Public and private providers
Norway has four regional health authorities (RHAs) named after their locations (Southeast, Central, North, and West). There are two types of radiology providers in Norway: private and public. Private providers operate as for-profit institutions that can have contracts with RHAs and deliver radiology services on public terms (Patients only pay the laboratory a patient co-payment, while the rest is paid by the state and the RHA.). Each RHA chooses a number of private radiology providers through a tendering process and by signing contracts with them for a specific number of services. This option is sometimes associated with wait times for patients. Private providers also deliver radiology services on private terms (when patients pay the full fee directly to the laboratory); this option is not associated with patient wait times.
The contracts with RHAs specify the volume of and reimbursement for examinations, the maximum number of services, and the total costs. Some contracts specify only an aggregated budget for services [27]. Other contracts are detailed and specify the budget for each type of service, such as ultrasound imaging (UI), magnetic resonance imaging (MRI), computed body tomography (CBT or CAT scans), and radiography (X-rays) [28].
Public providers are hospital radiology departments that deliver radiology services to the population on public terms; that is, they accept both patients from hospitals and outpatients referred to them by GPs and specialists. Visits to a public or private laboratory require a referral from a GP or a specialist to be covered by the National Health Insurance (NHI) [29]. In theory, radiology laboratories can decline to make an appointment, but in practice, this does not happen often because GPs already act as gatekeepers [30].
The 2008 reimbursement change
Reforms in the financing of specialist health care have been carried out since 1997 and activity-based funding (ABF) was introduced to encourage the achievement of activity targets ([31], p. 69). If these targets were not met, the RHAs lost income. If the activity levels were higher than targeted, then the costs would be only partially compensated. Hence, ABF was not intended to cover marginal costs or to encourage activity beyond the target ([32], p. 13).
For radiology services, ABF funding was first introduced 1 September 2005 to encourage RHA to take more responsibility for planning and prioritizing provision of radiology services [32,33,34], (p. 248 in [35]). Between 2005 and 2008, the proportions of activity-based and basic allocation were approximately equal. Figure 1 demonstrates that, prior to 2008, spending for private radiology continually increased.
The reimbursement change of 2008 changed radiology funding accordingly: from 1 January 2008 the ABF part decreased from 50 to 40%, and the basic allocation increased from 50 to 60% (from RHA) to compensate. The reimbursement scheme was changed to cut spending and to harmonise the financial scheme of radiology providers with the general system for financing outpatient medical services in Norway [32, 36]: ‘The aim was both to contain costs and to give providers sufficient flexibility to assure the best mix of services for patients’ [31].
In practice, introducing ABF meant that private providers had to enter into agreements with RHA to produce the agreed number of services and receive refunds. These providers would still receive ABF from NHI and co-payments from patients if they produced more services than agreed with RHA. The providers’ revenue thus included three components: the fee-for-service from the NHI scheme (or ABF), patient co-payments (the same for both private and public providers when received through NHI), and the invariable component (a basic allocation independent of the number of the services provided). The size of ABF was based on diagnostic related groups [31], while the size of the basic allocation was decided by several factors, including the number of inhabitants living in the region and the demographics of the population [32].
The public providers’ revenue also included three components equivalent to those of the private providers, but with a different reimbursement mechanism. The RHA and NHI did not reimburse public radiology laboratories directly; instead, they reimbursed the hospital affiliated with the laboratory. Thus, public outpatient providers were not as restricted by contracts as their private counterparts, so they had softer budget constraints than private providers did. Soft budget constraints are often related to a poor ability to balance budgets and providers with the tendency to increase activity or costs to a level above that preferred by the principle stakeholder [37,38,39]. In contrast, in the private sector, the number of services was controlled by hard budget constraints to maintain positive profits because contracts included specified volumes.
Data
Claims data were obtained from the Norwegian Directorate of Health. The dataset (aggregated at the municipality level) contained the number of radiology services (CAT scans, MRIs, X-rays, and ultrasounds) reimbursed per month by NHI from 2007 to 2010, the travel times from the patient’s municipality to the municipality with closest private or public provider, the number of inhabitants, the centrality of the municipalities, and the RHAs to which they belonged. Thus, 422 municipalities in 48 different periods (monthly observations from 2007 to 2010) were monitored for a total of 19,867 observations.
Variables
Travel times
Table 1 in Appendix contains an overview of the variables. The travel times were measured in hours according to driving time by car (provided by Info Map Norway [40]) between a patient’s residential municipality (approximated by the municipality of the patient’s GP) and the municipality of the public radiology provider (Pubtime) or the private radiology provider (Privtime). If patients had a radiology provider in their own municipality, then the travel time was set to zero by definition in the dataset. The difference in travel time between the nearest private provider and nearest public provider is represented by Time_difference = Privtime − Pubtime. The difference in travel time is included as the main independent variable because, when deciding between two providers in the settings of unevenly distributed providers, patients often choose a closer provider. Since private and public providers have different institutional settings, this choice affects the outcome.
Centrality/municipality level
Statistics Norway classifies every municipality in Norway by centrality. During the observation period, centralities were 0A and 0B, 1A and 1B, 2A and 2B and 3 [41]. In the data set, 0A and 0B is denoted by Centrality0, 1A and 1B by Centrality1, 2A and 2B by ‘Centrality2’, 3 by Centrality3 (Centrality0 through 3 are dummy variables), where Centrality3 represents the most central type of municipality (e.g., Oslo), and Centrality0 denotes the least central ones (e.g., small remote villages).
Centrality indicates the location of municipalities in relation to urban settlements of various sizes [42, 43] and reflects the travel time from an urban settlement to a centre with well-developed infrastructure, including banks and post offices, as well as the number of inhabitants and public services available (see [44, 45] for details). Since research indicates that residents of closely settled areas are much less willing than people in sparsely populated areas to travel to access a health care provider [17], centrality might not only reflect the type of municipality but may also be correlated with patients’ willingness to travel.
Regional health authorities
Region1 through 4 are dummy variables describing whether the municipality belongs to (1) the South East, (2) West, (3) Central, or (4) North RHAs.
Centrality0 through 3 and Region1 through 4 are time invariant. They are part of the fixed effects and are therefore cancelled out in the model, but they are used for descriptive statistics.
Number of services
The dependent variable is the number of services provided at private (Priv_Serv), public (Pub_Serv), or both types of providers (Total_Serv) per month. This variable was calculated by accumulating claims in every municipality. For example, if a patient from municipality A goes to municipality B to receive a radiology examination, that service is classified as a service delivered to municipality A. The measurement of this variable reflects the number of services per 1000 inhabitants in the municipality.
Hypotheses
Patients who live in the centres have better access to both public and private providers, while those who live remotely must travel up to several hours to reach a provider. The aim of the study is to investigate the interaction between patients’ travel times and the 2008 reimbursement change in terms of the number of services consumed. The Norwegian population was expected to be unevenly affected by the reimbursement change, depending on the distances to different types of the providers.
The hypotheses were based on two assumptions (A1 and A2): (A1) There is stream of patients who need services, and if one source reduces its offerings, the patients will switch to another more readily available source (both in terms of capacity and travel time); (A2) Public providers have softer budget constraints and can thus better stretch their capacity outside the limits set by budgets compared to private providers, which have hard budget constraints.
Thus, the following hypotheses were formulated.
Hypothesis 1
There will be a larger decrease in the number of private services than public services based on the differences in these services’ budget constraints.
Hypothesis 2
The stream of patients who move between providers and the effect on the total number of services will be different depending on the difference in the proximity of private and public radiology providers. The changes at private, public, and both providers will be following.
(2A)
Patients use private radiology more when these providers are relatively closer (i.e., Time_difference is negative or equal to zero), which means that, after 2008, the greatest reduction in the Priv_Serv will be in these areas. The reduction diminishes with the increase in Time_difference.
(2B)
The change for public providers consists of two effects. The first involves a reduction in the original public service users. The greater usage was before 2008; the greater reduction in the number of services will be after 2008. In general, patients use public radiology more when these providers are closer (that is, when Time_difference is zero or positive). The second effect relates to users switching from private radiology. These patients are more likely to switch the closer they live to a public provider compared to a private provider (i.e., the greater the value of Time_difference). Depending on what effect is greater, the change will be positive, negative, or equal to zero.
(2C)
Since private providers are more affected, the greatest reduction in the total number of services occurs in areas with negative Time_difference. This reduction will diminish with an increase in Time_difference because patients can more easily switch to a public provider.
Figure 2 represents a visual explanation of the hypotheses in terms of Time_difference—how the consumption of services would change when moving on the scale of Time_difference from negative to positive values. Figure 2 makes use of three states: negative, equal to zero, and positive values of Time_difference. The text boxes indicate what was expected in each of the three states and why.
Following Fig. 2, the first text box indicates that the closest radiology provider is private (Time_difference < 0). The total number of services is expected to decrease due to reduced offerings from private providers. Since there is a longer travel time to the public provider, fewer patients would move to the public provider due to time costs compared with the other two cases (when the public provider is closer or equally close). Therefore, more patients would rather not have the radiology examination at all or have the exam out of pocket. Thus, the total number of services would decrease more than if the closest radiology provider were public.
The second box indicates that the distance between private and public radiology providers is small (Time_difference \(\to\) 0). In this situation, patients can change providers more easily. The likelihood that patients will switch from private to public radiology provider is higher. Thus, a substantial drop in Priv_Serv and an increase in Pub_Serv is expected, while the total number of the services may not change.
In the third textbox, the closest radiology provider is public (Time_difference > 0). Patients are expected to use the public provider more than the private. Since public providers have softer budget constraints, the total number of services is expected to be less affected by the reimbursement change. However, some patients who used private providers before 2008 would move to public providers due to the private providers’ reduced offerings after 2008. Therefore, the total number of services is expected to stay the same, public services are expected to increase or stay the same, and private services are expected to decrease or stay the same.
Model
A model of how Time_difference would affect number of services for private, public, and both providers after the reimbursement change was estimated. Time-invariant heterogeneity is controlled for without observing it through the panel data. A fixed effects model was used because it is more robust and needs fewer assumptions fulfilled than a random effects model. The fixed effects model is based on the assumption that the errors are uncorrelated with the independent variables and that the errors are conditionally homoscedastic and not serially correlated [46].
The relationship between number of the services and the Time_difference was not expected to be completely linear. Thus, after trying several polynomial functions, a quadratic function was chosen. A regression model was estimated separately for each of the samples of private and public providers, as well as for the sample including both types of providers:
$$\begin{aligned} {\text{Y}}_{\text{it}}& = {\text{ B}}_{0} + {\text{ B}}_{ 1} post08_{\text{t}} + {\text{ B}}_{ 2} post08_{\text{t}} \cdot Time\_difference_{\text{it}} \hfill \\ & \quad + {\text{ B}}_{ 3} post08_{\text{t}} \cdot Time\_difference_{\text{it}}^{ 2} + {\text{ e}}_{\text{i}} + {\text{ u}}_{{ 1 {\text{it}}}} \hfill \\ \end{aligned}$$
where Yit denotes the number of services (Priv_Serv, Pub_Serv, Total_Serv) to municipality i (i = 1,…,422) in period t (t = 1,…,48), post08t is a dummy equal to 0 prior to 2008 and 1 after 1 January 2008, and Bk (k = 0…3) are the regression coefficients; ei is a provider specific fixed effect, and u1it is an error term.
Pubtimeit, Privtimeit, and Time_differenceit do not vary much over time for the same municipalities, ‘it’-indexes were still used to indicate even a small variation (although the variation is not enough to keep them as independent variables in the fixed-effects model without the interaction effect with post08).