The Causal Effect of Community Hospitals on General Hospital Admissions. Evaluation of a Natural Experiment Using Register Data

The Coordination reform of 2012

The Norwegian health care system is a two-tier organization: municipalities are responsible for providing primary health care, long-term care, and social services to everyone in need, while the central state is responsible for the specialist services [10, 11]. The central state took over the responsibility for specialist health care from the counties in 2002 and organized and funded the services through five, later reduced to four, regional health authorities [12]. The decentralized task structure of the primary care services is combined with central regulations of revenues and service standards. The central state also has the responsibility of exercising supervision and control. Thus, the Norwegian municipalities are more limited in their ability to prioritize and adapt services according to local preferences than suggested in the standard literature of fiscal federalism [13].

Although the Norwegian health care system is generally known for high quality services [14, 15, 16], a comprehensive reform package named the Coordination Reform was implemented from 2012. The reform came as a response to what the central government defined as three main challenges in the Norwegian health services [7]:

• – Patients’ needs for coordinated services were not sufficiently met.

• – There was too little initiative aimed at limiting and preventing diseases.

• – Population aging and the changing burden of diseases among the population.

The reform can be seen as part of a broader movement towards improving coordination and integration, and shifting of tasks and responsibilities from specialist to primary healthcare [17, 18]. The comprehensive policy package involved several measures to ensure successful implementation and commitment of the involved parties [7]. Two new pieces of legislation, The Norwegian Public Health Act (“Folkehelseloven”) of 2012 and the Act of Municipal Health and Care Services (“Helse- og omsorgstjenesteloven”) of 2011, were implemented together with three economic incentive measures:

• – An earmarked matching grant to the municipalities to stimulate investments in MAWs,

• – municipal co-financing of treatment in the hospitals’ internal medicine departments and outpatient clinics, and

• – municipal payment for patients ready for discharge and needing care from the municipalities [19].

The municipal co-financing had no significant effects on hospital admissions and was abandoned from 2014 [20].

The earmarked grant for the MAWs was intended to stimulate investments for a period of four years, from 2012 until 2016, when the MAWs became mandated. From 2016 the earmarked matching grant was included in the central state’s general grant to the municipalities. The municipalities could organize the MAWs either as a municipal service covering one municipality or as inter-municipal cooperations covering two or more (often smaller) municipalities [8, 21].

The MAWs – means to achieve care integration

The MAWs are organizations that aim to increase coordination between primary and specialist care by better triage and by treating selected patient groups at the primary care level. In this way the MAWs can be interpreted as an integrated care initiative with the aim to overcome fragmentation of the system and streamline the clinical pathways [22, 23, 24].

The MAWs resemble community hospitals known from high-income countries like Australia, Canada, the Netherlands, Sweden, UK, and the US [25, 26, 27, 28]. Winpenny et al’s [29] classification of community hospitals along several dimensions is fruitful also to describe the Norwegian MAWs. First, the MAWs must be distinguished from primary care services without inpatients beds like GPs and local emergency departments. Second, although the MAWs are often co-located with nursing homes, their responsibilities differ. Norwegian nursing homes are primarily occupied by patients in long-term care, yet with increasing amounts of short-term stays related to rehabilitation, respite care and palliative care. The MAWs, on the other hand, provide acute inpatient care for patients. Third, like most types of community hospitals, the MAWs are staffed with nurses and generalist doctors (GPs). Fourth, while community hospitals in many countries are located in rural areas, the Norwegians MAWs are present also in urban areas. Sixth, unlike in some countries, like Italy, where the focus of community hospitals are on post-acute care [30], the main purpose of the Norwegian MAWs is to reduce specialist care admissions. This means that patients in the MAWs are persons that otherwise would have been admitted to a hospital.

Many of the earlier studies of the effects of community hospitals have been small scale [31, 32] or studies with methodological shortcomings [33]. However, several of the case studies found that the introduction of community hospitals and other forms of intermediate care units resulted in a reduction of unnecessary hospital admissions among older adults [31] as well as reductions in prolonged hospital stays and reduction in delayed discharges from the hospitals [34]. A US-study of a post-acute intermediate-care geriatric unit [35] indicate that such units also may represent a potential alternative to acute hospitalization for selected older patients.

Also of importance are studies that have evaluated the effects of comparable initiatives as the MAWs in primary care. Some of these have evaluated the substitution effect of expanding GP and primary care emergency services on the use of hospital emergency capacity. Lowe et al. [36] found that patients’ use of emergency departments was reduced if GP practices were open during the evening hours. Krämer and Schreyögg [2] found negative effects of primary care emergency services on both ambulatory emergency visits and emergency inpatient admissions in a large-scale German study. Both these studies are in their approach similar to ours, but as already indicated, we have not found any quantitative studies that explicitly have evaluated the effects of introducing community hospitals/local acute wards on specialist care substitution rates.

Data and study population

Data on hospital admissions were obtained from the Norwegian Patient Register (NPR) and included patients admitted to hospitals between 1st January 2010 (two years before the start of the Coordination reform) and 31 December 2017, excluding psychiatric hospitals. The patients were grouped along four dimensions:

• – Contact types: Admissions or day stays

• – Urgency: Acute or elective treatment

• – Departments: Internal medicine or surgical

• – Age groups: <80 years or ≥80 years

Numbers of admissions for different types of services and age groups were allocated to the municipalities based on the patient’s place of residence, summed, and standardized per 1000 inhabitants. The cut-off at 80 years was pragmatically set but reflects that the MAWs were primarily targeting the population of older adults. As the life expectancy in Norway is above 80 years, a cut-off at 80 will make it possible isolate the effects of the MAWs on the main target group.

We acquired annual municipal data from Statistics Norway’s KOSTRA-database covering all Norwegian municipalities from 2010 through 2017. This database includes variables describing municipal supply of services, such as the number of nursing home beds, number of GPs, GP contracted hours per week in nursing homes, and variables describing the demand for services, i.e., total population and the population of specific age groups. These supply and demand side control variables were normalized per 1000 inhabitants.

We obtained monthly descriptive data for each MAW via structured telephone interviews with the managers of the MAWs or their deputies administered in three rounds taking place during summer 2013, summer 2015 and winter 2018. The data included among other variables describing month of the first MAW admission, available equipment, number of beds and availability of GPs on site. All the data from the data collection used in this article, were based on questions with pre-defined responses.

We linked data by municipality number, year and month followed by linear interpolation to approximate the monthly values for municipal demand and supply data that were reported annually. This gave us a dataset with panel structure. We included all municipalities except four which were amalgamated during the study period (N = 425). The unit of analysis is then municipality-month.

Statistical model

Our demand model was formalized within a generalized framework for policy analyses of panel data [41]. The coefficients of interest are b₃ in Equation 1 (Eq. 1) and b₄ in Equation 2 (Eq. 2).

where:

• – logDH_mt is the natural logarithm of variables describing use of hospital services for each municipality m at time t (for further specification, see Table 1),

• – logS_mt is a vector of municipal supply variables, i.e., nursing homes and GPs,

• – logD_mt is a vector of municipal demand factors, i.e., population and the age composition of the population,

• – MAW_mt is a dummy variable that takes the value of 1 in the first month a municipality implements a municipal acute unit and onwards, 0 else,

• – f are municipality-fixed effects to account for the time-invariant unobserved characteristics of municipalities,

• – y are year dummies to account for time specific variation,

• – t is a seasonal adjuster for month, and

• – u is the error term for municipality m at time t.

All variables except the dummies, were standardized by the total population in 1000 in each municipality.

After conducting the analyses of main effects, we dig further into the analyses of the patient group of most interest – the acute admissions among older adults (≥80 years) at internal medicine departments. We described variation in acute preparedness by the variable PREP_mt which is a dummy variable that takes the value of 1 if the MAW has a medical doctor on site 24/7 or is located less than 200 meters from a municipal emergency center, 0 else. The interaction term MAW_mt*PREP_mt will give us the additional effect of having a MAW with acute preparedness as compared to the effect of having a MAW without such a feature. This model can be summarized as:

Analyses

The use of fixed effects in data with panel structure allowed us to control for data heterogeneity and focus on the variable of interest. Of most interest, fixed effects for municipalities give us the within estimator of the MAW variables [42].

We regarded all variables except MAW as exogenous. MAW might have elements of endogeneity as municipalities with high admission rates have a slightly stronger incentive to implement MAWs than municipalities with lower admission rates. We handled this endogeneity problem via municipal fixed effects.

We used SAS, version 14.1, for all analyses.

Descriptive statistics

The most common way of organising the MAWs were within nursing homes, where approximately 60% of the units were localised. The size of the units varied from 1 to 72 beds in 2017. Due to the structure of Norwegian municipalities with half of the municipalities having less than 5000 inhabitants, the number of MAWS with less than 3 beds made up 30% of the total. Number of total MAW beds increased from 340 by the end of 2012 to 723 by the end of 2017. In the same period the number of MAW patients increased from app. 6600 to app. 40400 and the number of MAW bed days increased from 16 000 to 104 000 [8, 43]. 723 beds in total means that there are 0,13 beds per 1000 inhabitant. For further descriptive analyses, see [43, 44, 45, 46, 47].

The main target of the MAWs are the acute admissions at medical departments. These admissions are particularly high among the older adults ≥ 80 years with 32.2 admissions per 1000 inhabitants per year in 2010. Standardized by the total population (Table 1a), the average number per municipality and month is 1.76. Among the younger population (Table 1b) and reflecting the greater size of this age group, the comparable number is 6.04. The number of elective admissions at the medical departments are significantly lower than acute admissions in both age groups. At surgical departments, admissions are divided approximately equally between acute and elective admissions in both age groups. The number of day stays are comparatively low except for elective day stays at surgery departments. There are negative time trends in most of the dependent variables.

By the end of 2014, 221 (51.8%) municipalities had implemented MAWs (Figure 1) and by the beginning of 2016, the time when MAWs became mandatory, all except eight municipalities had implemented MAWs. The growth in MAWs was highest from summer 2012 and the next one and a half year and in the last months of 2015, just before MAWs became mandatory.

Figure 1

Reviewing descriptive statistics for the independent variables (Table 2), we observe a slight reduction in nursing home places over time, while the number of GPs is increasing. The share of population ≥ 80 years was 5.49 percent in 2010 and decreased until 2016. From 2016 to 2017 the relative share of the population ≥ 80 years increased. The population growth in the study period is approximately 1.1 percent per year. Observe that the number of inhabitants in Table 2 with this specification, describes the mean population at municipal level.

In line with Figure 1, we observe that the share of the population with access to MAW increases from 2012. In 2017, 97 percent of the population had access to MAWs. For approximately half of the population with access to MAW, their MAW either had a medical doctor on site 24/7 or was located less than 200 meters from a municipal emergency centre (the PREP variable).

Regression results

Our analyses indicate that implementation of MAWs decreased the number of acute admissions at medical departments for the age group ≥80 years by 3 percent (Table 3a). The demand variables yield reasonable effects. A one percent increase in the age group ≥80 years, meaning that the population is getting older, increased the number of acute admissions at hospitals by 0.56 percent. There are also significant and positive effects of number of inhabitants in general. For the variables describing municipal supply of services, effects were small and not always significant. For the elective admissions at medical departments, there were no significant effects of the MAWs. Also, for the other types of services the effects of the MAWs were either insignificant or very small in size (less than 1 percent).

For the age group <80 years (Table 3b), the effects of the MAWs were smaller than for the age group ≥80 years. There were, however, negative effects of the MAWs on both acute admissions and acute day stays at the internal medical departments. There were also a few positive effects of the MAWs, in particular on surgical admissions or day stays. In general, the level of these stays was very low (cf. Table 2). The effects of the demographic variables were less stable when analysing the lower age group.

Table 4 presents the result of the analyses where interaction terms between MAW and acute preparedness were included.

For the oldest age group, the effects of the introduction of MAWs on acute admissions in medical departments was minus 2 percent. There was an additional effect of minus 1 percent if the MAW was organized with a physician on site 24/7 or was located close to a local emergency centre. For the age group <80 years, the reduction in acute admissions could be found only among municipalities that had a MAW with the highest level of acute preparedness.