| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Family Studies, University of Kentucky, Lexington.
2 Department of Consumer Sciences, The Ohio State University, Columbus.
Address correspondence to Hyungsoo Kim, Department of Family Studies, School of Human Environmental Sciences, University of Kentucky, 316 FB Lexington, KY 40506. E-mail: hkim3{at}uky.edu.
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Using data from the Asset and Health Dynamics Among the Oldest Old study (1995 to 2002 waves), we conducted ordinary least squares regression analysis on wealth changes.
Results. We found that comorbidity leads to significant wealth depletion in later life, especially for single elders. Single elders with comorbidity depleted 20% to 22% of their wealth over a 2- to 3-year time period, especially those with the combination of heart disease and diabetes. The impact of comorbidity was disproportionately greater than the estimated impact of a single health problem. However, the impact of comorbidity did not appear to be significant among married people.
Discussion. We found that compounded health problems also create compounded financial problems in later life. For an accurate estimation of the financial consequences of health problems, it is important to consider comorbid health problems, as the effect of comorbidity is not equal to the sum of the effects of single health problems.
CHRONIC health conditions are leading causes of death for elders in the United States. Among others, cancer, heart disease, stroke, lung disease, and diabetes account for about 70% of all death among those aged 65 or older (National Vital Statistics System, 2004
). The prevalence of multiple conditions or comorbidity increases with age, resulting in increased needs for health care services and inflated costs. The proportion of elders with comorbidity is 35.3% among those between the ages of 65 and 79; this proportion increases to 70.2% among those aged 80 and older (Fried, Ferrucci, Darer, Williamson, & Anderson, 2004). In 1999, among Medicare beneficiaries, per capita annual expenditures were $1,154 for those with one condition, $2,394 for those with two conditions, $4,701 for those with three, and $13,973 for those with four or more (Wolff, Starfield, & Anderson, 2002). At an aggregate level, 96% of annual Medicare spending in 1996 was attributed to beneficiaries with comorbidity (Fried et al., 2004).
Elders spend a significant proportion of their income for out-of-pocket health care expenditures. Medicare beneficiaries spend 19% of their income for out-of-pocket health care expenditures, including health insurance premiums (Crystal, Johnson, Harman, Sambamoorthi, & Kumar, 2000; Gross et al., 1999). This proportion decreases with income and increases with age: Individuals in the lowest income group or those who live below poverty level spend 32% to 35% of their income for out-of-pocket health care, and, near the end of life, poor elders with annual incomes less than $12,000 spend up to 70% of their income on it (McGarry & Schoeni, 2005). However, Goldman and Smith (2001) have questioned Gross and colleagues' estimates, arguing that the percentages were overestimated due to underreporting of, and missing data on, income. Goldman and Smith's estimates of the proportion of income spent for out-of-pocket health care expenditures is 8% to 10% on average and 11% to 16% for poor elders.
Despite the significant financial consequences of health problems in later life, there have been only a few empirical investigations on the impact of health problems on individuals' wealth depletion (Adams, Hurd, McFadden, Merrill, & Ribeiro, 2003; Lee & Kim, 2003; Smith, 1999, 2003; Wu, 2003). Furthermore, no previous study has investigated the impact of comorbidity on wealth in later life, an issue that we address in the current study. Using data derived from four waves of the Asset and Health Dynamics Among the Oldest Old study (AHEAD), collected from 1995 to 2002, we examined the impact of the five major causes of death (cancer, heart disease, stroke, lung disease, and diabetes) on wealth. We looked particularly at how comorbidity of these chronic conditions influences wealth among older people in an effort to more accurately estimate the effect of compounded health problems.
Literature Review
The life cycle theory of consumption provides a theoretical framework for explaining wealth depletion in later life. The fundamental idea is that older people accumulate wealth before retirement and deplete that wealth after retirement in order to finance consumption over the remainder of their lives (Modigliani & Brumberg, 1954). The simple model of life cycle theory of consumption predicts that wealth will decline to zero coincident with the time of death, assuming that the date of death is known (Yaari, 1965).
This simple model, however, does not consider one important influencing factor of wealth depletion: unexpected shocks to the household. Haider, Hurd, Reardon, and Williamson (2000) proposed that actual changes in household wealth may differ from the intended changes due to unexpected expenditures stemming from unexpected shocks, such as acute health events or accidents. Other examples of unexpected shocks include widowhood and unexpected gains or losses in investments. Several studies (Haider et al., 2000; Hurd & Reardon, 2003; Smith, 1999) have provided empirical evidence that older people's financial portfolios significantly influenced wealth changes during the 1990s.
The existing literature on wealth changes in later life presents mixed findings. Diamond and Hausman (1984) and Hurd (1987, 2001) discovered that elders deplete wealth at an average annual rate of 1.2%–5.0%, and that the depletion rate was higher among those who were unmarried (3.9%–4.5%) than those who were married (0.8%–1.6%). On the other hand, Hurst, Luoh, and Stafford (1998) and Haider and colleagues (2000) reported wealth accumulation in later life, contrary to the life cycle theory.
So far, only a few studies have investigated the issue of unexpected health problems as an explanation for wealth depletion. Using data from the first two waves of AHEAD, Smith (1999) demonstrated that any new health problem results in about $10,000 of wealth depletion over the first 2 years, which comprises far more than just out-of-pocket medical expenses. Using the same data set, Haider and colleagues (2000) found that unanticipated health expenses from health decline led to a 10% depletion over 2 years. Smith (1999) explained that wealth depletion associated with ill health occurs not only through out-of-pocket medical expenses, but also because of other expenses, like transportation costs and reconfiguration of the home. Wu (2003) identified greater effects of ill health on wealth when the wife is ill, versus when a health problem strikes the husband. Such a difference arises from the fact that general living expenses generally increase when a wife becomes seriously ill. In contrast, Adams and colleagues (2003), using the first three waves of AHEAD, did not identify any causal relationship between new health problems and changes in wealth. Using data from the first and fifth waves of the Health and Retirement Study, Smith (2003) also did not find that new health events significantly induced onsets on wealth depletion.
Measurement Issues
These mixed results may stem from (a) the different time spans involved in the studies, (b) the different measures of wealth changes used, and (c) the different measures of health conditions used. Most of these studies used the data from two waves (Haider et al., 2000, Lee & Kim, 2003; Smith, 1999, 2003) or three waves (Adams et al., 2003) of the AHEAD. A recent study, however, reports that wealth was underreported in the first AHEAD wave and that researchers found a substantial, yet unreliable, increase of wealth during follow-up interviews, thereby concluding that opportunities to fix this error in a reliable manner were few (Rohwedder Haider, & Hurd, 2004). Given that the baseline of all previous studies was the first AHEAD data set (1993), the results may be biased. Therefore, for the purposes of this study, we dropped the 1993 data set and reexamined wealth changes in later life.
Second, different measures of change in wealth also may contribute to the mixed results observed in the literature. Whereas Smith (1999, 2003) and Adams and colleagues (2003) examined the dollar amount of wealth change, Haider and associates (2000) and Hurd and Kapteyn (2003) examined the percentage change in wealth, and Lee and Kim (2003) assessed whether wealth had been depleted. Given that the measures of wealth and the techniques used to gather data still are evolving, measurement errors in wealth change may be inevitable (Dynan, Skinner, & Zeldes, 2004; Juster, Smith, & Stafford, 1999). As a measure of wealth change, we employed percentage change in wealth, as the same dollar amount of wealth change represents a different relative magnitude depending upon wealth level.
Third, the literature has introduced different measures of health, such as self-rated health status (Hurd & Kapteyn, 2003), functional impairments (Haider et al., 2000), and acute or chronic health problems (Adams et al., 2003; Lee & Kim, 2003; Smith, 1999, 2003). Smith (1999, 2003) defined cancer, heart conditions, stroke, and lung disease as severe health problems and other health problems as mild based upon their threat to life and the magnitude of associated health care expenses. Along the same vein, Lee and Kim (2003) expanded this measure of severe versus mild health problems to include high blood pressure, diabetes, arthritis, and psychiatric problems. In investigating the incidence of health problems, Adams and colleagues (2003) examined incontinence, falls, and hip fractures in addition to cancer, heart conditions, and stroke. However, none of the previous studies examined the issue of comorbidity, which is the focus of this study.
Another methodological issue of importance when studying the relationship between health and wealth pertains to the endogenous characteristics of some independent variables. Most importantly, Heiss, Hurd, and Borsch-Supan (2003) argued for the endogeneity of living arrangement: Some elders move in with their children or to a nursing home when they become very frail or very poor. That is, wealth changes may significantly influence their choice of living arrangement. Consequently, in this study, we test the endogeneity of changes in living arrangement in our analysis of wealth change.
In summary, we propose to extend the existing literature in three ways. First, we examine the impact of comorbidity on wealth changes in later life. Second, we attempt to minimize measurement errors by excluding the potentially biased first data set of AHEAD, but including all other AHEAD waves (1995, 1998, 2000, and 2002). Third, we test any confounding effects of change in living arrangement while examining the impact of comorbidity on wealth change.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
). AHEAD reinterviewed these same individuals in 1995, 1998, 2000, and 2002. The AHEAD data provide in-depth information about the economic status of households, including assets and income, as well as comprehensive information about the health status of individuals, including the prevalence and incidence of major chronic conditions. Moreover, the longitudinal nature of the data set allows for estimation of the impact of health status on wealth changes. We constructed the data set for the present study from those who had participated in the AHEAD survey between 1995 and 2002. To be eligible for our analysis, a participant must have (a) participated in two or more consecutive waves, so that we could establish changes in wealth and health status; and (b) had wealth of at least $1,000 and not more than $5 million in 1995, so that we could minimize measurement errors by eliminating outliers. This led to 6,296 individuals in 1995. However, 1,214 of them (19.3%) had not been reinterviewed for one of a variety of reasons, which included death, relocation, and other loss to follow-up. Thus, the final data set consisted of 5,082 participants (representing 80.7% of the age-eligible respondents in 1995) and 11,175 respondent observations. Table 1 presents the baseline descriptive statistics for our sample of 5,082 and for those 1,214 who were not reinterviewed. Compared to the nonparticipants, our sample was more likely to be female, younger, better educated, more affluent, and healthier, and to have more health insurance coverage.
|
).
Comorbidity
In this study, comorbidity was a categorical variable that indicated the existence of health problems at baseline, identifying not only whether a respondent had a health problem but also whether comorbid conditions existed. Consequently, we categorized individuals as (a) having none of the five designated health problems (cancer, heart disease, stroke, lung disease, or diabetes); (b) having one of the designated health problems; (c) having two of the designated health problems; or (d) having three or more of the designated health problems. Subsequently, in order to identify impact of health relative to different combinations of comorbidity (e.g., heart conditions and cancer vs heart conditions and diabetes), we disaggregated those with two chronic conditions into one of five groups: heart conditions + cancer, heart + stroke, heart + lung, heart + diabetes, and some other combination. We aggregated the other combinations (e.g., cancer + diabetes) due to small cell size.
Other Variables
Existing literature suggests that the impact of preexisting health conditions and new health problems may differ, so we differentiated the effect of existing comorbidity from the onset of a new health problem. We created a binary variable (new health events) to indicate the onset of a new health problem at some time during the interval between two consecutive waves.
There is tremendous variability in the use of health services by older adults with specific chronic conditions or comorbidity. The use of these health services may influence wealth changes directly. Thus, we included data on seven different types of health services used between waves. We used categorical variables to indicate the number of nights of hospital stay (zero as a reference group, up to 10 nights, and more than 10 nights), days of nursing home stay (zero, up to 100 days, and more than 100 days), and doctor visits (zero, up to 10 times, and more than 10 times). We used binary variables for outpatient surgery, dentist visits, prescription drugs, and home care (1 = participant used each of these services, and 0 = otherwise).
Researchers have found that health insurance coverage significantly influences health care utilization and out-of-pocket expenditures that consequently affect wealth. Most elders have Medicare, but its coverage is limited. Thus, we captured additional health insurance coverage by including a categorical variable indicating whether a respondent had any additional health insurance (such as Medicaid, an employer-sponsored plan, or other supplementary plans; e.g., Medigap) versus having no additional health insurance. We considered those with a Medicare health maintenance organization to have additional health insurance coverage, given its generous coverage.
Along with the primary variables suggested by the life cycle theory, such as income and net worth, we also included some other variables that may affect wealth changes: living arrangement, wealth transfer, and wealth portfolio (Lee & Kim, 2003). As discussed earlier, living arrangements can be tied to wealth. Examined from two consecutive waves, change in living arrangement may be associated with a change in wealth. We included a categorical variable to identify changes in living arrangement: moving from community living (alone or together with children) to a nursing home; living alone to living together with children; or no change. Other changes in living arrangement affected only 3.2% of single elders and 2.3% of married elders.
For wealth transfer, we used a binary variable indicating whether elders had given financial gifts to family members between two consecutive waves. Given that widowhood leads to large declines in wealth, we also included a binary variable for widowhood between two consecutive waves (1 = a respondent became widowed, and 0 = otherwise).
The household portfolio influences change in wealth via different rates of return across different types of assets. We created a continuous variable to indicate the share of specific assets to total wealth: cash equivalent share (i.e., including cash and money in depository products) as a reference group; housing wealth share; stocks, bonds, and individual retirement accounts share; and other assets share.
As shown in Table 1, the remaining sample was healthier and more affluent than those who exited from the AHEAD interview. Thus, we included a categorical variable for attrition, indicating those who left in 1998, those who left in 2000, and those who remained through 2002.
Although wealth data from AHEAD were of very high quality, measurement errors still exist. In order to assess the impact of possible measurement errors caused by imputation, we added a categorical variable: wealth values imputed at both waves (e.g., in 1995 and 1998), at one wave (e.g., either 1995 or 1998), or at neither wave. In addition, we included binary variables to indicate if a proxy respondent or a financial respondent answered the questions related to income and wealth to see if this distinction mattered.
Analysis
We used a two-step procedure to test the endogeneity of the variable change in living arrangement (Green, 1993; Xu, 2002). In the first stage, we used a multinomial logit model to identify factors that predicted the probabilities of the three types of change in living arrangements. Using the predicted probabilities, we created three correction terms for changes in living arrangement, similar to an inverse Mill's ratio in a Heckman model. In order to estimate this model, we employed the number of female children as an instrumental variable. This variable likely was closely related to change in living arrangements of elderly parents, but not to change in the wealth of the parents, as female children are more likely to take care of frail elderly parents than male children are (Heiss et al., 2003). However, financial support of parents tends to be more limited from female children relative to male children (McGarry & Schoeni, 1995). In the second stage, we used ordinary least squares regression, including the three correction terms, along with other independent variables. We tested the endogeneity of the living arrangement variable by checking whether the estimates of the correction terms in ordinary least squares were zero, indicating that it was not an endogenous variable (Vella & Verbeek, 1999). Because the existing literature demonstrates a significant difference in wealth change across marital status, we conducted separate estimates for those who were unmarried versus those who were married.
The AHEAD data greatly improved wealth data quality by using brackets to cope with respondents' unwillingness to provide answers to questions pertaining to dollar amounts; exact amounts are then inferred from the brackets. However, this procedure can lead to significant errors. Also, creating a dependent variable from two consecutive waves potentially creates the problem of regression-to-mean errors, a common phenomenon in which individuals who mistakenly report lower wealth in one wave tend to report higher wealth in the other wave when using panel data. In addition, outliers overly influence estimation results. To mitigate this concern, we eliminated the top and bottom 1% (221 observations) of the data from multivariate analysis (the highest and lowest changes in household wealth). Despite the elimination of the most extreme 2% of observations, regression diagnostics identified some observations with a Cook's distance greater than the suggested 4/n cutoff—one of the most popular criteria used for dealing with outliers (Cook, 1977). Thus, we further excluded 260 influential observational outliers based upon Cook's cutoff criterion.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
|
|
|
Only two variables among the use of health services were significantly associated with wealth changes: nursing home stay of up to 100 days and the use of dental services. The elders who stayed in a nursing home for a short duration accumulated more wealth than those who did not use nursing home services. This positive association between a short-term nursing home stay and wealth accumulation is counterintuitive, but it is important to note that a long-term nursing home stay was not associated with wealth accumulation, and that moving into a nursing home was significantly associated with wealth depletion. One explanation for the positive effect of a short-term nursing home stay is that, because Medicare covers a nursing home stay of less than 100 days, and when one stays in a nursing home for a short period, he or she become aware of long-term care needs, and therefore reduce consumption and increase savings for these needs. This is only a speculation, and we need to undertake further investigation in order to explain this relationship.
The elders who received dental care increased their wealth by 6% compared to those who did not see dentists. It is interesting to note that the impact of new-onset health problems disappeared when we controlled for the use of health services. In addition, having only one health condition failed to significantly influence wealth changes.
In estimating the impact of comorbidity on married respondents, we included additional explanatory variables of spouse health status and marital status change from married to single. Unlike for single elders, we found that none of the health problems of married respondents and their spouses significantly influenced wealth. Only the use of dental care was significant among the use of health services variables. Although we found that having diabetes lowered wealth by 8.5% (data not shown) not controlling for the use of health services, when we controlled for the use of health services, this impact disappeared. Marital status change from married to single applied to individuals who had been newly widowed between waves; this widowhood led to about 9% wealth depletion after taking other factors into account.
Regarding the impact of various control variables, findings were similar for single and married respondents, with the exception of gender and health insurance. Single women depleted wealth by 11% compared to single men. Single elders with supplementary health insurance increased wealth by 11% compared to those without supplementary health insurance; however, among married respondents, the impact of supplementary insurance was not significant. Nonetheless, among both single and married respondents, Medicaid recipients lowered their wealth by 22% versus individuals with Medicare only. For both single and married respondents, age was not significantly associated with wealth change, despite prior evidence that wealth depletion rates increase with age. Blacks had a higher rate of wealth depletion, and more educated and higher income elders tended to accumulate wealth.
The coefficient of the wealth variable had a negative sign, indicating that elders with more wealth depleted wealth faster, which is consistent with findings from previous studies (Haider et al., 2000; Smith, 2003). Haider and colleagues (2000) explained this as the result of regression to the mean caused by measurement error. Financial transfer to children was positively associated with wealth depletion. Considering that more wealthy elders tend to give gifts to their children, this transfer may not have been enough to result in wealth depletion. Living arrangement and portfolio composition also had significant associations with wealth change. Attrition during 1998–2002 was not significantly associated with wealth change, indicating that there was no significant difference in wealth change between those people who remained alive and those who had died. Imputation of wealth values was significantly associated with positive wealth change, indicating that imputation of this data resulted in positive measurement errors. However, the use of financial and proxy respondents did not exhibit a significant effect upon results, indicating that the choice of who answered the questions did not significantly contribute to measurement error.
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
However, we did not find any significant impact of health problems, including comorbidity and new-onset health events, on wealth changes among married couples, even after controlling for health service use and spouse health status. This finding is consistent with Haider and colleagues' (2000) findings that the new development of functional limitations is not associated with wealth changes for married people, and Lee and Kim's (2003) findings that chronic health problems are not associated with married elders' wealth depletion. One explanation for this finding is that elders who are married may be less sensitive to the adverse effects of health problems than those who are single because, on the one hand, they have more resources and, on the other hand, they tend to be healthier than those who are single. In addition, our results showed that widowhood contributes to about 9% of wealth depletion. Given recent findings reported by McGarry and Schoeni (2005) that the majority of post-widowhood poverty can be explained on the basis of health care expenditures for the dying spouse, researchers should not overlook the importance of health problems among married people.
These findings have implications for health policy and retirement planning. From a policy perspective, the findings of this study highlight the importance of interventions to reduce the health-related financial burdens of single elders with comorbidity. Comorbidity imposes a serious burden on elders, both physically and financially. In an environment of ever-increasing health care costs, this finding is not surprising. What we learned from this study is that the financial consequences of comorbidity cannot simply be estimated by adding the individual contributions of a number of chronic conditions; the effects are more than additive. Albeit this study certainly does not present a complete picture. Nonetheless, we learned that the effects of comorbidity warrant further careful study, especially focusing on differences between single and married people and the impact of diabetes. For retirement planners who seek to help individuals better prepare for later life, our findings call for more careful estimation of health care costs among elders. In an age of ever increasing life expectancy, comorbidity will become even more prevalent in the future.
This study contributes to the existing literature by investigating the effect of comorbidity on wealth changes across 7 years of longitudinal data, which include four distinct data collection points gleaned from a large national database. However, this analysis does not provide any insight into the long-term effect of comorbidity. The field needs further studies in order to examine the cumulative effect of comorbidity over time, as its impact may increase over time, which consequently may deepen its financial impact upon elders. Although we minimized measurement errors caused by sample selection and imputation of missing data, we were not able to consider the impact of respondents who participated in 1993 but died between 1993 and 1995 (because they were no longer available in 1995). However, the bias caused by this attrition seems not to be serious; as shown in our regression analyses, the attrition that occurred between 1995 and 2002 had an insignificant effect upon our results. Finally, we did not deconstruct the pathways of how comorbidity leads to wealth depletion. Previous studies have addressed the fact that chronic conditions lead to more out-of-pocket medical expenses, including medications, home improvement cost, and general expenses. Further research is needed to examine the mechanisms by which comorbidity causes wealth depletion.
|
Footnotes |
|---|
Received for publication September 28, 2005. Accepted for publication March 15, 2006.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
