• Aucun résultat trouvé

3.4 Empirical study

3.4.1 Background

Formal home care, defined as care provided at home by paid professionals, has been expanding in Switzerland. In part, this expansion is a response to increasing demand for long-term care (LTC) and to people’s general preference to receive care in their homes rather than in nursing homes or other institutional settings. Besides, some policy makers believe that promoting home care services may help limit the growth in LTC costs by enabling people to remain in their homes. To date, there is no causal evidence to support that belief. Here, we investigate the causal effect of increasing home care policy generosity on canton-level rates of nursing home use in Switzerland.

The relationship between home care and nursing home care has received little attention in Europe. One study conducts a decomposition analysis of trends in LTC use in the Netherlands (Meijer et al., 2015). Its findings suggest that reform of the LTC system has helped more mildly disabled individuals to be treated at home, such that rates of nursing home use have decreased over time. Two randomized controlled trials among persons at high risk of nursing home placement find a positive effect of home preventive visits on nursing home use in Switzerland and no significant effect in Denmark (Stuck et al., 2000;

Vass et al., 2005).

Most evidence on the effect of home care policy on nursing home use comes from the US. The Channelling Demonstration expanded home care to frail elderly in ten states in the early 1980s. It reduced nursing home use to a small extent; a significant effect of expanded home care is found only for specific groups, such as unmarried individuals, and others find no significant effect (Greene et al., 1993a, 1993b, 1998;

Kemper, 1988; Pezzin et al., 1996). Similarly, other demonstrations either reduced or didn’t significantly impact nursing home use. The Cash and Counseling Demonstration gives persons a consumer-directed allowance to obtain Medicaid personal care or home-and community-based waiver services. Expenditures on those services increased but were partially offset by lower nursing home expenditures in Arkansas and New Jersey; no significant effects are found in Florida (Dale et al., 2003; Dale and Brown, 2007). The Managed Community Care Demonstration changed reimbursement of an Illinois home

care provider from fee-for-service to capitation. It increased home care use and had no significant impact on nursing home admissions, compared to fee-for-service patients in the same area (Hughes et al., 2003).

Non-experimental studies also find a negative or non-significant effect of home care on nursing home use. Two studies find lower Medicaid nursing home expenditures among participants in six home care programs in Arkansas and Florida. They address endogeneity using matched control and treatment groups (Felix et al., 2011; Shapiro et al., 2011).

Another study that tackles endogeneity by using an IV finds that use of Medicaid home care services significantly reduces nursing home use and expenditures (Guo et al., 2015b).

A Canadian study finds a negative effect of home care on the probability of living in a nursing home, using an IV and random-effects to account for endogeneity (Sarma and Simpson, 2007). A non-significant association between individuals’ expected Medicaid home health subsidy and the probability of residing in a nursing home is found in Hoerger et al. (1996). The change in reimbursement of Medicare home health (HH) providers in 1997 lowered average and marginal payments to HH providers and led to a decrease in HH services utilization. Using a difference-in-differences model, McKnight (2006) finds no significant impact of lower HH use induced by the reimbursement change on the probability of nursing home use or expenditures.

The literature reviewed so far measures home care and nursing home use at the individual level. Using variations across states and over time, some studies assess the relationship between state-level home care policy generosity and individual-level nursing home use. Home care policy generosity is proxied by home care expenditures per capita or the proportion of LTC expenditures allocated to home care. More generous state home care policy is associated with significantly lower nursing home use among unmarried older women (Burr et al., 2005) and childless seniors (Muramatsu et al., 2007); no significant effect is found in Pezzin and Kasper (2002). In analyses conducted entirely at the aggregate level, state-level nursing home use rates are measured by the number of nursing home residents divided by the 65+ population. More generous state home care policy is associated with significantly lower state-level nursing home use rates and per

capita expenditures (Buys et al., 2012; Kaye et al., 2009; Miller, 2011). In contrast, the number of home- and community-based services waiver participants is associated with higher state-level Medicaid nursing home expenditures (Amaral, 2010). These studies either don’t tackle the endogeneity of state home care policy generosity or account for it partially, using state fixed-effects but ignoring the possibility of simultaneity.

Overall, empirical evidence in the US is vast and seems to point towards some substitutability between home care and nursing home care. However, its generalizability to European contexts may be limited. For example, most studies focus on Medicaid and thus consider only the low-income population. In various European countries, such as Switzerland or the Netherlands, compulsory health insurance (CHI) pays for some home care services to sick or disabled persons regardless of income. Broadly, European countries differ from the US in terms of their health systems and cultural backgrounds. Another limitation of most existing literature is that home care policy is measured with error.

State home care policy generosity is a concept that is not directly observable. Using one or two observed indicators to capture it provides an incomplete view of generosity. Our analysis has two main contributions. First, we account for the fact that home care policy is unobservable by modeling it as a latent variable, using factor analysis. Second, we add to the limited causal evidence on the effect of home care policy generosity on nursing home utilization in Europe, by using Swiss data.

In Switzerland, CHI partly covers LTC by paying for medical care provided in nursing homes or at home for all residents. In this federal country, 26 cantons and more than 2,300 municipalities (the smallest administrative units) are responsible for LTC supply and financing. Besides CHI, LTC is financed by a mix of out-of-pocket, public subsidies to patients and providers, and other minor payers (accident and military insurances). CHI covers help with the activities of daily living (ADL), such as bathing and eating, as well as medical treatments, such as intravenous administration of medicine and monitoring of glucose levels. These services require a doctor referral to be reimbursed. Other services, including help with the instrumental activities of daily living (IADL) at home (e.g.

shopping, housekeeping) and food and board costs in nursing homes are paid out-of-pocket

or through complementary insurances. Public subsidies to patients may also be used to pay for non-covered services and their prices may depend on the patients’ income and wealth. Finally, cantons and municipalities directly subsidize home care providers and nursing homes. The relative importance of each payer differs for home care and nursing home care. For home care, the two largest payers are regional and local governments, through subsidies to providers, and CHI. For nursing home care, they are households (out-of-pocket) and governments through subsidies to patients (Weaver, 2011). Because of decentralization, there is large heterogeneity in the tools to finance LTC and amounts of subsidization of home care and nursing home care across cantons. This heterogeneity means that measuring empirically home care and nursing home policy is difficult.

In Switzerland in 2011, the rates of nursing home and home care use among the 65+

were 6.2% and 14.1%, the third and second highest rates among OECD countries (OECD, 2014). There are wide variations in LTC use across cantons as well as over time within cantons. For example in 2012, 4.8% of the 65+ population in Basel-Landschaft resided in nursing homes. The largest proportion of 65+ living in nursing homes was found in Appenzell Ausserrhoden (11.0%). Between 1997 and 2012, that proportion decreased in most cantons, by as much as 2.9 percentage points in Nidwalden. In Basel-Stadt, it increased by 1.8 percentage points. Globally, nursing home use rates tend to be lower in the French-speaking cantons —i.e. Western Switzerland (Figure 3.4.3). Turning to home care, in 2012 the proportion of the whole adult population using CHI-reimbursed home care ranged from 1.2% in Nidwalden to 3.3% in Jura. From 1997 to 2012, that proportion increased in most cantons, by as much as 2.0 percentage points in Ticino. In Obwalden, it decreased by 1.2 percentage points. Figure 3.4.4 shows home care policy generosity across cantons and over time, which includes the proportion of the population using CHI-reimbursed home care and other indicators of home care use (Section 3.4.2). Overall, home care policy tends to be more generous in the French-speaking cantons. Different rates of LTC use across cantons may be explained not only by cantons’ LTC policies but also by cantons’ different populations, including demographic structure, socioeconomic characteristics, health/disability status, culture, and preferences. For example, in the

German-speaking cantons, nursing homes have traditionally been perceived as ‘places to live’, whereas in the Latin cantons they are more commonly viewed as ‘places of care’.

In fact, most nursing homes in the Latin cantons are highly medicalized (Jaccard-Ruedin et al., 2006). Over the last few years, the situation has been slowly evolving toward medicalization in the German-speaking part of Switzerland as well.

Figure 3.4.3. Canton nursing home use rates in 1997 and 2012

(a) 1997

(b) 2012

Nursing home residents divided by the 65+ population. Darker gray represents higher nursing home use rates; the 1997-2012 distributions are cut into quartiles.

Whether home care substitutes for nursing home care is an empirical question. Home care may delay or eliminate the need for more intensive LTC in a nursing home, by providing patients with appropriate IADL, ADL help, and medical care in their homes.

Figure 3.4.4. Canton home care policy generosity in 1997 and 2012

(a) 1997

(b) 2012

Latent variable reflected by various indicators of canton-level home care use, constructed using factor analysis. Darker gray represents higher home care policy generosity; the 1997-2012 distributions are cut into quartiles.

On the other hand, in some cases home care providers may detect important needs that require higher levels of care available in nursing homes. Here, we assess the effect of canton home care policy generosity on the observed rates of nursing home use in each canton. Home care policy is a latent variable reflected by indicators of canton-level home care use, constructed using factor analysis. Implicitly, we assume that in cantons with greater aggregate home care use, a given person is more likely to use home care. Our analysis informs policy makers about the effect that may be expected from expanding or reducing home care policy generosity on aggregate nursing home use.

3.4.2 Data

We use data from the Swiss Home Care Survey, which covers all public and private non-for-profit home care providers.5 Those data are merged with other canton-level variables available from various surveys of the Federal Statistical Office. The dataset includes the 26 cantons over the period 1997-2012. Excluding incomplete observations gives a final sample size of 378 canton-year observations.

The dependent variable, y, is the rate of nursing home use, defined as the number of nursing home residents in a given canton at the end of the year divided by the canton’s 65+ population in that year. This ratio is multiplied by 100 to represent percentages.

The denominator includes only the 65+ population, as the vast majority of nursing home residents belongs to that age group.

The variable of main interest is the latent variable ‘home care policy’, denoted by F in equation (3.3). This is the only latent explanatory variable in the regression. There are three indicators in W used to measure home care policy. They are CHI-reimbursed home care users per capita, home care users per 65+ persons, and home care users per LTC user (i.e. home care users plus nursing home residents) receiving home care.6 This latent variable is in fact the Participation dimension of canton home care policy from Appendix 2.A. It makes sense to look at this home care policy dimension because the dependent variable, nursing home use rate, is also a measure of participation in the nursing home care market. Although it might also be interesting to look at the Intensity dimension of canton home care policy, a relevant instrument for this dimension could not be found.

Observed explanatory variables, X, include measures of the cantons’ socioeconomic, demographic, and geographic characteristics, health policy controls, region and year fixed-effects. Cantons’ characteristics include GDP index, measured by the ratio of each canton’s GDP with respect to the Swiss GDP, proportion of 65+ aged 80 and older, proportion of 65+ who live alone, rate of home ownership, population density,

5Private for-profit providers have only been surveyed since 2010. They represent a minority of the home care market; about 15%.

6Home care users includes some double counting, as a user of both CHI-reimbursed home care services and other non-reimbursed home care (e.g. IADL help) in a given year is counted twice.

and proportion of German-speaking population. Health policy controls are nursing home bed occupancy rate, doctors per capita, and hospital beds per capita. There are seven regions in Switzerland made of one to seven neighboring cantons. Region fixed-effects are included as an alternative to canton fixed-effects to reduce multicollinearity.

Canton home care policy is likely to be endogenous (i.e. E[F µ]6= 0). First, there is risk of simultaneity, as cantons may coordinate their home care and nursing home policies.

Second, canton home care policy and nursing home use rates are both influenced by health policy (e.g. supply of acute care and other LTC services, subsidies to different providers), needs, and preferences of the population. Nursing home use may itself be seen by policy makers as an indicator of aggregate need for LTC. The regression includes health policy controls and variables that partly capture needs and preferences. In addition, region fixed-effects control for time-invariant differences in health policy, needs, and preferences across regions. Year fixed-effects account for changes that impact all cantons’ home care policies and rates of nursing home use similarly over time. Yet, there may be some degree of canton-level unobserved heterogeneity left. To address canton home care policy endogeneity, we use an instrument.

The instrument, Z, is the proportion of seats in the cantons’ legislative assemblies occupied by women. The political science literature suggests that women with political power may give priority to public policies related to women’s traditional role as caregivers (Besley and Case, 2000). If this is true, politically engaged women may support the expansion of home care policy generosity to alleviate the caregiving burden of their gender.

Thus, in cantons where women have more political power, home care policy is likely to be more generous. The validity of that instrument is undermined if it influences the rate of nursing home use in other ways besides through home care policy. For example, politically engaged women may also promote the expansion of other health services, such as nursing home, doctor, or hospital care. The risk that the instrument impacts nursing home care deserves special attention, because many women face LTC decisions at some point; e.g.

institutionalize an elderly parent or provide informal care. To investigate this, we look at the annual growth rate in the number of nursing home beds, which is rather stable around

zero in most cantons over the period of our study (Figure 3.E.1, Appendix 3.E). In fact, in some cantons for some years, expanding the number of nursing home beds was prohibited by law (i.e. moratorium). Overall, politically involved women likely have limited power to influence the expansion of nursing home care. The numbers of hospital beds and doctors per capita are stable over time within most cantons and thus also unlikely to have been impacted by the proportion of women in the cantons’ legislative assemblies. This provides reassurance regarding the validity of our instrument. Besides, the regression includes controls for nursing home, doctor, and hospital care supply, so the instrument is likely to be exogenous conditional on the set of control variables. With one instrument (i.e. just-identified case), it is not possible to test statistically the validity of the exclusion restriction. Other instruments were considered: e.g. proportion of women in the cantons’

governments, proportion of left-wing politicians in the cantons’ legislative assemblies and governments, being a canton that borders another country, introduction of patient cost-sharing for CHI-reimbursed home care in some cantons in 2011. Unfortunately, they were not relevant in explaining canton home care policy generosity in this study. To further investigate the conditional exogeneity of the instrument, we check the sensitivity of our results to the exclusion of health policy controls and region fixed-effects from the regression. We also exclude from the sample the canton Appenzell Innerrhoden, which has large changes in the number of nursing home beds over time. Lastly, all standard errors are clustered at the canton level.

3.4.3 Results

The summary statistics of all variables can be found in Table 3.4.1. Variations in nursing home use rates are discussed in Section 3.4.1. Here, we focus on the home care indicators, which vary considerably across cantons over the sixteen years. The proportion of the population that uses CHI-reimbursed home care services varies from about 0.2% in Ticino in 1998 to 3.6% in Glarus in 2010. The proportion of 65+ persons that use any home care varies from about 1.8% in Ticino in 1998 to 32.5% in Vaud in 2010. Between 22.9%

of LTC users in Ticino in 1998 received home care, compared to 87.3% in Jura in 2002.

There is also considerable variation in the instrument: the proportion of women in the cantons’ legislative assemblies varies between 10% and 37%.

Table 3.4.1. Summary statistics (N = 378)

Mean Std. Dev. Min. Max.

Dependent variable

Rate of nursing home use (%) 7.33 1.63 4.72 12.90

(BL, ’11) (GL, ’99) Home care policy

CHI-reimbursed home care users per capita (%) 1.94 0.57 0.17 3.62 (TI, ’98) (GL, ’10)

Home care users per 65+ persons (%) 15.98 4.65 1.82 32.53

(TI, ’98) (VD, ’10)

Home care users per LTC user (%) 73.30 7.14 22.85 87.28

(TI, ’98) (JU, ’02)

Home care policy (factor scoresa) 0.03 0.85 -2.71 3.06

(TI, ’98) (VD, ’10) Covariates

GDP index 96.90 29.73 64.05 221.54

(UR, ’08) (BS, ’12)

Prop. of 65+ aged 80 and older (%) 27.59 2.28 21.74 34.98

(BL, ’99) (BS, ’12)

Prop. of 65+ who live alone (%) 30.48 5.32 17.98 48.82

(FR, ’07) (BS, ’97)

Prop. of German-speaking pop. (%) 69.12 33.41 3.93 96.06

(GE, ’00) (NW, ’12)

Pop. density (pop./km2) 444.89 981.39 26.07 5,217.65

(GR, ’01) (BS, ’97)

Rate of home ownership (%) 40.70 10.44 12.12 61.40

(BS, ’97) (VS, ’00)

Nursing home occupancy rate (%) 96.02 2.59 86.46 100.00

(AI, ’97) (NE, ’12)

Doctors per 1,000 pop. 1.74 0.62 0.89 4.02

(NW, ’97) (BS, ’11)

Hospital beds per 1,000 pop. 5.26 2.29 1.70 14.60

(SZ, ’12) (BS, ’98) Instrument

Prop. of women in legislative assemblies (%) 24.14 6.75 10.00 37.00 (GL, ’00) (BS, ’11)

aFactor scores are standardized to have mean zero. Pop. is short for population; prop. is short for proportion. Canton and year in parentheses. Canton abbreviations: AI = Appenzell Innerrhoden; BL = Basel-Landschaft; BS = Basel-Stadt;

FR = Fribourg; GE = Geneva; GL = Glarus; GR = Graub¨unden; JU = Jura; NE = Neuchˆatel; NW = Nidwalden; SZ = Schwyz; TI = Ticino; UR = Uri; VD = Vaud; VS = Valais.

The measurement model of canton home care policy generosity is displayed in Table 3.4.2. It is a CFA model derived based on theory, exploratory and confirmatory factor analysis (Chapter 2). All three indicators have significant and large factor loadings.

The square of the factor loadings gives the R2, the proportion of the variation in each indicator accounted for by the latent variable, canton home care policy. Thus, home care policy accounts for 74% of the variation in CHI-reimbursed home care users per capita

(0.8612), 99% of the variation in home care users per 65+ persons (0.9972), and 58% of the variation in home care users per LTC user (0.7612). We estimate the factor scores

(0.8612), 99% of the variation in home care users per 65+ persons (0.9972), and 58% of the variation in home care users per LTC user (0.7612). We estimate the factor scores

Documents relatifs