Design of the study
The study is designed as a retrospective analysis of all hospitalizations in Switzerland for the years 2008–2010. In 2002, the OECD initiated the Health Care Quality Indicator Project to measure and compare the quality of health care provision across countries and to develop a set of health care quality indicators [
15,
19,
20]. Within this framework, the OECD uses AH as a measure of quality for prevention and management of chronic diseases in primary care. For comparability reasons, these indicators, including hospitalizations with a principal ICD10 code of asthma, chronic obstructive pulmonary disease (COPD), diabetes complications, congestive heart failure (CHF), and hypertension [
19], are used to define AH for this study. Based on OECD eligibility criteria, we included only patients aged 15 and up (15+) and patients not transferred from other hospitals. Detailed ICD codes and additional information on inclusion and exclusion criteria of these conditions as defined by the OECD are given in the Additional file
1.
Data
We used data from multiple sources. Inpatient care data, including patient demographics, regional data of patient residency, characteristics of hospitalizations (length of stay, type of discharge, referral pathways, health insurance status) and diagnostic and treatment data (ICD10, procedure codes and All Patient Diagnosis Related Groups (APDRG’s)) were extracted from the “Medizinische Statistik der Krankenhäuser”, housed at the Swiss Federal Statistical Office (SFSO) [
21]. These data cover all hospitalizations of acute care hospitals in Switzerland. One patient can be hospitalised multiple times (multiple cases). Anonymized unique patient identifiers allow tracking patients across hospitals in case of multiple hospitalizations. Data on structural attributes of acute care hospitals (localization, type, size and specialization of hospitals) are also available from the SFSO (Krankenhausstatistik).
Demographic data at the community level, including age and gender distribution, were available from the SFSO (census data) [
22]. In 2010, a new federal population census was introduced by the SFSO (SHAPE project) and the population aged 15+ was used to build the denominator to calculate regional rates of AH for 2010. SHAPE data were also used to directly standardize rates by sex and age groups.
Geographic unit
Utilization-based health service areas (HSA) of acute care hospitals were the unit of geographic analyses. HSA’s were constructed by analysing discharge data of all acute care hospitals in Switzerland for the period of 2008–2010 [
23,
24]. We used HSA’s and aggregated zip-code areas of patient residence as the smallest geographic unit (MedStat areas). HSA’s were constructed by cross-tabulating the sum of discharges of every zip-code cluster with all possible hospital regions, and then these regions were merged into an HSA by assigning to the hospital region in which the highest number of patients were treated [
25]. Using HSA’s has the advantage of describing where patients actually receive care, without regard to cantonal or other administrative borders [
21,
23,
24,
26,
27]. This approach is well established and has become an indispensable source of information for current US healthcare reformers [
28].
In 2008, the SFSO modified the concept of MedStat areas to make them compatible with other geographic classification systems. However, these changes were not equally implemented in the data collection procedures of all Swiss cantons during the course of the study. In consequence, geographic classification of hospitals and of patient residence was inconsistent in some cantons. Data from cantons Appenzell Innerrhoden, Appenzell Ausserrhoden, Schaffhausen, St. Gallen, Thurgau and Zurich were therefore discarded from the small area analysis.
Statistical procedures
All hospitalizations corresponding to the list of AH published by the OECD were included [
15]. Statistical analysis of the data was performed in two steps. The first included a descriptive nationwide analysis of AH; the second step identified determinants of regional differences of AH-rates in cantons with eligible data. Descriptive procedures documented overall rates, demographic characteristics, comorbid conditions of patients (Charlson index [
29]), length of stay, APDRG cost-weights, and inpatient mortality of all AH in Switzerland. APDRG cost weights that accounted for outliers of length of stay were calculated according to version 6.0 of the specifications of APDRG-Suisse [
30].
Regional rates of AH were calculated at the level of zip-code clusters; the number of AH admittances in the numerator and the total regional population aged 15+ were the denominator. Direct standardization of rates by sex and age was performed at the level of zip-code clusters and used the total 15+ Swiss population of 2010 as the reference.
For geographic analyses we summarized the data at two levels (utilization-based health service areas [n = 59] and aggregated zip-code clusters [n = 436]) and developed statistical models to explore the relationship between rates of AH and characteristics of regional supply of medical care. For each zip code cluster, we determined the density of primary care physicians and of specialists in own practice (physicians per 10000 population). Physician groups were defined based on definitions established by the Swiss Medical association [
31]. We also calculated the proportion of the population living in rural communities [
32]. At the level of HSA’s we calculated the number of acute care hospital beds per 10000 people as a measure of regional hospital supply.
We used a multilevel poisson regression model with the natural logarithm of the age and sex standardized number of AH as the outcome. We used the log number of the population of zip-code clusters as a fixed offset term in the regression equation. We added information on regional supply of ambulatory care and of population characteristics at the level of zip-code clusters, and added predictors at the level of HSA to estimate effects related to hospital supply. The final set of explanatory variables was obtained after a series of preliminary analyses that explored bivariate associations between the outcome and various measures of physician’s supply, including full time equivalents, and other methods to classify patient residency geographically. We eventually included explanatory variables which explained the largest variability of AH admissions.
These are the explanatory variables included in the final model:
Level 1 (436 zip code areas)
-
Year (2008, 2009, 2010)
-
Number of primary care physicians per 10000 population
-
Number of specialists per 10000 population (all medical specialists with office based practice)
-
Proportion of the population living in rural areas
-
Type of hospital reimbursement system (APDRG vs. other systems)
Level 2 (59 utilization based health service areas)
We added random intercepts at the level of HSA’s and zip-code clusters to allow for unexplained variation around the respective means. We used the same model to analyse effects associated with the overall rate, and for condition specific rates of AH’s. In order to explore linear relationships between continuous explanatory variables and AH, we additionally defined a second model and replaced the continuous data with quintiles of the respective variables. Differences to the first quintile were documented as incidence risk ratios. SAS 9.3 (proc GLIMMIX) was used for multilevel modeling and ArcGis 10 to create maps, the level of significance was set to p < 0.05 throughout the study.