Ageing Problems and Adaptive Performance: The Effect of Ageing on Hospital Performance in the Netherlands

(1)

Ageing Problems and Adaptive Performance

The Effect of Ageing on Hospital Performance in the Netherlands

Isaya Wullings s2690292

Master Thesis Public Administration, Leiden University Environmental Dynamism and Organizational Performance

Supervisor: Petra van den Bekerom Second reader: Valéry Pattyn

June 9, 2020 Word count: 11598

(2)

Abstract

The present study seeks to analyze the effect of ageing on hospital performance, measured as the performance in the Intensive Care Unit. The hypothesis is that an increase in the amount of elderly negatively affects hospital performance. The hypothesis is tested on a dataset by the National

Intensive Care Evaluation foundation (n = 70), using a multi-level longitudinal analysis on the years 2013-2018. The yearly percentage change in the amount of elderlies is calculated and shows a significant effect for the mortality rate and the readmission to the ICU. Moreover, a three-level analysis shows that there is more variance within hospitals than between the different hospitals, confirming that the percentage of elderlies is clustered in hospitals and COROP-regions, rather than an individual explanatory variable.

(3)

Since the beginning of the COVID-19 pandemic, it became clear that older people are at a higher risk of getting infected (National Intensive Care Evaluation [NICE], 2020). Not only were they a bigger risk group, they were also more likely to suffer severe consequences and were more likely to face admission to the Intensive Care Unit (ICU). However, for older people, the chances of surviving an ICU admission are small and they are often to fragile to survive the oxygen admission in the ICU (Buurman et al., 2020). However, this is not only the case of COVID-19 patients, but for many other diseases as well, which poses a new challenge for the acute care departments.

One of the reasons for the increasing pressure on the acute care department is the ageing of the Dutch population (Nederlandse Zorgautoriteit [NZA], 2018). Two simultaneous effects can be distinguished: not only do people become older nowadays, due to modern medicine and better insights in healthy lifestyles, but also a relatively big share of the population is 65 years of older, as a consequence of the baby boom after the Second World War. Not only do older people require more care, they also visit the doctor more often, and they tend to visit the Emergency Room more often, 1.4% more in 2017 compared to the year before. Thus, the Dutch population has a double ageing development, that can lead to increasing pressures on the hospital care (Post et al., 2018).

Further, a shortage on the labor market has led to shortage in the number of doctors, nurses and other hospital workers. This can be explained by the increasing demand for care, combined with little inflow of workers, too much outflow and more absenteeism, due to many extra hours, more intensive workweeks, and more patients compared to the number of doctors and workers. This puts an enormous pressure on people working in the hospital section and therefore requires a lot of dedication to the job. For some workers, this workload is too heavy, which leads to a bigger outflow from the health sector. The biggest shortages are found in the acute care departments and as a consequence, the Emergency Room is overcrowded, ambulances do not have a place to bring new patients and people need to find a place in the ICU abroad (Nederlandse Vereniging van

(5)

Previously, research has mostly been done from the field of Nursing Studies, focusing mainly on job satisfaction, stress, burnout, drop out, and other related factors. A lot of research has been focusing on the pressure on nursing and the occupational stress (Farquharson et al., 2013; Piko, 2006; Vahey et al., 2004). However, all of these are based on conducted surveys, in which nurses are self-reporting their experiences. Even though these factors can influence the performance of nurses and thus the hospital performance, they are not the only measure of quality in the hospital context.

It is important to also provide more insight in more objective measures of hospital performance, such as the length of treatment patients receive, the mortality rate of hospitals, the number of readmissions and so forth. Once there is more knowledge on the double ageing problem and its effects on hospital care, hospitals can prepare their staff and create better services to adapt towards a more efficient system with less occupational pressure for nurses and doctors.

Moreover, the context in which hospitals are operating is always changing, which makes it difficult for organizations to predict future developments. The turbulence of their environment is putting more challenges to their performance, which can lead to negative outcomes. Previously, research on environmental turbulence and organizational performance has shown that these are negatively associated in the context of school systems in the United States or local governments in the United Kingdom (Andrews et al., 2013; Boyne & Meier, 2009; Meier & O’Toole, 2009; Van den Bekerom et al., 2016).

The present study tries to provide a little more insight in hospital performance as a consequence of environmental turbulence, focusing on the objective measures, rather than the subjective measures from previous research. The focus will be on Dutch hospitals from 2013-2018 and the double ageing problem that the Netherlands in facing in that period. The environmental turbulence of ageing will be used to measure the effect on intensive care performance to provide a better knowledge of what to expect and if this problem is something hospitals should anticipate for.

(6)

The central question in this research is thus: “What is the effect of the double ageing problem in 2013-2018 on hospital performance in the Netherlands?”.

The National Intensive Care Evaluation (NICE) foundation has a dataset in which they obtain the objective measures that will be used in this research to try and provide more insight in the central question. Before doing so, the research context will be shortly discussed. Then, I will elaborate on previous research, the central concepts, and the hypothesis. In the 4th_{chapter the}

datasets and the methods will be discussed, all before analyzing the hypothesis and discussing its results.

(7)

(8)

2. Research Context

In this chapter the research context of hospital will be addressed. I will look at the Dutch Healthcare System in section 2.1 from a public management perspective.

2.1 The Dutch Healthcare System

In 2013 and 2019, the Netherlands counts respectively 92 and 79 hospital organizations, both including 8 university medical centers (UMC’s) (Volksgezondheidenzorg, 2018a). All of these are responsible for over 1.5 million hospitalizations (887,8 per 10.000 inhabitants) (Centraal Bureau voor Statistiek [CBS], 2020a).

From the perspectiveof public management, hospital performance has not been thoroughly researched before, meaning that there is very little research on hospital performance as a public event, but rather on hospital performance measured as the individual performance of nurses and doctors. However, a public management perspective is interesting, since the Dutch healthcare system is publicly organized by the Dutch government and hospitals thus operate as public organizations, even though they are private care providers, which makes their performance interesting to other public organizations. The Dutch healthcare system rests on five basic acts: the Health Insurance Act, the Long-term Care Act, the Social Support Act, the Public Health Act and the Youth Act. These form the foundation of the health care system (Ministry of Health, Welfare and Sport [MHWS], 2018).

The Health Insurance Act was enacted in 2006, in which everyone is entitled to have a basic health insurance. The provision of care is the task of private, competitive health insurance

companies and health care providers. All residents of the Netherlands are responsible for choosing their own insurer, and insurers are required to accept all individuals, and provide equal premiums. Further, they can supplement their basic insurance. It can thus be seen as a triangle between providers, insurers and individuals (MHWS, 2018).

(9)

The central government is responsible for the healthcare system and the quality

requirements that the system must guarantee (MHWS, 2018). Finally, to guarantee the quality of care, hospitals will receive a set amount of money for each completed treatment, based on the demand for care. This is meant to incentive hospitals and health insurers to create better quality of care against better prices. In short, hospitals will receive a block grant funding that is not subject to competition or changes, a share of money open to competition or changes in for example the performance funding, and a share of money that is regulated with maximum prices (MHWS, 2011).

(10)

3. Theoretical Framework

In this chapter I will look at the concepts of organizational environment and environmental dynamics in section 3.1. I will then apply these consecutively to hospital performance in section 3.2 and 3.3. Finally, in section 3.4, I will look at the conceptualization as well as previous research on these topics.

3.1 The Organizational Environment and Environmental Dynamics

An organization’s environment can be conceptualized as “the total set of interstructured activities in which it is engaged at any one time and over which it has discretion to initiate,

maintain, or end behaviors” (Pfeffer & Salancik 2003, p. 32). In the context of Dutch hospitals, this means that the environment not only consists out of hospital workers and patients, but also all other organizations, institutions and events it is related to and can possibly influence the performance of the hospital. A hospital’s environment can thus provide the hospital with resources but can also impose constraints on its organizational behavior (Pfeffer & Salancik, 2003).

Another definition is given by Daft (2010, p. 220): “all elements that exist outside the boundary of the organization and have the potential to affect all or part of the organization”. Here, a link to dynamism is made, since all activities both outside an organization are inherently dynamic and can lead to turbulence within the organization that has not been anticipated for (Aldrich, 2008). With dynamism, changes in the organization’s environment are meant (Van den Bekerom, et al., 2018). Changes in the environment can be shocks or can be more of a background ‘noise’. A shock, or a crisis, is something that happens more abruptly, seen as “Any sort of disruption emanating from outside the administrative system and buffeting the core organization”, and is more difficult to prepare for (Meier & O’Toole, 2009, p. 487). Turbulence, or background ‘noise’, is not just a dynamic environment, but is a change of which the extent is unpredictable: “In sum, turbulence is unpredictable change in the munificence and complexity of an organization’s environment.” (Boyne & Meier, 2009, p. 803)

(11)

Such dynamics be different in frequency, size and turbulence and can thus be a challenge depending on the scope of the change (Aldrich, 2008; Emery & Trist, 1965). The first two refer to how often elements in the environment happen and the scope of the change in a given period, for example if the Dutch government would implement quality standards every other month, this will more likely affect hospitals then if one law is implemented each decade. Further, the size of the change is important, because environments are constantly changing. However, a big change is expected to have a bigger effect on organizations, since bigger changes in the environment require bigger changes in the organization. However, according to Boyne and Meier (2009) these elements are important, but if they are known, organizations or managers can anticipate these changes. Therefore, the unpredictability of changes and the interconnectedness is important to add, because they create the biggest challenges to cope with (Aldrich, 2008; Boyne & Meier, 2009).

A distinction should be made between the internal environment, e.g. the organization itself, and the external environment, e.g. the activities an organization is involved in, but is outside the boundaries of an organization (Aldrich, 2008; Daft, 2010). Here, environmental dynamism will refer to such external changes, because they are produced by obscure forces and therefore are difficult to predict or plan for (Aldrich, 2008, 69).

A second distinction should be made between positive and negative shocks (Meier & O’Toole, 2009). Positive shocks, such as an unexpected change in the financial situation of an organization, can influence its core as well, but are expected to have either a positive influence on the organizations, since they can provide organizations with more or better resources, or can affect an organization negatively, but the disruptive effects are often only in the short run. Negative shocks are more likely to affect the organization’s stability negatively in the long run and can be potentially dangerous (Meier & O’Toole, 2009; O’Toole & Meier, 2011).

A third distinction should be made between types of environmental changes. There are six dimensions in which an organization’s environment can be divided, the so-called PESTEL-framework (Johnson & Scholes, 2002, pp. 65-69): “political (e.g., the role of governments),

(12)

economic (e.g. differential growth rates), social (e.g., changing demographics), technical (e.g., technological innovations such as the internet), environmental (e.g., green issues such as pollution), and legal (e.g. health and safety legislation).”. These dimensions are interconnected and vary in terms of dynamism.

According to Meier and O’Toole (2009), absence of fluctuations in an organization’s

environment could show a subtle signal of managerial success and achievement. So, hypothetically, absence of change means that an organization has a stable environment and that it can develop routines to deal with environmental elements (Aldrich, 2008).

The general hypothesis is that environmental shocks affect the organization’s performance negatively. The expectation is that an environmental shock will affect the core of the organization and will disrupt the performance of managers and other people working in an organization since they have to divide their time and attention to the occurring problem and can therefore devote less to the organizations performance as they would otherwise do.

Previous research on organizational performance has provided evidence for this hypothesis. Boyne and Meier (2009) have confirmed this hypothesis by researching how the performance of public organizations is affected by unpredictable changes in their external environment. They have used panel data from 8 years in Texas’ school districts to analyze the performance of the school districts.

For the independent variable, they have included five individual indicators to measure the turbulence of the munificence. They also included measures for the complexity of the environment. To measure organizational stability, they use two measures: vertical stability, referring to the constancy in overall hierarchical roles and horizontal stability, referring to the specialization and differentiation of the process. The dependent variables are the percentage of students that pass the state standardized exam and the percentage of students that have been named ‘college ready’ by the state (Boyne & Meier, 2009).

(13)

They have found that organizations in a turbulent environment are experiencing more difficulties to perform well but can be mitigated if organizations have a structural stability. However, these benefits are also dependent on the specialization of a school (Boyne & Meier, 2009).

More specifically, the research by Meier and O’Toole (2009) focusses on economic shocks, showing that budget shocks and pupil enrollment negatively affect student scores on standardized tests in school districts in the state of Texas, United States. They examine the relation between budget shocks and pupil enrollment on performance by looking at the impact of managers and their management style.

The independent variable that they use are budget cuts, to measure economic shocks. The dependent variable, indicating performance, is measured across ten different indicators, among which the overall pass rate of the Texas Assessment of Academic Skills (TAAS). They find that if public organizations experience significant budget shocks, they will not experience much

performance decline in general on the short-term, because mangers appear to be successful at generating desired outcomes, while absorbing the event. However, if the organization experiences repeated budget cuts, the performance will affect performance negatively (Meier & O’Toole, 2009).

In the context of the Dutch primary education system, the moderating effects of internally and externally oriented managerial networking activities on environmental turbulence and

organizational performance has been examined (Van den Bekerom et al., 2016). They expected that a percentage change in the number of pupils will affect school performance negatively and that the intensity of networking styles attenuates the negative effect of pupil percentage change on school performance, regardless of the networking style.

The dependent variable they use to measure school performance is the Cito test score, taken in the final semester of the last year of primary education. To measure environmental turbulence, the independent variable, they use the percentage change in the number of pupils. The networking

(14)

style is measured by a survey of the school principals on the interaction with each of the external organizations and actors (Van den Bekerom et al., 2016).

They have found that environmental shocks do negatively affect organizational

performance, as was expected based on the research by Meier and O’Toole (2009). They also find evidence for the attenuating effects of school principals’ downward-oriented and sideward-oriented networking on the negative effect of external shocks on organizational performance. The internally oriented networking activities seem to neutralize the negative effects and there is no evidence for a reinforcing effect of school principals’ upward-oriented networking or outward-oriented networking on the expected cause-effect relation. In this research, they find that the negative impact of

turbulence in the Dutch school environment is moderated by internally oriented networking activities (Van den Bekerom et al., 2016).

Research by Andrews and colleagues (2013) focusses on another external change, namely a population increase due to migration. To study the effect of economic immigration on the local government capacity and performance, they look at six local government services focusing on children and young people, adult social care, environment, housing, libraries and leisure, and benefits.

They have used Comprehensive Performance Assessments by the Audit Commission to measure the dependent variable of service performance and a standardized service expenditure to measure the dependent variable of money value. To measure the independent variables they have used the cumulative annual allocation of worker migration numbers to measure the economic migration, the expenditure on central administration per capital to measure the stock of

administrative capacity, and the number of community, social, and personal services organizations per 1,000 capital registering for value added tax to measure the capacity of the community. They find that the unexpected arrival of large numbers of economic migrants has a significant negative effect on performance, but this can be limited by the capacity of the administration and the community (Andrews et al., 2013).

(15)

3.2 The Organizational Environment and Hospital Performance

The organizational environment of Dutch hospitals can be mapped out by looking the six categories of the PESTEL framework by Johnson and Scholes (2002). First, there is the political context, meaning the role of the government. This is important, because hospitals in the Netherlands have to comply with the quality standards, rules and regulations set by the Government: every citizen needs to have access to the necessary medical care of good quality and it should be

affordable for the entire population (Van den Berg et al., 2014a). The government can for example change its taxation policy or its social welfare policies, which may lead to more or less funding for hospitals (Johnson & Scholes, 2002).

Second, government funding is also part of a hospital’s economic environment, since the changes in funding influence the money supply of hospitals. Finally, hospitals are also subject to the regulated competition between insurance companies, which has consequences for the market where insurance companies ‘buy care’. Here, at the zorginkoopmarket, insurance companies and all care providers including hospitals, negotiate care policies, which decides the quality, volume and price of care (Van den Berg et al., 2014a). This means that economic changes can lead to different care policies, which can influence the quality, volume and price that people or their insurance company pay for hospital care (Johnson & Scholes, 2002).

The latter relates to the third category, namely the legal environment of a hospital. Changes in the legal environment of a hospital refer to competition law, employment law, health and safety and product safety. This is where changes in the law on quality and safety of care, including the products used and the care provision, will affect the organization (Johnson & Scholes, 2002).

Fourth, the social category shows an important part of the environment, namely the changing demographics: population growth and ageing being the most important pressures on the quality of care (NVZ, 2018). For example, more people in a country leads to a bigger demand for care, which creates increasing pressures for hospitals, especially if bigger parts of society are at a higher risk of needing care (Johnson & Scholes, 2002).

(16)

Fifth, changes in the technological environment of a hospital can lead to new discoveries, changing the way people work, their living standards and their lifestyles (Johnson & Scholes, 2002). For example, the use of internet and virtual communication appears to have a negative impact on physical and mental health of people, which can lead to more hospital visits (Farhud, 2015).

Sixth, environmental changes can be summarized as environmental protection laws, waste disposal and energy consumption. According to research on the hospital waste management in Pakistan, hospital waste needs to be segregated to avoid disease spreading. For hospitals unaware of this quality measurement, trainings need to be provided, as well as purchasing sanitary products. Further, it can also mean that more time should be invested in waste disposal, meaning that they need to invest more money in waste disposal, but also attract and train new people for this hospital branch (Anwar et al., 2013).

So, in a hospital setting, negative shocks can involve political turbulence, budget cuts, changes in the law, technological changes, environmental developments or changes in number of patients seeking care (Van den Berg et al., 2014a; NVZ, 2018). However, changes often are the combined effect of the separate factors. In this research, I will look at the social environment of hospitals, since this category captures major changes from all categories: income distribution, lifestyle changes, consumerism, levels of education, mobility and demographics (Johnson &

Scholes, 2002). Since people are becoming older nowadays due to new insights in healthy lifestyles, but also due to developments in modern medicine and high quality of care, it is interesting to

examine the effect of changes in the amount of elderlies on hospital performance (Post et al., 2018). In this research, I will thus look at changes in the amount of elderlies seeking acute care in the Emergency Room, and eventually the ICU, on hospital performances in those departments.

3.3 Environmental Dynamics and Hospital Performance

Previously, research has shown that stressors at work can lead to negative individual experiences for nurses, that have led to diminished quality of care (Cleland, 1965; Deckard et al.,

(17)

1988). Cleland (1965) researched the performance of nurses under conditions of stress. For the independent variable, she used the need for social approval and the situational stressors, meaning that there is an interaction between the subjective motivation of the job and the field situation. For de dependent variable, she used the scores of the Nursing Achievement Test and the Social

Interaction Test. She found that increased environmental turbulence can lead to deteriorated performance. However, if the nurse has a lower need for social approval, they seem to perform better under a high situational stressor level than the other need-for-approval groups, so the subjective stress measure seems to have a moderate effect on nurse performance (Cleland, 1965, 298).

Deckard and colleagues (1988) examined the effect of work experience and organizational environment may influence nursing productivity. They have used the Multiple Affect Adjective Check List (MAACL) to measure sensitivity to the short-term stress-generating or stress-alleviating conditions and therapy. To measure the work stress or negative psychological reactions to work factors, the Index of Organizational Reactions (IOR) is used. For the dependent variable, the Maslach Burnout Inventory (MBI) is used to measure three dimensions of burnout, e.g. nursing productivity. Their research shows that positive environmental influences can lead to a better performance and productivity of nurses.

The study by Salyer (1995), shows that the number of admissions relatively to the measures of discharges from a unit in the period of 24 hours, negatively affects the performance of nurses. However, the performance measure is a survey based on their own reports.

In the Netherlands, the government National Institute for Health and Environment offers research publications focusing on the overall quality, accessibility and costs (Van den Berg et al., 2014a). In 2014, they stated that the health care is of high quality and that there are a lot of favorable trends in the long-term caretaking, such as drops in the number of cases of malnutrition and pressure ulcers. This assumes that there is a positive relationship between environmental trends and performances. However, they also show that there is a bigger need for care among elderlies and

(18)

that high pressures decrease the invested time per patient (Van den Berg et al. 2014a). High pressure on nurses has been previously linked to negative outcomes for both staff ant patients (Farquharson et al., 2013). Farquharson and colleagues used a real-time repeated measures design with nurses from a general teaching hospital in Scotland to make a better theoretical understanding of occupational stress for nurses. This is important to assess, because the demand for nursing is increasing with the ageing of the population and thus puts an increasing pressure on their services and the quality of care. The expectation is that an increase in the amount of elderlies negatively affects hospital performance.

A cross-sectional survey of nurses and patients in the United States, showed that nurses who experienced working in their units as something positive, with adequate staff, support and good relationships between doctors and nurses, also satisfied the patients of those units more and showed a lower burnout number (Vahey et al., 2004).

Another, yet similar conclusion has been drawn in a research on Hungarian health care staff by Piko (2006), in which she showed that burnout is strongly related to job dissatisfaction, which lead to the conclusion that the work environment is an important factor for performance.

In general, it can be said that people become older and the older they get, the more often they appeal to care, which can increase pressure for hospitals. Not only do older people more often need to consult a doctor, use medicines and need other types of help, they are also more often in need of acute care in the Emergency Room (Post et al., 2018). In 2017, 47% of the people older than 65 years used the Emergency Care once or more and 30% of visits the Emergency room on a yearly basis (Vektis, 2019). However, there are big regional differences, but in general older people are more vulnerable and are more likely to fall. Yet, falling incidents are not putting an increased pressure on the hospital performance, but an increasing amount of elderlies also increases the chance of hospital stays and follow-up caretaking (NZA, 2018). This will lead to an increase in the demand of care.

(19)

Furthermore, an increase in 65+ers also leads to more pressure on ambulance workers, since older people are more likely to use an ambulance due to a lack of transport for example (NZA, 2018). The Dutch Care Authority (NZA) monitored that the acute care is sufficiently accessible, but that the pressure on this part of the care increases, due to increase in demand of care (NZA, 2018). Therefore, hospital performance in this article is measured in terms of acute care, using extensive data on the ICU.

Hypothesis: a change in the amount of elderlies negatively affects hospital performance in the Intensive Care Unit.

3.4 Central Concepts

3.4.1 Acute Care

Acute care can be defined as care that is urgent, also called emergency care (NZA, 2018). It is a situation in which someone needs to be treated as quickly as possible to avoid death or severe consequences (Raad voor de Volksgezondheid en Zorg [RVZ], 2003). As quickly as possible means that care should be provided within several minutes or hours, depending on the severity of the situation.

Emergency care is defined as a broader term, namely care as a response to something the patient considers an emergency (Nederlands Huisartsen Genootschap, 2013). This does not mean that the situation in itself needs to be life threatening, but it is seen from the perspective of the patient. This can also be called ‘acute care demand’, when a patient or bystander asks for immediate care as a consequence of a potential severe or possibly life-threatening situation (RVZ, 2003).

Two perspectives need thus to be distinguished, namely the perspective of the patient and the perspective of the care provider. However, often the patient or the care demander forms the central perspective. Every care that cannot wait until the first work-day possibility to see a doctor is considered acute care. This does not always mean that the patient is directly send to the hospital, it

(20)

can also be a phone call offering comfort (Van den Berg et al., 2014a). In this research, acute care will be any situation that requires immediate care in the hospital.

3.4.2 Hospital performance

Defining hospital performance can be done along the requirements set by the National Institute for Public Health and the Environment (Van den Berg et al., 2014a). They define hospital performance based on quality, accessibility and affordability. These three measures generally coincide with other performance measurement systems across OECD countries (Beazley et al., 2019).

In this research, I will focus primarily on the quality component of hospitals, using a general definition by the Institute of Medicine (IOM) (2001, pp. 39-40): “doing the right thing, at the right time, in the right way, for the right person, and having the best possible results”. This definition has also been used by the National Institute for Public Health and the Environment, because it covers a number of concepts that are considered important in the literature on health care quality, namely effectiveness, safety, timeliness and responsiveness (Arah, 2005). To the Dutch government, effectiveness, safety and patient-centeredness appear most important to measure the quality of hospital performance (Van den Berg et al., 2014a).

3.4.3 Environmental turbulence and Acute Care

In this research, environmental turbulence refers to the percentage change in the amount of elderlies. Since people over 65 years demand the most care and due to a double ageing development also becomes a bigger group, this will put a big pressure on the Dutch Healthcare System, since older people are more at risk to have health issues and to experience severe consequences of those (NZA, 2018; Post, et al., 2018). In general, the Dutch Healthcare System is of high quality and develops mostly positively. Nonetheless, in the elderly care several less positive developments are distinguished: there is little time invested in their caretaking, due to a shortage in available

(21)

caretakers (Van den Berg et al., 2014a). This shortage is not only visible in the Netherlands, but also internationally, due to increased environmental pressures (Adams & Bond, 2003).

4. Data and Methodology

In this chapter, I will elaborate on the methodology that is used in this thesis to analyze the effect of a change in the amount of elderlies on the hospital performance in the ICU. First, I will discuss the research design. Second, in section 4.2 the data collection method will be discussed as well as the datasets that I will use. Third, the central concepts will be operationalized, providing details on how they will be measured. And fourth, I will elaborate on the validity and reliability of the research.

4.1 Research Design

In this research a multilevel regression will be done to test the hypothesis. In order to do so, a longitudinal design is used, meaning that the data on all variables is obtained for multiple time periods (Menard, 2008). This allows for a measurement of changes in quality over time, because it covers a broader time period and is therefore more suitable to explain development rather than a measurement in a single time period. In this research, demographic transitions are used as explanatory variable and therefore a longitudinal design will be more suitable to show this demographic development.

The design will cover almost the entire hospital population from 2013 until 2018 in the Netherlands. Even though some hospitals might have merged or do not exist any longer, there is a substantial overlap of hospitals across time. In total there are six moments of measurement, with a total of 420 observations, measured by the NICE foundation. All hospitals with an ICU are included in their data collection, but in this research, I only use the hospitals that have agreed to publish the numbers on the NICE website, coming to a total amount of 70 hospitals.

(22)

In 2013, there were 84 general hospitals and 8 UMCs in the Netherlands (NZA, 2017). In 2018, there were 71 general hospitals and 8 UMCs in the Netherlands (Volksgezondheidenzorg, 2018a). In 2013 and 2018, there were respectively 89 and 78 ICUs in the Netherlands (NICE, 2013-2018a).

4.2 Data collection

To test my hypothesis, the panel data on ICUs of al hospitals in the Netherlands by the NICE foundation is used. On a yearly basis, all hospitals provide detailed information through an online input module, in which the hospitals can add their numbers. The participating hospitals can use different input modules, all answering the same quantitative questions by adding the numbers into the database. All hospitals answer in absolute numbers, that the NICE foundation later uses to provide percentages or calculate durations of oxygen admission, duration of treatment, or the number of readmissions (NICE, 2013-2018a).

All hospitals are required to at least provide core variables if they want to be included in the data collection of the NICE foundation, such as the number of patients, the reason of the visit, the amount of deaths and the duration of treatment (NICE, 1996-2020). Further, they possess a wide range of other variables that I will use as control variables in this research, such as size measures and data on the type of patients. Some hospitals do not have a contract for specific variables, such as the duration of oxygen, resulting in a lower number of observations. Some hospitals only started uploading their data from a later period then 2013, resulting in some missing years as well. Each hospital has a unique identification number, assigned by me. After a hospital merge, a new number has been created.

Second, the Dutch population register has been used. To measure a change in the hospital environment, a yearly percentage change in the amount of elderlies (65 years and older) in a COROP-region between 2013-2018 will be used to see if the share of elderlies increased in this period of time. A COROP-region is term for regional classifications, created in 1970, in which each region follows the commuters flow (CBS, 2020c). This means that each COROP-region exists out

(23)

of multiple municipalities that are part of the same catchment area, and thus are likely to use the same hospital. The Netherlands knows 40 COROP-regions and all hospitals have been sorted to know how many elderlies approximately live in the area that is most likely to make use of this hospital (CBS, 2020b). Therefore, since each year is nested in hospitals and hospitals are nested in COROP-regions, a three-level regression will be done to control for hospitals operating in the same COROP-region. Each COROP region therefore has an identification number to run this regression, assigned by me.

Finally, for the remainder of the variables, such as the location of hospitals (Randstad dummy and COROP-region) and the type of hospital (whether academic or not), I used all websites of the hospitals and collected the data myself.

4.3 Measures

Hospital performance. The dependent variable in this research is the hospital performance over the report years 2013 until 2018. It is measured by a hospital performance in the ICU, since they are meant to treat any emergencies, whether caused by an accident or by acute illnesses such as heart failure (NZA, 2018). As previously mentioned, the functioning of an Acute Care Department is representative towards the hospital performance in unexpected situations (Aldrich, 2008; Van den Bekerom et al., 2018). This is because even though the Acute Care Department is prepared for unexpected inflow of patients, they deal with turbulence all the time and will influence the performance later in the caretaking process (Zorginstituut Nederland, 2020).

Often, hospital performance is measured through patient surveys (Beazley et al., 2016; Van den Berg et al., 2014b). However, in this research I will focus on five objective indicators to measure the quality of the ICU, and I will run five separate regressions. I have included all quality measures from the NICE foundation. The quality indicators that I have used are (NICE, 2013-2018b): first, the Standard Mortality Ratio, the total observed deaths divided by the total expected deaths in a hospital. Second, the median duration of treatment in days, calculated by subtracting the date and time of admission from the date and time of discharge from the hospital. This indicator has

(24)

been divided into two groups: people that died in the ICU and people that have recovered

sufficiently to get out of the ICU. The fourth indicator, is the median duration of oxygen in days, subtracting the time and date of oxygen admission from the time and date when oxygen admission has been stopped. And finally, the fifth is readmission, any ICU admission in the same period as the previous admission, calculated as the sum of the number of readmissions in a population divided by the sum of expected readmissions.

I believe these measures draw an overview of the performance in the ICU, since they cover important quality measures of the performance in general. However, there might be other quality indicators that are not excluded in this research, and therefore this small selection of quality measurements might influence the quality of the results.

Further, since almost all hospitals in the Netherlands have been included, all academic hospitals and the majority of the general hospitals, I believe the data is representative towards the entire hospital population. Therefore, I believe this variable is externally valid. However, it is unsure whether it is representative towards other countries, since health systems greatly differ, including their quality standards.

Environmental turbulence. To measure the environmental turbulence of hospital

environments, I use the data from the population register from the Centraal Bureau voor Statistiek (CBS) in 2012 until 2018 and used the absolute number of people of 65 years and older each year and its previous year to calculate the yearly percentage change in the number of people of 65 years and older, both measured on the first of January. So, for the quality measures in 2013, the

percentage change of elderlies between the first of January in 2012 and 2013 is calculated.

According to the Zorgthermometer Ouderenzorg, 1 in 5 people is 65 years or older, meaning that a normal amount of elderlies in a region should be around 20% of the entire population and the expectation is the share of 65+ers will be 25% of the entire population by 2030 (Post et al., 2018). The increase in people over 65 years is also visible in the dataset (see appendix 9.1).

(25)

However, not all COROP-regions have been included, since some regions do not have a hospital and it is unclear what other hospital they would use or because the hospital in a specific region is one of the hospitals that is not included in the dataset. In total, there are seven regions not included, which means that the internal validity of the research might be affected, because this can change the percentage change of the explanatory variable for some hospitals.

Controls. To make sure that no important variables are omitted, I control for a number of variables that could influence the cause-and-effect relationship or have an effect on the variables used in the dataset. I control for the size of the hospital in terms of the amount of beds, nurses and intensivists in the ICU, because larger ICUs are likely to cope with patient changes. Further, I control for hospitals operating in the same COROP-region, since they are likely to cope better with patient changes here, since they can share the pressure of an increasing amount of ageing.

I control for several institutional characteristics of the hospital, such as its location, since hospitals in the Randstad area, the urban area of the Netherlands, are more populated. However, the big cities of the Netherlands often attract more younger people (Volksgezondheidenzorg, 2018b). Another characteristic that is controlled for are Academic hospitals, because they have a great number of students that cooperate in the hospital, they are more likely to adapt to changes since they engage in many researches on complex problems (Nederlandse Federatie van Universitair Medische Centra, 2020).

The last control variable controls for the amount of people in high risk groups in the ICUs, because in case of high-risk people, the chances of passing away during the stay in the hospital are bigger. It is also more likely that older people belong to those high-risk groups, which makes it interesting to control for.

Even though there always is a chance of omitted variable bias, I believe that the important variables that are likely to influence the cause-and-effect relationship have been included. Further, the percentage change in the amount of elderlies have been preceded the performance data, which also contributes to increase the internal validity. This research is reliable, since the used datasets are

(26)

publicly accessible, the method is transparent, and the results should thus be reproducible in future research.

Table 1 provides summaries of the descriptive statistics. The descriptive panel data shows that there are 70 hospitals included in the dataset and they all have six moments of observation, adding up to a total of 420 observations.

Table 1

Descriptive Statistics for all Variables in the Analyses.

Variable N. Obs Mean Standard Deviation Range

Year 420 (overall) 70 (between) 6 (within) 2015.5 1.71 (overall) 0 (between) 1.71 (within) 2013 – 2018 (overall) Hospitals 420 (overall) 70 (between) 6 (within) 43.23 25.51 (overall) 25.66 (between) 0 (within) 1 – 85 (overall) 1. Mortality 388 .75 .18 .29 – 1.47

2. Duration of Treatment of People that

died in days. 388 2.17 .63 .8 – 4.2

3. Duration of Treatment of People that

survived in days. 388 1.22 .39 .3 – 2.9 4. Duration of Oxygen admission in days. 275 1.18 .67 .2 – 4.1

5. Readmission to the hospital in total. 388 .95 .28 .1 – 1.8

6. Difference in share of elderlies (65+)

per year. 420 2.47 .84 .95 – 5.92

7. Number of ICU beds 348 15.58 11.56 3 – 84

8. Number of Intensivists 344 6.86 4.96 1 – 42

9. Number of Nurses on the ICU 346 52.51 44.21 1 – 375

10. Risk of dying on the ICU <30% 387 784.05 630.55 70 – 3037

11. Risk of dying on the ICU 30%-70% 387 115.45 76.23 13 – 386

12. Risk of dying on the ICU >70% 387 54.69 45.55 1 – 217

13. Urbanization level (dummy) 420 .46 .50 0-1

14. Academic Hospital (dummy) 420 .11 .32 0-1

Corop (three level analysis) 384 23.16 10.04 (overall) 10.11 (between) 0 (within)

3 – 39 (overall)

All numbers have been rounded up to two decimals.

However, not all hospitals were part of the dataset in each individual year, meaning that there are some missing values for each variable. In table 1, I have removed the missing values per variable and have included the number of observations per variable to make sure the scope of the descriptive table is clear. Further, there is variation between hospitals and there is variation over time in the panel data. The range of hospitals (1-85) is a little bigger than 70, because some

(27)

numbers have been deleted since the hospitals were not part of the dataset. Further, there are 40 COROP-regions in the Netherlands, and seven of those regions are not included in the dataset at all. However, some of the remaining 33 regions only have a hospital that they share with another region, and therefore those regions are marked as missing values, to avoid any irregularities in the dataset due to skewed results.

Table 2 provides the correlations between the variables. A pairwise correlation test is done to see provide an overview of the Pearson’s r correlation between variables. Pearson’s r measures the strength of the linear relationship between two variables (Lane, 2003). The values of the variables will be between -1 and 1, in which any coefficient closer to -1 or 1 indicates a more perfect relationship between variables, and any coefficient closer to 0 means that there is no linear relationship between the variables. As we can see, many variables are significantly correlated, but many only have a weak or moderate relationship. However, size measures (number of beds, number of intensivists and the number of nurses) are strongly correlated with each other, as well as with the risk groups, meaning that they have a strong positive linear correlation.

(28)

Table 2

Correlations for all Variables in the Analyses.

Variable 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1. Mortality 1.000

2. Duration of Treatment of People that died in days.

-.104** 1.000

3. Duration of Treatment of People that survived in days.

-.082 .278*** 1.000 4. Duration of Oxygen admission in days. .049 .060 .400*** 1.000 5. Readmission to the hospital in total. .138*** -.030 -.132*** -.161*** 1.000 6. Difference in share of elderlies (65+) per year. -.153*** -.000 -.029 .037 -.020 1.000 7. Number of ICU-beds .072 .218*** -.047 -.358*** .335*** -.071 1.000 8. Number of Intensivists .120** .216*** -.121** -.298*** .359*** -.111** .886*** 1.000 9. Number of Nurses on the ICU .044 .205*** -.096* -.352*** .252*** -.038 .878*** .876*** 1.000 10. Risk of dying on the

ICU <30% .025 .169*** -.218*** -.475*** .300*** -.042 .776*** .747*** .789*** 1.000 11. Risk of dying on the

ICU 30%-70% .135*** .207*** .028 -.345*** .332*** -.050 .792*** .785*** .750*** .795*** 1.000 12. Risk of dying on the

ICU >70% .060 .206*** -.012 -.324*** .281*** -.000 .734*** .701*** .699*** .724*** .879*** 1.000 13. Urbanization level (dummy) .051 .089* -.131*** .074 .211*** -.294 -.000 .020 -.011 -.008 -.031 .031 1.000 14. Academic Hospital (dummy) .084* .218*** -.051 -.411*** .2401*** -.061 .562*** .578*** .516 .422*** .621*** .580*** .121** 1.000

(29)

(30)

5. Results

In this chapter, I will discuss the results of the regressions by analyzing the statistics. I will start with a basic longitudinal regression adding the control variables (section 5.1). A longitudinal analysis is used to explain the effect of the difference in the share of elderlies of 65 years and older on hospital performance. In the following sections I will discuss the relationship of a yearly percentage change in the amount of elderlies on five separate dependent variables. After analyzing these five regressions, I will do a separate three-level analysis to control for hospitals operating in the same COROP-region in section 5.2.

5.1 Effect of a Percentage Change in the Amount of Elderlies: a Longitudinal analysis

To measure hospital performance, five dependent variables are used. Table 2 presents the five results of a multilevel analysis testing whether a percentage change in the number of elderlies negatively affects hospital performance. Each dependent variable has its own model, respectively: the mortality rate, the median duration of treatment in days for people that died in the ICU, the median duration of treatment in days for people that have recovered

sufficiently to get out of the ICU, the median duration of oxygen in days, and the readmission to the ICU.

In table 3 all control variables are added for each of the dependent variables. Each regression is done twice, one to measure for random effects of the variables and one to measure for fixed effects of the variables. In panel analysis, fixed effects omit time-invariant measures, e.g. variables that do not change over time, in this case both the Randstad dummy and the UMC dummy, because they are a constant characteristic of the hospital and therefore they are correlated with the dependent variable. A random effect regression assumes that the time invariant measures, so the hospital’s specific effect is uncorrelated with the independent variables, so in this regression, all variables will be included (Zhang, 2015).

(31)

To decide what regression should be included in the analysis, I have run a Hausman test, which identifies the presence of endogeneity in the explanatory variables. If endogeneity is present, it means that the explanatory variable, which is de percentage change in the yearly amount of elderly, is correlated with the error term, simply put that there is a chance of omitted variable bias. The Hausman test thus identifies this and then shows whether the fixed model or the random model is more useful (STATA, 2013). Before running the Hausman test, it is important to understand its hypotheses: the null hypothesis is that there is no correlation between the variables in the panel data model, so the random effects model can be used. And hypothesis one is that there is a correlation between the time-invariant variables and the dependent variable, so the fixed effects model should be used. The Hausman test provides a probability measure, which is also presented in table 3.

Other measures included in table 3 are two variance levels, σe is the variance at the

year level, explaining the differences between years within one hospital and σu is the variance

at the hospital level, explaining the differences between hospitals.

Finally, to measure for systematic changes in the spread of the residuals, meaning that the variance between variables is not constant, vce robustness checks are added to the

regressions to control for such heteroscedasticity.

The first model is the mortality rate model. The Hausman test shows that there is a significant effect (.0007 < a p-value of .05), so the null hypothesis is rejected and thus the fixed effect regression model should be used. As shown in table 3, model 1, there is a significant effect of the percentage change in the number of elderlies of -.056, meaning that for each percentage change in the amount of older people, the mortality rate decreases with .056. This means that the rate calculated as the amount of observed deaths divided by the expected deaths in a hospital increases.

(32)

Table 3

Multilevel regression analysis: including all controls. (longitudinal analysis)

Model 1 Model 2 Model 3 Model 4 Model 5 Year level (level 1)

Percent change in number

of elderlies per year -.056 (.012)*** [.010]*** .057 (.040) [.046] -.020 (.017) [.022] .057 (.037) [.043] .042 (.016)*** [.021]**

Hospital level (level 2)

Size: number of beds Size: number of intensivists

Size: number of nurses Risk: low risk groups Risk: mid risk groups Risk: high risk groups Location: Randstad (dummy) Type: UMC (dummy) -.001 (.002) [.002] .017 (.006)*** [.005]*** -.002 (.001)*** [.001]*** -.000 (.000)*** [.000]*** .001 (.000)** [.000]*** -.001 (.001)* [.000]** omitted omitted .003 (.007) [.004] -.004 (.019) [.012] .001 (.002) [.001] -.000 (.000) [.000] .000 (.001) [.001] .001 (.002) [.001] .148 (.115) [.111] .206 (.226) [.168] .005 (.003) [.004] -.015 (.009)* [.009]* .000 (.001) [.001] -.000 (.000) [.000] -.000 (.001) [.001] .001 (.001) [.001] omitted omitted .007 (.008) [.009] .012 (.028) [.026] -.006 (.002)*** [.002]*** .000 (.000) [.000] -.004 (.002)** [.002]** .001 (.002) [.002] omitted omitted .006 (.003)** [.002] ** .011 (.008) [.007] -.002 (.001)** [.001]*** .000 (.000) [.000] .001 (.001) [.001] -.000 (.001) [.001] .128 (.045)*** [.047]*** .041 (.089) [.091] Constant .837 (.078)*** [.059]*** 1.805 (.169)*** [.191]*** 1.300 (.114)*** [.123]*** 1.448 (2.64)*** [.278]*** .640 (.068)*** [.083]*** Hausman test prob>chi

at 5% significance level .0007 .7097 .0099 .0000 .0827 Variance level 1 σe .134 .490 .195 .361 .197 Variance level 2 σu .132 .402 .332 .568 .154 N level 1 342 342 342 256 342 N level 2 70 70 70 53 70 R2_within R2_between R2_overall .204 .013 .065 .012 .127 .054 .027 .136 .092 .097 .064 .078 .027 .339 .200 *p<0.10 **p<0.05 ***p<0.01. (two-tailed). Standard Error between brackets (…) and Robustness checks between […]. All

numbers have been rounded up to three decimals, except for the Hausman test.

In other words, two possibilities could be that the hospitals might have expected more deaths, as I expected based on the literature and the hypothesis, but the amount of observed deaths did not increase as much. Or second, they expected the same amount of deaths, but the observed amount actually decreased, which is an opposite effect of the hypothesis, since less people died, which could indicate that the hospitals perform well, or that they have less severe cases in the ICU.

(33)

However, both of these scenario’s show that the turbulence did not have its expected working, since the mortality rate decreased or that hospitals have already adapted their policies and methods towards expected ageing problems. Further, we can see significance effects of almost all variables, except for the number of beds, however, these coefficients are very small, so they rarely have any effect on the mortality rate. The robustness checks have been added to control for heteroscedasticity, but as shown, the standard errors slightly

decrease and the mid risk and high-risk groups become a little more significant, however, the effect remains small.

The overall model has an explanatory power of .065, which means that the effect of the number of elderlies including all control variables explains the mortality rate for 6.5%. Yet, this number might not adequately show the explanatory power of the regression but might be due to the fact that older people have a higher mortality rate than younger people, so the rate automatically changes once there are more older people. As previously mentioned, the variance levels of both the year level and the hospital level show that the proportion of variance at the hospital level can be calculated as p= .132 / (.132 + .134) ≈ .496, which means that 49.6% is explained by variance in mortality rate between hospitals, and the other 50.4% can be ascribed to differences between one hospital over time.

The second model is the duration of treatment model for people that died in the ICU. The Hausman test shows that there is no significant effect (.7097 > a p-value of .05), so the null hypothesis is accepted, and thus the random effect regression model should be used. However, there is no significant effect, so there is nothing this model can explain in terms of ageing and hospital performance.

The third model is the duration of treatment model for people that came out of the ICU, using the fixed effects model again (.0099 < a p-value of .05), but only the number of

(34)

intensivists has a significant effect at the 10% level, meaning that it is difficult to predict an effect using this model.

The percent change in the number of elderlies per year has no significance in the fourth fixed regression model (.0000 < a p-value of .05), and therefore, nothing can be

explained using the duration of oxygen admission as a dependent variable. We further see that this is the only model with a smaller N, since not all hospitals have a contract for providing this data.

Finally, in the fifth model, the percentage change in the number of elderlies has a highly significant effect on the readmission in a hospital. As we can see from the random regression model (.0827 > a p-value of .05), the overall explanatory power of this model is 20%, even if we look at the explanatory power between hospitals, it explains almost 34% of the number of readmissions, which is the biggest of all models. The variance levels show that as p= .154 / (.154 + .197) ≈ .439, which means that almost 44% can be explained by variance in readmission between hospitals, and 56% can be explained by variance within hospitals over time. This means that the variance is more likely to be subject to changes in time, than

changes in hospitals, so the explanatory power of time variance is bigger than those of

hospitals. As we expected, the amount of elderly increases, so the variance in time proves that this effect is stronger than differences between hospitals.

The independent variable here is positively and significantly associated with hospital’s standardized readmission percentage, meaning that each percentage increase in the number of people over 65 years, the readmission percentage increases with .042. Thus, hospitals have a higher readmission once the population ages, as we expected from the literature. The risk of being in the ICU increases with age, and thus, the expectation is that there is more pressure in the ICU (Post et al., 2018; Vektis, 2019). Two scenarios come to mind when seeing this result: firstly, there are more people at the ICU and thus less time can be invested in each

(35)

patient, leading to faster discharges from the ICU. Secondly, it can be possible that the patient’s situation becomes worse outside of the ICU and needs to be readmitted quickly. Both of these scenarios are related to the ageing of the population. However, some hospitals also have a medium care department, which can skew the results, since this is not taking into account in the readmission ratio, and some hospitals therefore have a lower readmission percentage (NICE, 2013-2018b).

A final note about the fifth model is that the Randstad dummy has a significant and positive effect of .128, meaning that hospitals within the Randstad are more likely to have a bigger readmission ratio. Interestingly enough, the amount of 65+ers increases less over time than in rural areas, so it is an opposite effect from what was expected in the hypothesis. This can possibly be due to a bigger population density or a higher occupational pressure.

In general, this analysis proves it difficult to explain the effect of the amount of elderlies on ICU performance. Two out of five regressions prove a significant effect,

however, many variables are significantly associated with the dependent variables, but have very little effect.

The hypothesis that a percentage change in the amount of elderlies negatively affects hospital performance can therefore not be accepted on the basis of these regressions.

However, it is also difficult to reject the hypothesis based on these results, since there are two significant effects of the percentage change in the amount of older people on the mortality rate and the readmission percentage. On the basis of the five longitudinal regressions, the effect of an increase in the amount of elderlies cannot be used as a significant measure on hospital performance, but still provides two substantial insights.

5.2 Effect of a Percentage Change in the Amount of Elderlies: a Three-Level Analysis

In this section the final control variable has been added. Since many hospitals operate in the same COROP-region as others, the expectation is that they possibly experience less

(36)

turbulence, since they can share an increase in the amount of elderlies among themselves. To avoid any skewed results, all hospitals that operate in multiple COROP-regions have been marked as a missing value, coming to a total of 64 hospitals, operating in 29 COROP-regions and 313 observations in four models, and a total of 48 hospitals, operating in 25 COROP-regions. In table 4, the results of the analysis are shown. We can see that the range of

hospitals in a COROP-region varies between one and six hospitals. Another measure shown in table 4 are the variance estimates at the COROP-region level and the hospital level, as well as the variance of the residuals. The significance levels are not present in the STATA output, so I have calculated their z-score and then found their p-value. However, these cannot tell us much, since this part of the output does not take into account that the sampling distributions are positively skewed, meaning that they have a long right tail, rather than being normally distributed (Leckie, 2013). It makes therefore more sense to use the Likelihood ratio (LR) test, which is also presented in the table, because they do not rely on normal distributions of the sample.

The first model shows a negative significant effect between the effect of elderlies per year of -.039, which is a smaller coefficient than the coefficient in table 3 (-.056). This could mean that there if we control for COROP-regions, the effect of turbulence on the mortality rate becomes smaller, and thus we can assume that it matters in which COROP-region a hospital operates. However, the effect is very small, so it is difficult to predict the effect of the independent variable in this model. We further see other significant effects, but they hardly show any effect.

Model 2, 3 and 4, show no significant effects on the independent variable, and therefore do not differ much from the previous table with only the panel results.

(37)

Table 4

Multilevel regression analysis: including COROP-region. (three level analysis)

Model 1 Model 2 Model 3 Model 4 Model 5 Variables

Percent change in number of elderlies per year Size: number of beds Size: number of intensivists Size: number of nurses Risk: low risk groups Risk: mid risk groups Risk: high risk groups Location: Randstad (dummy) Type: UMC (dummy) Constant -.039 (.011)*** -.001 (.011) .014 (.005)*** -.001 (.001)** -.000 (.000)** .001 (.000)*** -.001 (.000)** -.005 (.033) -.024 (.063) .794 (.049)*** .032 (.041) .003 (.007) -.005 (.019) .001 (.002) -.000 (.000) .001 (.001) .001 (.002) .109 (.120) .187 (.214) 1.879 (.173)*** -.015(.017) .005 (.003)* -.018 (.008)** .001 (.001) -.000 (.000)*** .001 (.001) .001 (.001) -.072 (.092) -.080 (.141) 1.301 (.096)*** .056 (.036) .002 (.007) .021 (.021) -.004 (.002)** -.000 (.000) -.001 (.001) .000 (.002) .119 (.157) -.472 (.253)* 1.260 (.196)*** .031 (.017)* .006 (.003)** .010 (.008) -.002 (.001)** .000 (.000) .001 (.001) -.000 (.001) .108 (.048)** .045 (.091) .681 (.072)***

Hospital level (level 2)

Variance .012 (.004)*** .128 (.048)*** .076 (.023) .183 (.062)*** .025 (.007)*** COROP-region (level 3) Variance .000 (.003) .009 (.038) .014 (.023) .021 (.051) .000 (.003) Variance residual .017 (.002)*** .231 (.021)*** .036 (.003)*** .121 (.013)*** .038 (.003)*** LR test 66.75*** 53.85*** 192.30*** 84.27*** 65.66*** N level 2 Range 64 1 – 6 64 1-6 64 1-6 48 1-6 64 1-6 N level 3 Range 29 2 – 37 29 2-37 29 2-37 25 1-37 29 2-37 Number of observations 313 313 313 230 313

*p<0.10 **p<0.05 ***p<0.01. (two-tailed). Standard Error between brackets (…). All numbers have been rounded up to

three decimals, except for the LR test.

The fifth model shows a small significant effect, where the percentage change in the number of elderlies is positively associated with the readmission percentage. This parameter is smaller than in table 3, but it remains a small difference and therefore it is difficult to make any decisive remarks about this.

(38)

For all the models, a significant LR variance estimate is shown, which means that there is a variance between hospitals within a COROP-region of respectively .012, .128, .076, .183 and .025. This means that if we compare the three-level model to the longitudinal panel data model with no COROP-effects, the three-level model offers a better fit to interpret the data than the longitudinal panel data model and is thus preferred. In other words, years are not seen as independent observations, but rather that they are clustered by hospitals and COROP-regions.

(39)

(40)

6. Conclusion

This research tried to provide insights into hospital performance from a public administration perspective, in which objective measures were used, rather than focusing on self-reporting studies on occupational stress by nurses and doctors. These self-reports showed similar results to the governmental reports from the Netherlands, namely that there is

increasing pressure on hospitals (Cleland, 1965; Deckard et al., 1988; Salyer, 1995). This increasing pressure is due to high quality standards to the hospital’s performance, but also because of a growing number of patients. The latter seems a consequence of the double ageing problem that the Netherlands is facing, since people are becoming older and the baby boom generation is now of the retirement age (Van den Berg et al., 2014a). Therefore, the central question in this research is: “What is the effect of the double ageing problem in 2013-2018 on hospital performance in the Netherlands?”.

This research follows the reasoning of previous theories on environmental turbulence and organizational performance. In these researches, turbulence in the environment, measured as an increase in the number of students or immigrants, negatively affects the performance of public organizations such as schools or local governments (Andrews et al., 2013; Boyne & Meier, 2009; Meier & O’Toole, 2009; Van den Bekerom et al., 2016). Therefore, in this research I expected that a change in the amount of elderlies negatively affects hospital performance in the Intensive Care Unit.

Based on the analyses, a number of conclusions can be drawn. First of all, a longitudinal panel analysis proves that there is a significant effect of a yearly percentage change in the amount of elderlies on the mortality rate and the readmission percentage. However, it is difficult to determine that this provides prove for the hypothesis in terms of hospital performance. Based on these two rates, which show only small effects, the hypothesis cannot be easily accepted. Nevertheless, more insights can be provided from a public

(41)

administration perspective on hospital performance, focusing on objective measures, rather than self-reported quality measures.

The second analysis showed that there is more variance between hospitals within COROP-regions, and therefore a three-level model proves that the years are more likely to be clustered by hospitals and COROP-regions, than seen as individual measures. So, this study implicates that there is a significant effect of environmental turbulence measured as the amount of elderlies in a COROP-region on hospital performance in terms of mortality and readmission but does not provide a concluding answer to the central question.

Ageing Problems and Adaptive Performance: The Effect of Ageing on Hospital Performance in the Netherlands