Balancing supply and demand for dementia care in the Netherlands
Roel van der Sanden*
Hayke Everwijn?
Etiénne Rouwette*
Jan Gubbels?
1 Methodology Department, Nijmegen School of Management, Radboud University Nijmegen, The
Netherlands
? Achmea, The Netherlands
Corresponding author:
Etiénne Rouwette
Methodology Department, Nijmegen School of Management, Radboud University Nijmegen, Thomas
van Aquinostraat 1.2.31
PO Box 9108, 6500 HK Nijmegen, The Netherlands
tel +31 24 3611468, fax +31 24 3611599, E.Rouwette@fm.ru.nl
Abstract
Health care in the Netherlands presents a unique mix of governmental and private responsibilities.
Costs for long-term care, expensive treatments and uninsurable care for the complete Dutch
population are covered by the Exceptional Medical Expenses Act (AWBZ). The administration of care
services under the AWBZ is the responsibility of 32 regional offices. In each region the insurer with
the highest number of clients is charged with planning and administration of AWBZ care services. At
present, health care services are provided by not-for-profit organizations such as hospitals and
nursing homes. Every health care provider operates under a contract with the regional administration
office and needs to stay within an estimated maximum capacity. Once contracted services are
available, insurers are obliged to reimburse providers for these services even if they are not used by
clients. In the coming years part of the Dutch health care will be deregulated and several types of
care will be offered under market conditions. Whereas costs for care capacity are at present
reimbursed by the government, this situation might change in the future. Regional care offices
generally have adequate insight into short term developments in demand and supply for care services.
Long term developments are however more difficult to predict. In this case a health care insurer
feared that current expansions of nursing home capacity might overshoot their target. In effect the
insurer wanted to know whether they were constructing nursing homes that would be left unoccupied
in a few years’ time. This paper describes a system dynamics study on demand and supply for a
specific type of nursing care, dementia, in a single care region. The model shows how feedback
between waiting lists and volume of different types of demand for care, leads to fluctuations in
required nursing capacity. The feared overshoot in long term nursing capacity did not materialize in
model runs under a range of environmental scenarios.
Introduction
Health insurance and health care in the Netherlands are to a great extent regulated by the central
government. Health care costs are covered by three categories of insurance that provide access to the
following services (Ministry of Health, Welfare and Sports, 2005):
1. The Exceptional Medical Expenses Act (Algemene Wet Bijzondere Ziektekosten or AWBZ)
which covers long-term care, expensive treatments and uninsurable care such as
institutionalized care for elderly or disabled people, home care, and psychiatric care. This form
of insurance covers all citizens of the Netherlands.
2. The Sickness Fund Insurance (Ziekenfondswet or ZFW) which covers acute medical care,
private health insurance and specific care for civil servants, including hospital admittance,
medical care, medical aids and pharmaceuticals, maternity care and dental care. This form of
insurance applies to youths under the age of 18, employed or self-employed persons with
lower incomes, certain groups of retired persons, and unemployed or welfare recipients.
3. Private Supplemental Insurance which Dutch citizens may opt to take for other health care
costs not covered under the first two categories.
Two thirds of the population are members of a sickness fund; one third carries private health
insurance. Direct patient payment and government subsidies play a minor role in covering health
expenditures.
AWBZ health care is administered in 32 regional offices. A regional office is managed by the health
care insurance company that insures the largest percentage of people in the region. These
organizational entities are called ‘care offices’. Care offices are responsible for balancing demand and
supply of AWBZ care, which is a challenging task. Health care providers contract with the care office
and are obliged to remain within an estimated maximum capacity. The care office reimburses a care
provider for its total capacity of available services, regardless of the extent to which services were
actually used by clients. At present balancing demand and supply of care services is primarily
important from the perspective of efficiently spending government money. In the coming years part of
the Dutch health care will be deregulated and several types of care will be offered under market
conditions. Whereas costs for care capacity are at present reimbursed by the government, this
situation might change in the future. In the situation when excess capacity is not reimbursed by the
insurer, care providers will be more reluctant to increase capacity if future demand is uncertain. For
the insurer on the other hand, capacity that is contracted under present regulations but is not used by
clients in the future will lead to costs that are not covered by client fees. At present different scenarios
for introducing market conditions in the Dutch health care sector are still being discussed. Even
without changing regulations, long-term developments in health care are difficult to foresee.
Understanding future developments of demand and supply is however necessary since decisions on
health care services have long term consequences. The trajectory from planning to deliverance of new
infrastructure for example takes about seven years. Traditional spreadsheet-based planning
techniques seem to offer little support for long term planning. Supply and demand of care services
are influenced by a lot of factors that are often interrelated. These factors come from a variety of
areas like demography, culture and public policy. Interrelations are often nonlinear and subject to
different time delays.
Achmea is a large health insurance company that administrates care offices in the Netherlands. In
December 2003 the company started a research project with the goal to increase understanding of
long term developments in health care supply and demand. The research project explored the
possible added value of system dynamics for long term planning of AWBZ care. In order to estimate
the added value of system dynamics, the project focused on supply and demand for a specific type of
care in a single care region. Dementia care was chosen as the topic of interest, since planning around
this type of care had been particularly difficult in the past. Dementia is a prevalent disease among the
elderly and accounts for 4.9 percent of Dutch health care costs (estimate for 1999, Polder and
Takken, 2002). In addition the number of people suffering from dementia is expected to increase by
40 percent between 2000 and 2020 (De Lange, Poos and Gijsen, 2004). The project was restricted to
the geographical area Kennemerland in the north west of the Netherlands. Achmea administrates the
care office in Kennemerland, which facilitates access to data. In the remainder of the paper we
describe previous applications of system dynamics to health care, the process of building the model,
the structure of the resulting model, baseline behavior, validity tests and policy experiments. We close
with conclusions and recommendations.
System dynamics and health care
Health care has been an active area of research in system dynamics, as is evidenced by a special
issue of System dynamics review on health and health care dynamics (Fall 1999), over 120 references
to health care in the most recent system dynamics bibliography and plenary presentations at last
year’s conference (Wolstenholme et al., 2004; Homer et al., 2004). Group model building projects on
problems related to health care are reported by Huz (1999), Royston et al. (1999) and Cavana et al.
(1999). In the Netherlands system dynamics has been applied to health care related problems by
Vennix et al. (1990), Bronkhorst et a. (1991), Post et al. (1992), Verburgh (1994), Vennix and
Gubbels (1994). Most of the Dutch studies focus at the national level. The single exception is the
study of Verburgh which builds on the conceptual model of national health care costs developed by
Vennix et al. The client in Verburgh’s study, a health insurance organization, adapts the model and
participates in policy experiments. The number of applications seems to suggest system dynamics can
offer useful insights into health care related problems. The present study contributes to research
conducted in the Netherlands by focusing at a regional level and one particular type of health care
service.
Is system dynamics a useful method in the situation introduced above? On the basis of criteria
formulated by Meadows (1980), Lyneis (1980), Post et al. (1992) and Vennix (1996) we feel that this
appears to be a the case:
- There seem to be feedback effects; policy makers involved indicated that demand increases when
the amount of beds for elderly increases.
- There are delayed relationships, e.g. the construction time for nursing homes;
- The model is aimed at long term prediction.
- There seem to be non-linear effects.
- Variables of multiple areas are taken into account in the model: examples are the processes of
building health infrastructure, assigning nursing care to patients and influences of waiting time on
decision makers.
- Qualitative variables play an important role; e.g. the action of decision makers will influence the
behavior of the model.
Besides checking if system dynamics is an appropriate method in this case, it is important to take a
closer look at the way system dynamics is used. Post et al. (1992) described a number of conditions
that improve the usefulness of constructed system dynamics models. These conditions are:
- make ‘small’ models;
- model a problem and not a system;
- from the start keep the implementation of the results into account;
- let the future user of the model participate in the model-building process;
- use the model not only for making predictions, but also for structuring a policy problem and
creating insight in this problem;
- provide good documentation of the model.
According to the researchers these conditions were met in this project. There are, however, some
things that can be improved, as will be revealed by the remainder of the paper.
Process of model construction
The study on supply and demand for dementia care in Kennemerland was conducted in two phases.
In the first phase, from J anuary to August 2004, a conceptual model was built. Information sources in
this part of the study were documents related to dementia care and expert interviews. On the basis of
research papers, policy reports, Internet publications and secondary data provided by Achmea a first
conceptual model was constructed. This model was then shown to experts in the field in a series of
interviews. Experts were chosen to represent the following fields: care offices, health care insurance
organizations, infrastructure, the Regional Assessment Boards (responsible for assigning clients to
different forms of AWBZ care), health care training, nursing homes, research and consultancy. The list
of interviewees was to include representatives from client organizations and the government.
However, in the time period of the research the experts who were contacted turned out to be unable
to participate. Experts’ comments were used to revise the model and complete the final report of this
phase of the research (Bergman and Everwijn, 2004). In October 2004 the experts involved in the first
phase were invited to a workshop, in which the report and revised conceptual model were presented.
This workshop was also used to present a preliminary quantitative model.
In the second phase of the project the quantitative model was revised on the basis of data provided
by the experts, additional documents and databases on health care services provided by Achmea.
Data from the expert group were gathered by using a workbook tailored to each participant's role in
health care planning. After using the information gathered with these workbooks a final presentation
was organized in which the participants could provide feedback on the final model structure and the
behavior of the model in various scenario’s. This resulted in the final report of the second phase of
this model building process (Van der Sanden, 2005) which was provided to Achmea to support them
in the use of the model.
Description of the causal structure of the model
In this section the causal structure of the model is described. We will also outline the the results of
the data collection of the most important variables. The model is divided into three partial models on
clients, infrastructure and personnel. Next, the main feedback loops in the model are described.
Clients
The first part of the model shows the flow of aged people who develop a need for AWBZ care. At the
left side of the model people diagnosed with dementia are flowing into the model. Dementia is a
slowly developing disease. Depending on the strictness of the criteria applied, the diagnosis ‘dementia’
is made at a different point in the development of this progressive disease. The incidence of dementia
is highly dependent on the age of people. When a person experiences problems due to dementia, he
or she can either choose to ask for care, or to try to care for him- or herself, eventually with help from
family or other informal care. In the first case, the person flows into the stock of self-supplying aged
people. In the second case the person makes a request for indication of AWBZ care. When this
request is approved, a person in the majority of cases receives care in a nursing home. However,
nursing home care is often scarce because of shortages in infrastructure (buildings, beds) and staff,
so people often are placed on a waiting list. Most of the time people receive temporary care to bridge
the time they are on the waiting list. When capacity is available, people are assigned to nursing home
care. When the nursing care ends, people flow out of the system. Mostly this happens when people
die.
When people place a request for AWBZ care, other care than nursing home care may also be
assigned. The distribution of nursing home care compared to other AWBZ care is an important
variable. It is influenced by the policy of the responsible Regional Assessment Board and by the
preferences of the clients. If the waiting time for care in a nursing home increases, clients’ perceived
waiting time increases and it grows less attractive to ask for this type of care. As a result fewer clients
will ask for nursing care and the distribution will change. Formally, the Regional Assessment Board
does not take the waiting lists into account in its decision, but there are indications that the perceived
waiting time does have an influence on the decision process of the boards.
There are a couple of important time delays. First, there is an information delay in the perception of
policy makers regarding the demand for nursing home care, which can cause planning problems.
Second, it takes time for the Regional Assessment Board to process a request. Also, when capacity is
available, it takes some time to formally assign a person to a nursing home.
Since the people in this system are relatively old, it must be taken into account that in every stock a
certain percentage of the aged people will pass away. This is displayed by the flows of mortality. The
model is shown in figure 1.
Figure 1. Partial model of client flow
During the data collection process only a number of variables could be filled with data directly (see
appendix 1). However, we were not able to quantify a lot of variables in the data sources available.
There was simply no data collected on the specific topics that we needed for our model. For these
variables we had to make some assumptions before we could use them in the model. In Appendix 2
you find the variables in the client flow with the assumptions we made. The main sources for out data
we could use directly were earlier studies from Timmermans and Woittiez (2004) and publications of
other organizations (Bureau Lindenhoek and the SCP). The main assumption we had to make for the
data which couldn't be used immediately, was the fact that elderly which are being treated for
dementia are (almost) equal to the people receiving psychogeriatric care. This assumption was based
on the fact that the RIVM reported that dementia compromises 93 % of the total amount of costs
made for psychogeriatric care. For the details of data used in this model we refer to appendices 1 and
2.
Infrastructure
In the second part of the model the processes regarding the construction of health care infrastructure
(buildings, beds and care facilities) are depicted. Based on the expected demand for health care
services, policy makers estimate an expected demand for care infrastructure. The discrepancy
between supply and demand of health care infrastructure determines the desired amount of
infrastructure to be realized. In its planning process the care office tries to take into account the
infrastructure that is under construction. However, estimations of when new volumes of infrastructure
become available is not always accurate. In this process of estimating the desired infrastructure to be
realized, an information delay is at work.
When additional infrastructure is taken into planning, it takes some time before the plans are
approved by the responsible policy makers. If approved, it takes on average seven years to construct
the infrastructure. Client demands might change over time and influence the perceived quality of
buildings. In the model a distinction is made between modern and aged infrastructure. Clients do not
prefer to live in aged nursing homes, and thus in some care regions waiting lists exist although (aged)
nursing homes are not used to full capacity. When technically depreciated, the infrastructure flows out
of the system.
Iie expectancy care
infastncture
time to old care uation outlow care
inrastruchue isirsineture
approval time
construction time
care infestrctue ueder |_ A supply new care supply old care
nimi infrastructure inirstuctue
approval plans on = completion care |_ 5. veroutering outfow care
‘( infastcture infrastructure ap ifrastrctre
desired change in care |
infastncture
perception delay
demand
perception total supply car’ total effective supply Tati new / oll care
discrepancy \inrastucture care infrastuchre infastncture
doce change in perception |
jemand - supply discrepancy demand -
supply |
Xe . ratio empty old care
infastctue
discrepancy expected demand
demand - supply cares gt
+ prediction people
aged 75+
infastuctue per
peson | ——————p» desired care for
+ people aged 75+
.
umber of people
aged 75+
Figure 2. Partial model of infrastructure
For this part of the model we could find a part of the variables directly in the data sources used (see
appendix 3). For other variables we had to make assumptions again (appendix 4). Data directly used
in the model most often came from experts in the field which were interviewed. The indirect data was
calculated with both data from a central database and knowledge from one of the experts.
Personnel
In the third part of the model the processes regarding the workforce in nursing home care are shown.
Based on the number of people receiving care in nursing homes and average productivity, the total
desired workforce is calculated. Depending on the current workforce there can be a shortage or an
excess of workforce. To resolve this discrepancy, employees can be hired or fired. Because of scarcity
on the labor market, it can be hard for nursing homes to employ appropriate workforce. This is
included in the model by the effect of total demand of workforce for other health care on the hiring
time.
Depending on workforce discrepancy of, employees perceive a certain workload. This workload has an
effect on the turnover percentage as well as on productivity. So when workload is high, the capacity
of personnel is negatively influenced in two ways.
number of weeks
averaged
standard
productivity
capaciy personel. ne ae
¢ productivity
productivity _
= +
perceived work ===
pressure poe personnel discrepancy
work pressire
ie
effect work pressure
total demand on productivity
effect total other demand
forcareonbirngtine = S
hiring time
inflow personnel I outlow personnel
quiting time
Figure 3. Partial model of personnel
In this part of the model we again had to make some assumptions for a number of variables while
others could be directly filled with data (appendices 5 and 6). Directly used data came from the
workbooks send to the experts, while assumed data was based upon the LISA-database.
Feedback
In the model, different feedback processes are at work. The most important loops are shown in figure
4. There are three balancing loops involving the waiting time for nursing homes.
If waiting time increases, perceived waiting time will also increase. As a consequence fewer clients will
place a request for nursing home care. As described above, it seems that the decision-makers at the
Regional Assessment Board informally adapt their assignments to the waiting lists. This leads to fewer
indications for nursing home care and consequently to a shorter waiting time. A longer perceived
waiting time will also have an effect on the behavior of aged people that receive other AWBZ care.
When the perceived waiting time is long, there will be fewer requests for nursing home care. In
effect, the waiting list will grow shorter, so over time perceived waiting time will decrease. The
perceived waiting time will also positively influence the time that aged people will be self-supplying.
As a consequence, fewer people will be assigned to nursing home care and waiting time will be
shorter.
Another set of feedback loops concerns the health care infrastructure. When there is a perceived
discrepancy between capacity and desired capacity, new infrastructure will be taken into planning.
With a material delay, this will lead to more capacity and the discrepancy will decrease.
Finally there are three feedback loops concerning supply of workforce. The effects of an increase in
workload cause two reinforcing loops. When workload increases too much, productivity decreases. By
consequence, the number of employees needed increases and so does the shortage in workforce.
This makes the workload increase. Another consequence of an increase in workload is the effect on
turnover percentage. More people will look for another job if workload is too high. This increases the
shortage in workforce and consequently increases workload. A final, self-evident balancing feedback
loop concerns the hiring and firing of workforce. Existence of a shortage of workforce will, over time,
lead to hiring of employees while an excess of workforce will lead to firing of employees. In both
cases the discrepancy will decrease over time.
Tunover
percentage
NL Toot dese
! watloe
3 “— 2
top) nn =
a \ pac anand —* Dacmpaney st
y yoflae itech
Neng eng of stn
wie Ne 4 ty
/ | ‘
orcived demand in le day
Noone LAs)
k 3) \ J
—— sf ey \ 4) State Piadh
_- metic an —
ee of pe da —
4 f salmemagempary —— Nuurbauae net
a / | ealth care
\ tap)
/ x fl
| nN Ave, wai me
Avg. perciwad.a~ ©
No of people assipod o otter
AWBZ carw makin requast for
‘change of indication
Figure 4. Feedbackloops
Baseline behavior
In this section we describe the behavior of the model under the condition that structural, policy and
contextual parameters do not change. Figure 5 shows the incidence of dementia from 2000 to 2020,
which is an exogenous input to the model. (ZOiS-database, 2004)
80
hei Batis person Wesk
Figure 5. Baseline Incidence of dementia (exogenous input)
Policy makers in health care planning often focus on the number of people on waiting lists and the
number of people in nursing homes as the main indicators for the field. Figure 6 shows the behavior
of these two indicators.
10
2000 2002 2004 2006 2005 2010 2012 2014 2016 OIE m0
jaa
sao oudurec op wack en overbraguorg : Bact, ——————— person
sue euderen fs werpieoginins : Basis ————————— pesos
Figure 6. Baseline waiting lists and number of people in nursing homes
The figures show that the increase in incidence rate does not directly result in an increase in waiting
lists and people in nursing homes. Two processes are responsible for this. First, the long waiting lists
keeps people at home (more people remain self-supplying) and the Regional Assessment Board is less
likely to refer people to nursing care. Second, the large construction volumes at the beginning of the
simulation run limit the waiting lists and increase the number of people in nursing homes from 2010
on. The volume of infrastructure is shown in figure 7.
200 400
700 300
400 200
09 200
800 °
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
ime jee
‘verwackos wrag morgbfecmucnas “Basis bed monsiin Bais bed
Figure 7. Baseline expected demand for care infrastructure and infrastructure under construction (in
number of beds)
Because the capacity of infrastructure is expanded in the first years of the simulation run, waiting lists
grow shorter (figure 6). Together with an increase in incidence, this leads to more requests for
nursing care. Over time waiting lists start increasing again as capacity cannot be increased
immediately, since planning and construction take a number of years.
Validation tests
This model has been subjected to the ten basic tests formulated by Forrester and Senge (1980): the
structure verification test, the parameter verification test, the extreme condition test, the boundary
adequacy test, the dimensional consistency check, the behavior reproduction test, the behavior
anomaly test, the behavior sensitivity test, the changed behavior prediction and the policy sensitivity
test.
11
The model passes most of the tests without any problems. Four tests deserve special attention. The
tests that are inconclusive are the structure verification test and the boundary adequacy test. In the
first expert workshop, some participants criticized both the structure of the model and the model
boundary. In response to these criticisms the model has been changed or choices have been
explained. The second workshop didn’t give much new criticism. So most feedback was dealt with in a
correct way according to the respondents.
The tests that cast some doubt on the model's validity were the parameter verification test and the
extreme condition test. This was caused by a known problem in system dynamics (Verburgh 1994;
Wolstenholme 2004): problems with the collection of the required data. The researchers have tried to
identify the likely range over which these parameter values could vary, and conduct a sensitivity test
on the basis of minimum and maximum values. For example, the incidence of dementia is a
parameter that is unsure and has large consequences for the behavior of the model. This uncertainty
comes from the disease of dementia itself. It is hard to say when people exactly have dementia,
because it’s a slowly surfacing disease. The difference in the data sources concerning the amount of
people that have dementia comes from this characteristic of the disease. This doesn’t improve the
validity of the model and needs to be kept in mind while looking at the results the model produces.
However with all the uncertainties in the parameter values, the behavior reproduction test gave a lot
of confidence in the model. The historic behavior was matched closely, which is shown in figures 8
and 9. The left part of figure 9 shows the historic and simulated waiting list that seems to develop
along a quite different path. However, in the opinion of several experts historic data on waiting lists
are suspect and it may be better to consider the delayed perception of the waiting list. This is
depicted in the right part of figure 9.
1,000 2,000
as
300
2000 2001 2002
jaa
ana] ondaren i verplooginizec : Basis ———————persoas_Pgrconms! wothznam in veplbeginsssorg Basis
RMOB beddenpatiamsa’: Basis ———— persons RMOB perveces!: Basis, ———— ban
Figure 8. Historical data and model behavior for number of people in nursing homes and personnel
12
200 er 2002 2003 (2000 2001 2002 2003
jee jen
anal eueren op wackukr en eatin Batis perceaz
MOB waht: Bass. parsax
Figure 9. Historical data and model behavior for waiting list and perceived waiting list
The extreme conditions test showed a problem with a part of the mode: the workforce. Due to the
adjustments made to the model in the quantification phase, the influence of this part of the model
was diminished so much the behavior shown at the extreme condition test was not plausible anymore.
When there is no personnel in nursing homes, the model still allows new patients to get assigned to
places in nursing homes.
While building the model the researchers chose not to build the model of the workforce in detail,
because it was so complex and not strictly necessary looking at the goal of the model. Why was this
part of the model not removed entirely from the model? This was done for two reasons: first it gives
insight into the workforce that is needed in different scenarios. Second it keeps the attention to an
important part of the real system and might give future researchers a good starting point for new
research.
All together it can be said that the tests provide enough confidence in the model to continue the
research with the model.
Policy experiments
With the model a number of policy experiments have been conducted. The experiments were a
combination of changes in policy with different developments of the environment. These experiments
were set up in consultation with the clients. The following table shows the combinations that have
been run with the implications to the model. We chose these combinations as they came out of our
consultation with Achmea as scenario’s which could give them additional insight into the system, With
these combinations the policy makers expected certain problems to appear. People at Achmea assume
that a proper way to deal with overcrowded nursing homes (for example because people in nursing
homes live longer) is to give them nursing care outside the nursing homes (i.e. at home) longer. With
the second policy-environment combinations (see table 1) it was assumed that there would be a large
overcapacity at the nursing homes, because the feedback loop that usually fills up empty spaces gets
less strong. The final combination is interesting, because the lower incidence leads to a lower demand
in nursing home care. It was assumed that the system would react more efficiently (less
overproduction) when construction time was reduced.
13
Policy Environment Changes to the model
People longer outside the nursing Standard Standard ratio:
care homes 0.1 > 0.09
No of people in nursing care homes | Standard ratio:
live longer 0.1 > 0.09
Average nursing time:
101 > 120
Build more health infrastructure Standard Extra inflow:
10*PULSE(261, 60)*
Less strong feedback effects | Extra inflow:
waiting time 10*PULSE(261, 60)*
Help:
13.2?
Shorter building time Standard Construction time:
364 > 290
Lower incidence of dementia Construction time:
364 > 290
Percentage incidence:
Till week 260 it stays the same.
After that it declines till 80 % of the
standard scenario.
Table 1. Overview of policy experiments
By running the above scenarios it was possible to give an answer to the question Achmea was asking:
“Are we building the empty nursing homes of the day after tomorrow?” Below the results for the first
scenario are shown.
261 780 100
sustal oudaces op waist /overbrugataong” bass —————— pecvous
susal ondacen op wachtict/overbrogsisapry’: Langer extramezal —— psccoos
“tl deen op wacktjat vertraggizancy’ Langer exramral~ ogareperscutas
' This “Pulse” -function produces from week 261 an extra inflow of 10 beds a week for 60 weeks. This gives a
total of 600 extra beds.
? In comparison to the baseline parameter setting, the feedhack effect is now reduced to .5. In the most extreme
circumstance (with no waiting time) the number of people assigned to nursing home care grows with only 10
percent.
14
Figure 10. Results of scenario ‘People longer outside the nursing care homes’ for number of people on
waiting list and people in nursing homes (1. baseline; 2. longer at home; 3. longer at home and live
longer)
400 2000
300 Le
200 1400
00 ue
‘ - or
261 780 1040 1 261 x 70 1040
m ‘im zachown : bask bed tomslasubed sorgnfestructar : basis ——————_________—. bed
- ‘in emaboum : Langer "7 bed tomalambed Lanse ! bed
sorapfartuctaz i eadsun Langer euramrzal— begets veblifidas —— ved tomalambed mrgofesmictar Langer euruarzal~ anaes vabiSdar — bed
Figure 11. Results of scenario ‘People longer outside the nursing care homes’ for infrastructure under
construction and total infrastructure (number of beds, 1. baseline; 2. longer at home; 3. longer at
home and live longer)
From figure 10 and 11 it appears that the policy of increasing extramural care (people stay at home
longer) is almost completely offset by an increase in life expectancy.
None of the scenarios shows long-term empty nursing homes. Even when we artificially build an extra
health infrastructure capacity of 600 beds, they are all occupied by the end of the simulation period.
The main reason for this is the feedback effect of waiting time on the duration of self supply, the
delay for request for change of indication and the ratio assignment nursing home care compared to
other AWBZ care. These influences make sure that when the waiting list gets short and empty homes
become a risk, the shorter waiting times make sure people who first took care of themselves now
send a request for an indication sooner. People who already receive AWBZ care will sooner ask for a
change of the received indication. This leads to a larger number of people requests an indication, so
more people will be assigned to nursing homes. In addition the shorter waiting lists will encourage the
Regional Assessment Boards to assign people to nursing homes. As described before in the case of
long waiting lists this is less likely to occur, because a long waiting list has the effect that people will
not get a place in a nursing home anyway. With a greater number of people with a request for
indication and a larger percentage of people being assigned to nursing homes, no nursing homes will
become empty. It needs to be said however, that definite data on the strength of the feedback effect
are not available. The scenario with the weaker feedback effect shows that short term emptiness can
occur. Nevertheless the historical behavior pattern suggests that the feedback effect exists and that
its strength is considerable.
A policy change that can cope with threatening empty nursing homes is shortening the construction
time. The model now responds slowly to changes in demand. So when it becomes apparent that there
15
is adequate capacity, it takes a long time before this is seen in the supply of health infrastructure.
With a shorter construction time the system responds faster to changes in demand, because the stock
‘health infrastructure under construction’ is smaller. This change in policy was also tested with a
decline in incidence of dementia by 20 percent and it responded in such a way that no emptiness in
nursing homes occurred.
Another way of coping with the uncertainty of the future number of people that needs nursing care is
to provide nursing care outside of the nursing homes. New technologies provide alternative ways of
doing this. The model shows that when less people get assigned to nursing homes by the indication
organization the pressure on the nursing homes gets considerably lower. That this policy will lead to a
reduction in problems infrastructural viewpoint does not mean it will be beneficial from other
perspectives. In this new form of nursing care personnel is needed as well, so this change in policy
can cause a huge pressure on the labor market. To look into this effect further an expansion of the
model is required.
All in all it can be said that it is highly unlikely that emptiness will occur with nursing homes. Within
certain scenario’s it is possible to create this emptiness, but with a shorter construction time these
problems will be dealt with. It needs to be said that Achmea can’t change this policy on its own, but it
needs to cooperate with other organizations in the field to get these results.
Conclusions and recommendations
During this project a number of problems occurred which make using system dynamics in this
situation more difficult.
First of all there was the problem with the collection of the data needed for running the model. In
summary, the use of system dynamics needs a different kind of data then the data currently collected
(which is needed for the current planning methods). This meant that a number of uncertain
assumptions had to be made while filling the model with data. The validity tests showed that despite
these uncertain assumptions there is reason for some confidence in the results of the model.
However, to make the use of system dynamics easier in the future the researchers recommend a
change in the sort of data that is collected. This will further increase the trust put into the constructed
models.
Secondly there were problems with defining the concepts used in the model. This problem is twofold.
First there are the changes in the health care system that makes the indication of nursing home care
decrease. To an increasing extent, intensive care for elderly is delivered independent of the home
people live in. Although there is a lot of discussion, professionals agree more and more on that
distinguishing in the delivery of care and living space increases quality and decreases costs.
There are also problems with definitions of concepts that influence the data. Different definitions lead
to different measures and different data as a consequence. An example is the difference in the
incidence of dementia that was found in two different sources. Another example is the existence of
16
self-supplying people that was disputed by different experts that were interviewed during the model-
building process.
During the model-building process choices had to be made about what to include in the model and
how to include it. These choices can negatively affect the trust people have in the results of the
model. However, this also emphasizes a great advantage of system dynamics: the effects of different
assumptions can be tested.
While describing the validity tests several recommendations were made about future research for
expanding the model. Expanding the following parts of the model would greatly enhance its
usefulness: taking other health care into account more detailed, and more precisely modeling the
workforce process.
However, there are other, maybe more important questions that can be addressed and which this
study cannot answer. One important issue is the quality of care. It would be interesting to know what
quality outcomes the different scenario’s would have, e.g. with varying waiting lists and varying
distribution of types of care. Another important issue concerns the costs of care. One of the criticisms
of the AWBZ care is that it is too expensive. Efficiency indicators are not included in the model. It
would be very interesting to be able to simulate different policy scenarios, e.g. to calculate what
distribution of types of care is the most cost efficient.
A final criticism; the implementation of the model in supporting planning at Achmea is somewhat
delayed. Therefore the model has not proven its value yet in supporting the planning process. It is
recommended that Achmea would appoint someone who will manage the future use of the system
dynamics models.
Literature
Barlas, Y., ‘Formal aspects of model validity and validation in system dynamics’, In: System dynamics
review, Vol. 12, no. 3, 1996, pp. 183-210.
Berg Jeths, A. van der, R. Gijsen (1998), ‘De gezondheidstoestand van ouderen: een epidemiologisch
overzicht’, de Psycholoog, (9), pp. 370-375.
Berg Jeths, A van der et. al. (2004), Ouderen nu en in de toekomst: Gezondheid, verpleging en
verzorging 2000 - 2020, Rijksinstituut voor Volksgezondheid en Milieu, Den Haag.
Bronkhorst EM, Wiersma T, Truin GJ. 1991. Using complex system dynamics models: an example
concerning the Dutch dental health care system. Proceedings of the 1991 International System
Dynamics Conference. Bangkok, Thailand, 155 - 163.
Calsbeek, H., L. Hingstman (2002), ‘Verpleging en verzorging samengevat’ In: Volksgezondheid
7
Toekomst Verkenning, Nationaal Kompas Volksgezondheid, RIVM: Bilthoven,
<http://www.nationaalkompas.nl> Zorg\Verpleging en verzorging, 18 maart 2002.
Campen, C. van, E. van Gameren (2003), Vragen om hulp, vraagmodel verpleging en verzorging,
Sociaal en Cultureel Planbureau: Den Haag.
Cavana RY, Davies PK, Robson RM, Wilson KJ. 1999. Drivers of quality in health services: different
worldviews of clinicians and policy managers revealed. System Dynamics Review 15(3): 331 - 340.
Coyle, G., D. Exelby, The validation of commercial system dynamics models’, In: System dynamics
review, Vol.16 No. 1, 2000, pp. 27-41.
Emans, B. (1990), Interviewen: theorie, techniek en training, derde druk, Wolters Noordhoff:
Groningen.
Everwijn, H., Bergman, C. (2004), Strategisch plannen van de zware zorg voor ouderen: een
systeemdynamicamodel van de processen rond vraag en aanbod in de verpleeghuiszorg, Radboud
Universiteit Nijmegen.
Forrester, J.W., P.M. Senge (1980), Tests for building confidence in system dynamics models’. In:
A.A. Legasto, J.W. Forrester, J.M. Lyneis (red.), TIMS Studies in the Management Sciences 14, pp.
209-228.
Francke, A.L. (2001), ‘Verpleeghuis- en verzorgingshuiszorg samengevat’. In: Volksgezondheid
Toekomst Verkenning, Nationaal Kompas Volksgezondheid. RIVM: Bilthoven,
<http://www.nationaalkompas.nl> Zorg\Verpleging en verzorging\Verpleeghuis- en
verzorgingshuiszorg, 5 december 2001.
Geurts, J.L.A., J.A.M. Vennix (1989), Verkenningen in beleidsanalyse : theorie en praktijk van
modelbouw en simulatie, Kerckebosch: Zeist.
Heeren, TJ. et al., ‘A community based study of survival in dementia’, in: Acta Psychiatric Scand,
1992a; 85: 415-418.
Huz S. 1999. Alignment from Group Model Building for Systems Thinking: Measurement and
Evaluation from a Public Policy Setting. Unpublished doctoral dissertation. SUNY: Albany, New York.
Lange J. de, MJJ.C. Poos , R. Gijsen, ‘Neemt het aantal mensen met dementie toe of af?’ In:
Volksgezondheid Toekomst Verkenning, Nationaal Kompas Volksgezondheid. Bilthoven: RIVM, 2004.
Linstone, H., M.Turoff (1975), The Delphi method: techniques and applications, Addison-Wesley:
Londen.
18
Lyneis, J.M. (1980), Corporate planning and policy design: a system dynamics approach, Pugh-
Roberts Associates Inc., Cambridge, Massachusetts.
Meadows, D.H. et al. (1972), The limits to growth, Universe Books: New York.
Meadows, D.H. (1980), The unavoidable apriori’, In: J. Randers (red.), Elements of the system
dynamics method, MIT press: Cambridge.
Meulen-Arts, S. van der (2002), ‘Wat is woonzorg?’ In: Volksgezondheid Toekomst Verkenning,
Nationaal Kompas Volksgezondheid. RIVM: Bilthoven:, <http://www.nationaalkompas.nl>
Zorg\Verpleging en verzorging\Woonzorg, 23 september 2002.
Meulen-Arts, S. van der en A.L. Francke (2002), ‘Wat is verpleeghuis- en verzorgingshuiszorg?’ In:
Volksgezondheid Toekomst Verkenning, Nationaal Kompas Volksgezondheid, RIVM: Bilthoven,
<http://www.nationaalkompas.nl> Zorg\Verpleging en verzorging\Verpleeghuis- en
verzorgingshuiszorg, 28 mei 2002.
Post, D., J.A.M. Vennix et al. (1992), De gezondheidszorg in model : ons complex zorgsysteem
gemeten en gewogen, Bohn Stafleu van Loghum, Houten / Zaventem.
Polder, J.J., et al. (2002), Kosten van ziekten in Nederland. De Zorgeuro ontrafeld, Themarapport van
de Volksgezondheid Toekomst Verkenning, RIVM, Bilthoven.
Rouwette, E. A.J. A., Vennix, J. A. M., & Mullekom, T. v. (2002). ‘Group model effectiveness: A review
of assessment studies’, In: System Dynamics Review, 18(1), pp. 5-45.
Richardson, G.P. (1991), Feedback thought in social science and systems theory, University of
Pennsylvania Press: Philadelphia.
Richardson, G.P, D.F. Andersen (1995), Teamwork in group model building, System Dynamics Review
11 (2), pp. 113-137.
Richmond, B., ‘The strategic forum: aligning objectives, strategy and process’, In: System dynamics
review, Vol. 13, no. 2, 1997, pp. 131-148.
RIVM (2004), Volksgezondheid Toekomst Verkenning, Nationaal Kompas Volksgezondheid, RIVM:
Bilthoven <http://www.nationaalkompas.nI> versie 2.7, 19 mei 2004.
Royston G, Dost A, Townshend J, Turner H. 1999. Using system dynamics to help develop and
implement policies and programmes in health care in England. System Dynamics Review 15(3): 293
- 313.
Ruysbroek, J.M.H. (2001), ‘Aan welke ziekten en aandoeningen wordt het geld besteed?’ In:
Volksgezondheid Toekomst Verkenning, Nationaal Kompas Volksgezondheid, RIVM: Bilthoven,
19
<http://www.nationaalkompas.nl> Zorg\Verpleging en verzorging, 24 september 2001.
Ruysbroek, J.M.H. (2002) ‘Wat is verpleging en verzorging?'In: Volksgezondheid Toekomst
Verkenning, Nationaal Kompas Volksgezondheid. RIVM: Bilthoven,
<http://www.nationaalkompas.nl> Zorg\Verpleging en verzorging, 27 mei 2002.
Sterman, J.D. (2000), Business dynamics: systems thinking and modeling for a complex world,
Irwin/McGraw-Hill: Boston.
Timmermans, J.M., |. Woittiez (2004) Verklaringsmodel verpleging en verzorging:
onderzoeksverantwoording, Sociaal en Cultureel Planbureau: Den Haag.
Timmermans, }.M., |. Woittiez (2004) Advies ramingen verpleging en verzorging, Sociaal en Cultureel
Planbureau: Den Haag.
Vennix, J.A.M., ‘Building consensus in strategic decision making: insights from the process of group
model-building’, In: Group decision and negotiation, 4 (4), 1995, pp. 335-355
Vennix, J.A.M. (1996), Group model building : facilitating team learning using system dynamics,
Wiley: Chichester.
Vennix, J.A.M., E.AJ.A. Rouwette, C.M. Thijssen, Systeemdynamica: methode voor strategie-
ontwikkeling, interne uitgave Faculteit der Beleidswetenschappen aan de Katholieke Universiteit
Nijmegen: Nijmegen.
Verburgh, L. D. (1994), Participative policy modelling: applied to the health care insurance industry,
Benda, Nijmegen.
VWS (2003g), Zorgnota 2003, Ministerie van Volksgezondheid Welzijn en Sport: Den Haag.
Wolstenholme, E., et al. (2004), Using system dynamics in modelling health and social care
commissioning in the UK, OLM Consulting, Middlesex.
Zorgkantoor Kennemerland, J aarverslag 2002, Achmea.
http://www.achmea.nl/upload/20040416/achmea% 20corp% 20Brochure_3.pdf, 20-12-2004
Jaarverslag Achmea holding N.V.
2001,http://www.achmea.nl/upload/20030530/Achmea% 20} aarverslag% 202001_3.pdf, 20-12-2004
20
Appendix 1
Variabel Source Value
No. Of people 75+ IPB oktober 2002 (uit: Zorgweb Graph, pictured in figure 2
Haaglanden, Bureau Lindenhoek,
2004)
standard delay request for change SCP-rapport “Vragen om hulp” 104 weeks
of indication (2004)
Standard ratio Verpleging en Verzorging verklaard, | 0.1 (=10 %)
2004, Timmermans & Woittiez,
Tabel 2.1, Blz 12
Assignment percentage Verpleging en Verzorging verklaard, | 0.95 (=95 %)
2004, Timmermans en Woittiez
Delay of indication Congress “Indicatiestelling op de 4 weeks
helling”
Table 1: Variabels in the client flow directly filled with data
35000
34000 oe
33000
32000 ae
31000 eal
30000 a
29000
28000
27000
2000 © 2005S 201020152020
Figure 2: No. Of people 75+ in the Kennemerland-region
Appendix 2
Variabel Source Value Assumption
No of people with dementia | Centraal Administratie | 835 elderly No. PG? beds in VPH? in Kennemerland in
in nursing homes (in 2000) Kantoor (uit: ZOIS 2003 is 877 (of a total of 1520 VPH
database) beds). No. beds in 2000 is 1448. If the
share of PG patients stays the same, this
means: 835 beds.
Life time expectancy self CBS 330 weeks The life expectancy of pleople with
supply care bevolkinsstatistiek, dementia is 1,9 times as low as that of
Heeren e.a., 1992a those without dementia. (Heeren e.a.,
1992a). Of the people over 90 years old
only 10 % lives at home (Piet van der
Lei) Of the people over 75 years old 90
% lives at home. So the life expectance
of this age group is used. This is: 12,074
years (CBS). For people with dementia
this means 6,35 years.
Lifetime expectation Arcares, CBS 220 weeks Average caring time is 3,71 years for
VZH> (Arcares). Also people of 75-79
years old (corrected for dementia live
4,71 years (CBS). Life expectancy will be
in this range (in a VZH are the relative
heavier cases of dementia which live
3 Psychogeriatric
“ Nursing homes
° Less intensive nursing homes
21
relatively short). The average life
expectancy of these two groups is 220
weeks.
Avarage nursing time
Arcares, RIVM
101 weeks
The average nursing time for
psychogeriatric care is according to
Arcares 710 days. We assume
psychogeriatric care equals care for
dementia, because dementia makes up
93 % of the total amount of costs for
psychogeriatric care (RIVM).
No of people on waiting list
and receiving temporary
health care (in 2000)
Hoeksma, Homans &
Menting (HHM)
260 elderly
Tota waiting list for VPH is 450 elderly in
2000. In a VPH 877 people out of 1520
are PG. 877/1520*450 = 260. Again the
assumption PG = dementia.
Table 2: variable in client flow with assumptions
Appendix 3
Variable Source Value
Contruction time Interviews Piet van der Lei and Marcel 364 weeks
Garritsen
Average lifetime care http://www.woonzorg.nl/publicaties/media | 2600 weeks
infrastructure /corpbrochure2003. pdf
average aging time health Interview Piet van der Lei, and 1300 weeks
infrastructure
www.woonzorg.nl
Expected no. Of people 75+
IPB oktober 2002 (from: Zorgweb
Haaglanden, Bureau Lindenhoek, 2004)
Graph, seen in figure 2.
Approval delay Interviews Marcel Garritsen and Piet van 52 weeks
der Lei
Percentage of aged infrastructure | workbooks 0%
being uninhabited
Table 3: variables in infrastructure model directly filled with data
Appendix 4
Variable Source Value Assumption
Supply of old and new care Centraal Administratie | New: 417 beds The total supply of infrastructure in 2000
infrastructure (in 2000) Kantoor, interview Old: 418 beds was 835. Mr. Van der Lei said at the
Piet van der Lei
moment about half of the infrastructure
is old and half is new.
Table 4: Variable in infrastructure model with assumptions
22
Appendix 5
Variable Source Value
Standard hiring time Workbooks 10 weeks
Quiting time Workbooks 8 weeks
Table 5: Variables in personnel model directly filled with data
Appendix 6
Variable Source Value Assumption
Standard productivity ETIN/ Landelijk 0.56 elderly per The division of jobs over PG en SOM®
InfoSysteem person in the care in nursing homes is equal to the
Arbeidsplaatsen workforce division of beds. This means in 2003
(LISA) 1520 persons in the workfoce took care
for 877 patients.
Personnel in nursing homes | ETIN/ Landelijk 1344 people In 2003 there were 2330 people
(in 2000) InfoSysteem employed in VPH. We assumed PG and
Arbeidsplaatsen SOM patients required the same amount
(LISA) of personnel. Therefore we assume
877/1520 * 2330= 1344 people were
employed to nurse the PG patients
Table 6: Variables in personnel model with assumptions
° Somatic
23
