|Home | About | Journals | Submit | Contact Us | Français|
Our objective in this study was to compare assistance received by individuals in the United States and Sweden with characteristics associated with low, moderate, or high 1-year placement risk in the United States.
We used longitudinal nationally representative data from 4,579 participants aged 75 years and older in the 1992 and 1993 waves of the Medicare Current Beneficiary Survey (MCBS) and cross-sectional data from 1,379 individuals aged 75 years and older in the Swedish Aging at Home (AH) national survey for comparative purposes. We developed a logistic regression equation using U.S. data to identify individuals with 3 levels (low, moderate, or high) of predicted 1-year institutional placement risk. Groups with the same characteristics were identified in the Swedish sample and compared on formal and informal assistance received.
Formal service utilization was higher in Swedish sample, whereas informal service use is lower overall. Individuals with characteristics associated with high placement risk received more formal and less informal assistance in Sweden relative to the United States.
Differences suggest formal services supplement informal support in the United States and that formal and informal services are complementary in Sweden.
The effectiveness with which support is provided to community-dwelling older adults is important in preventing or delaying costly institutional placement. Research points to the role of both formal and informal care systems in this process. One of the major cross-national differences is how these resources are allocated (e.g., Esping-Anderson, 1999). Policies in the United States place primary responsibility on the contributions of family, with formal services often playing a supplemental role. In some countries, notably in Sweden and the other Scandinavian countries, formal services are more likely to have a primary role in providing care. Despite increasing strain on services in Sweden, Swedish policies have retained an explicit focus on tying formal services tightly with need, thereby improving allocation of resources. Recent work suggests there is more effective targeting of formal services in Sweden than in the United States for individuals with the least access to the informal care system (e.g., Johansson, Sundström, & Hassing, 2003; Shea et al., 2003; Sundström, 2003). Because institutional placement represents the most extreme level of the need for assistance, it provides a useful comparison of support networks in the United States and Sweden, and we address this issue in the current article.
Previous research has identified a number of predictors of institutionalization using logistic regression and event history models (e.g., Greene & Ondrich, 1990; Jette, Branch, Sleeper, Feldman, & Sullivan, 1992; Wolinsky, Callahan, Fitzgerald, & Johnson, 1992). Across the diversity of designs, samples, analytic techniques, and nations surveyed, several consistent predictors emerge, with the following characteristics being among the most important and replicable. The oldest old (usually defined as those aged 85 years and older), gender, individuals with greater impairment on instrumental and physical activities of daily living (IADLs and PADLs, especially the latter), and those without available family or friends to provide informal assistance (e.g., Carriere & Pelletier, 1995; Greene & Ondrich, 1990; Jette et al., 1992; Kleibsch, Sturmer, Siebert, & Brenner, 1998; Wolinsky et al.) are at greatest placement risk. Early work by Shapiro and Tate (1988) highlighted the importance of considering institutional risk as the interaction of patterns of individual characteristics even when predictors do not themselves interact, owing to the nonlinear characteristics of modeling these probabilistic transitions. This approach has been echoed by a number of researchers since then (Greene, Lovely, & Ondrich, 1993; Jette et al., 1992) and elaborated upon to compare higher and lower institutional risk groups (e.g., Stuck et al., 2000). However, much less research has examined the potential role of the formal support system in terms of helping to maintain older adults in the community, either by reducing or delaying institutionalization in conjunction with informal supports.
Contributions within this area suggest that the patterns are complex. For example, in Greene and colleagues’ (1993) analysis of Channeling data, they found that utilization of community-based formal services reduced nursing home use, but only when services were appropriately tied to needs (e.g., personal care aide or housekeeping help for individuals with limitations in instrumental activities of daily living). A variety of other studies have found that use of formal services is often positively associated with institutional placement (Jette, Tennstedt, & Crawford, 1995; Kliebsch et al., 1998; Romøren, 2003), however, some of these authors have not always paid as much attention as Greene and colleagues (1993) to the fact that need, rather than utilization, is the causal agent. A more macrolevel analysis by Miller and colleagues (1998) suggests that states spending more money on home- and community-based services delay nursing home placement.
Given the importance of both functional limitations and formal and informal supports provided to individuals in predicting institutional placement, we address this issue next. Despite considerable interest in the interface between formal and informal care systems (Anderson, 1995; Patsios & Davey, in press), there remains a “dearth of theoretical analysis of the relationship between informal and formal care in a welfare state of the Swedish kind” (Sundström, Johansson, & Hassing, 2002, p. 350). Initial interest in the relationship between informal and formal support grew out of social gerontological inquiry. Cantor’s (1975) hierarchical compensatory model suggested elderly individuals would prefer to be cared for first by their spouse, then children, other family members, friends, and lastly formal caregivers (Cantor, 1975, 1980). Each group successively provides assistance when a preferred source of care is either not available or unable to meet the needs of the care recipient. This model assumes the substitutability of one service for another but within a preferred ordering. Although the literature is consistent about older adults’ preference for caregivers, there is little evidence to support the compensatory nature of the informal care network (Denton, 1997; Penning, 1990).
In contrast, a supplementation model predicts that it is more common for functionally dependent or disabled elders to receive both informal care and formal services (Soldo, Agree & Wolf, 1989; Tennstedt et al., 1990), particularly if their care needs are extensive (Kemper, 1992; Tennstedt, Sullivan, McKinlay, & D’Agostino, 1990). It is clear that informal caregivers play a vital role in maintaining functionally dependent elders in the community (Cantor, 1980; Wenger, 1990), however, research showed also that the formal system supplements care provided by informal caregivers (Davey & Patsios, 1999), particularly when the needs of the older person exceed the resources and capacity of the informal network (Edelman & Hughes, 1990; Stoller & Pugliesi, 1988). Supplementation assumes that kin caregivers are the major helpers and use service providers to augment their efforts or to provide temporary relief (i.e., respite care; Edelman & Hughes; Noelker & Bass, 1989; Stoller, 1989; Stoller & Pugliesi).
The complementarity model took aspects of both the compensatory and supplementary models described by George (1989), Chappell (1985) and Chappell & Blandford (1991) and put forward that formal care is mobilized when crucial elements of the informal network are lacking or when there is substantial need. In short, formal services provide for those tasks which informal caregivers are unable to provide. Instead of informal caregivers providing care in isolation of formal services, there is an overall sharing of care tasks, that is, they complement one another (George, 1987). This was supported by past research showing that both support networks provide the necessary care when elements of the informal system cannot do so alone (Bone, 1995; Chappell & Blandford; Denton, 1997). To the extent that the policy context explicitly targets individuals for formal services based upon greatest need or risk (Bergmark, Parker, & Thorslund, 2000; Thorslund, Bergmark, & Parker, 1997), we should expect these differences to be more pronounced in a country such as Sweden, which has a more extensive and accessible formal system than does the United States. As a result, we would also expect that rates of unmet or undermet need (having one or more limitations for which no support is provided or for which support is provided but not sufficient) would be lower for those with the highest risk and that this finding would be truer in Sweden than in the United States.
Our study involved a cross-national comparison of individuals in terms of the receipt of formal and informal help based on characteristics associated with high, moderate, and low predicted risk of 1-year institutional placement in the United States. Because previous research suggested that informal services should be most useful with low to moderate care needs, (e.g., Chappell & Blandford, 1991; Davey et al., 1999; Denton, 1997; Patsios & Davey, in press), we expected that rates of informal support would be higher in the United States than in Sweden, although these differences should be smallest at moderate levels of institutional risk. If universally available long-term care services were targeted to those with the greatest need in Sweden, as recent policy analyses suggest (e.g., Sundström et al., 2002; Thorslund et al., 1997), then receipt of formal services, in particular formal services alone, should be higher in Sweden and should be greatest among those with the greatest need, and thus the corresponding greatest risk of placement. To address our research questions, we proceeded in two steps. First, we developed a logistic regression model predicting placement in the United States. From there, we identified individuals with characteristics corresponding to individuals with low, average, or high placement risk in each country and compared the mix of formal and informal support that they received with results verified using log-linear models.
We drew upon two data sets for the current analyses. In the United States, we used the 1992 and 1993 waves of the Medicare Current Beneficiary Survey (MCBS), a nationally representative longitudinal panel survey of Medicare beneficiaries conducted by the Center for Medicare and Medicaid Services (CMS). MCBS includes information on use of community-based long-term care services for a representative sample of older people in the United States and collects extensive information on individuals’ use and expenditures for health services, sources of payment, type of health insurance, access to care, health and functional status, and socioeconomic and demographic characteristics (see http://www.hcfa.gov/surveys/mcbs/Overview.htm).
Our MCBS analytic sample consisted of 4,579 community-dwelling individuals aged 75 and older who were interviewed for the 1992 MCBS merged with corresponding 1993 data indicating whether the individual was currently living in a facility or had filed a skilled nursing facility claim in the intervening year. Normalized sampling weights were applied to make the sample representative of the corresponding U.S. population.
Swedish data were taken from the 1994 survey, Hemma På Äldre D’ar—Aging at Home (AH; Socialstyrelsen, 1994), which examined health, functional status, utilization of formal, community-based services, assistance from informal providers, and sociodemographic characteristics. AH is a nationally representative sample of 1,378 people aged 75 and older living in the community, based on a sampling frame of the complete population. The sample was stratified to yield approximately equal numbers of men and women in three age groups: 75–79, 80–84, and 85 and older, oversampling men and the oldest old. Normalized sampling weights were again applied to the data to ensure that the data were representative of the Swedish population aged 75 and above.
Both surveys included comparably worded items for the critical domains of demographics, perceived health, limitations in ADLs and IADLs, and formal and informal service use, which were used to construct equivalent measures in each data set. Descriptive statistics for each country are shown in Table 1.
We obtained information on age, gender, and living arrangements. For age, it was possible to classify respondents into categories aged 75 through 79, 80 through 84, and 85 and older with sufficient numbers of respondents at these ages available in each survey. We examined several indicators for comparing socioeconomic status between the two samples. In the end, the variable that seemed most comparable between the two countries was formal education. It was possible to categorize people in the samples as having: (a) some grade school education (grundskola in Sweden), (b) high school education (gymnasium and equivalents in Sweden), and (c) college level education. Although education does not tell us how much economic resources are available to a person, it is correlated with income. Education also affects service use. Although education and health are positively correlated, individuals with more education are more likely to seek out services (Anderson, 1995; Dunlop, Manheim, Song, & Chang, 2002; Powell-Griner, Bolen, & Bland, 1999).
Current functioning is measured by need for assistance with IADLs and PADLs. IADL and PADL disabilities indicate the need for assistance and are important to consider when examining patterns of service use. IADLs common to both surveys were housework, meal preparation and shopping. PADLs common to both surveys were receipt of assistance with bathing, dressing, transferring from a bed or chair, walking, and toileting. Preliminary analyses indicated that the presence or absence of any need for assistance across these domains was a better predictor of placement than the actual number of limitations present, so we retained this coding in subsequent analyses.
MCBS and AH both ask a short series of questions (2–4 questions) about each PADL and IADL. The translation of the wording for the individual activity questions that were provided varies slightly in the Swedish survey, but each question generally begins by asking respondents whether they were able to manage the activity or needed help with it. A follow-up question for the IADLs alone asks whether they were unable to manage the activity without assistance. For the Swedish sample, a limitation was defined as needing help or being unable to manage the activity without assistance, regardless of whether assistance was actually provided.
The MCBS asked respondents whether they had any difficulty performing the activity or did not do it. If they did not do it, they were asked whether that was because of a health reason. Individuals in the MCBS were considered to have a limitation if they had any difficulty or were not able to perform the activity because of a health reason.
Prior to conducting the analysis, we compared different ways of constructing comparable measures from the MCBS to the AH. Measures used in the analysis, described above, were those that appeared to be most comparable in need for assistance, given the language differences (Shea et al., 2003). Because so few MCBS respondents reported not doing PADL tasks for nonhealth reasons, potential for bias introduced by the difference in survey format is likely to be quite small, both conceptually and empirically. We recorded individuals as having an ADL limitation regardless of whether they used an assistive device because this information was only available in the MCBS.
Following Habib and colleagues (1993), an index of potential availability of help was constructed as follows: (a) elder lives with a spouse and has living children, (b) elder lives with spouse and has no living children, (c) elder lives with a child, (d) elder lives alone but has children, and (e) elder lives alone and has no children. This categorization reflects normative expectations in both countries for providing help. There were a small number of persons who were living with both a spouse and adult children. These persons are included in Group A. Small groups of elders in both countries had other living arrangements (e.g., living with a boarder who provides some help), but the groups were too small and heterogeneous to be included in the analysis. We also could not take into account the residential propinquity of living children outside the home.
Each survey determined whether a respondent received help with a PADL or IADL and who provided that help. Helpers were classified as informal (from family, friends, or volunteers) and/or formal (provided by paid helpers or agencies). We classified people into 4 groups, regardless of their functional limitations: (a) receiving neither formal nor informal help, (b) receiving informal help only, (c) receiving formal help only, and (d) receiving both informal and formal help. It is important to note that because of the way ADL questions were asked in each country, it was possible to disentangle functional limitations from whether assistance was actually provided, regardless of limitations, as well as the source or sources of assistance.
Our primary outcome measure in this study was experience of admission to a nursing home or skilled nursing facility during the year after initial interview. This dichotomous variable was coded as 0 if an individual did not report a Medicaid/Medicare claim in this period and 1 if they did report a claim. This variable was only available for individuals in the United States, and 197 (4.2%) respondents had at least one institutional stay in the year following the interview.
To address our research question, we first conducted a logistic regression analysis predicting 1-year probability of institutional placement in the United States. From the model, we then constructed the logistic regression equation using the predictor variables age group, gender, IADL limitations, PADL limitations, and living arrangements. We also included educational attainment and urinary incontinence in initial models. However, because they were not associated with placement risk in the United States, and completely comparable variables were not available in the Swedish data set, we did not retain them in the final model. In any case, much of the influence of these variables is likely to be mediated through health variables already included in our model. Urinary incontinence was associated with increased placement risk at the zero-order level but was not significant in the multivariate model ( p = .34). Table 2 presents the results of this analysis. Relative to individuals aged 75 to 79, those 85 or older were more than twice as likely (2.22) to experience institutional placement in the following year. Controlling for other variables, most especially potential availability of kin support, women were roughly half as likely (0.57) as men with comparable characteristics to experience institutional placement. Having IADL limitations was associated with twice the likelihood of placement in the subsequent year (2.03), as was having PADL limitations (1.90). Relative to being married with living children, living alone either with or without living children was associated with a 2.26 and 2.31 greater risk of institutional placement, respectively. Our initial model also included a dichotomous index of cognitive impairment (“Has anyone ever told you that you have Alzheimer’s disease?”). Not surprisingly, this question was answered in the affirmative by 2.9% of the U.S. sample. This variable did predict greater probability of placement. However, it was not retained in the final model because cognitive impairment was measured using a performance-based test in Sweden and indicated much higher levels of cognitive impairment, as would be expected. Before reaching the decision to exclude this index of cognitive impairment from our model, we first confirmed that it did not alter the results presented in the next section.
In the next step, based upon this model, we used individuals’ characteristics on the predictor variables to estimate a predicted probability of 1-year placement for all individuals in the sample. The probability of placement was equal to (ecutoff)/(1 + ecutoff), where cutoff = −4.25 + 0.22 × (age 80–84) + 0.79 × (age 85+)−0.55 (female) + 0.71 × (any IADLs) + 0.64 × (any PADLs) + 0.28 × (married, no children) + 0.19 × (lives with children) + 0.82 × (lives alone, has children) + 0.84 × (lives alone, has no children). In the United States, 4.2% of individuals (197) in the sample experienced a stay in a skilled nursing facility in the subsequent 1-year period. We present example profiles for individuals with a predicted “low” risk (predicted probability less than half of the average risk of placement, or below 2.1%), “moderate” risk (intermediate to the low and high placement risk groups), or “high” risk (predicted probability at least twice the average risk of 8.4% or above). The number of individuals in each group in the United States and Sweden was quite comparable and is shown in Table 1. Below, we present some prototypical characteristics of individuals in each risk group:
Of course, many other combinations are possible within each risk profile, and these are merely presented as illustrations of the characteristics of individuals at each level of risk. Having divided the sample in each country into three groups, based upon their predicted risk, we next compared each group on the basis of the type of support they received.
When comparing the combination of support received from formal and informal sources, clear differences emerged between the United States and Sweden as a function of predicted risk of institutional placement (Figure 1). We consider each type separately below.
As can be seen in Figure 1, the results reveal substantial differences in care mix between the two nations as a function of placement risk. These differences were borne out in a log-linear analysis. A log-linear analysis is an extension of a contingency table to examine associations among more than two dimensions, in this case, country (two levels), predicted placement risk (three levels), and care mix (four levels), where no variable is an “outcome” in the traditional sense. Results from the loglinear model are presented in Table 3 and show that the three main effects, three two-way interactions, and the one three-way interaction were all significant at p < .0001. This analysis suggests that the association between placement risk and care mix differs significantly between the United States and Sweden. The rest of this section is devoted to fully characterizing the nature of these differences.
In the United States, the proportion of individuals receiving assistance from informal sources alone is higher than in Sweden at all levels of risk for institutional placement (U.S.: 45.0%, 57.6%, and 69.4% for low, moderate, and high placement risk, respectively; Sweden: 5.0%, 34.3%, and 23.3% for low, moderate, and high placement risk, respectively) and increases slightly with greater risk for institutionalization. Contrastingly, in Sweden, the largest number of individuals receive informal assistance at the moderate level of risk, with the proportion of individuals receiving informal support alone being lower in the high-risk group, and quite negligible in the lowest risk group. This is most compatible with the complementarity perspective in Sweden.
In the United States, a very small proportion of individuals receive formal supports alone, regardless of their predicted risk of placement. In Sweden, a very different picture emerges, with an increasing proportion of individuals receiving formal assistance alone in the group with characteristics associated with the greatest risk of placement in the United States. In fact, in the highest risk group, this is the modal form of assistance in Sweden (U.S.: 1.3%, 1.4%, and 0.3% for low, moderate, and high placement risk, respectively; Sweden: 2.2%, 14.8%, and 42.1% for low, moderate, and high placement risk, respectively).
Across both countries, the proportion of individuals receiving both formal and informal assistance is greater for groups with the greatest predicted risk of institutional placement. In the United States, this proportion is approximately equal to those receiving informal supports alone, and in Sweden it is not much less common than receipt of formal supports alone for this group (U.S.: 1.9%, 11.2%, and 21.8% for low, moderate, and high placement risk, respectively; Sweden: 1.0%, 9.9%, and 33.5% for low, moderate, and high placement risk, respectively).
The very different interface between formal and informal care systems results in observed differences between the two nations in the numbers of individuals receiving neither formal nor informal assistance. At low levels of risk for placement, typically groups with few limitations of IADLs or PADLs, a large proportion of individuals receive no assistance from any source (U.S.: 51.8%, 29.9%, and 8.5% for low, moderate, and high placement risk, respectively; Sweden: 91.8%, 41.0%, and 1.1% for low, moderate, and high placement risk, respectively). This is appropriate in the low-risk group because most of these individuals did not have functional limitations (92.4% in Sweden and 87.2% in the United States). However, in the highest risk group, almost all individuals in Sweden received one or more forms of assistance, whereas nearly 1 in 10 individuals in the United States, almost all of whom had limitations on IADLs (100%) or PADLs (92%), received no assistance from either formal or informal sources (e.g., Shea et al., 2003). Clearly, this speaks for the highly effective targeting of assistance to those with the greatest need in Sweden relative to the United States (Sundström et al., 2002).
We investigated formal and informal supports to individuals in the United States and Sweden with characteristics that placed them at a low, moderate, or high level of risk for nursing home placement using a model derived from longitudinal data in the United States. Our expectations were that rates of informal assistance would be higher in the United States than in Sweden, but that these differences would be smallest at moderate levels of risk for nursing home placement. As well, we predicted that more individuals would receive formal services alone in Sweden than in the United States and that greater emphasis on needs-based targeting in Sweden would result in higher rates of assistance to those whose characteristics placed them at greatest risk of nursing home placement. Overall, our expectations were largely borne out in our data.
Our risk-based analysis also shed light on theoretical approaches to the study of the interface between formal and informal care. We found greatest support for a supplemental model in the United States and for complementarity in Sweden. This was most clearly evidenced by the very low levels of formal supports in the absence of informal support, as well as by the high levels of unmet or undermet need (e.g., Davey & Patsios, 1999; Shea et al., 2003). In contrast, the evidence from Sweden was most consistent with complementary and substitution perspectives, as evidenced by high levels of formal assistance regardless of the presence of informal support. There is also clear evidence for more effective targeting of supports on the basis of need in Sweden compared with the United States; whereas almost all individuals in Sweden whose characteristics are associated with placement in the United States are receiving some form of assistance, nearly 1 in 10 comparable individuals in the United States receive no assistance from any source. This finding is balanced by an unexpected finding of high levels of informal assistance in the United States regardless of placement risk. In fact, rates of informal assistance alone are barely associated with risk in the United States, whereas they are highest (34%) in the moderaterisk group in Sweden and relatively lower in the low (5%) and high (23%) placement risk groups. Further, of those in the United States reporting neither IADL nor PADL limitations, 38.4% received at least some informal assistance, and 7.6% received some formal assistance. A corollary is that among those in the United States with either IADL or PADL limitations, fully 21% report receiving neither formal nor informal support. Sensitivity analyses indicated that these cross-national differences were not explained by the excluded ADL domains that differed between the countries, although we acknowledge that we cannot rule out the possibility that differences in wording of the IADL and PADL items (i.e., needing help to perform a task in Sweden versus being unable to perform a task without assistance in the United States) may contribute in part to this difference. We note, however, functional disability is only one aspect of how risk groups were defined, with advanced age, gender, and availability of kin support also playing a potential role.
While this study has shed light on a new aspect of cross-national comparison of community-based support to older adults, there are, of course, several limitations. The most obvious among these is clearly the lack of comparable outcome data in Sweden; ideally, we could have developed a prediction model including individuals in both countries simultaneously. However, given the comparability of findings across the diversity of methods, models, and nations where placement risk has been studied, in practice the effects of this shortcoming are likely to be fairly small. In this study, we have tried to address this issue by focusing our attention on individuals with similar characteristics in each country, as opposed to reaching beyond the limitations of our data. Another limitation is the possible exclusion of additional predictors, which would have improved our prediction model. Based on previous research, the most important omissions likely included prior institutionalization history (Jette et al., 1992; Wolinsky et al., 1992), regional characteristics (e.g., urban or rural location, and availability of local services; c.f., Greene & Ondrich, 1990; Miller et al., 1998), and (when applicable) caregiver characteristics (Fisher & Lieberman, 1999; Freedman, 1996; Tsuji, Whalen, & Finucane, 1995). Given the relative independence of predictor variables suggested by previous analyses (e.g., Greene et al., 1993), this is unlikely to have been a substantial source of systematic bias in our analyses. Additionally, we were able to verify that the exclusion of cognitive impairment and urinary incontinence did not significantly alter our results, the first because it did not change the other regression coefficients in our model, and the second because it did not make an independent contribution in predicting placement.
Future research should attempt to track these changes longitudinally at both an individual and systemic level (i.e., as resources and policies evolve over time, as well as individual needs change). Work on more efficient allocation of existing resources, as well as greater needs- and means-testing would aid in this process, as there is, at least in theory, the potential for significant reductions of nursing home placements through more effective allocations of existing resources (Greene, Lovely, Miller, & Ondrich, 1995). With all Western industrialized nations feeling the strain of aging populations on social services, these issues become increasingly important to address.
This research was supported by a grant from the National Institute on Aging (R03AG023301) and the AARP Andrus Foundation.