes, for example.²⁹ Future work should consider this relationship, as well as the quality of care and quality of life among individuals with ADRD to put spending into context.

4.2. Conclusions

Our results add to a growing literature on the costs of ADRD faced by the health care system⁷ and families.^9–12,28 We highlight the burden of OOP costs for people living with ADRD resulting from a gap in insurance for LTC services, which is especially high for women. While the financial burden of OOP health care spending among our Black/Hispanic cohort is relatively lower, additional research should seek to determine the underlying causes.

Supplementary Material

supinfo

Supplementary Table S1. Summarizing Out-of-Pocket Costs and Respondent Characteristics at Baseline, by Dementia Status Notes: All variables are at the respondent level and unweighted. The baseline period is defined as the wave prior to the wave in which the respondent received a dementia-qualifying score. Insurance categories are not mutually exclusive.

Supplementary Table S2. Period-specific incremental costs by Category, Full Sample Notes: Month 1 is the month in which the respondent received a dementia-qualifying score. 95 percent confidence intervals reported in parentheses.

Supplementary Text S1.

Key Points.

• OOP costs attributable to dementia are $8,751 over the first 8 years after onset.

• Nursing homes comprise the largest component of OOP spending, and are a major driver of gender and racial differences.

Why does this matter?

The gap in insurance coverage leaves a large financial burden on those with dementia.