The influence of preference for control on the
relationship between experienced
health-related needs and health care expenses
Business Administration – Change Management University of Groningen
Faculty of Economics and Business
Master Thesis Author: G.IJ.S. ter Bruggen Student number: 2402564
g.i.s.ter.bruggen@student.rug.nl
First supervisor: M.A.G. van Offenbeek Second supervisor: M.L. Hage
2
“Let food be thy medicine and medicine be thy food.”
Hippocrates
Abstract
This paper analyses the influence that preference for control has on health care expenses of older adults, in combination with a segmentation approach based on the patient’s experienced health related needs (EHRN). Research on the impact of a patient’s preference for control on health care expenses is lacking in current literature. In order to fill this gap in the existing literature, an empirical research was done. A sample of 2019 older adults was used, from which longitudinal data were gathered over a three-year period. Experienced health-related needs segmentation was used consisting of five segments, namely: vital, physical and mobility problems, psychosocial coping problems, multi-domain problems, and extremely frail. Their preference for control was measured by checking their preferences for involvement in the health care process, which resulted in three preference construct: shared decision-making, information exchange, and healthy behavior. These two independent variables are regressed against the cost for cure (ZVW), care (AWBZ), and the total health care expenses for the respondents. The analyses only found a direct influence of the healthy behavior preference on both the care expenses and the total expenses. No mediating effect was found, however, the healthy behavior was found to negatively influence the relationship between the extremely frail segment and all three types of health care expenses. The preference for shared decision-making also influence three relationship between EHRN segments and health care expenses. Information exchange did not regress significant. The paper, finally, provides health care policy makers with possible solutions for the ever-soaring health care expenses, based on the study’s findings.
Keywords: Health care, preference for control, healthy behavior, shared decision-making,
4 Table of contents 1 Introduction ... 5 Practical necessity ... 5 1.1 Knowledge gap ... 6 1.2 2 Theoretical framework ... 8
Preference for control ... 8
2.1 Preference for shared decision-making (short: shared decision-making) ... 9
2.1.1 Preference for information exchange (short: information exchange) ... 10
2.1.2 Preference for healthy behavior (short: healthy behavior) ... 10
2.1.3 Experienced health-related needs (EHRN)... 10
2.2 Health care expenses... 12
2.3 Other relevant variables ... 12
2.4 Proposed relationships ... 13
2.5 3 Methodology ... 15
Population and sample ... 15
3.1 Measures ... 16 3.2 Dependent variable ... 16 3.2.1 Independent variables ... 17 3.2.2 Control variables ... 18 3.2.3 Data reduction ... 19 3.3 Data analysis ... 19 3.4 4 Results ... 22
Factor and reliability analyses ... 23
4.1 Descriptives ... 23 4.2 Linear regression ... 26 4.3 5 Discussion ... 37
Discussion of the results ... 38
5.1 The relationship between preference for control and health care expenses ... 38
5.1.1 The relationship between EHRN segments and health care expenses ... 38
5.1.2 The influence of the EHRN segments on health care expenses, through the preference for control ... 39
5.1.3 The influence of preference for control on the relationship between EHRN and health care expenses ... 39
5.1.4 Managerial implications ... 40
5.2 Limitations and future research ... 42
5.3 6 Conclusion ... 42
Acknowledgements ... 43
References ... 44
Appendix 1: Correlation tables ... 48
Appendix 2: Q-Q plots ... 52
Appendix 3: Factor and reliability analysis ... 54
5
1 Introduction
Practical necessity 1.1
As the population is aging and the costs of health care keep rising as a result, finding new avenues for organizing healthcare becomes increasingly important (Airoldi, 2013; Ewings, 2013). The costs of prolonged care in the Netherlands are extremely high in comparison to those costs in other developed countries (Ministerie van Volksgezondheid, Welzijn en Sport, 2012). If this trend of cost growth in prolonged care continues at the current rate, it will be far greater than the cost of prolonged care in countries like Germany and France (7,5% of GDP vs. 2,1-2.2% of GDP) (Ministerie van Volksgezondheid, Welzijn en Sport, 2012). To reduce the health care costs, it is most important to target these costs that are financed through the AWBZ, ZVW, and WMO.
Internationally, policy makers have tried a large number of incremental fixes, but none of the fixes have had a large impact on the cost of health care (Porter and Lee, 2013). Porter and Lee (2013) ascribe this lack of impact to the way the health care sector is organized, revolving around the professionals instead of the patient. To organize the decision-making closer to the patient, the Dutch government has changed the division of tasks in regulating care provisions (Transitie Bureau Begeleiding in de WMO, 2012). In The Netherlands part of the distribution of responsibilities for the homecare for older adults has shifted. The state delegated part of its responsibilities to municipalities. Of interest for this research is the so-called shift from AWBZ to WMO. This means that the municipality becomes responsible for something it has never dealt with before. This transition in responsibilities entails the allocation of resources towards the WMO in the municipality, the allocation of funds for individuals that apply for care, and the control of the usage of these funds (Transitie Bureau Begeleiding in de WMO, 2012). Furthermore, changing regulations will require older adults to live independently in their own home for a longer time, as this is expected to reduce the health care expenses.
6 keeping the quality of care unchanged. In this I take the position that quality of care requires it to be demand-oriented (Rijckmans et al., 2007), that I translate to patient-centered and value-driven care in this thesis (Porter and Lee, 2013).
Knowledge gap 1.2
The reason that this research targets older adults is that the older adults require substantially higher levels of care. Also the proportion of older adults in society will increase in the upcoming years (Airoldi, 2013; Ewings, 2013; McNamee, 2004; ZIP, 2009). In the Netherlands, the proportion of the population that is aged 65 – the retirement age until 2013 – or higher will increase from 15% in 2009, towards an expected 22% in 2025 (ZIP, 2009). To ensure that care focusses on their needs, a segmentation of the older adults is adviced (LaFortune, Béland, Bergman, & Ankri, 2009). In the past, several authors have already divided older adults into more homogeneous groups (McNamee, 2004; LaFortune et al., 2009). The main advantage of dividing the older adults into segments is that it allows for making predictions about the costs of healthcare for these segments. LaFortune et al. (2009) and Eissens van der Laan et al. (2014) have already researched in what way these different segments consume healthcare.
The patient’s rights – of which patient-centered care is a part – belongs to the current key debates in health care (Tayler & Hawley, 2010). It is interesting to see how these groups relate to the subject of patient-centered care, as there has not been any research on this part yet. Taylor & Hawley (2010) argue that the input of patients is necessary to ensure cooperation and compliance with the health care process. They do also make the link to patients’ needs in the way that services have to be tailor-made. Different people will, however, have different preferences for their involvement in the health care process itself. Part of this involvement is their preferred control over the health care process (Ong, de Haes, Hoos, & Lammes, 1995; Auerbach, 2001). As the segmentation and preference for control are both aimed at providing patient-centered care, the link between these concepts is very interesting. This is especially true in relationship to the health care expenses, as multiple scholars have indicated (Stewart et al., 2000; Porter & Lee, 2009).
7 evaluation of the current health care system. It will provide health care policy makers and care givers practical advice on how to address and curb possible inequalities in health care expenses. Porter and Lee (2013) touch the subject of cost; they are astounded that in an industry with such high costs. The costs associated with specific health care outcomes are largely unknown. They continue by saying that it is impossible to reduce costs when you don’t know what causes them. In addition it makes eliminating inequality difficult as well. A large Dutch project on the introduction of the WMO showed that a majority of the people dependent on WMO support do not dare to ask for help (Aandacht voor iedereen, 2014). What does that mean for them in terms of their health care consumption? Does segmentation influence the cost that these people make? Is there inequality in the system? To get a better understanding of these issues the analysis of the usage and the expenses of health care provision within these groups is important. As literature on parts of this topic is scarce, it was necessary to make some assumptions. When I make assumptions this is clearly stated and I will reflect upon these assumptions in the discussion section of this research.
The theoretical contribution of this paper lies in the fact that it, to my knowledge, is the first empirical study comparing the cost of health care, dependent on their preference for control within patient segments. Therefore, the knowledge gap this research addresses is whether the older adults’ involvement in the health care process, by means of their preference for control, will increase the health care expenses of older adults sharing similar experience health-related needs.
This gap translates to the following overall research question:
In what way does the preference for control of older adults influence health expenses of older adults with similar experienced health related needs?
8 limitations – and future research directions are also indicated. Finally, section 6 presents the conclusion of this research.
2 Theoretical framework
In this section the available literature for this research is examined. First the variables are explained, followed by the expected relationships between the different variables. From these expected relationships, hypotheses are developed that guide the data analysis. The dependent variable of the research are the health care expenses of older adults, to be more precise the AWBZ, ZVW and total expenses of the respondents. The independent variables are the experienced health-related problems (latent-class older adults segments), older adults’ preference for control (three-dimensional), and a specified range of control variables, as can be seen in figure 1.
Preference for control 2.1
The central variable in this research is the older adults’ preference for control. In modern medicine there is an increased emphasis on involving the patient’s preference in the health care process, using more two-way communication. This phenomenon of involving patients in health care as mentioned above contributes to patient-centered care (Porter & Lee, 2013). The premise of patient-centered care lies in that it tries to accommodate the patient’s preferences in the area of information exchange, shared decision-making and the patient’s own preference for living healthy behavior (Ong et al., 1995; Auerbach, 2001). The combination of these three factors will form the variable ‘preference for control’ in this research. The three components will be further elaborated below.
9 Lippman, & Dangelo, 1989; Hack, Degner, & Dyck, 1994). Kiesler & Auerbach (2006) found that the preference for control of older adults is generally lower than that of younger people. Patients that do ask questions and voice their opinion tend to understand the prescribed treatment better and, therefore, experience better health outcomes. They might for instance act faster when they experience side effects and, therefore, get treatment modifications (Beisecker, 1996).
Preference for shared decision-making (short: shared decision-making) 2.1.1
10 Preference for information exchange (short: information exchange)
2.1.2
The concept of the exchange of information is defined as the verbal communication between the doctor and patient (Ong et al., 1995). The exchange of information is two-fold, namely information seeking and information giving (Ong et al., 1995). In the past, the doctor provided the patient with information related to the disease, and gave recommendation, the patient voiced symptoms and concerns (Epstein & Street Jr., 2007). Nowadays the patient’s needs, opinions, and beliefs are given more attention, and patients are better prepared when they go to the doctor and bring disease-related information (Epstein & Street Jr., 2007). Patients can be active of passive in information exchange, but a passive patient can still make decisions autonomously (Street Jr., Makoul, Arora, & Epstein, 2009).
Preference for healthy behavior (short: healthy behavior) 2.1.3
Hayes & Ross (1987) argue that people with an internal locus of control believe that they are responsible for and in control of their own health. These people are, therefore, more inclined to live healthy and belief that they can avoid getting sick by doing this (Hayes & Ross, 1987). They also argue that people with an external locus of control belief that they have no influence on their health and, therefore, do not feel the necessity to live healthy. Mirowsky & Ross (2003) adhere to this view, in that they claim that people that want to control their own health have better health outcomes. They link people that want to control their own health to being people that have an internal locus of control. Self-efficacy is also an indicator for more healthy behavior, whereas helplessness was found to decrease healthy behavior (Mirowsky & Ross, 200). This control is linked to the person’s lifestyle, but also to the how well someone adheres to treatment and willingness to change one’s lifestyle when necessary (Hayes & Ross, 1987; Mirowsky & Ross, 2003). I, therefore, expect people that have a higher preference for healthy behavior to have lower health care expenses.
For this research three preference for control constructs will be used, namely: the preference for shared decision-making, the preference for information exchange, and the preference for healthy behavior.
Experienced health-related needs (EHRN) 2.2
11 these changes has however proven to be have a great deal of effect on these rising costs (Porter & Lee, 2013). Therefore, Porter & Lee (2013) propose a more radical solution for the problem by shifting from a supply-driven – physician-centered – approach to health care towards a demand-driven – patient-centered – approach to health care. In order to provide better value to patients, Lafortune et al. (2009) have divided older adults in four more homogeneous patient groups, ranging from vital to extremely frail. Their segmentation was, however, still based on medical indications like diseases. With this research in mind, Eissens van der Laan, Van Offenbeek, Broekhuis, & Slaets (2014) have adopted more patient-centered indicators for such a segmentation, and found five groups of older adults that are considered to have some limited homogeneous health-related needs. The term needs is used in this research. because it is not focused on objective health-related problems (e.g. diseases), as for instance the research of Lafortune et al. (2009) is. The research of Eissens van der Laan et al. (2014) is new in the way that they focus on what the person experiences as a satisfactory level of health, based on their biopsychosocial needs. This leads to a difference in fulfilled and unfulfilled needs, where unfulfilled needs will lead to decreased experienced health (Eissens van der Laan et al., 2014). Human motivation theory entails that everybody has these needs, and realization of these needs will determine one’s well-being (Maslow 1943, Alderfer 1969, Acton & Malathum 2000, Tay & Diener 2011). As stated by Eissens van der Laan et al. (2014; p6): “this basic need concept does not constitute a ‘need for care’, but a need for achieving a satisfactory level of biopsychosocial functioning”. The five segments found in the research are: ‘Vital’, ‘Psychosocial coping problems’, ‘Physical and mobility problems’, ‘Problems in multiple domains’, and ‘Extremely frail’. As it is assumed that patients within one segment experience certain percentage of shared health care needs, it can be expected that physicians can better tailor their services to satisfy the needs of the different EHRN segments, possibly resulting in better utilization of resources, and therefore increased effectiveness and efficiency. Insurers can also benefit from segmentation of older adults, as this can help them align the supplies in a region with the experienced health care problems among the patients (Eissens van der Laan et al. 2014).
12
Health care expenses 2.3
As already mentioned before, the cost of health care is constantly increasing. This is due to the fact that consumers want the best care for themselves, therefore disregard the prices of this care, and as a result the costs soar (Hall & Schneider, 2012). The cost of health care is a highly relevant societal variable, as with the current growth of the costs states will in the future have to lower spending on other fundamental areas as education, infrastructure, and consumer goods (Sommers, 2010). Therefore, health care expenses will be the dependent variable for this research. Carpenter (2012) draws attention to the fact that the term health care costs has taken many different meanings. The five ways used interchangeably are: costs, charges, prices, payments, and expenditures, although their meanings are actually different (Carpenter, 2012). For this research we define the cost of health care as the per capita expenditure on health care services (Martin Martin, del Amo Gonzalez, & Garcia, 2011). This definition is used as it is aimed at the patient (Martin Martin et al., 2011; Carpenter, 2012). This can also be called payment for care, as this is measured by the expenditure of insurers on a specific patient (Carpenter, 2012). These payments are the basis for comparison of health care cost in this research.
In the Netherlands, the health care expenses can be separated in cure and care. The Zorgverzekeringswet (ZVW) covers cure, regular care is covered by the Algemene Wet Bijzondere Ziektekosten (AWBZ), additional care is covered by the Wet Maatschappelijke Ondersteuning (WMO). Elective cure and care are covered by additional insurance or paid by the patients themselves. Because of the aging population, the cost of health care in the Netherlands is growing excessively (ZIP, 2009). As a result, there is a steep increase in chronic diseases, which are very costly, as the intensity and duration of treatment are high (RIVM, 2006).
For this research I will use both the AWBZ and ZVW expenses, as well as the combined total expenses, as this will paint a more comprehensive and realistic picture.
Other relevant variables 2.4
13 strongly dependent on age (Airoldi, 2013; Ewing, 2013). Gender is used because of the differences in certain conditions between genders that have proven to have a significant influence on health expenditure (Bertakis & Azari, 2010). Education has been proven to have a high influence on the lifestyle of people, which has been found to have a direct negative relationship to the cost of health care of individuals (Cutler and Lleras-Muney, 2010). Living arrangements account for a large part of the difference in health care cost (L. Dorland, personal communication, February 13, 2014).
Proposed relationships 2.5
This paragraph describes the relationships between the different variables through a main hypothesis and multiple sub-hypotheses. From these relationships a conceptual model is constructed that is depicted in figure 1. The main hypothesis of the research is as follows:
Main hypothesis: Older adults’ preference for control influences the relationship between
elderlies experienced unfulfilled health-related needs and their health care expenditure.
Eissens van der Laan et al. (2014) have already shown that the only the extremely frail segment has significantly different preference for shared decision-making, all other segments do not significantly differ. I expect the relationship between the EHRN segments to be positive in relation to the preference for shared decision-making, this is supported by the findings of the Dutch Centre for Ethics and Health (CEG) (2013). From this I expect that this will be the same for preferred information exchange and the preference for healthy behavior. From this the following hypothesis is created:
14 adults, so the within group consumption of health care should differ less than with the grouping on more practicle segmentation. The individual characteristics in each group are however still present, e.g. differences in the control variables – living arrangements, age, education, and gender – that influence health care expenditure heavily are, although less exstensive, still present (Eissens van der Laan et al., 2014). With specific diseases come certain costs, but this seemingly homogenous grouping is incomplete in the way that this way of grouping people disregards the unobservable variations between people that account for a large part of the health care expenses (Lafortune et al., 2009). This is supported by McNamee (2004), whom also argues that people with different frailty states have different health care expenses, e.g. people that are very frail have a higher cost of health care then people that are healthy. So there is a theorized positive relationship between the EHRN and the health care consumption, and thus, presumably health care expenses.
Hypothesis 2: Different EHRN-segments have different health care expenses
Visser, Westendorp, Cools, Kremer, and Klink (2012) have concluded that current health care is concerned with delivering more care, instead of better care. However, when the patient is involved in the decision-making, care can be expected to fit better to the patient’s condition, but does this influence the cost of health care in a positive way? Charles et al. (1999) found that the increase in patient involvement in decision-making was a reaction to counter the increase in cost of health care. In more recent years however Hall & Schneider (2012) argue that consumerism will not prevail in health care, as the prerequisites for consumerism are not met in health care. According to the CEG (2013) involving the patient does not lower cost, moreover shared decision-making does take more appointments and time to get to the decision, and physician time is very expensive, they base this on the findings of their own research (CEG, 2013). From this I argue that a patient that has a high preference for shared decision-making will have higher health care expenditures as a result. Whether this relationship is direct, manifested through mediation, or by moderation is explored through the next hypotheses:
Hypothesis 3a: ‘Preference for control’ influences the cost of health care
Hypothesis 3b: ‘Preference for control’ governs the relationship between EHRN and the
15 Hypothesis 3c: ‘Preference for control’ influences the relationship between EHRN and
the health care expenses.
Figure 1: Conceptual model
3 Methodology
In this section, the research design of this thesis will be discussed. Subsequently the population and sample, the measures, the data reduction, and the data analysis techniques for this research will be reviewed.
Population and sample 3.1
16 and Slaets, 2014). This does limit the generalizability of the research, as certain groups might be under or over represented.
The sample is derived from a database that contains a partially cleaned dataset derived from a longitudinal survey study on this sample that included the Groningen Frailty Indicator (GFI), the INTERMED, and a questionnaire on patient preferences. The database was later supplemented with the health care expenditures of part of the respondents. The total sample size for the survey study was 2019 older adults, whom all have given consent for the use of the anonymous data (on T0 (2010) n=2019 were included, on T1 (2011) n=1507, and on T2 (2012) n=1131). Next to this database, the largest insurer in the Northern region of The Netherlands (and other agencies) have provided the health care costs of part of the respondents. The cost of health care used in this research are the AWBZ, ZVW and total health care expenses of the older adults included in the sample. For this research the health care expenses of the respondents for 2010, 2011, and 2012 were used, as the longitudinal design of the study will cover the three years of the cleaned dataset of the Eissens van der Laan et al. study (2014).
Measures 3.2
Table 1 provides an overview of the different measures that were used in this research. Dependent variable
3.2.1
3.2.1.1 Cost of healthcare
The cost of health care is measured using both the ZVW and AWBZ expenses, as well as the combined total expenses of the older adults. Health care expenses are a widely used measure for cost of health care (Carpenter, 2012). By using both expenses, a complete picture of an older adults’ cost in the health care process can be obtained. This ensures the content and construct validity, as all costs of health care that are available were taken into account and a complete picture could be provided. As the study embraces a longitudinal design, the expenses for T0=2010, T1=2011, and T2=2012 were used. The expenses of the older adults can be classified as
17 medicine, and contribution related. However, these more fine-grained distinctions were not used in the analyses presented.
Table 1: Overview of measures used in this research
Variable Measure Definition Description
Cost of health care
Health expenses Individual expenses of AWBZ & ZVW, ratio scale
The AWBZ expenses (care), as well as the ZVW expenses (cure), and all expenses of the patient together are the dependent subject of this research. Experienced
health-related needs
GFI,
INTERMED, adapted for self-assessment
Both the individual level, as heterogeneous groups of older adults. Segmentation based on latent-class analysis, interval scale
Multiple scales, leading to a latent-class calculation for grouping. Each segment gets a score, running from .00 to 1.00, where the sum of the scores of all segments equals 1.00.
Preference for control
Patient
preference scale
Preference regarding the involvement in the health-care process, Interval scale
Divided in three constructs; preference shared decision-making, preference for doctor-patient information exchange, and the preference for healthy behavior. Scale ranges from 0 = No preference to 4 = High preference
Control variables
Age Age during the measurement, differs between measurements, interval scale
Age in years during the measurement, calculated by date of the survey – date of birth / 365.25
Living arrangement
The living arrangement during measurement, nominal scale
The measure runs from living independently alone (0) to living in a nursing home (7)
Gender Gender of the respondent, ordinal scale
Male (0) or Female (1)
Education Highest finished education, ordinal scale
Level of education. 7 levels, starting with not finishing pre-school (0) and going up-to finished higher education (6)
Independent variables 3.2.2
3.2.2.1 Preference for control
The preference for control will be constructed through three multi-item scales that were also included in the questionnaires, and are derived from Auerbach’s (2001) research on patient preferences. The patient’s preference for control data can be classified as interval data as multi-item 5-point answering scales were used. These data were only collected for T1 and T2. For the
18 the factor analysis can be found in Appendix 3, where the rotation matrix for the concepts loading towards preference for control can be found.
3.2.2.2 Experienced health-related needs (EHRN)
The measurement for EHRN in this research was the latent-class patient segmentation, as developed and validated by Eissens van der Laan et al. (2014), could be viewed as nominal data, as it is not possible to rank the segments. Every patient gets a percentage based likelihood of belonging in a certain class using latent-class analysis, based on their self-reported GFI and INTERMED scores. The different segments and respondents’ scores on these segments have already been determined using latent-class cluster analysis (LCCA) during the research by Eissens van der Laan et al. (2014). Eissens van der Laan et al. (2014) found 5 different segments in their latent-class analysis, as explained earlier. For certain analyses the segment for which the individual had the highest likelihood score was used, as the analysis was not capable of coping with the latent-classes itself. The segmentation was performed on T0, T1, and T2, so it fits the
longitudinal nature of this study. Control variables
3.2.3
3.2.3.1 Age
Age is the main control variable in this research, as with age the cost of health care normally rises (Porter and Lee, 2013; Airoldi, 2013; Ewing, 2013). The age was measured in years, and calculated for each measurement individually, using the date of birth and the date the survey was completed.
3.2.3.2 Living arrangement
Living arrangements explain a large part of the costs of health care (L. Dorland, personal communication, February 13, 2014) and, therefore, needs to be used as a control variable in this research. The data are nominal as there is no rank in the living arrangements. For this variable 7 dummies will be formed for the data analyses, as indicated in table 1.
3.2.3.3 Gender
19
3.2.3.4 Education
Education has been linked to different health outcomes, i.e. higher education is associated with better health outcomes and higher life expectancy (Cutler and Lleras-Muney, 2010). Therefore, this can be seen as a relevant control variable. See table 1 for the scaling.
Data reduction 3.3
For the data analysis SPSS v20.0.0.2 will be used, as this is the universities primary data analysis software and the provided databases were SPSS compatible. First, data reduction needs to be performed by mode of data cleaning.
The dataset that was provided for this research consisted out of older adults in general, so also of older adults that did not use health care services. Therefore, the data set had to be cleaned in order to provide a sample with older adults that do use health care services. This was done through a log-transformation that provided all respondents with no health care costs with a missing value. Following this, the normality was proven by checking the Q-Q plots, as the large sample clouded the regular normality analyses (Thode, 2002; Field, 2009). The Q-Q plots for this research can be found in Appendix 3.
The start of the data reduction was the development of different preference for control scales. Explanatory factor analysis was used to define the different components of the concept (Floyd & Widaman, 1995; DeVellis, 2012), combined with a reliability analysis. The results of this factor analysis and reliability analysis can be found in the next section in Appendix 3. Finally, data from respondents from whom data on one or more of the included variables was not present were excluded from the analysis.
Data analysis 3.4
Descriptive analysis was used to summarize findings and describe the sample. Finally, inferential statistics were used to draw conclusions from the data. From earlier analyses (Laura Dorland, personal communication, February 20141) it is already know that the cost data are only normally distributed if people that do not use health care services are excluded. A logarithmic transformation was used to exclude cases where costs are 0, and to provide a normal distribution
20 (Aitchison & Brown, 1969). To assess the relationship between the dependent variables and the independent variables, multiple linear regression was found to be appropriate (Draper & Smith, 1998). To test hypothesis 1, the following models were used:
Model a, b, and c: Preference of control = b0 + b1 x Vital + b2 x Physical problems + b3 x Psychosocial problems + b4 x Multi-domain problems + b5 x Extremely frail + b6 x Age + b7 x Gender + b8 x LA dummy 1 + b9 x LA dummy 2 + b10 x LA dummy 3 + b11 x LA dummy 4 + b12 x LA dummy 5 + b13 x LA dummy 6 + b14 x LA dummy 7 + b15 x Education + ε
Where a is the model with shared decision-making as the dependent variable, b is the model with information exchange as the dependent variable and c is the model with healthy behavior as the dependent variable. The dependent variables for all models are shown in table 2.
For researching hypothesis 2, the following models where used:
Model 1, 4 and 7: expenses = b0 + b1 x Vital + b2 x Physical problems + b3 x Psychosocial problems + b4 x Multi-domain problems + b5 x Extremely frail + b6 x Age + b7 x Gender + b8 x LA dummy 1 + b9 x LA dummy 2 + b10 x LA dummy 3 + b11 x LA dummy 4 + b12 x LA dummy 5 + b13 x LA dummy 6 + b14 x LA dummy 7 + b15 x Education + ε
Table 2: Dependent variable for every model
Model Dependent variable
a Shared decision-making b Information exchange c Healthy behavior 1, 2, 3, u, x AWBZ 4, 5, 6, v, y ZVW 7, 8, 9, w, z Total expenses
21 care that are involved in the direct relationships and mediation, numbers are used. Letters from the end of the alphabet (u through z) are used for the moderation analyses.
To research hypothesis 3a, the following models where used:
Models 2, 5 and 8: expenses = b0 + b1 x Decision making + b2 x Information exchange + b3 x Healthy behavior + b4 x Age + b5 x Gender + b6 x LA dummy 1 + b7 x LA dummy 2 + b8 x LA dummy 3 + b9 x LA dummy 4 + b10 x LA dummy 5 + b11 x LA dummy 6 + b12 x LA dummy 7 + b13 x Education + ε
For hypothesis 3b multiple models where used. This was done using the widely excepted mediation analysis method, as developed by Baron & Kenny (1986). Using their method, models 1 through 9 were used. Models a, b, c, 1, 3, 4, 6, 7, 8 and 9 were used as described above, models 2, 5 and 8 was as follows:
Models 3, 6 and 9: expenses = b0 + b1 x Vital + b2 x Physical problems + b3 x Psychosocial problems + b4 x Multi-domain problems + b5 x Extremely frail + b6 x Decision making + b7 x Information exchange + b8 x Healthy behavior + b9 x Age + b10 x Gender + b11 x LA dummy 1 + b12 x LA dummy 2 + b13 x LA dummy 3 + b14 x LA dummy 4 + b15 x LA dummy 5 + b16 x LA dummy 6 + b17 x LA dummy 7 + b18 x Education + ε
The possibility of mediation was tested following the criteria of Baron & Kenny (1986). There is mediation when a) the relationship between the EHRN and preference – model a, b and
c – was significant. b) The relationship between that preference and the health care expenses –
model 2, 5 and 8 – was significant as well. c) The relationship between the EHRN and the health care expenses – model 1, 4 and 7 – was significant too. And finally, d) the relationship between the EHRN and the health care expenses becomes insignificant when it was in the same model –
3, 6 and 9 – as the preference. When the mediation is partial the relationship between EHRN and
the health care expenses in d) did not fall out of significance, the relationship was only weakened.
22 preference and EHRN segment. The model was compared against a second model – x, y and z – with an interaction variable. The latter was composed by centering the data of both the concerning preference and the EHRN segment and then multiplying these two scores.
Model u, v and w: expenses: b0 + b1 x preference + b2 x EHRN segment + b3 x Age + b4 x Gender + b5 x LA dummy 1 + b6 x LA dummy 2 + b7 x LA dummy 3 + b8 x LA dummy 4 + b9 x LA dummy 5 + b10 x LA dummy 6 + b11 x LA dummy 7 + b12 x Education + ε
Model x, y and z: expenses: b0 + b1 x preference + b2 x EHRN segment + b3 x interaction variable + b4 x Age + b5 x Gender + b6 x LA dummy 1 + b7 x LA dummy 2 + b8 x LA dummy 3 + b9 x LA dummy 4 + b10 x LA dummy 5 + b11 x LA dummy 6 + b12 x LA dummy 7 + b13 x Education + ε
Moderation was confirmed when the interaction variable was significant and the EHRN segment was significant as well, and furthermore, the models adjusted R2 had to be higher than that of the model without the interaction. Simply a higher R2 of the model with the interaction variable included does not automatically prove moderation (Baron & Kenny, 1986).
The longitudinal aspect of the analysis entailed a lagged analysis. A lagged analysis can be used to check for the influence of independent variable a at Tn on the dependent variable b at
Tn+1 or Tn+2 (Hamilton, 1994). In this research the time of the independent variables was Tn and
the dependent variable and control variables were Tn+1 or Tn+2. The lagged analysis were denoted
as T0-T1, T1-T2 or T0-T2 in this research. The research this type of longitudinal analysis was used,
was to find out whether the health care expenses were influence by independent variables from previous years.
4 Results
23
Factor and reliability analyses 4.1
The results of the factor analysis suggest acceptable construct validity for this specific population. As can be seen in Appendix 3, the construct patient’s preference for control can be subdivided in three different components in line with the three constructs described in the theoretical framework section. Factor analysis was an appropriate technique as the Kaiser-Meyer-Olkin Measure of Sampling Adequacy and Bartlett’s Test of Sphericity are both sufficient. The three subscales are also reliable in the way that Cronbach’s alphas for components 1, 2, and 3 are all acceptable. The alpha scores are all well above the 0,7, which is deemed sufficient for research, the scores can also be found in Appendix 3 (Nunally, 1978). This suggests that the three subscales are useful constructs for measuring the preference for control concept.
Descriptives 4.2
Tables 3 and 4 comprise the descriptive statistics for this research. The descriptives are given for the database in total and for every separate measurement (Tn).
Table 3: Descriptives table for interval variables
Variable Tn N Mean (Log) S.D. (Log) Min Max
Total health care expenses All 972 30010 (9.9461) 22515 (.96197) 2411 143491
0 446 31543 (10.0293) 23668 (.90560) 2411 143491
1 312 29645 (9.9125) 22295 (.99429) 2413 124282
2 214 27346 (9.8219) 20018 (1.01493) 2453 87124
AWBZ expenses All 877 27922 (10.0291) 17771 (.73683) 1781 124282
0 421 26720 (9.9595) 16945 (.75619) 1781 124282 1 280 29484 (10.0933) 19331 (.73741) 2030 124282 2 176 28579 (10.0935) 16952 (.67523) 2017 79188 ZVW expenses All 1769 8643 (8.4064) 11814 (1.22652) 129 141985 0 779 11003 (8.6292) 14682 (1.26923) 129 141985 1 572 7141 (8.2994) 8731 (1.14288) 129 85650 2 418 6301 (8.1376) 8136 (1.18457) 131 76368 Complexity All 4656 12.87 6.967 0 44 0 2019 13.36 7.165 0 44 1 1506 12.78 6.875 0 40 2 1131 12.12 6.656 0 36 Frailty All 4650 4.65 3.223 0 15 0 2015 4.36 3.006 0 14 1 1504 5.01 3.401 0 15 2 1131 4.69 3.305 0 15 1 Vital All 4647 .2506 .33617 .00 .97 0 2013 .2462 .33657 .00 .97 1 1504 .2446 .33363 .00 .97 2 1130 .2666 .33617 .00 .97
2 Psychosocial coping problems All 4647 .1938 .22703 .00 .95
0 2013 .1855 .22706 .00 .95
1 1504 .1940 .22432 .00 .95
24 3 Physical and mobility problems All 4647 .2891 .29028 .00 1.00
0 2013 .2970 .29829 .00 1.00
1 1504 .2833 .28773 .00 1.00
2 1130 .2828 .27885 .00 .99
4 Multi-domain problems All 4647 .2261 .31719 .00 .99
0 2013 .2342 .32227 .00 .99
1 1504 .2330 .32136 .00 .99
2 1130 .2026 .30118 .00 .99
5 Extremely frail All 4647 .0403 .16213 .00 1.00
0 2013 .0371 .15295 .00 1.00
1 1504 .0451 .17412 .00 1.00
2 1130 .0395 .16146 .00 1.00
Decision making preference All 2571 1.2081 .91519 .00 4.00
1 1464 1.1610 .91854 .00 4.00
2 1107 1.2703 .90174 .00 4.00
Information exchange preference All 2571 3.2421 .72996 .00 4.00
1 1464 3.2056 .75116 .00 4.00
2 1107 3.2904 .69834 .00 4.00
Preference for healthy behavior All 2571 3.1846 .64262 .00 4.00
1 1464 3.1685 .66133 .00 4.00 2 1107 3.2057 .61666 .00 4.00 Age All 4656 79.14 7.752 64 102 0 2019 78.86 7.806 64 100 1 1506 79.35 7.774 65 101 2 1131 79.34 7.637 66 102
Table 4: Descriptives table for nominal and ordinal variables
Nominal/ordinal variables Tn N # of N % of N
Gender All 4656 Male: 1814 Female: 2842 Male: 39.0% Female: 61.0% 0 2019 Male: 785 Female: 1234 Male: 38.9% Female: 61.1% 1 1506 Male: 575 Female: 931 Male: 38.2% Female: 61.8% 2 1131 Male: 454 Female: 667 Male: 40.1% Female: 59.9% Education All 4650 351 (0), 1023 (1),706 (2), 570 (3), 1473 (4), 207 (5), 320 (6) 7.5% (0), 22.0% (1), 15.2% (2), 12.3% (3), 31.7% (4), 4.5% (5), 6.9% (6) 0 2016 164 (0), 457 (1), 312 (2), 253 (3), 618 (4), 83 (5), 129 (6) 8.1% (0), 22.7% (1), 15.5% (2), 12.5% (3), 30.7% (4), 4.1% (5), 6.4% (6) 1 1504 113 (0), 336 (1), 223 (2), 182 (3), 479 (4), 67 (5), 104 (6) 7.5% (0), 22.3% (1), 14.8% (2), 12.1% (3), 31.8% (4), 4.5% (5), 6.9% (6) 2 1130 74 (0), 230 (1), 171 (2), 135 (3), 376 (4), 57 (5), 87 (6) 6.5% (0), 20.4% (1), 15.1% (2), 11.9% (3), 33.3% (4), 5.0% (5), 7.7% (6)
Living arrangement All 4656 1876 (0), 1034 (1), 73 (2), 163 (3), 113 (4), 985 (5), 385 (6), 27 (7) 40.3% (0), 22.2% (1), 1.6% (2), 3.5% (3), 2.4% (4), 21.2% (5), 8.3% (6), .6% (7) 0 2019 784 (0), 434 (1), 29 (2), 65 (3), 54 (4), 433 (5), 203 (6), 17 (7) 38.8% (0), 21.5% (1), 1.4% (2), 3.2% (3), 2.7 (4), 21.4% (5), 10.1% (6), .8% (7) 1 1506 610 (0), 334 (1), 24 (2), 50 (3), 36 (4), 332 (5), 119 (6), 1 (7) 40.5% (0), 22.2% (1), 1.6% (2), 3.3% (3), 2.4% (4), 22.0% (5), 7.9% (6), .1% (7) 2 1131 482 (0), 266 (1), 20 (2), 48 (3), 23 (4), 220 (5), 63 (6), 9 (7) 42.6% (0), 23.5% (1), 1.8% (2), 4.2% (3), 2.0% (4), 19.5% (5), 5.6% (6), .8% (7)
25 tables for ΔT0-T1, ΔT1-T2 and ΔT0-T2 are given. Table 5 summarizes the findings from the
correlation matrices that are most important for the regression analysis. From this it becomes apparent that all five latent-classes correlate significantly with the total expenses. The only segment that does not correlate with the AWBZ expenses is the physical and mobility problems segment. The ZVW expenses do not correlate with the psychosocial problems segment, and in
ΔT0-T1 and ΔT0-T2 the extremely frail segment does not correlate with the ZVW expenses.
Furthermore, the correlation between the ZVW expenses and the extremely frail segment is even when it is significant fairly weak and negative instead of a positive correlation with the AWBZ and total expenses. The vital and psychosocial segments have a, when they correlate, negative correlation with the expenses. The Physical and mobility problems segment and the multi-domain segment have a positive correlation with al expenses, when they correlate.
What stands out with the preference for control concepts is the lack of correlation between the ZVW and the concepts. Only the healthy behavior preference correlates significant at the 95% level in Tall. The decision-making preference and healthy behavior preference do
correlate significant with the AWBZ and total expenses, where the correlation with decision-making is positive and with the healthy behavior preference is negative. So the cost of health care increases when people have an increased shared decision-making preference and the costs decline when a patient’s healthy behavior preference increases.
In general the correlations with the total health care expenses are stronger than the correlation with the AWBZ and ZVW expenses.
Table 5: Correlations important for the regression
Tn Vital Mobility problems Psychosocial problems Multi-domain problems Extremely frail Shared decision-making Information exchange Healthy lifestyle AWBZ Tall -.256** .029 -.254** .127** .232** .210** -.048 -.218** ΔT0-T1 -.240** .001 -.251** .214** .175** ΔT1-T2 -.212** -.008 -.282** .152* .274** .247** -.131 -.277** ΔT0-T2 -.369** .076 -.196** .210** .201** ZVW Tall -.218** .152** -.023 .155** -.097** -.014 .003 -.081* ΔT0-T1 -.274** .153** -.003 .189** -.065 ΔT1-T2 -.232** .128** .035 .191** -.176** .007 .043 -.060 ΔT0-T2 -.249** .126** .030 .176** -.050
Total expenses Tall -.498** .216** -.242** .355** .221** .237** -.073 -.264**
ΔT0-T1 -.531** .216** -.187** .392** .169**
ΔT1-T2 -.493** .240** -.206** .390** .218** .252** -.004 -.231**
26
Linear regression 4.3
The linear regression provides us with the results for sub hypotheses 1, 2, 3a, 3b and 3c. A summary of results concerning theses sub hypotheses can be found at the end of this section in table 13. The first thing that stands out with the regression for hypothesis 1, is the regression, or better the lack of regression, for the vital and psychosocial problems segment, as the psychosocial problems segment does not correlate significant with any of the preference for control components and the vital segment is excluded from the analysis due to multicollinearity. A reason for the multicollinearity might be that the vital segment correlates strong with the other EHRN segments. As for the other EHRN segments in hypothesis 1, the decision-making and healthy behavior preference for control constructs regress significant with the physical and mobility problems segment, the multi-domain segment and the extremely frail segment, as can be found in models a, and c in table 6. The relationship is negative in the case of healthy behavior, whereas the relationship with decision-making is positive. So people that score higher on these segments have a higher decision making preference, but a lower preference for healthy behavior. Information exchange only regresses significant with the multi-domain segment in the Tall, but does not regress significant with any other segment in any other situation.
Model 1, 4, and 7 were used to provide the outcome for hypothesis 2, where the relationship between the EHRN segments and the cost of health care is hypothesized. From the five EHRN segments, 3 regress significant with the AWBZ, namely; Vital, psychosocial coping problems and extremely frail, where it has to be said that extremely frail only regresses significant in Tall. ZVW regresses significant with all EHRN segments except for the vital
27 For the hypotheses 3a, 3b and 3c, about the regression of preference for control on the cost of health care, models 2, 5, and 8 are used for hypothesis 3a on the direct relationship, models 1 through 9 and models a, b, and c are used for hypothesis 3b on the mediation relationship, finally models u through z are used for hypothesis 3c. For the decision making preference and information exchange preference there is no direct relationship found with any of the health care expenses. Healthy behavior preference regresses negatively significant to all three types of health care expenses for Tall. It is also significantly negative for the AWBZ and total
expenses in T1-T2, it does however fall out of significance for the ZVW expenses. So sub
hypothesis 3a can only partially be accepted, on the account that healthy behavior regresses significantly negative to all health care expenses.
For the possibility mediation, the values of the EHRN segments should drop in significance in models 3, 6, and 9, whilst the regressions for the other models should be significant. There is an indication for mediation of the healthy behavior preference for the extremely frail segment in Tall for the AWBZ costs, as the significance levels of the extremely
frail segment drop in model 3, whilst the healthy behavior preference stays significant, albeit on p<.05 instead of p<.01. The drop of significance for healthy behavior preference in ΔT1-T2 of
model 3 might however counter this finding, but the drop in significant in ΔT1-T2 might also be
due to the lower N of AWBZ in T1-T2.
28 negative, meaning that the expenses get lower when decision-making is higher. Information exchange does only have a positive moderating interaction on the relationship between the multi-domain segment and the AWBZ expenses. As the moderation analysis entails are great amount of large tables, the tables of the moderation analysis for the AWBZ and ZVW expenses can be found in Appendix 4 for the sake of the readability of this paper.
29
Table 6: Regression of EHRN segments on preference for control - models a, b and c
Predictor Preference for shared decision-making Preference for information exchange Preference for healthy behavior
(a) (b) (c)
Segments Tall T0-T1 T1-T2 Tall T0-T1 T1-T2 Tall T0-T1 T1-T2
Vital Excluded# Excluded# Excluded# Excluded# Excluded# Excluded# Excluded# Excluded# Excluded#
30
Table 7: Regression for AWBZ expenses - models 1, 2 and 3
Predictor AWBZ
(1) (2) (3)
Tall T0-T1 T1-T2 T0-T2 Tall T1-T2 Tall T1- T2
Segments Vital -.570** (.121) -.655** (.201) -.402 (.247) -.862** (.237) -.503** (.175) -.376 (.244)
Physical and mobility problems Excluded# -.245
(.152)
Excluded# Excluded# Excluded# Excluded#
Psychosocial problems -.608** (.128) -.930** (.225) -.818** (.272) -.441 (.293) -.731** (.191) -.785** (.266) Multi-domain problems .106 (.084) Excluded# .245 (.187) .109 (.193) .068 (.121) .167 (.186) Extremely frail .274* (.111) .122 (.259) .415 (.274) .085 (.294) .175 (.167) .413 (.307)
Preference for control
Preference for shared decision-making .054 (.038) .077 (.064) -.016 (.038) .010 (.066) Preference for information
exchange -.071 (.052) -.063 (.084) .057 (.051) -.093 (.082)
Preference for healthy behavior -.194**
31
Table 8: Regression analysis for ZVW expenses - models 4, 5 and 6
Predictor ZVW
(4) (5) (6)
Tall T0-T1 T1-T2 T0-T2 Tall T1-T2 Tall T1- T2
Segments
Vital Excluded# Excluded# Excluded# Excluded# Excluded# Excluded#
Physical and mobility problems 1.382**
(.122) 1.250** (.203) 1.207** (.229) 1.218** (.235) 1.478** (.155) 1.155** (.231) Psychosocial problems .455** (.144) .667** (.240) .637* (.266) .687** (.293) .585** (.182) .677* (.269) Multi-domain problems 1.293** (.107) 1.385** (.180) 1.394** (.203) 1.066** (.204) 1.448** (.136) 1.318** (.207) Extremely frail 1.003** (.212) .621 (.443) .860 (.473) 1.919** (.618) 1.598** (.268) .640 (.494)
Preference for control
Preference for shared decision-making -.007 (.045) .081 (.071) -.056 (.042) .039 (.069) Preference for information
exchange .078 (.058) .143 (.086) .034 (.054) .084 (.083)
Preference for healthy behavior -.235**
32
Table 9: Regression analyses for Total expenses - models 7, 8 and 9
Predictor Total expenses
(7) (8) (9)
Tall T0-T1 T1-T2 T0-T2 Tall T1-T2 Tall T1- T2
Segments
Vital Excluded# Excluded# Excluded# Excluded# Excluded# Excluded#
Physical and mobility problems 1.635** (.141) 1.539** (.249) 1.583** (.287) 1.363** (.319) 1.834** (.184) 1.555** (.289) Psychosocial problems .257 (.169) .451 (.290) .141 (.347) .245 (.351) .364 (.223) .245 (.354) Multi-domain problems 1.744** (.125) 1.733** (.214) 1.918** (.270) 1.723** (.266) 1.944** (.168) 1.844** (.280) Extremely frail 1.729** (.220) 1.648** (.446) 2.347** (.588) 1.265** (.716) 1.725** (.299) 2.380** (.660)
Preference for control
Preference for shared decision-making .074 (.055) .124 (.098) -.014 (.050) .003 (.092) Preference for information
exchange .114 (.073) .212 (.118) .053 (.066) .074 (.109) Preference for healthy
33
Table 10: Moderation analysis preference for shared decision-making on the EHRN segment - Total expenses relationship
Predictor Total expenses – preference for shared decision-making
Vital Physical and mobility
problems
Psychosocial problems Multi-domain
problems
Extremely Frail
Model (w) (z) (w) (z) (w) (z) (w) (z) (w) (z)
Variables
34
Table 11: Moderation analysis preference for information exchange on the EHRN segment - Total expenses relationship
Predictor Total expenses – Preference for information exchange
Vital Physical and mobility
problems
Psychosocial problems Multi-domain
35
Table 12: Moderation analysis preference for healthy behavior on the EHRN segment - Total expenses relationship
Predictor Total expenses – Preference for healthy behavior
Vital Physical and mobility
problems
Psychosocial problems Multi-domain
problems
Extremely Frail
Model (w) (z) (w) (z) (w) (z) (w) (z) (w) (z)
Variables
36
Table 13: Results of the sub hypotheses
Hypothesis
Dependent variable
Independent
variable(s) Cross-sectional results Longitudinal results (Δ)
1 Accepted Rejected
Preference for shared decision-making
EHRN Physical, multi-domain, extremely frail
(all +)
Vital, psychosocial Physical and mobility problems not significant in longitudinal analysis Preference for
information exchange
EHRN Multi-domain (+) Vital, physical,
psychosocial, extremely frail
No significant results in longitudinal analysis Preference for healthy
behavior
EHRN Physical, multi-domain, extremely frail
(all +)
Vital, psychosocial Extremely frail not significant in T0-T1
2 Accepted Rejected
AWBZ EHRN Vital (-), psychosocial (-)
Extremely frail (+)(partially)
Physical and mobility, multi-domain
Vital not significant in T1-T2, psychosocial not
significant in T0-T2,
extremely frail only significant in Tall
ZVW EHRN Physical and mobility (+), multi-domain
(+), psychosocial (+), extremely frail (+)
Vital In T0-T1 and T1-T2 extremely
frail is not significant
Total expenses EHRN Physical and mobility (+), multi-domain
(+), extremely frail (+)
Vital, psychosocial Similar to cross-sectional
3a Accepted Rejected
AWBZ Preference for
control
Healthy behavior (-) Shared decision-making,
information exchange
Similar to cross-sectional
ZVW Preference for
control
Healthy behavior (-)(partially) Shared decision-making,
information exchange
Healthy lifestyle not significant in T1-T2
Total expenses Preference for
control
Healthy behavior (-) Shared decision-making,
information exchange
Similar to cross-sectional
3b: Mediation Accepted Rejected
AWBZ Preference for
control and EHRN
- Rejected for all cases N/A
ZVW Preference for
control and EHRN
- Rejected for all cases N/A
Total expenses Preference for
control and EHRN
- Rejected for all cases N/A
3c: Moderation Accepted Rejected
AWBZ Preference for
control and EHRN
Information exchange/multi-domain (-), Healthy lifestyle/ extremely frail (+)
All others N/A
ZVW Preference for
control and EHRN
Shared decision-making/multi-domain (-), Healthy lifestyle/extremely frail (+)
All others N/A
Total expenses Preference for
control and EHRN
Shared decision-making/mobility problems (-),
Shared decision-making/psychosocial problems (-),
Healthy lifestyle/vital (-),
Healthy lifestyle/psychosocial problems (-), Healthy lifestyle/extremely frail (+)
All others N/A
37 for 29.3% of the AWBZ expenses in the Tall and for 35.0% of the expenses in the longitudinal
variant. Model 6 can account for 23.6% of the ZVW expenses in the Tall variant and for 29.5% of
the ZVW expenses in the longitudinal version for this sample. This is surprising as the total expenses is built of from the AWBZ and ZVW from the concerning year. The time lag variants of the analysis account for more of the expenses than the normal analysis. This can be due to the fact that the Tall analysis is build-up from data that are from subsequent years. From the control
variables, living arrangements and gender appear to be the best predictors of health care expenses for the total expenses. Gender and living arrangement dummies 6 – living in a nursing home alone – and 7 – living alone in a care home – predict the most variance for the ZVW. For the AWBZ the control variables that predict the most variance are the living arrangements. Furthermore the decrease in the maximum of all three cost variables is interesting, just as the seemingly decreasing average cost of health care for the population. This might be due to new regulations, or insurance changes.
5 Discussion
38
Discussion of the results 5.1
The relationship between preference for control and health care expenses 5.1.1
Healthy behavior is the only preference for control that has a direct effect on the health care expenses, where older adults that behave healthier are less expensive. This is in line with the hypothesis for this preference, Mirowsky & Ross (2003) already argued the benefits of healthy behavior in relation to the health care expenses. Interestingly more decision-making control does not lower the health care expenses, unlike the findings of Charles et al. (1999). The findings is more in line with the more recent findings of the CEG (2013) and Hall & Schneider (2012) that increased shared decision-making does not lower expenses on its own, but the hypothesis that the cost would increase was not supported as well. The lack of effect from the EHRN on the information exchange might be due to already extensive information exchange between the physician and the patient.
The relationship between EHRN segments and health care expenses 5.1.2
39 nursing home expenses are part of AWBZ in The Netherlands. Eissens van der Laan et al. (2014) also found this low usage of care by the psychosocial coping problems segment. Maybe part of these nursing home costs are appointed to the ZVW, as the psychosocial segment uses more ZVW expenses, which can also be explained by the utilization of specialist and the general practitioner services (Eissens van der Laan et al., 2014). The combination of the positive relationship with the ZVW expenses and the negative relation with the AWBZ expenses explains the lack of relationship with the total expenses for the psychosocial segment. The extremely frail segments results are not surprising, as the extremely frail segment covers people that get palliative care, which makes it the only segment that has higher AWBZ expenses (Eissens van der Laan et al., 2014).
The influence of the EHRN segments on health care expenses, through the 5.1.3
preference for control
The health care expenses did not change as a result of the EHRN segments influencing the preference for control, although the EHRN segments do have an influence on preference for control itself. According to Zhao, Lynch & Chen (2010), this does not mean that the expenses are not influence through the preference for control, as these might be hidden in the direct influence that EHRN segments has on the health care expenses. The mediation analysis can, therefore, be deemed unreliable, as it does not always find the mediation effect. Thus, it is not relevant for the remainder of this research.
The influence of preference for control on the relationship between EHRN and 5.1.4
health care expenses
40 due to already extensive information exchange, or due to the older adults’ experience with the old professionalism, resulting in less information exchange (Taylor & Hawley, 2010).
Managerial implications 5.2
The most substantial contribution of the research will be for managers and policy makers in health care, where its premise lies in possible cost reduction in older adults health care. To be more specific, it gives managers a clear indication what one of the manageable factors – i.e. the preference for healthy behavior – for health care expenses is. Furthermore, this paragraph aims to provide them with an advice on how to potentially cope with this specific patient preference below. Finally, this section provides health care professionals with an overview of what patient segments influence what costs, so that they can target these specific patient groups in their health care cost improvements plans. As only one of the preference for control aspects influences the ZVW in one occasion, it does not deserve any specific attention in the remainder of this research. The ZVW is however very interesting for further research, as it still accounts for a large part of the total costs of health care.
41 this specific situation, a large role should be reserved for the general practitioner (GP), as the GP is the gatekeeper for secondary care in the Netherlands, and he is the most utilized source of care among older adults in the Netherlands (Eissens van der Laan et al., 2014). Furthermore the GP already has a task for monitoring the health in his region, for which he is compensated (Van Weel, Schers & Timmersmans, 2012). The GP could help with identifying older adults that could change their healthy behavior preference. The exact content is outside of the scope of this research and should be part of change process itself, a prevention assistant, as can be seen in the Dutch dental care, might however be taken into consideration when looking for possible solutions for this problem (Bruers, Van Dam & Den Boer, 2008).
42
Limitations and future research 5.3
Just as most researches, this research has some limitations, in some of which lie further research directions. The main limitation of this research lies in the fact that it only involves people that actually made health care costs. Although these people are crucial in the search for a solution of the cost problem, this does not make the research a perfect representation of the reality. This representation was only possible using multilevel analyses, as the data were not normally distributed. However, that type of research fell outside the scope and scale of this master thesis. In this limitation lies the basis for the first future research directive. As the data are available, a multilevel research would paint a more accurate picture of the health care expenses in the older adults community, making the research more relevant for policy makers. Furthermore, as the data gathered for this research serve a practical purpose, the sample of this research was carefully constructed using stratified sampling, with the objective that all EHRN segments were represented in the research, which is beneficial for the internal validity. The disadvantage is that the sample will not be a perfect representation of the overall population of older adults, sacrificing the generalizability of this research. This is reinforced by the data cleaning as a result of which the sample decreased in size for the different measures. Also the data were gathered in three provinces in the Netherlands. The effect of this is that the external validity becomes lower, as the specific conditions for this research are hard to reproduce (Cronbach, Nageswari & Gleser, 1963). Therefore, comparable research in more geographic locations, with randomly selected older adults would be necessary to confirm the findings of this research. A final limitations lies in the preference for control component, of which only two measurement moments were available. To enhance the longitudinal nature, a follow-up study with more measurements over time would be beneficial.
