• Aucun résultat trouvé

The analyses are conducted on data from the Swiss Health Survey (SHS), which focuses on the 15+ population residing in Switzerland. The SHS is a repeated cross-section conducted every five years since 1992. Phone interviews are conducted with one random member of selected households. The respondent is also invited to answer a written questionnaire.

Because not everyone fills in this supplemental questionnaire, some variables have large proportions of missing values, e.g. specific medical conditions and supplemental health insurance.

This study uses the four waves for which home care data are available: 1997, 2002, 2007, and 2012. We keep individuals aged 20 and older and drop LOS beyond 180 days (eight observations). The main results are based on the German-speaking cantons (19 cantons out of 26), for which our instrument is relevant (Sections 1.5.3 and 1.6.2).

Those cantons include between 37,791 observations for specialist visits and 38,658 for hospitalizations.

Canton-level home care data come from the Home Care Survey, which covers all public and private non-for-profit home care providers. Information on patient contributions in each canton (i.e. the instrument) was provided by the Swiss home care association. Other canton-level variables used in sensitivity checks come from various surveys of the Federal Statistical Office.

1.5.1 Dependent variables

First, we look at the number of days spent in the hospital, which includes at least one night. We consider different LOS to approximate different types of hospitalizations: up to 5, 10, 15, 30, 60, 90, or 180 days (acute hospitalizations are LOS up to 30 days).3 The other dependent variables are the number of doctor visits of any kind, GP, and specialist visits. The recall period is the 12 months preceding the interview.

1.5.2 Home care use

The varying degrees of generosity in home care policy across cantons and over time are captured by aggregate home care use, as this is the result of various canton policies that are difficult to measure (e.g. different regulation of supply and diverse subsidies across cantons; Section 1.2). The most common measure of aggregate home care use is home care expenditures per capita (e.g. Kemper et al., 2008; Muramatsu and Campbell, 2002;

Muramatsu et al., 2007; Pezzin and Kasper, 2002; Stabile et al., 2006). As our focus is on CHI-covered services and we cannot distinguish between expenditures on CHI-covered and non-covered home care, we use hours. CHI-reimbursed home care hours per capita is defined as the number of CHI-reimbursed home care hours provided in a given canton and year divided by the canton’s population in that year. We take the natural logarithm of this ratio (ln) to account for non-linearities and reduce the influence of outliers.

1.5.3 Instrumental variable

We instrument home care use with a binary indicator that captures the introduction of patient contributions for CHI-reimbursed home care in fifteen German-speaking cantons in 2011: Appenzell Innerrhoden, Appenzell Ausserrhoden, Basel-Stadt, Graub¨unden, Luzern, Nidwalden, Obwalden, St. Gallen, Schaffhausen, Solothurn, Schwyz, Thurgau, Uri, Zug, and Z¨urich. Four cantons, representing almost 30% of the sample in 2012, did not introduce patient contributions: Aargau, Bern, Basel-Landschaft, and Glarus.

3The data do not allow us to identify multiple hospitalizations in all waves. LOS includes the total number of days spent in the hospital in the previous 12 months.

Both groups include cantons with different socioeconomic, demographic, and geographic situations and are likely to be comparable. Before 2011, there was no patient cost-sharing for home care in Switzerland. The contribution scheme and amount varies across cantons, but the maximum corresponds to 20% of the price of one hour of home care per day (15.95 CHF). Using variations in the monetary contribution itself did not provide strong enough instruments. A dichotomous measure captures the fact that overall, home care became more costly to the patients in cantons that introduced patient contributions, as before it was fully reimbursed by CHI.

The use of this instrument forms a difference-in-differences first-stage regression. We assume that the cantons that didn’t implement patient contributions form an accurate counterfactual; i.e. they represent what would have occurred in the other cantons, if they hadn’t introduced patient contributions. The German and Latin regions of Switzerland may differ in preferences and health policy dimensions beyond what is captured by the canton fixed-effects. Taking the seven Latin cantons into account would potentially undermine the accuracy of the counterfactual, because only Geneva introduced patient contributions —i.e. all but one Latin canton would be in the control group. Thus, our main analyses focus on the German-speaking cantons, for which the instrument works well. Those cantons represent two thirds of the Swiss population. Figure 1.5.1 shows the evolution of home care hours per capita in the German-speaking cantons that introduced patient contributions in 2011 and those that didn’t (i.e. where home care continued to be fully covered by CHI and public subsidies to providers). Until 2010, the two groups of cantons show increasing trends. After the introduction of patient contributions in 2011, the trends clearly diverge. In sum, for the German-speaking cantons, the instrument is expected to predict home care use —i.e. be relevant.

Besides being relevant, the instrument must be uncorrelated with hospitalizations and doctor visits, except through home care use. The introduction of patient contributions is unlikely to have a direct impact on hospitalizations and doctor visits, but it could have an indirect impact through unobserved health policy. With only one instrument, it is not possible to test the exclusion restriction statistically. However, we look at whether the

Figure 1.5.1. Home care hours per capita in cantons with and without patient contributions

Legend: BLACK: cantons without patient contributions; GRAY: cantons with patient contributions since 2011. Only German-speaking cantons included.

introduction of patient contributions modified the trends in some proxies of health policy that may be related to hospitalizations or doctor visits, such as hospital beds, doctors, and nursing home beds per capita. The absence of changes in the trends of those indicators points towards the validity of our instrument (Figures 1.A.1-1.A.3 in Appendix 1.A).

1.5.4 Individual-level explanatory variables

Individual-level covariates include informal care availability, health status and behaviors, socio-demographics, and pressure to remain at home. Informal care availability is captured by whether the person lives with another adult. This measure excludes potential help from outside the household. In Switzerland, as in most western countries, the main source of informal care is the spouse, who typically lives in the same household, followed by children. Unfortunately, the number of children is not available in the dataset. Health status is measured by self-assessed health, number of symptoms (e.g. back pain, diarrhea), whether the person has ADL limitations, and whether she is unable to walk for at least 200 meters. Health behaviors are being obese, a smoker or ex-smoker, and drinking more than two ‘standard drinks’ (10 grams of alcohol) per day for men and one drink for

women. Socio-demographics include age groups and gender interacted, Swiss nationality, residential area, education, and income. Pressure to remain at home is captured by having children interacted with gender, a full-time or part-time job (Weaver and Weaver, 2014).

Documents relatifs