Regional Variation of Avoidable Hospitalisations: An Observational Study


 Background Primary health care is subject to regional variation, which may be due to unequal and inefficient distribution of services. One key measure of such variation are avoidable hospitalisations, i.e., hospitalisations for conditions that could have been dealt with in situ by sufficient primary health care provision. Particularly, avoidable hospitalisations for ambulatory care-sensitive conditions (ACSCs) are a substantial and growing burden for health care systems that require targeting in health care policy. Aims Using data from the Swiss Federal Statistical Office (SFSO) from 2017, we applied small area analysis to visualize regional variation to comprehensively map avoidable hospitalisations for five ACSCs from Swiss nursing homes, home care organisations and the general population. Methods This retrospective observational study used data on all Swiss hospitalisations in 2017 to assess regional variations of avoidable hospitalisations for angina pectoris, congestive heart failure, chronic obstructive pulmonary disease , diabetes complications and hypertension. We used small areas (MedStat), utilisation-based hospital service areas (HSAs), and administrative districts (Cantons) as geographic zones. The outcomes of interest were age and sex standardised rates of avoidable hospitalisations for ACSCs in adults (>15years). Our inferential analyses used linear mixed models with Gaussian distribution. Results We identified 46,479 hospitalisations for ACSC, or 4.3% of all hospitalisations. Most of these occurred in the elderly population for congestive heart failure and COPD. The median rate of avoidable hospitalisation for ACSC was 1,080 (IQR 893 – 1,274) per 100.000 inhabitants. We found substantial regional variation for HSAs and administrative districts as well as disease-specific regional patterns. Conclusions Differences in continuity of care might be key drivers for regional variation of avoidable hospitalisations for ACSCs. These results provide a new perspective on the functioning of primary care structures in Switzerland and call for novel approaches in effective primary care delivery.

Switzerland offers a unique opportunity to explore regional variations in avoidable hospitalisations. Because of its status as a confederation, Swiss health law includes relatively high levels of regional autonomy, allowing regions to establish their own care structures and approaches (14). Mandatory health insurance with varying deductibles and primary care-mostly provided by general practitioners with freedom of choice for patients-provides the basis for all administrative districts (14). Activity-based funding to reimburse hospitals has been used since 2012 (15).
Overall, the Swiss primary health care system is stable and fully functional. Switzerland is the world's second-highest per capita spender on health care (2,(16)(17)(18)(19), and considering that income is relatively equally distributed across the country, administrative districts are offered similar preconditions to establish their primary care structures. A recent study of ACSC Swiss contexts showed a 12-fold level of variation among small regions. That particular study was restricted to certain regions of Switzerland and provided limited insight into regional patterns regarding diagnoses and primary care provision in different settings (2).
Employing small area analysis with data from the Swiss federal statistical o ce, we sought to establish the rst complete epidemiological map of avoidable hospitalisations for our ve selected ACSCs from Swiss nursing homes, home care organisations and the general population. The results offer a new perspective on regional variation concerning the distribution, the evaluation and planning of primary health care services for the chronically ill population.

Design and sample
This retrospective analysis used routine health care data of all Swiss hospitals (20) from 2017, as provided by the Swiss Federal Statistical O ce (SFSO). To assess the suitability of eligible diagnoses we used a set of quality indicators provided by the OECD to compare the quality of health care provision between countries. As the OECD Health Care Quality Indicator Project uses avoidable hospitalisations for ACSCs as a quality measure for chronic disease management in primary care (3,21), we included four prominent chronic conditions from their list of indicators: congestive heart failure, chronic obstructive pulmonary disease (COPD), diabetes complications and hypertension (3,21). A fth indicator, angina pectoris, is commonly used in similar settings (6, 7). ICD 10 codes for these conditions can be found in table 1. Consistent with the OECD criteria, we included all hospitalisations in the population aged 15+ and not referred from other hospitals or rehabilitation clinics (3,21).

Data sources
Data for this study were extracted from the SFSO's annual census report and from the medical statistical data collected by Swiss hospitals on all hospitalisations. Census data included age and gender distribution for each small area (22). All Swiss hospitals collect medical data continuously in compliance with Swiss federal law and provide them annually to the SFSO (23). Patients hospitalised multiple times were assigned to multiple cases with unique anonymized patient identi ers to allow us to track their hospital admissions throughout the year.

Geographic areas
We used the Swiss acute care hospitals' and administrative districts' (cantons) utilisation-based hospital service areas (HSA) for geographical analyses. HSAs and cantons are compatible with MedStat's geographical units used in Switzerland to provide anonymized data on patients' residences based on small geographical areas. Census data and data on patients' residences were provided by the SFSO. Each MedStat unit is home to approximately 10,000 inhabitants and is contained within a single HSA or canton (24). Based on discharge data from the Swiss acute care hospitals, HSAs are de ned and maintained by the SFSO (25).
Using HSAs and MedStat areas is an established approach to analysing area-speci c medical data (2). It ensures compatibility of the medical statistical data for hospitals with census data across all three levels (MedStat, HSA, canton).

Variables and measurements
All variables used are described in the SFSO's variable speci cations for medical statistical data for hospitals and are applicable to the 2017 dataset (26). Variables include data on diagnosis, locational and sociodemographic characteristics, as well as determinants of hospital stay, admission and discharge (table 2).

Statistical procedures
All analyses were performed using R 3.5.2 (27). The SFSO provided the dataset without missing data of the relevant variables. We performed descriptive analyses on the dataset after ltering out hospitalisations that were referrals from other hospitals or rehabilitation clinics or were patients under the age of 15. The dataset provided information on whether and when each patient was rehospitalised. With these data we calculated rehospitalisation rates within the given year. We also calculated comorbidity scores for each patient using the Elixhauser comorbidity score and the "comorbidity 0.5.3" software package (28,29). Used to provide a condensed score for all de ned comorbidities for each patient, the Elixhauser comorbidity score theoretically ranges from -19 to +89, with higher scores indicating more comorbidities (28, 29).
To determine the number of hospitalisations for each stratum, data were aggregated to each level (MedStat, HSA, canton), each diagnosis group and care structure. Rates were calculated using the number of admissions for avoidable hospitalisations as numerator and the population of each small area (MedStat) over the age of 15 as denominator and multiplied by 100,000. The rates for each small area were standardized for sex and age using direct standardisation based on the 2013 standard population for the European Union (EU) (30). We then calculated median rates of avoidable hospitalisation per 100,000 adult inhabitants, as well as interquartile ranges (IQR) for all ACSCs and care structures for all HSAs (n=61) and cantons (n=26). Outliers -datapoints 1.5 times the IQR above the upper or below the lower quartile -, were assessed individually and kept in the dataset.
For inferential statistical analysis we used linear mixed models with Gaussian distribution to assess regional variation using the "lme4" software package (31). ICC 1s were calculated for the HSA and canton levels using the package "RptR" with bootstrap set at 2,000 (32). ICC 1 values above 0.05 were considered meaningful (33). Models were calculated with random effects for HSAs and cantons.
For geographic visualisation we used SFSO-provided geodata. The "sf 0.8.1" and "tidyverse 1.3.0" software packages were used to merge the geodata with the dataset and compute spatial visualisations (34,35).

Characteristics of avoidable hospitalisations
The data from 2017 included 287 hospitals and specialized clinics that reported to the SFSO. This included all Swiss hospitals (20).
In 2017, SFSO medical statistical data recorded 1,468,245 hospitalisations. Excluding paediatric hospitalisations (<15 years) and those resulting from referrals from other hospitals or rehabilitation clinics left 1,076,716. From this number, we identi ed and included 46,479 with main diagnoses corresponding with one of our selected ACSCs, possibly indicating avoidable hospitalisations. Figure 1 illustrates the sample selection process. Our sample amounted to 4.3 % of all hospitalisations from primary health care in the adult population. We observed a median length of stay of 6 (IQR 2-10) days for avoidable hospitalisations for ACSC and a median Elixhauser comorbidity score of 4.5 (IQR 4.0-5.0). In 46.8 % of hospital admissions for ACSC in 2017, physicians referred cases to the hospital. Overall, 78.2 % of admissions for were referred as emergencies and 21.5 % were scheduled. Mortality rate in ACSC cases was 4.1 % and the rehospitalisation rate was 30.0 %. The age distribution regarding hospitalisations for ACSCs is illustrated in gure 2. We found that 90.2% of such hospitalisations for ACSCs came from home, while 2.1% were patients using home care services. Cases from nursing homes amounted to 4.7 % of avoidable hospitalisations for ACSCs and 3 % of cases came from psychiatric, penal, other or unknown institutions. For more details on sample characteristics see Tables 3 and 4.

Regional Variation
For this study, Switzerland was divided into 705 small areas (MedStat) with a median population size of 10,665 (IQR: 8,261-14,356). These occupied 61 hospital service areas (HSA) with a median population of 95,353 (IQR: 64,748-163,939). The 26 Swiss cantons provided the highest level for this analysis, with a median population of 234,857 (IQR: 75,384-393,331).
Overall, the median unadjusted rate of avoidable hospitalisation was 489 (IQR 396-592, min. 102, max. 1,677) per 100,000 adult inhabitants. The overall sex-and age-standardized median rate of avoidable hospitalisation for ACSC was 1,080 (IQR 893-1.274, min. 317, max. 3.033) per 100.000 adult inhabitants. On the HSA level, we found median sex-and age-standardized rates of 1,113 (IQR: 961-1,208, min. 754, max. 1,581) avoidable hospitalisations for ACSC per 100,000 adult inhabitants. On the cantonal level we calculated a median sex-and age-standardized rate of 1,062 (IQR 978-1,208, min. 800, max. 2,208) avoidable ACSC hospitalisations per 100,000 adult inhabitants. On the HSA level the ICC 1 was 0.16 (95% CI: 0.09-0.24); on the cantonal level, it was 0.20 (95% CI: 0.09-0.31). Table 5 describes unadjusted and direct age-and sex-standardized rates per 100.000 inhabitants for the observed ACSC and respective ICC 1 values for the HSA and cantonal levels. Table 6 describes unadjusted and direct age-and sex-standardized rates per 100,000 inhabitants for the observed settings and respective ICC 1 values for the HSA and cantonal levels. A geographical representation of total sex-and age-standardized rates of avoidable hospitalisation for ACSC per 100,000 adult inhabitants of all 705 small (MedStat) areas is provided in gure 3. Additional maps for the three settings (home, nursing home and home care) and for the various diagnostic groups (angina pectoris, congestive heart failure, COPD, diabetes complications and hypertension) are available in the supplementary materials.

Discussion
This study provides the rst complete mapping of avoidable hospitalisations for ACSCs in Switzerland in 2017. Using small area analysis to determine regional variation for various ACSCs and primary care structures, we found substantial regional variation with distinct disease-speci c regional patterns. Standardized for sex and age, the overall degree of regional variation was higher than in other European countries i.e. Denmark, England Portugal, Slovenia and Spain (36).

General characteristics
Our results suggest that 4.3% of all hospitalisations in 2017 were avoidable. Furthermore, we observed a gradual increase in avoidable hospitalisations for ACSCs in the population above 65 years of age, peaking at the 80-84-year age group. This pattern is consistent with results from a similar study that investigated avoidable hospitalisations for ACSCs in France (6).
More speci cally, consistent with previous studies in Germany, we found that hospitalisations for congestive heart failure and COPD account for a substantial fraction of avoidable hospitalisations (7,10). About half of the identi ed cases were referred to hospital by physicians, with roughly three quarters of patients admitted to hospital as emergencies.
Our sample's Elixhauser comorbidity scores were rather high at 4.5 (IQR 4.0-5.0) compared to those measured by van Walraven et al.
(2009), who recorded a median score of 0 (IQR 0-8). This indicates, that our population had more comorbidities present, than a regular hospital cohort. Aditionally, the high number of emergency admissions suggest that the admitted patients had experienced a profound deterioration of their already fragile health prior to admission with multiple comorbidities. Interestingly, about 30% of cases are readmitted to hospital within the year 2017, indicating challenges in primary health care provision, especially regarding self-management and monitoring of early warning signs.
When addressing overall unadjusted rates of avoidable hospitalisations for ACSCs, we found similar results for four of our diagnoses of interest (congestive heart failure, COPD, diabetes complications and hypertension) also used in a similar study by Berlin et al. (2014) in the Swiss context (2). Compared with that study's ndings, our overall unadjusted rates of avoidable hospitalisations for ACSCs indicate an increase of 2.7% over a seven year period (2). However, compared to similar studies in Swiss, French and German contexts (2, 6, 7), this increase is actually quite low.
Standardized rates of avoidable hospitalisations for ACSCs in Switzerland were high in the European context; likewise, regional variations for the ACSCs of interest were considerably higher in Switzerland than in other European countries (36).
The ndings suggest a high degree of variation amongst HSA and cantons regarding avoidable hospitalisations for all ACSCs.
Moreover, we found pronounced geographical patterns based on both diagnosis and setting. Most prominently, variation in the management of angina pectoris shows substantial variation in Switzerland's northern and north-eastern regions. Interestingly, when assessing rates of congestive heart failure, these patterns shift towards the southwest. Regarding hypertension, though, we found consistently lower rates in the southwest. While these patterns indicate some of the challenges HSAs and cantons face in providing specialized primary health care for different diseases, they also underscore the importance of differentiating between diseases and visualizing results to address issues in primary health care provision. The broad regional variation for the various diagnoses may re ect speci c regional (cantonal) approaches to primary care provision. Evidence supports the possibility that socioeconomic, demographic and provider speci c determinants contribute to the emergence of avoidable hospitalisations for ACSCs (13,17,(37)(38)(39)(40). Switzerland's primary health care system is stable and functional, and its income inequality quite low (2,(16)(17)(18)(19). Rather than access to affordable medical advice, contributing factors might include regional differences in compensation structures for both primary and hospital care, as well as the proximity and density of hospitals, primary health care networks and regional public health programs. While it remains unclear just how these factors affect the rates of ACSC-related hospitalisation, minimizing those rates will demand an understanding of the contributing factors.

Contributing Factors and Impulse for Health Policy
Evidence suggests that physician density, healthcare accessibility, resources for primary health care and continuity of care are all related to rates of avoidable hospitalisation for ACSCs (2,(41)(42)(43)(44)(45). In Switzerland, except in some isolated alpine regions, accessibility to primary health care is consistently high (46), i.e., resources for primary health care were reinforced in 2014, and physician density is su cient. More physicians might actually lower healthcare e ciency: several studies suggest that high physician density can in ate demand for health care services (2, 40, 47).
The issue of regional variation and the high rates of avoidable hospitalisation for ACSCs despite high accessibility to and adequate resourcing for primary care might be an indication that continuity of care plays a crucial role. Continuity of care should focus on a team-based approach to reduce fragmentation of care and improve patient safety and quality of care (48). Focussing of chronic care management might provide useful guidance to improve continuity of care. Chronic care management in Switzerland is still predominately provided by primary care physicians. With Switzerland's primary care physician workforce aging this may eventually lead to a shortage of general practitioners and disrupt chronic care management (16). Chronic care management for speci c populations might therefore require novel roles in care delivery for the chronically ill (49). Interventions to reduce hospitalisations for ACSC include specialized home care, promotion of self-management and the integration of primary and secondary care (50). Complementary to established primary care models nurse-led models can be part of such interventions. Advanced practice nurses already play crucial roles in specialized and primary care delivery, improving outcomes in chronic care management, e.g., reduced hospitalisations or improved blood pressure management, as compared to established models of care (51,52). Advanced practice roles such as nurse practitioners can perform many tasks of chronic care management such as promoting self-management and care coordination. However, such roles are currently underdeveloped in Switzerland and limited to a small number of collaborative efforts (53,54). Hence, nurses can only provide limited resources in Switzerland. Collaboration with different primary care providers such as physiotherapists, dietitians, occupational therapists is crucial to address self-management and ultimately reduce hospitalisations for ACSC. Swiss health policy makers could address these challenges by promoting these approaches to bridge the gap in chronic care management and to improve continuity of care.
The geographical representation and small area approach differentiated by diagnosis and care structure highlight the various Swiss regions' relative success at minimizing avoidable hospitalisations. There is a need to understand the speci c context and its impact on continuity of care. Health policy makers should address these regional variations with a distinct focus on strengthening continuity of care for the chronically ill.

Strengths And Limitations
This study offers the rst complete map of avoidable hospitalisations for ACSCs in Switzerland. There are, however, several limitations. Selection criteria for ACSCs differ in the literature and interpretations differ regarding the preventability of certain ACSCrelated hospitalizations (5). Moreover, we cannot discriminate between clinically avoidable or necessary hospitalisations beyond the information provided within the routine dataset. Further, this study did not account for sociodemographic or socioeconomic differences such as education and income. Nor did it examine behavioural or cultural factors affecting the use of hospitals and primary care or account for the distribution of healthcare structures, e.g. the number of nursing homes within an area. Still, the study offers a new perspective on regional variation of avoidable hospitalisations for ACSCs in Switzerland. On the other hand, one of this study's strengths is the inclusion of the most prevalent ACSCs for chronic conditions. This will help rst to identify well-functioning primary care services in regions to inform and enable health policy adjustments.

Conclusion
This study identi ed substantial regional variation in and comparably high rates of avoidable ACSC-based hospitalisation in Switzerland. We suspect that differences in continuity of care are predominantly responsible for these regional variations. As ACSCs account for an increasing number of hospitalisations in Switzerland, indicating a need for multidisciplinary care models of care that allow increased continuity of care, they should be dealt with speci cally at the health policy level. Further research is needed to model and assess the impact of different primary care models on ACSCs.

Consent for publication
Not applicable.

Availability of data and materials
The data of this study are available from the Federal Statistical O ce of Switzerland, Departments of Health Services and Public Health ( "Sektionen Gesundheitsversorgung, Gesundheit der Bevölkerung") upon application. Further information are available here: https://www.bfs.admin.ch/bfs/de/home/statistiken/gesundheit/erhebungen/ms.html

Competing interests
The authors declare that they have no competing interests.

Funding
This study did not receive external funding. Expenses for data delivery from the SFSO were covered by internal Institute of Nursing Science funding.
Authors' contributions MS, FZ and NG developed the idea of the study. NG prepared and analyzed the data. NG, MS and FZ interpreted the results. NG drafted the manuscript while MS and FZ provided critical feedback and revision of the manuscript. All authors read and approved the nal manuscript. 54. Burke RE, Guo R, Prochazka AV, Misky GJ. Identifying keys to success in reducing readmissions using the ideal transitions in care framework. BMC Health Serv Res. 2014;14:423. Tables   Table 1   ICD 10 Codes and Sources for Analysis   Condition  ICD-10 Codes  Source Angina pectoris I20 I24.0 I24.8 I24. 9 Purdy et al.   *Some admissions were coded as "other" and "unknown" thus percentages do not add up to 100 %

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.