Economic incentives and survival probability for chronic and multimorbidity patients - Do the relationships depend on patient pathways?

The Norwegian Coordination Reform (CR) in 2012 introduced new economic incentives aimed at weaknesses in the way primary /social care and specialist care interacted. This paper studies the association of a new co-payment scheme on 30-day survival probability for chronic and multimorbidity patients. We also analyse whether or not admission types - planned or emergency matters for survival rates. Furthermore, we examine the importance of patient pathways. Several different pathways are possible, depending on where patients came from before being admitted to hospital and their destination after discharge from hospital. The study uses data from three different registers for the period 2010 to 2013. We consider 30 common chronic conditions for which administrative data are available (n=563,096). We look at three mutually exclusive pathways, pathways that are the important ones in terms of the number of patients dependent on co-operation and co-ordination between health care providers. Using a quasi-experimental design—the difference-in-differences approach—we estimate the associations between the co-payment scheme and survival probability by admission type and by patient pathway. For subgroup analysis the reform, behind the increase in the survival rate. after analysing emergency planned admissions separately, that the change in the survival probability significant for emergency admissions only. There is no significant change for planned admissions.


Abstract Background
The Norwegian Coordination Reform (CR) in 2012 introduced new economic incentives aimed at weaknesses in the way primary /social care and specialist care interacted. This paper studies the association of a new co-payment scheme on 30-day survival probability for chronic and multimorbidity patients. We also analyse whether or not admission types -planned or emergencymatters for survival rates. Furthermore, we examine the importance of patient pathways. Several different pathways are possible, depending on where patients came from before being admitted to hospital and their destination after discharge from hospital.

Methods
The study uses data from three different registers for the period 2010 to 2013. We consider 30 common chronic conditions for which administrative data are available (n=563,096). We look at three mutually exclusive pathways, pathways that are the important ones in terms of the number of patients dependent on co-operation and co-ordination between health care providers. Using a quasiexperimental design-the difference-in-differences approach-we estimate the associations between the co-payment scheme and survival probability by admission type and by patient pathway.

Results
We find that the change in survival probability is significant and positively associated with the copayment scheme for emergency admissions, but no significant association is found for planned admissions. A positive and significant relationship is found for emergency patients for two specific pathways-patients coming from home and discharged to social care institutions after hospitalization, and patients coming from home and discharged to other health care institutions after hospitalization.
For planned admissions, the survival probability is significantly and negatively associated for patients coming from home and later discharged to social care institutions. Multimorbidity subgroup analysis shows that the negative association with survival is significant for only planned admissions for the patients coming from home and later discharged to home after hospitalization.

Conclusion
We conclude that the 30-day survival probability is positively associated with the new economic incentives but the result depends on admission type, patient pathway and multimorbidity status.
Without modelling admission type, pathway and multimorbidity explicitly, one may overlook important relationships associated with the economic incentives. Future policy evaluations in any pertinent context should envisage these aspects.

Background
There is a rising prevalence of chronic conditions and multimorbidity worldwide. Patients with chronic conditions and multimorbidity are the main users of health care services. In a global context, chronic diseases are the largest cause of death (Yach et al., 2004;Abegunde et al., 2007). In the US, chronic health conditions are the leading cause of death and disability and represent the largest component of health care costs (Buttorff et al, 2017;Centers for Disease Control and Prevention, 2019). In Europe, the significance of chronic diseases in terms of deaths and health care costs are also well established (Reinhardt, 2010). Recent studies also show that multimorbidity patients account for up to 64% of total health care expenditure (Orueta et al, 2014). Evidently, a substantial part of these costs could be avoided, in particular by reducing inefficiencies in health care organizations and delivery, and by increasing the substitution of expensive care (e.g. specialist hospital care) with less expensive care (primary care and home/social care).
Chronic conditions and multimorbidity affect a diverse range of individuals, and they require different kinds and combinations of health care services. There is recognition in most health care systems that health and social care delivery to these patients is sub-optimal, which has led many health authorities the world over to redesign their care provision. A concern is the lack of co-ordination between providers within the health and social care sectors. In particular, payment systems are often designed in a way that does not incentivize desired co-ordination, and thus creates a barrier for the desired patient flow, which should provide high-quality health and social care in a cost-efficient way.
Ideally, treatment of patients with chronic conditions and multimorbidities should occur through some kind of bundled payment (Quinn, 2015;Stokes et al., 2018). In reality, an efficient form of bundled payment is difficult to implement because it will require expensive data and sophisticated systems of income sharing between units and independent systems, over time. However, as illustrated by Stokes et al. (2018), in practice there are different payment systems that incorporate elements of bundling, for example, income sharing between treatment levels (primary care, social care, specialist care) and a specific time horizon that a particular remuneration covers.
It is of interest to study changes in the quality of care following policy interventions encompassing elements of bundling and co-ordinated payment systems, in particular, for patients with chronic conditions and multimorbidities because they are likely to demand services from different types of providers. For example, elderly people receiving social care services at home or in a nursing home, often delivered by municipalities, may suddenly suffer a worsening of their physical functioning. Then, there is the question of whether further treatment should take place at the same location as previously, or whether the person should be transferred temporarily to another care unit. If this decision has relevance for the final health outcome, the payment system in effect should support decisions that have the best health effect, implying incentives for treatment at the optimal care level.
Thus, our research question is whether payment systems that increase the degree of payment coordination through some degree of bundling affect health outcomes, in particular for patients with chronic conditions and multimorbidities.
Trying to accomplish a reorientation of care, the health authorities in Norway implemented the Coordination Reform (CR) in January 2012 (White Paper, 2008;St. meld. Nr. 47 (2008-2009). The reform aimed to identified weaknesses in primary, social and specialist care interaction highlighted by the large number of patients not treated at 'the most effective level of care'. The White Paper argued that far too many people were admitted to hospital when they could have been taken care of by primary and social care services, and far too many patients were staying in hospital for too long while waiting for follow-ups to the same services. An important component of the reform was a change in the payment system.
Patients with chronic conditions and/or multimorbidity constitute a substantial share of the patients targeted by the CR. The provision of primary and social care services, nursing homes and home care services, in particular, is the responsibility of local governments (i.e. municipalities in Norway), whereas the provision of hospital services is the responsibility of state-owned hospitals. Thus, the responsibility for patients changes as patients move between different service providers. For instance, the responsibility for discharged multimorbidity patients in need of follow-up services, medical and otherwise, is carried over from the state-owned hospitals to local governments. Referrals to specialist care are largely the responsibility of general practitioners (GP) under contract with local governments acting both as gatekeepers and patients' advocates.
In our data, we have identified several different pathways depending on where patients came from before being admitted to hospital and their destination after discharge from hospital. We will look at three mutually exclusive pathways, pathways that are the important ones in terms of the number of patients dependent on co-operation and co-ordination between the health care providers. Although the empirical literature has highlighted the limitations of performance measures based on health outcomes, we still find it relevant to analyse changes in mortality rates following a system reform, such as the CR. The 30-day mortality rate after admission to hospital is among the indicators considered to be effective for assessing the quality/outcome of hospital care (Borzecki et al., 2010).
One advantage of using mortality rates is that they are unequivocal. A higher mortality rate implies, everything else constant, poorer quality of care (Cashin et al., 2014;Milstein and Schreyoegg, 2016) [1] . We hypothesized that the new incentives may influence the composition of patients allocated to the specific pathways, which in turn influences survival probability.
Our paper contributes to the literature in several ways. For a better understanding of how changes in financial incentives influence survival rates for patients with chronic conditions, we use rich registry data with a battery of contextual (i.e. municipality-level primary and social care attributes) and patient-level control variables. Using a quasi-experimental design-the difference-in-differences (DiD) approach-we estimate several models to test the robustness of the results. In particular, patient pathways are modelled explicitly and to our knowledge, this aspect is missing from the earlier literature. Our subgroup analysis adds an additional dimension to examine any differential effects by multimodality status for alternative in-patient admissions as well as pathways.
Our DiD results show that the survival probability for chronic patients positively and significantly associated following the introduction of the CR. However, analysing emergency and planned admissions separately, we find that the change in survival probability is significant and positive for emergency admissions, but not significant for planned admissions. Our pathway-specific analyses indicate a positive and significant association with survival probability for emergency patients for two

The Coordination Reform and economic incentives
To facilitate better follow-up of chronic and multimorbidity patients specifically but not exclusively, the CR reform introduced new economic incentives consequential for both specialist care and social care. Rather than changing ownership structures and imposing more vertical or horizontal integration of service provision, the local authorities should be incentivized to internalize some of the costs associated with both hospitalization and 'bed blocking'. The CR included the following economic incentives designed by the Government to affect the degree of co-ordination between health care providers and patient flows between them: i. Implementation of a co-payment scheme in which a municipality internalize part of the costs of hospitalization for patients living in the municipality.
ii. Penalize municipalities if patients in need of social care follow-up and with a 'ready for discharge' status are hospitalized beyond the discharge date [ 2 ] .
iii. Transfers (or subsidies) to municipalities establishing 24/7 emergency bed capacity (EBC) within their primary/social care facilities. EBC should be available by 1st January 2016 at the latest.
The co-payment scheme was introduced as payment per admitted episode, equal to 20 per cent of the national average cost for the specific diagnoses-related groups (DRGs). However only medical diagnoses, not surgical diagnoses, were covered. To incentivize municipalities to lower the demand for hospital services, the co-payment scheme [3] reduced allocations to the state-owned hospital sector by 20 per cent and re-allocated the financial resources to municipalities. Municipalities received their share based on a per capita remuneration and dependent on pre-reform consumption of the relevant medical DRGs.
Critics of the CR scheme argued that people with chronic conditions, comorbidities and/or elderly patients-patient groups most likely to be affected if municipalities adapted to the incentive by providing more services locally-faced the prospect of receiving lower quality services. One of the main concerns expressed by those opposing the reform or those suspicious of it was how older patients and those with complex needs (multimorbidity) would be affected. Some of those opposing the reform argued that the chronic and multimorbidity patient group would receive lower quality of care than in the pre-reform situation. Hospitals have incentives to shorten the LOS, and municipalities have incentives to avoid hospitalization relying instead on the medical resources available locally.

Previous studies
Several aspects of the CR have been evaluated previously. Askildsen et al. (2016) concluded that the use of specialist somatic health care services has not changed since the introduction of the municipal co-payment system in 2012. Melberg and Hagen (2016) found that the LOS in hospital is somewhat reduced because of the new penalty scheme for patients waiting to be discharged. Ambugo and Hagen (2019) found little evidence of the new penalty scheme has on the municipal level death rates and readmissions for COPD/asthma, heart failure, hip fracture, and stroke. Islam and Kjerstad (2019) found negative associations between EBC and changes in emergency admissions to hospitals.
Whether the quality of health services has changed, however, is an issue that has been sparsely addressed in the literature.
In a recent study, Bruvik et al. (2017) analysed the changes in survival following the CR with a small sample of patients living at a 35-bed short-term ward in a nursing home. They find that the number of patients that died in the nursing home after hospitalization doubled (27% versus 13%) following the CR while procedures and staffing were unchanged between pre-and post-CR. Regarding patients' composition, particularly the age distribution, they find that during pre-CR, admitted patients were older (median 88 years, range 77-103 years) than post-CR (median 85 years, range 77-99 years). Bruvik et al. (2017) concluded that shorter hospitalization periods for older patients led to an increase in the number of patients transferred to nursing homes. Referring to Gautun and Syse (2013) and Abelsen et al. (2014), they further concluded, 'The transfer of responsibility for treatment from hospitals to the municipalities entails that the patients being transferred suffer from more serious, complex and treatment-intensive conditions, when compared to the situation before the introduction of the co-ordination reform.' Obviously, the change in other aspects of the composition of patients may have contributed to the higher death numbers reported by Bruvik et al. (2017).
Meacock et al. (2017) made a similar argument in their study of mortality rates among emergency patients admitted to hospitals on weekends. They found fewer deaths but higher mortality rates than for those admitted on weekdays arguing that patients admitted on weekends are a selected group of patients. Because of the reduced availability of primary care services and higher threshold for admissions to accident and emergency facilities during weekends, fewer but sicker patients are admitted. Kahlon et al. (2015) studied the association between frailty and 30-day mortality outcomes after discharge from hospital. They concluded that frailty at the time of admission is associated with a substantially increased risk of early readmission and death after discharge from medical wards. Monkerud and Tjerbo (2016) is also relevant in our context as they study the effects of the municipal co-financing regime. Although Monkerud and Tjerbo (2016) are not concerned with changes in survival rates, their study is pertinent because they show that municipalities seem to react to changes in incentive structures. Rather than expanding their own rehabilitation capacity, municipalities seem to take advantage of the fact that private institutions charge less than the prices they are facing under the co-payment regime. They conclude that municipalities shift away from specialist rehabilitation services towards the use of rehabilitation services in private institutions. The goal as emphasized by policymakers was to incentivize a shift towards more municipal rehabilitation and less use of specialist rehabilitation.
The rest of the paper is organized as follows. Section 2 describes the data and variables, while Section 3 describes the estimation strategy used in the analyses. Section 4 provides the descriptive and analytical results for the full sample, by emergency and planned admission, by alternative pathways and for the multimorbidity subsample, and examines the robustness of the results. Section 5 concludes the paper.
[1] Mortality is perceived negatively, therefore, the Norwegian Knowledge Centre for the Health Services (NOKC) suggests using 30-day survival probabilities as routinely reported quality indicators for Norwegian hospitals (see e.g. Lindman et al., 2014).
[2] Municipalities are obliged to reimburse hospitals NOK 4,000 per day after the discharge date, which in principle is determined unilaterally by the hospital at which the patient was admitted. The penalty scheme does not make a distinction between medical and surgical diagnoses.
[3] The co-payment regime was abolished in early 2015, the main arguments being that the scheme did not work as envisaged and that it placed too much risk on the municipalities, many of them small ones.

Data and variables
Our patient data come from the Norwegian Patient Registry (NPR), from which we extracted information on different in-patient hospital admissions (emergency or planned admissions), patient age, gender, diagnoses, admission and discharge dates, death date, patient pathways, DRG codes etc. The primary NPR dataset contains only the main diagnosis of patients admitted to hospital (available until the year 2014 at our disposal). We obtained another dataset from the NPR with not only the main diagnosis but also the secondary diagnoses involving important chronic conditions. This information is useful for defining the multimorbidity status (not comorbidity) of the patients. One limitation of the data set is that the data with secondary diagnoses is presently available only through 2013. However, the effects of the CR on survival are likely to be stronger for chronic illness with complex needs than for the majority of in-patient admissions [4]. Thus, the study uses subset of NPR data (and two other registry data sets) for the period 2010 to 2013 (two years pre-reform and two years post-reform).
During our study period, there were roughly 3.18 million in-patient admissions with length of hospital stay (LOS) greater than or equal to one day, and where 67 per cent of admissions were emergencies and the rest were planned. This study considers 30 important chronic and long-term conditions for which administrative data are available, and these conditions are also used in the previous literature Finally, to determine whether discharged patients are in need of social care services, we used information describing the patients' pathway after discharge from hospital, and place of residence before and after hospitalization. Theses variable were collected from the NPR data.
Norwegian municipalities are responsible for providing primary health care and social care, therefore controlling for municipality relevant attributes (time-varying) describing these services may be pertinent. These attributes are obtained from Statistics Norway's KOSTRA dataset. In particular, to control for the magnitude and quality of primary care services, we include the number of patients per GP and percentage of the GPs with open lists (as an indicator of effective supply of primary care in municipalities) (Islam & Kjerstad, 2017). Municipality level per capita net operating expenses for nursing and care services also are controlled. Proxy variables for social care service capacity in municipalities are the number of home care receivers relative to the number of inhabitants aged 80 years and over ('home care') and the number of institutional care receivers relative to the number of citizens aged 80 years and over ('institution care'). We further include a municipality-level attribute describing the 'quality of services' offered at social care institutions, such as physician hours per week available for the residents in nursing homes (Phy_Inst). A priori, we believe a higher number of physician hours in nursing homes indicates higher quality of services associated with higher survival probability because the care of discharged patients is better facilitated than with municipalities with lower physician capacity. We also control for the share of inhabitants living alone at home, as an indicator of formal care needs. Table 1 describes the variables used in the analyses.

All chronic and multimorbidity patients
We examine whether economic incentives provided under the CR in Norway associated with the 30day survival probability for chronic and multimorbidity patients. In particular, we explore the associations of introducing a municipal co-payment of 20 per cent of the national DRG price, paid by the home municipality of the patients when they are admitted to hospital. Most notably, the scheme was introduced for medical DRGs only, not for surgical diagnoses. We use this difference as an identifying restriction by assuming that the expected change in 30-day survival probability for a patient admitted for surgical DRGs (control group) is the same as it would have been for the medical DRGs (treatment group) in the absence of the 20 per cent co-payment (the reform). The quasi-natural experiment design in this context is one where we observe outcomes (30-day survival probability) for two groups for the pre-and post-reform periods. The treatment group is exposed to a treatment in the second period (after the reform), but not in the first period (before the reform), whereas the control group is not exposed to the treatment during either period. Using a quasi-experimental framework, we estimate a standard DiD equation, which is expressed as: where is the 30-day survival probability after in-patient hospital admission of a patient-episode i, We performed the analysis at the individual patient's spell-level using a linear regression model (i.e. linear probability model, because our dependent variable is dichotomous). To control for observable patient differences, comprises important patient-level observable characteristics that could influence survival probability. In particular, patients' socio-economic attributes and patient-level attributes reveal patients' needs and severity, as mentioned in Data and variables. To control for observable contextual differences given where a patient lives, we also control for their time-variant municipal level observable characteristics, (see Data and variables). In Equation (1), the vector x includes patient-level attributes and vector m includes municipality-level characteristics. The estimated interaction term, the DiD estimate, measures the effect of CR reform.
Municipalities make decisions for their inhabitants that may not be observable in our data, but that may lead to changes in patients' health and social care use, and in turn their survival probability (i.e. the source of potential unobservable confounding factors). Given the heterogeneity amongst municipalities in the coverage of different health and social care services, in Equation (1) we include municipality unobservable fixed effects ν j , to capture municipality differences that are constant over time. To capture survival differences over time that are common to all municipalities we also include yearly fixed effects, μ t · ε ijt is the idiosyncratic error term.
There might be circumstances when the co-payment affects different types of hospital admission episodes differently because of the selection of patients. For example, one may deduce that the financial incentives may not affect emergency admissions and in turn the survival probability but may influence planned admissions and the associated survival probability. To capture this sort of heterogeneity in the patient groups and to understand the mechanisms, we estimate separate regressions for emergency and planned admission episodes.

Multimorbidity sub-analysis
As mentioned above, it is important to study whether there is any differential relationship of the economic incentives for multimorbidity patients. To achieve this, we add additional interaction terms to examine any differential associations of multimorbidity status. Equation (1) can be rewritten as: The estimated coefficient for the triple interaction term (difference-in-differences-in-differences (DiDiD)) in Equation 2), measures the differential relationships of the reform on multimorbidity patients compared with non-multimorbidity patients. We also estimate Equation (2) for alternative pathways and alternate admission types.

Parallel trends assumption
A key assumption of the DiD analysis is that of parallel trends. This states in general terms that in the absence of intervention/reform, the average outcomes of the treatment group and the comparison groups are the same in the pre-intervention period (Angrist & Pischke, 2009). This assumption has been tested graphically and statistically. To test the assumption statistically, we define the pre-reform period as 2010 (placebo pre-reform) and the post-reform period as 2011 (placebo post-reform) as a way 'artificially' to scrutinize the parallel trends assumption. This is tested by evaluating the significance of the coefficients and in regression models (1) and (2), respectively. Statistical insignificance of the coefficients implies that the parallel trends assumption is valid (Dimick & Ryan, 2014).

Robustness checks
To examine the robustness of our results, we performed a variety of analyses. The size of the Norwegian municipalities is diverse. There are many small municipalities in Norway with few inhabitants. The CR reform outcome may be different for smaller municipalities (or create noise in the estimation) than for larger ones. In the first set of robustness checks, we re-estimate (using Equations 1 and 2) our baseline DiD and DiDiD models (for all admissions; emergency and planned admissions separately) by dropping small municipalities (i.e. municipalities with total number of inhabitants less than 10,000). Additional DiDiD models are estimated by using an alternative definition of multimorbidity status of the patients, namely, if the number of chronic comorbidities is greater than or equal to 3.
In the second set of robustness checks, we include all in-patient admissions (chronic and/or nonchronic) during our study period 2010-2013. First, using Equation (1) for this large sample, we estimate our baseline DiD models for all admissions, and emergency and planned admissions separately (because of a lack of information on multimorbidity status we cannot estimate the DiDiD models for this sample). Second, using the same sample and dropping small municipalities, we reestimate the models (for all admissions, and for emergency and planned admissions separately) by dropping small municipalities (total number of inhabitants less than 10,000).
[4] Moreover, assessing the impact on the 30-day survival rates, using data on all in-patients' categories and all diagnoses in a regression specification may produce 'noise' because not all patients or diagnoses are equally important in terms of the CR.
[5] Patients often appear multiple times within the in-patient records since these are predominantly recorded at the episode level. Sometimes we use the terms "admission" and "episode" interchangeably.
[6] To check robustness of our results, we alternatively constructed multimorbidity status (multi=1) if the number of chronic comorbidities greater or equal to three.
[7] DRGs provide a flat per-discharge payment that varies based on diagnoses, severity, and procedures were performed. A standard year-specific monetary conversion factor, used to convert DRG weights into base payment rates for each DRG. Table 2 describes the composition of patients for the medical DRGs and surgical DRGs, in both the pre-and post-reform periods. The survival rates (surv_30) are comparable across the pre-and postreform periods for all admissions, emergency admissions and planned admissions. As expected, the survival rates are lower for emergency admissions than for planned admissions for all groups of patients. We observe a slight decline in the share of admissions for people multimorbidity patients (multi) between the pre-and post-reform period. The age and gender compositions are rather stable too. However, post-reform, those aged 67-79 years (age79) constitute a larger share of both emergency and planned admissions than in the pre-reform period. The share of youngest patients is higher in the medical group than in the surgical group. The composition of patients in the elderly age groups are comparable across these two groups. Regarding education level, the differences between groups and over periods are small. We have argued that the pathways that patients take between service providers might affect survival rates. Most of the municipality attributes are found be remain relatively constant between the pre-and postreform periods. However, it is evident that the Net_Exp and the Phys_Inst attributes increased after the reform for both the intervention and comparison groups (Table 2) and within the pathways (Table   3).

Main results and results by admission type for all chronic patients
As discussed in the descriptive section of the paper, there are some noticeable differences between the pathways with regard to the age composition of patients and the share of patients taking the different pathways pre-and post-reform. Thus, a more detailed analysis is required. The estimated main DiD results are presented in Tables 4. Table 4 shows positive and significant the CR associations with survival rates for all admission episodes (0.35%) and emergency episodes (

Alternative pathways: overall results and results by admission type
The results from the analyses of the three most important pathways are reported in likely to be rather healthier compared with patients discharged elsewhere. There is also a negative age effect because younger patients have higher survival rates than older ones.
Patients following the H o -H-S c pathway experienced a significant association with their survival rate on average, but there is a substantial difference between emergency and planned admissions. The survival probability increases on average by 1.75 per cent for all admissions, with emergency admitted patients experiencing a significant improvement in their survival rate (1.62 per cent), however, there was a 7.14 per cent decrease in the survival rate of planned admissions. Table 3 shows the descriptive statistics that more patients followed this pathway after reform and that the age compositions changed towards larger shares of older cohorts. The LOS also decreased quite substantially. Although not reported in Table 2, in our data we found that the average LOS decreased from 9.4 days to 7.4 days for emergency admissions but increased for planned admissions from 7.2 days to 9.1 days for patients following the H o -H-S c pathway. The reduced survival rates for planned admissions indicate that the chronic and multimorbidity patients are older and frailer than in the prereform period, while emergency patients are younger and stronger leading to an increase in their survival rates for this pathway. The CR is likely to be the cause of this shift in the composition of patients.
The survival rate for emergency admitted patients following the H o -H-O h pathway also positively associated, while there is no significant association for planned admissions. Finally, LOS seems to matter significantly for emergency admissions, but not for planned ones.

Multimorbidity subgroup: overall results and results by admission type and pathways
Using Equation (2) for multimorbidity patients, the summary results (analogous to those in Table 5) are reported in the first row of Table 6 (i.e.). No significant association is observed for all admissions and alternative admission types-emergency admissions or planned admissions, for these multimorbidity patients.

Testing the parallel trends assumption
The statistical tests for placebo reform effects (test of the parallel trends assumption) for chronic patients (DiD estimates) and for the multimorbidity subgroup results (DiDiD estimates) are presented in Table 7. Our graphical (Fig. 1)  Notice that for all other placebo DiD and DiDiD estimates (reported in Table 7), we do not find any positive and significant relationships with any of the alternative admissions and pathways. Intuitively, this result suggests that the pre-reform trend in the survival rate was either 'negative' or 'no trend', but was not 'positive'. Table 8 presents the DiD and DiDiD estimates for chronic and multimorbidity subgroup patients with small municipalities omitted. Compared with the DiD coefficients reported in Table 5, we find positive and significant associations with CR (and larger) for all admissions (0.50%) and emergency admission (0.87%) and no association for planned admissions. Analogously, for this restricted sample, as before, null effect is observed for the H o -H-H o pathway; positive and significant associations are observed for all admissions for the two other pathways (see Table 5) [8].

Robustness checks
For this restricted sample, our DiDiD estimates for multimorbidity subgroup patients also produce similar results to those reported in Table 6. Moreover, an alternative definition of multimorbidity status of the patients, that is, if the number of chronic comorbidities is greater than or equal to 3, also yield similar results.
Using Equation (1) H-S c pathway (not reported here but results can be available upon request). As a further robustness check, we repeat the same analyses for all in-patient admissions, but excluding small municipalities, and obtain similar results as those when including all municipalities.
[8] Moreover, we have also estimated the models excluding municipality-level time-varying characteristics, and find similar results (in respect of the magnitude of coefficients and their directions) as reported in Tables 4 & 5. In particular, positive and significant CR effects for all admissions, emergency admission, and null effect for planned admissions.
[9] For this large sample, in our data we do not have information about the patients' multimorbidity status.

Discussion
In Norway, the provision of primary and social care services is the responsibility of the municipalities, Our difference-in-differences main results show that the survival probability for patients with chronic conditions significantly and positively associated with the CR. The overall findings of a positive association are stable across the majority of robustness checks. This is a promising result in terms of the longer-term effects of the reform, although at a glance our analysis cannot provide specific mechanisms behind the increase in the survival rate. However, after analysing emergency and planned admissions separately, we find that the change in the survival probability is significant and positive for emergency admissions only. There is no significant change for planned admissions.
Patient pathway-specific analyses offer some diverse results. We find no significant difference for patients that are admitted to hospital from home and return home (H o -H-H o ), the largest group of patients. Interestingly, we do find positive and significant associations with survival for emergency patients for two specific pathways; patients coming from home and discharged to either a social care

institution or to another health care institution after hospitalization (H o -H-S c and H o -H-O h ). The former
result potentially contradicts the results reported in Bruvik et al. (2017), as we noted in the introduction. They find, based on a case study, that the number of patients who died in the nursing home after hospitalization doubled following the CR, while procedures and staffing were unchanged between pre-and post-CR. They report, as we do, that the composition of patients changes, particularly the age distribution: patients were older pre-reform than post-reform. Bruvik et al. (2017) concluded that shorter hospitalization periods leading to an increase in the number of patients transferred to nursing homes must be a result of a path shifting mechanism: patients that stay longer in hospital would be in better health when discharged thus avoiding the need to be admitted to social care institutions. As far as we know, Bruvik et al. (2017) do not distinguish between patients admitted to hospital from home or from a social care institution, which would potentially be an indicator of health status upon admission. Their argument is that patients are frailer when discharged sooner, thus, the survival rates for the patients will drop. We find that for planned admissions of patients coming from home and subsequently discharged to social care institutions (a small number of patients), the survival probability is reduced significantly.
Multimorbidity subgroup analysis shows that this negative survival relationship is also consistent across all pathways, and most evidently for the patients following the H o -H-O h pathway. However, based on our analysis we cannot support the notion that the CR led to a reduction in survival rates more generally, at least not for all chronic patients.
It could be the case that the co-payment system might induce selection of patients, i.e. for a given level of effort and quality a municipality could be induced to hospitalize the relatively more demanding patients while taking care of the less demanding patients. Thus, one could argued that the co-payment system might lead to a situation in which hospitals would have to take care of the comparatively more sick or frail patients. The fact that the average LOS is reduced quite substantially may at first contradict the selection hypothesis, i.e. if patients were more ill, they would be expected to stay longer in hospital. However, the penalty scheme may also come into play. The penalty scheme in which hospitals can unilaterally decide the discharge date, at least in principle, may have induced hospitals to bring forward discharge dates compared with the pre-reform period for several interlinked reasons. First, the understanding that some patient groups did stay too long in hospital in the prereform period from a medical perspective and now the municipalities have an economic incentive to discharge patients sooner. Overall, our results indicate that the CR has influenced municipalities and hospitals to change both the number of patients following the different pathways and the composition of patients. It is likely that the new economic incentives have played a role.
The study has a number of weaknesses. A concern regarding the estimation is the parallel trends assumption that the DiD approach relies on. Although we controlled for the patient and municipal level attributes, and municipality fixed-effects, our main weakness is that we have only limited data points both pre-and post-CR to validate whether this assumption is well satisfied. Our statistical evidence is not totally convincing for comparable pre-intervention trends for all alternative admissions and pathways, except for the H o -H-S c pathway. However, our placebo reform estimates (Table 7) suggests that the pre-reform trend in the survival rate was either 'negative' or 'no trend', but was not 'positive'.
An issue we briefly mentioned in the introduction is that the CR introduced several policy instruments simultaneously. Although municipalities could have implemented EBC 'at will' as long as they had met the deadline, the interplay between the co-payment scheme and the use of discharge fees in relation to delay from hospitals are not clear-cut. Nevertheless, Ambugo and Hagen (2019) found little evidence of the discharge fees in relation to delay instrument has an effect on the municipal level death rates and readmissions for some specific diagnosis. Moreover, we have tried to disentangle one of the links by analysing EBC and non-EBC municipalities separately. We do not find any noteworthy differences in our main results reported here for patients from municipalities with EBC and those without.

Conclusions
Three concluding remarks can be made. First, on average the majority of chronic patients either positively associated or experience status quo in terms of survival probability following the introduction of the CR. However, the relationship is not evident for all multimorbidity patients.
Second, the survival rates for patients with chronic conditions and multimorbidity patients depend both on their admission type and on the specific pathway they follow. In general, the positive survival relationship with the CR is consistent mainly for the Home-Hospital-Social care (H o -H-S c ) patient pathway. Moreover, the CR seems to have altered the relative importance of the specific pathways.
Therefore, we accomplish that without modelling pathways explicitly, one may overlook important changes following the CR. Future policy evaluation studies in any pertinent context should consider these aspects.
Finally, emergency admissions represent the largest group of patients and this group has increased survival rates for two specific pathways. However, the survival probability for a smaller group of patients, planned admissions, did decrease.
Policy recommendations must be made with some reservations. There is no consensus in the economic literature concerning the efficacy, applicability and optimal implementation of alternative incentive schemes under different institutional arrangements. Obviously, thorough theoretical and empirical studies are required to make a valid assessment of any new incentive schemes. By estimating the cost and benefits of the CR across pathways and alternate admissions, a comprehensive welfare analysis in the future would be beneficial.

Authors' contributions
MKI and JEA conceived this study. MKI and EK designed the analysis plan. MKI conducted the analysis.
MKI and EK wrote the first draft of the paper, and all authors revised the paper and approved the final version.

Funding
The SELFIE (Sustainable intEgrated care modeLs for multi-morbidity: delivery, FInancing and performancE) project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 634288. The content of this work reflects only the SELFIE groups' views and the European Commission is not liable for any use that may be made of the information contained herein. The funding body had no role in the design of the study; in the collection, analysis, and interpretation of data; and in the writing of the manuscript.

Availability of data and materials
The data used in this study are available from The Norwegian Directorate of Health (www.helsedirektoratet.no/) and Statistics Norway (www.ssb.no), but restrictions apply to the availability of these data, which were used under licence for the current study, and so are not publicly available. However, data are available from the authors upon reasonable request and with included permission from The Norwegian Directorate of Health, Statistics Norway and Norwegian Data Protection Authority.

Ethics approval and consent to participate
Not applicable since all data is anonymous.
Cluster standard errors at municipality level are in parentheses. The numbers of observations are in brackets.
Cluster standard errors at municipality level are in parentheses. The numbers of observations are in brackets.
'**' and '***' represent significance at the 5% and 1% levels, respectively. All models control for the covariates included in Table 4 combined with year and municipality fixed effects.
Cluster standard errors at municipality level are in parentheses. The numbers of observations are in brackets.
'**' and '***' represent significance at the 5% and 1% levels, respectively. All models control for the covariates included in Table 4 combined with year and municipality fixed effects.
Cluster standard errors at municipality level are in parentheses. The numbers of observations are in brackets.
'**' and '***' represent significance at the 5% and 1% levels, respectively. Figure 1 Survival probability of surgical and medical diagnostics, and by emergency and planned admissions for 30 chronic conditions