Sunteți pe pagina 1din 29

Health Care Manag Sci (2012) 15:385412

DOI 10.1007/s10729-012-9204-0

Modeling the demand for long-term care services


under uncertain information
Teresa Cardoso & Mnica Duarte Oliveira &
Ana Barbosa-Pvoa & Stefan Nickel

Received: 23 November 2011 / Accepted: 31 May 2012 / Published online: 11 July 2012
# Springer Science+Business Media, LLC 2012

Abstract Developing a network of long-term care (LTC)


services is currently a health policy priority in many
countries, in particular in countries with a health system
based on a National Health Service (NHS) structure. Developing such a network requires proper planning and basic
information on future demand and utilization of LTC services. Unfortunately, this information is often not available
and the development of methods to properly predict demand
is therefore essential. The current study proposes a simulation model based on a Markov cycle tree structure to predict
annual demand for LTC services so as to inform the planning of these services at the small-area level in the coming
years. The simulation model is multiservice, as it allows for
predicting the annual number of individuals in need of each
type of LTC service (formal and informal home-based,
ambulatory and institutional services), the resources/services that are required to satisfy those needs (informal caregivers, domiciliary visits, consultations and beds) and the
associated costs. The model developed was validated using
past data and key international figures and applied to Portugal at the Lisbon borough level for the 20102015 period.
Given data imperfections and uncertainties related to predicting future LTC demand, uncertainty was modeled
through an integrated approach that combines scenario
T. Cardoso (*) : M. D. Oliveira : A. Barbosa-Pvoa
Centre for Management Studies of Instituto Superior Tcnico,
Universidade Tcnica de Lisboa,
Avenida Rovisco Pais 1,
1049-001 Lisbon, Portugal
e-mail: teresacardoso@ist.utl.pt
S. Nickel
Discrete Optimization and Logistics, Karlsruhe Institute of
Technology, Institute of Operations Research,
Kaiserstr. 12,
76131 Karlsruhe, Germany

analysis with probabilistic sensitivity analysis using Monte


Carlo simulation. Results show that the model provides
information critical for informing the planning and financing of LTC networks.
Keywords Long-term care . Demand . Markov model .
Simulation . Uncertainty . Portugal
Abbreviations
AAGR Average annual growth rate
AC
Ambulatory care
ADL
Activity of daily living
CD
Chronic diseases
CI
Confidence intervals
CS
Chronic patients with symptoms
CWS
Chronic patients without symptoms
D
Dependent individuals
FHC
Formal home-based care
GDP
Gross domestic product
IADL
Instrumental activity of daily living
IC
Institutional care
IHC
Informal home-based care
LOS
Length of stay
LTC
Long-term care
NC
Non-chronic individuals
ND
Non-dependent individuals
NHS
National health service
NHSur National health survey
NS
Non-single individuals
NVLI
Not very low income
OC
Other causes
PE
Point estimates
PSSRU Personal social services research unit
RNCCI Rede nacional de cuidados continuados
integrados

386

S
TC
VLI

T. Cardoso et al.

Single individuals
Total cost
Very low income

1 Introduction
Long-term care (LTC) comprises a set of non-medical and
medical services delivered to individuals who have lost
some capacity for self-care due to chronic illness or disability [1]. This type of care aims at ensuring that an individual
who is not fully capable of long-term self-care can maintain
the best possible quality of life, with the greatest possible
degree of independence, autonomy, participation, personal
fulfilment, and human dignity[1].
Developed countries have been establishing LTC programs within the scope of health and welfare systems, and
many developing countries are currently in the initial stages
of development of this type of program [2]. This is the case
in Portugal where a National Network of Long-Term Care
Services (Rede Nacional de Cuidados Continuados Integrados, RNCCI) has been implemented since 2006 [3]. Nevertheless, there is no single paradigm for the organization and
provision of LTC services [2]: different countries are in
different stages of economic development and consequently
can afford packages of services with different levels of
comprehensiveness; countries have diverse sociodemographic and epidemiological patterns, cultures and
values, giving different importance to various types of
LTC services; and services are provided at different costs
in different contexts. One of the most important dimensions
of LTC delivery is the setting in which these services can be
provided [2]: i) in home-based settings (provided either by
formal caregivers, such as nurses, doctors or social workers;
or by full-time or part-time informal caregivers, such as
family members and friends); ii) in ambulatory settings;
and iii) in institutional settings.
Available evidence shows that the organization of LTC
delivery clearly varies across countriesfor instance, in
Sweden most families do not feel obliged to provide informal care, whereas this is not the case in most Mediterranean
countries (such as Italy and Spain) where informal care play
a key role [2]. Additionally, in Northern Europe institutional
care make up to 12 % of the total provision of care, whereas
in Southern Europe much lower levels of institutional care
are typically offered (e.g., around 3 % and 4 % of older
people in Portugal and Italy receive institutional care, respectively, while in Greece this figure has been reported to
be well below 1 %) [4, 5]. As a result, many Southern
European countries are expanding the formal delivery of
care, while Northern European countries are targeting a
reduction in the number of existing LTC beds while promoting home-based care [4].

Furthermore, it is widely recognized that the population


needs for LTC are changing on a global scale, being highly
influenced by factors such as [1]: an increasing prevalence of
chronic diseases that are becoming the leading cause of disability; an increasing proportion of elderly people in developed countries; and a growth in the participation of women in
the labor market. These factors are expected to contribute to
both an increase in the demand for LTC services and in the
proportion of the gross domestic product (GDP) spent on
LTC. According to Comas-Herrera et al. [6, 7], the proportion
of GDP spent on LTC is projected to more than double
between 2000 and 2050 in some countries (e.g., 145 % increase in the UK and 155 % increase in Portugal). In this
context, the development of a network of LTC services ranks
highly on the health policy agenda of many countries. In
particular, in countries with a National Health Service
(NHS) structure the development of such a network requires
proper planning and information on the future demand for
LTC services. Nevertheless, this information might not be
available [8], as observed in the Portuguese health system.
Previous studies have proposed methods to predict future
demand for LTC, with most of them making use of information on current levels of utilization. This is not always a
desirable feature since utilization data might be highly constrained by the supply of services. According to Roberfroid
et al. [9], when the current supply of services is inappropriate for meeting all the associated needs, any approach based
on the current utilization of services is unsuitable. This is
often the case for LTC supply since available supply might
far from meet the expected demand for services (e.g., in
Portugal [10]). Accordingly, for many countries, it is useful
to develop alternative methodologies for predicting LTC
demand that depart from a needs assessment. Nevertheless,
one should note that converting those needs into indicators
of demand should also account for supply-side related factors. For instance, the predefined choice of an LTC delivery
model (which might be informed by evidence on the past
supply of services) and aspects related with the interaction
between demand and supply might be accounted for.
In addition, previous studies used scenario analysis to
study the sensitivity of LTC demand to changes in assumptions on demography and on levels of dependency (among
others). Nevertheless, these studies did not consider, within
an integrated approach, other important uncertainties that
might influence LTC demand, such as the uncertainty that
surrounds key determinants of LTC demand (e.g., uncertainty of future incidence rates for chronic diseases). Moreover,
an analysis of the literature in the area has shown that few
studies have applied methods to the small-area level, which
is an important level for health care planning (for instance,
health care determinants are known to vary geographically).
This study develops methods for predicting the demand
and informing the planning of an LTC network. It builds up

Modeling the demand for long-term care services

a simulation model based on a Markov cycle tree structure


so as to predict the annual number of individuals in need of
each type of LTC service (formal and informal home-based,
ambulatory and institutional services) and the resources/
services (informal caregivers, domiciliary visits, consultations and beds) that are required to satisfy those needs and
associated costs. The model is comprehensive as it takes
into account the population needs for LTC by considering a
wide range of health and socio-economic-related characteristics, instead of making use of information on the current
levels of utilization (utilization data is often not available or
might be highly influenced by services supply). Those needs
are then converted into indicators of demand for services
and resources based on a predefined LTC delivery model.
By considering a wide range of health and socio-economic
characteristics, the proposed model provides results with a
higher level of detail when compared with previous studies
(e.g., it gives information on the influence of each type of
chronic disease on the demand for LTC).
Within the developed simulation model, uncertainty is
dealt with through the use of a Monte Carlo simulation
integrated approach that combines scenario analysis with
probabilistic sensitivity analysis. To the authors best
knowledge, an integrated approach for analyzing uncertainty has not yet been used in the LTC literatureprevious
studies have mostly used scenario analysis as the preferred
approach for analyzing the uncertainty surrounding predictions for LTC demand. The developed model is applied to
Portuguese context (for which there is a critical lack of data
on LTC demand) and used to predict demand at the smallarea level.
The remainder of this paper is structured as follows. A
literature review on the methods used to predict LTC demand is provided in the second section. In the third section
the simulation model developed is presented along with a
description of the integrated approach used for modeling
uncertainty. Afterwards, the results obtained are presented
and discussed. In particular, the results from applying and
validating the model to the county level in Lisbon are first
presented, followed by the results obtained by applying the
model to each borough from the Lisbon county. A comparison between predicted demand and current supply is also
provided in the results section. The final section draws some
conclusions and possible lines for future work.

2 Literature review
Predicting demand for LTC services is essential when planning a network of services able to meet the populations
needs for care. According to Chung et al. [11], simulation
models represent the most widely used technique in this
area. To our knowledge, the first models ever developed to

387

predict LTC demand were carried out by Lane et al. [12] and
by Rivlin and Wiener [13]. In 1985, Lane et al. proposed a
Markov model to predict the transition of patients through a
variety of LTC services. A few years later, Rivlin and
Wiener proposed the Brookings-ICF Long-Term Care Financing Model, which projected, among other figures, the
number of elderly individuals who will need to enter a
nursing home per year during the 19862020 period. A
decade later, Robinson [14] proposed a different Markov
model to determine the LTC demand in the US using information collected from national LTC surveys reporting the
level of disability and utilization of care of elderly Medicare
enrollees.
One of the most comprehensive, policy-relevant and
widely known simulation model developed is the Personal
Social Services Research Unit (PSSRU) model. This model
was proposed by Wittenberg et al. [15, 16] and was constructed as part of a PSSRU long-run project at the London
School of Economics. One of its aims was to predict levels
of LTC demand, departing from current patterns of utilization in England, and disaggregated by type of service. This
model was then regularly updated and expanded [17, 18]
and several authors have also adapted and applied it to
different contexts [11, 1922], with some being used at the
small-area level [2325]. Other applications to predict LTC
demand at the small-area level rarely appeared in the literature. Exceptions are the works of Xie et al. [26, 27], which
built a Markov model to study the pattern of length of stay
(LOS) for elderly people that flow within and between
different types of LTC services. This model was applied to
the London borough of Merton and was recently updated
and applied to 26 London primary care trusts with the
objective of determining how many patients are expected
to stay in each service and for how long [28]. The PSSRU
model was also employed to investigate trends of LTC
expenditure and demand in four European countries (Germany, Italy, Spain and the UK) [6, 7]. For that purpose, four
adapted models were used to obtain projections on the LTC
demand by people aged over 65 over the period 20002050
(with sensitivity analysis carried out on some model
assumptions related to demography, dependency, informal
care, formal care and cost indicators). The Modelling Ageing Populations to 2030 (MAP2030) research group also
used the PSSRU model to investigate (among other things)
how different trends among drivers of demand for LTC
might affect future demand [29].
Several authors have developed alternative simulation
approaches to the PSSRU model. These differ mainly either
by considering the preferences of individuals in need (e.g.,
Hostetler [30]), by using different dependency measures
(e.g., Karlsson et al. [31]), or by using frailty along with
the level of dependency as a proxy for the need for LTC
(e.g., Zuttion et al. [32]). Other approaches have dealt with

388

specific contexts and project objectives in LTC, for example: Leung [33] developed a Markov model to project the
LTC demand in Australia; the Economic Policy Committee
and the European Commission proposed a simulation model
to investigate the future LTC demand for all the member
states of the EU [34]; and Lagergren [35, 36] developed the
ASIM III (ldre Simulering, which means Elderly Simulation) model to forecast the likely consumption of LTC
services in Sweden.
In addition to these studies, other recent European projects have shown the policy relevance of developing research
into LTC demand. Some examples are the Future Elderly
Living Conditions in Europe (FELICIE, 20032006) project
[37], which was centered on a 20002030 forecast of LTC
demand in nine European countries; the INTERLINKS
project (20082011) [38], a Europe-wide project that
resulted in a framework for LTC that provided tools for
the assessment of the LTC demand in several European
countries (among other things); and the Assessing Needs
of Care in European Nations (ANCIEN, 20092012) project
[39], which is still ongoing and is concerned with the future
of LTC for the elderly in Europe, in particular with the
question How will need, demand, supply and use of LTC
develop?. It is worth noting that the demand for LTC in
Portugal was addressed within the FELICIE project, and is
currently being assessed within the ANCIEN project.
Although different in several aspects, all the models
described above have some common features that are summarized in Table 1 (note that some of the characteristics
used to compare studies could not be analyzed, e.g., Lane et
al. [12] did not carry out any type of uncertainty analysis).
Analysis of these studies shows that: i) they based their
predictions on information about the current (or past) levels
of service utilization; ii) most of them explored the influence
of different scenarios on demand and present estimates
concerning the costs of providing LTC; iii) most studies
have used activities of daily living (ADLs) and/or instrumental activities of daily living (IADLs) as a proxy for
individuals level of dependency; and iv) the type of LTC
services under analysis has not been the same across the
studies, since it depends on the organizational structure of
the current LTC system in each country (as discussed in the
Introduction section).
In this article we present a simulation model that differs
from the reviewed studies in several ways. First, it does not
rely on the demographic extrapolation of current levels of
service utilization. Instead, the proposed model relies on
health and socio-economic characteristics of the population
that are relevant for defining the need for LTC. Accordingly,
as discussed above, this feature makes the proposed simulation model suitable for application in countries with both
appropriate and inappropriate levels of supply. Secondly, by
considering a wide range of health and socio-economic

T. Cardoso et al.

characteristics of the population, our model provides results


with a higher level of detail and with potential relevance for
the organization, planning and finance of the LTC sector.
For instance, the model gives information on the influence
of each type of chronic disease on the LTC demand, and also
on the income distribution among those needing LTC. It also
provides information on how demand for LTC varies across
age and gender groups. To our knowledge, previous studies
report predictions for the older age groups, but not for the
younger groups. Thirdly, an integrated approach for assessing different types of uncertainty that might influence these
predictions is usedin particular, our model uses scenario
analysis combined with a probabilistic sensitivity analysis to
analyze the uncertainty surrounding key parameters and the
impact of major assumptions on model results. This appears
to be a more comprehensive approach in comparison to the
scenario analysis used in previous studies (including the
PSSRU model).
Some other application-related features of the simulation
model proposed in this article should be noted. Specifically,
the proposed model is applied to Portugal, for which there is
a critical lack of data on the demand for LTC. Although two
European projects were/are expected to predict LTC demand
for Portugal, so far only the FELICIE project has been
concluded, providing some global estimates for the demand
of individuals aged over 75 by gender and marital status.
These demand forecasts, however, do not consider the impact of chronic diseases on future demand and have as
reference the national geographic level [37]. The model
proposed in this article predicts demand disaggregated by
health and socio-economic characteristics and to the smallarea level. Applications at this level have not been typical in
the area of LTC demand prediction and require a prior
collection of demographic information (e.g., number of
inhabitants disaggregated by age and gender) for small areas
(such as boroughs), and thus small samples need to be used
to compute the probabilities of individuals belonging to
different groups. Consequently, the estimation of model
input parameters requires special attention related to the
use of small samples since it could lead to inaccurate estimations [40]. A Monte Carlo simulation was considered to
be appropriate in this context, as it can handle large uncertainties in input parameters [41]. Even if Monte Carlo simulation is widely used in health care research, most studies
predicting demand for LTC have not used it.

3 Methodology
This section provides a comprehensive description of the
simulation model developed to predict future LTC demand.
The general structure of the model is first described. Afterwards, the data set used for applying the model along with a

Markov model
Markov model

Monte Carlo
simulation

Robinson [14]
Leung [33]

Rivlin and Wiener


[13]

Difficulties in performing ADLs

Measure of dependency

Macrosimulation Age and gender


model

Microsimulation
model

Karlsson et al. [31]

Hostetler [30]

Age

Macrosimulation Age, gender and marital


model
status

Lagergren [35, 36]

Scenario
analysis

Difficulties in performing ADLs

Problems related to locomotion, reaching and stretching, Scenario


dexterity, personal care, continence, seeing, hearing,
analysis
communication, behavior, intellectual functioning,
consciousness, eating, drinking and digestion
Scenario
analysis

Problems related to functional capacity, mobility,


incontinence, anxiety and dementia

Scenario
analysis

Scenario
analysis
Scenario
analysis

Difficulties in performing ADLs and IADLs


Difficulties in performing ADLs

Scenario
analysis

Scenario
analysis
Scenario
analysis

Institutional
services

1995-2031

1986-2020

50-year span

1978-1982

Time period

Pennsylvania

50-year span

Until 2040

2000-2030

2001-2051

Germany, Italy,
2000-2050
Spain and UK
All the Member
2004-2050
States of the EU

England

US

Institutional
Italy
and noninstitutional
services
Institutional
Sweden
and noninstitutional
services
Informal and UK
formal care

Institutional
and noninstitutional
services
Institutional
and noninstitutional
services
Informal and
formal care
Informal and
formal care

Institutional
British Columbia
and noninstitutional
services
US
Australia

Uncertainty Application domain


analysis
Services
Country

Difficulties in performing ADLs and IADLs

Difficulties in performing activities related to


self-care, mobility and communication
Age, gender, marital status, Difficulties in performing ADLs and IADLs
income and assets

Age and gender


Age and gender

Age and gender

Level of disaggregation

Macrosimulation Age, gender, dependency,


model
marital status, household
composition and housing
tenure
Comas-Herrera et al. Macrosimulation Age and gender
[6, 7]
model
Economic Policy
Macrosimulation Age and gender
Committee and
model
European
Commission [34]
Zuttion et al. [32]
Macrosimulation Age, gender, household
model
composition and housing
tenure

Markov model

Lane et al. [12]

Wittenberg et al.
[15, 16]

Methodology

Authors

Simulation model

Table 1 Simulation models proposed in the literature to predict future demand for LTC

Modeling the demand for long-term care services


389

390

brief description of the LTC system currently operating in


Portugal, are presented. The approach used for modeling
uncertainty is explored at the end of the section.
3.1 Model structure
The simulation model was built to predict future LTC demand for a planning horizon divided into annual periods.
The model is multiservice as it allows for predicting the
annual number of individuals in need of the following types
of LTC services: formal and informal home-based, ambulatory and institutional services. According to Coyte et al.
[42], institutional and home-based care (both formal and
informal) represent the main typologies of LTC services.
Yet, as noted by Evashwick [43], ambulatory care provided
in formal outpatient units (e.g., in hospitals) is also an
important LTC component for those requiring ongoing monitoring for chronic conditions and not requiring institutional
or home-based care. Accordingly, these three types of
service are considered in our study so as to ensure the
continuum of care. In addition, the model also provides
information on the resources/services to be provided for
the individuals in need (informal caregivers, domiciliary
visits, consultations and beds) and the costs associated with
health care delivery.
The proposed simulation model has a Markov cycle tree
structure, including a short-term decision tree and a longterm Markov model [44, 45]. This structure is particularly
well suited to modeling key features of the demand for LTC
(see [46, 47] on the key features of Markov models), namely: individuals need to be organized into cohorts (i.e., groups
of individuals sharing similar characteristics); individuals
from different cohorts do not interact between themselves;
individuals naturally progress to other states/cohorts according to epidemiological patterns, and there is a repetition of
epidemiological patterns over time (this precludes the use of
decision trees [46, 47]); also, demand for care should be
disentangled for several subtypes of chronic diseases,
including ischemic heart diseases, oncological diseases,
cerebrovascular diseases, arthritis, chronic obstructive pulmonary disease, dementia and depression [48], and these
subtypes of care should be accounted for in a long-term
perspective [49]; and, as explained below, the use of a
Markov structure with two levels seems adequate to understand the dynamics of the model. The model allows the
demand for LTC services to be predicted, using as a starting
point an assessment of the absolute need for care. This need
is considered to depend on a wide range of health and socioeconomic characteristics of the population (and not on
supply-related factors such as prices and the current supply
of services), this being critical to inform which population
groups have the greatest need and should be targeted by the
health care system. Also, the model informs on the

T. Cardoso et al.

influence of different types of chronic diseases on the


demand and on the groups that might not be able to afford
to pay for the services. Based on the literature in this area,
the following set of health and socio-economic characteristics are considered within the model (to our knowledge,
no study has comprehensively considered all these factors
simultaneously):
a) Age and genderthe proportion of people in need of
LTC rises markedly with age and differs between men
and women [15]. For instance, according to the literature, women at older ages typically have higher levels of
demand than men of the same age group, largely because
of the larger number of women in the eldest age groups
(as a result from womens larger life expectancy). On the
other hand, previous studies have not reported the
behavior of LTC demand at younger ages;
b) Type of chronic diseaseschronic patients are one
of the main groups of patients requiring LTC services [48];
c) Level of dependencyWittenberg et al. [15] argued
that dependency is a crucial factor in considering
future needs of elderly people for long-term care.
According to the literature in the area, ADLs and IADLs
have been the most widely used proxies for the level of
dependency. For the purpose of our study, a specific
ADL-based dependency measure was used, namely, a
dependent individual is considered to be always in bed,
or always sitting in a chair, or limited to his/her own
house;
d) Household compositionevidence indicates that people living with a partner are much less likely to be
institutionalized than single people, since they can receive care from a companion [15];
e) Level of incomeWittenberg et al. [15] argued that the
receipt of LTC services is influenced by socio-economic
factors, including [] income.
Figure 1a and b depict the two components of the Markov cycle tree: the short-term decision tree, which predicts
the number of individuals in need of LTC in the first year
(Fig. 1a); and the long-term model, which defines how LTC
needs evolve over time (Fig. 1b). This Markov structure
separated into two levels seems adequate for better understanding the LTC determinants and the dynamics of the
model. The estimates of need in the first year are made
differently from the remaining years due to the epidemiological data in use. For instance, LTC demand in the first
year depends on prevalence rates for each chronic disease,
whereas for the remaining years it depends dynamically on
incidence and mortality rates. Underlying the structure of
the Markov model is an epidemiological model which considers population demography, prevalence and incidence of
chronic diseases, and mortality rates.

Modeling the demand for long-term care services

391

Fig. 1 a Short-term decision tree built to predict the initial number of


individuals in need of LTC. Legend: VLI: Very Low Income individuals; NVLI: Not Very Low Income individuals; CWS: Chronic patients
Without Symptoms; CS: Chronic patients with Symptoms; NC: NonChronic individuals; D: Dependent individuals; ND: Non-Dependent

individuals; S: Single individuals; NS: Non-Single individuals. b


Long-term model built to predict how the number of individuals in
need of LTC will evolve over time. Legend: CD: Chronic Diseases;
OC: Other Causes; see also legend in a

The short-term decision tree (see Fig. 1a) is composed of 18 branches corresponding to different health
and socio-economic population groups and defines the
number of individuals across those groups in the starting year. The distribution of the population across these
groups is defined similarly to the following case, taken
as example: the number of individuals that belong to
the ND|CS VLI branch (for a specific age, gender
and disease-specific group) depends on the probability
of individuals being non-dependent on others to perform
their daily activities (ND) given that they are chronic
patients showing some recent symptoms (CS) and with
very low income (VLI) (note that these abbreviations
are explained in the legend of Fig. 1a and b). Once the
distribution of the population among these 18 branches
is determined, it is mapped into a specific need for
LTC, according to Table 2. Table 2 should be read as
follows: non-dependent individuals (ND) that suffer
from a chronic disease showing some recent symptoms
(CS) and have very low income (VLI) (ND|CS VLI
branch) will only need to receive ambulatory care (with
100 % probability of needing ambulatory care and 0 %

probability of requiring institutional or home-based


care).
After identifying the individuals in each group for the starting period, individuals progress through different (Markov)
states over time. This evolution is defined by the long-term
model depicted in Fig. 1b (bold rectangles and arrows are used
to simplify the graphical representation of the model). This
model is composed of 20 states (represented by circles), 18 of
which correspond to the branches of the short-term decision
tree and thus can also be mapped into a specific need for LTC
using information from Table 2. The remaining two states
correspond to absorbing states representing death due to chronic diseases (CD) and other causes (OC). Patients transit between these states, except for the following cases: states
involving chronic conditions (CS and CWS) cannot evolve
to states with non-chronic conditions (NC) given that chronic
diseases are long-lasting diseases of a non-reversible nature
[50]; and transitions between non-single (NS) and single (S)
states are not considered, this being an assumption used to
simplify the model (although the model can be easily changed
to include these transitions). The match between the 18 states
of the short-term decision tree and of the long-term model

392

T. Cardoso et al.

Table 2 Branches and states of the short-term decision tree and long-term model (see legend in Fig. 1a and b), respectively; and LTC services that
should be provided for individuals belonging to these branches/states
Branches/states

LTC service

Evidence from previous studies

ND|NC VLI
ND|NC NVLI
ND|CWS VLI
ND|CWS NVLI

No need for LTC

Non-chronic and non-dependent individuals do not need to receive LTC, and the
same applies to those with no dependency and without chronic symptoms [48].

ND|CS VLI
ND|CS NVLI

Ambulatory care

Chronic (with symptoms) and independent patients will need to receive


ambulatory care [67].

Institutional care

The type of care that should be provided for dependent individuals depends on
the availability of informal care single people are much more likely to be
institutionalized than non-single people because non-single people can receive
care from a companion (ex. spouse) and are much more likely to have children,
another important source of informal care [15].

S|
S|
S|
S|
S|
S|

D NC VLI
D CWS VLI
D CS VLI
D NC NVLI
D CWS NVLI
D CS NVLI

NS| D NC VLI
NS| D CWS VLI
NS| D CS VLI

Formal and informal home-based care

NS| D NC NVLI
NS| D CWS NVLI
NS| D CS NVLI

should be read as follows: individuals that belong to the ND|


NC VLI branch in t00 will start their evolution in the
corresponding ND|NC VLI state in the long-term model.
Afterwards, the annual evolution of these individuals (individuals in the ND|NC VLI state) within the long-term model
will depend on predefined probabilities for:
&
&

&
&

Staying in the same state;


Being diagnosed with a chronic disease and changing to
the ND|CS VLI (if they continue to be nondependent [ND] on others), to the S|D CS VLI
(if they are single [S] and become dependent on others
[D]) or to the NS|D CS VLI (if they are non-single
[NS] and become dependent on others [D]);
Becoming dependent on others (D) and changing to the
S|D NC VLI (if they are single [S]) or to the NS|
D NC VLI (if they are non-single [NS]);
Dying due to causes other than chronic diseases and
changing to the state Death due to OC.

Once the number of individuals in each state is predicted,


indicators of demand for services and resources (number of
informal caregivers, domiciliary visits, consultations and
beds) are calculated according to information about a predefined LTC delivery model, as identified in Table 2. Implicit to
a delivery model is information provided by literature in the
area and by previous experience in delivering LTC services
(note that this experience might be context-dependent, and
may depend on policy and institutional choices, on prices and

on the past supply of services). According to the chosen LTC


delivery model, the type and number of services for each type
of individual are set. For example, the average number of
informal caregivers (regardless of whether they provide care
on a full- or part-time basis), domiciliary visits and consultations that are required per patient, as well as the average LOS
that characterizes institutionalizations, might be settled. Afterwards, the total cost required to deliver those services can be
computed using estimates of unit costs i) per domiciliary
patient, ii) per consultation, and iii) per bed.
The mathematical formulation and the main assumptions
of the model are presented in Appendix A. The mathematical formulation shows that each branch of the short-term
decision tree and Markov state has an associated age, gender
and disease-specific probability and makes clear the epidemiological structure of the model.
The approach proposed in this study has the advantage of
creating information for paying purchasers when the system
is based on the principle money follows the patient (meaning that money follows each patient when using a service).
Nevertheless, it has the disadvantage of using a definition of
need and demand that disregards the existence of resource
limitations that have to be accounted for by policymakers
when planning and funding a network. Furthermore, the
proposed approach uses information on health and socioeconomic determinants which are typically collected from
surveys, and many issues arise about the reliability and
validity of survey-based data (e.g., there might be problems

Modeling the demand for long-term care services

such as sampling and reporting bias [51]). Yet surveys are


usually a rich source of information, providing detailed and
diverse data. The proposed model was developed with a
generic structure, and might be applied to most developed
countries as it makes use of routinely collected data available
for most countries. Nevertheless, it might need to be adjusted
according to the services that need to be planned for a
country, which may vary (as described in the Introduction
section), and also to the type of cost data available.

393

and of ensuring further funding so that the network can


answer the populations needs for LTC [10]. Nevertheless,
government funding and private demand for LTC services
have not motivated the development of supply so as to meet
the demand for services [52]. In addition, information on the
expected demand for LTC is lacking for Portugal, and it
represents essential information for planning an LTC network able to meet associated demand in such a challenging
environment. It is thus relevant to apply the proposed model
to the Portuguese context.

3.2 Case study and data set used


3.2.2 Data set used
The simulation model was implemented in Excel and applied to Portuguese data. This section starts by describing
the LTC system currently operating in Portugal, followed by
a presentation of the data set used for applying the model to
the Portuguese context.
3.2.1 LTC in Portugal
The implementation of the National Network of Long-Term
Care Services (RNCCI) in Portugal started in 2006 and is
expected to be concluded in 2013 with the national territory
being fully covered [3, 52]. This network is responsible for
coordinating the delivery of a wide range of health and
social care services. Health services are provided in the
context of an NHS-based system, which is instituted to
guarantee universal coverage and nearly free access at the
point of use, to be tax-financed and where public and private
providers coexist. On the other hand, social services are
provided within a distinct legal framework, with the state
having a subsidiary role in the provision of social services
and access to these services being sometimes means-tested
[3, 53]. The RNCCI is a multiservice and multi-provider
network that comprises home-based, ambulatory and institutional services. Providers might be families and a wide
range of public and private entities (for-profit and nonprofit) that include hospitals and primary health care providers. Local and district social security services and municipalities are also directly or indirectly involved in the
financing and/or delivery of LTC services [52].
Similarly to many other European countries, the LTC
sector in Portugal is facing several challenges [1, 52, 54]:
i) the population need for LTC is changing on a global scale;
ii) the current supply of LTC services is not adequate to
meet this increasing demand; and iii) severe budget cuts are
affecting the health care sector and reducing the public
funding to extend the LTC network. This is creating extra
difficulties for the development of the RNCCI. In fact,
although an important share of LTC services in Portugal
has been provided by private not-for-profit providers (namely Misericrdias), since 2006 the government has assumed
the responsibility of developing the network of LTC services

Table 3 presents the data set used for applying the model to
the Portuguese context so as to predict future LTC demand
for the 20102015 period.
Demographic data for past years on the number of inhabitants in each Lisbon borough of the county of Lisbon and in
the Lisbon and Tagus Valley region was collected from the
Portuguese National Institute of Statistics [55]. The administrative city of Lisbon (which corresponds to a British
ward) is composed of 53 boroughs (Fig. 2), has a higher
proportion of women (currently representing 52 % of the
total population) and a proportion of elderly people above
the national average (people aged over 65 currently represent 24.2 % of its total population, compared to 18.2 % of
the national average). Santa Maria dos Olivais and Mrtires
are the most and the least populous boroughs, with around
43,200 and 320 inhabitants, respectively. Lisbon is also
characterized by high levels of hospital supply, although
this supply is also used to serve large catchment areas in
the Centre and South regions of Portugal [10].
The 4th Portuguese National Health Survey (NHSur,
20052006) was used to estimate the following figures: prevalence, incidence and dependency rates; and to build population groups according to household composition and level of
income. As an example, Table 4 shows the numerical values
obtained for the probabilities of a non-single (NS) individual,
given that he/she is dependent on others (D), is non-chronic
(NC) and has/has not very low income (VLI/NVLI) (NS|D
NC VLI and NS|D NC NVLI states), computed
using data from the 4th NHSur (one should note that these two
probabilities should not add up to 100 % since they are
conditional probabilities; see Appendix A).
Finally, the recommendations for the provision of LTC
announced by the Mission Unit for Integrated Continuing
Care [56], which represent best practice according to
Portuguese LTC professionals, were used to estimate the
services and resources required to deliver care to individuals in need. Past unit costs, with reference to the provision of LTC within the RNCCI, were used to estimate
the costs for meeting the estimated LTC demand. These unit
costs represent health care provider costs, not including other

394

T. Cardoso et al.

Table 3 Dataset used for the application of the model to the Portuguese context
Data

Source

Year

Comments

Demographic data

[55]

2010

Mortality rates

[55]

2005/2009

Prevalence and incidence


of chronic diseases

4th National Health


Survey (NHSur) [68]

2005/2006

Level of dependency

4th National Health


Survey (NHSur) [68]
4th National Health
Survey (NHSur) [68]
4th National Health
Survey (NHSur) [68]

2005/2006

Number of inhabitants aged over 15 in each Lisbon borough, in the county


of Lisbon and in the Lisbon and Tagus Valley region, by age and gender.
Data aggregated and disaggregated per type of chronic disease, age and
gender.
The following chronic diseases that mostly affect the Portuguese
population were considered: (1) oncological diseases, (2) chronic
obstructive pulmonary diseases, (3) cerebrovascular diseases and
(4) ischemic heart diseases.
Individuals were defined as dependent if (1) are always in bed,
(2) are always sitting in a chair, or (3) are limited to their own house.
Two types of households were used as a proxy for the availability of
informal care: single people; and non-single people.
Two groups were set: very low income individuals, defined as those
living below the poverty line; and not very low income individuals
(the remaining ones).
These recommendations concern: the average number of informal
caregivers, domiciliary visits (provided by nurses and doctors) and
consultations per patient; and the average LOS for institutionalizations.
Three types of unit costs were used: unit costs per domiciliary patient;
unit costs per consultation; and unit costs per bed.

Household composition
Level of income

2005/2006
2005/2006

Recommendations for the


provision of LTC

[60]

2009

Unit costs

[52]

2006-2009

Fig. 2 Total number and percentage of individuals in need of FHC, IHC, AC and IC in 2010 (PE and 95 % CI); and the corresponding AAGR
predicted for the total number of individuals in need until 2015

Modeling the demand for long-term care services

395

Table 4 Conditional probabilities by age and gender of being a nonsingle individual (NS) given that he/she is dependent on others to perform
their daily activities (D), is non-chronic (NC) and has/has not very low
income (VLI/NVLI) using information from the 4th NHSur
State

NSjD \ NC \ VLI

NSjD \ NC \ NVLI

Age group

Probabilities (%)
Men

Women

[15;29]
[30;64]

0
50

0
0

+65
[15;29]
[30;64]
+65

73.69
0
66.67
81.82

33.33
0
100
42.86

components of costs, such as costs related to the provision of


informal care or to patients transportation to providers of care.
3.3 Uncertainty analysis
Many factors, including advances in medical technology
and changes in epidemiological profiles [57], are difficult
to foresee with confidence and may affect the future demand
for LTC services. This implies that a high level of uncertainty is often associated with the prediction of LTC demand, and therefore its effect should be accounted for.
Within the context of this study, two key types of uncertainty were considered in an integrated approach: parameter
uncertainty, which arises due to incomplete information
(such as probabilities estimated with missing or low numbers by age and gender groups using data from the NHSur);
and model or structural uncertainty, which arises from simplifications and assumptions in use [46].
Parameter uncertainty was introduced by treating as uncertain the following input parameters of the model: i) the incidence of chronic diseases, since it is not clear whether
expected longer life expectancy will result in higher incidence
rates for older or younger groups [58]; and ii) mortality rates,
given that longevity has been improving in an unpredictable
way and there is no clear understanding on how mortality rates
will evolve in the future [59]. The uncertainty surrounding
these parameters was analyzed by running a probabilistic
sensitivity analysis using a Monte Carlo simulation. Based
on literature [46], a beta probability distribution was assigned
to both parameters (the beta distribution is phenomenological
ideally suited for data on a proportion scale, and it is the only
continuous distribution that is bounded 01) and no dependency was considered between them. The incidence of chronic diseases was modeled through a beta distribution
betaa1ijk ; Total ijk  a1ijk where a1ijk represents the number
of incidents (first diagnosis of a given chronic disease) for age

group i, gender j and borough k, and Totalijk represents the


total number of individuals from age group i, gender j and
borough k. Mortality rates were modeled using a similar beta
distribution betad 1ijkt ; Totalijkt  d 1ijkt , where d 1ijkt represents
the number of individuals from age group i, gender j and
borough k who died during time period t. As an example,
Table 5 provides male mortality rates (mean values and 95 %
confidence intervals [CI]) used in the simulation model. These
rates were generated by using the beta distribution with the
parameters Total ijkt and d 1ijkt calibrated using the data set in use.
With regard to model or structural uncertainty, the impact
of changing some model assumptions was analyzed. For this
purpose, assumptions related to the use and supply of homebased and institutional services (linked to the choice of a
predefined LTC delivery model) were selected. In particular,
we assumed that these services can be seen as substitute
services and also that the LOS for institutionalizations might
vary significantly (as shown in the literature [60]). Within
this context, six different scenarios were defined (Table 6),
with scenario 2 being taken as the base scenario.
In order to propagate the parameter uncertainty in each
scenario, we have run a Monte Carlo simulation with 2,000
iterations (this number has guaranteed the stability of the
results) using the @RISK software version 5.7 for Excel.
Monte Carlo simulation was used since it can handle large
uncertainties in the input parameters [41], with this being
particularly relevant when small samples and numbers, such
as those characterizing small-area level applications, are used.

4 Results
This section begins by presenting the results obtained in the
model validation when applying it to the county of Lisbon.
Later, selected results obtained from applying the model to
Table 5 Mortality rates for chronic and non-chronic men (mean and
95 % CIs) used for the application of the model to the Portuguese context
Data

Mortality rates
(chronic men)

Mortality rates
(non-chronic men)

Age group

2010-2015
Mean (%)

95 % CI (%)

[15;34]
[35;54]
[55;74]
+75
[15;34]

0.4
3.8
6.8
20.7
0.1

[0;1.7]
[2.1;5.9]
[5.6;8.2]
[17.2;24.3]
[0;0.2]

[35;54]
[55;74]
+75

0.2
0.7
5.6

[0.1;0.3]
[0.6;0.9]
[4.6;6.7]

396

T. Cardoso et al.

Table 6 Scenarios under analysis


Scenarios

LOS in
institutions

Substitutability between home-based


and institutional services

1
2
3

Minimum LOS
Average LOS
Maximum LOS

25 % of the people requiring


institutional care will receive
home-based care (both formal
and informal), while the remaining
ones (75 %) will receive
institutional care

4
5
6

Minimum LOS
Average LOS
Maximum LOS

All the people requiring


institutional care will receive
institutional care

the 53 Lisbon boroughs for the 20102015 period are portrayed. The number of individuals in need of LTC during that
period is presented at first, followed by the results on the
volume of resources/services required to meet those needs
and associated costs. These results are shown for the five most
populated boroughs in Lisbon (Santa Maria dos Olivais, Benfica, Marvila, Lumiar and So Domingos de Benfica), and we
have selected for detailed analysis the most populous borough
in Lisbon (Santa Maria dos Olivais). Afterwards, LTC demand
in the Lisbon and Tagus Valley region (one of the five regions
of Mainland Portugal) as predicted by the proposed simulation
model is compared with the current supply of services in that
region. Key results are then discussed. One should bear in
mind that the results for the number of individuals in need of
LTC are the same for the six scenarios under analysis, while the
volume of resources/services to be provided for these individuals in need and the associated costs vary across scenarios.

national data could not be carried out since no such information was available for Portugal.
Comparing model predictions for the number of Lisbon
inhabitants for the 20052009 period with real data has shown
that there are no significant deviations (see maximum deviation gaps in Table 7). The figures in Table 7 should be read as
follows: the number of women aged over 65 for the 2005
2009 period, as predicted by the simulation model, has a
deviation of 0.38 % when compared to real demographic data.
Deviations in Table 7 can be partly explained by several
factors. First, mortality rates disaggregated by type of chronic
disease are only available for 2008 and 2009 [55]. We have
thus assumed that mortality rates for previous years were the
same as those collected for 2008. Secondly, the impact of
migrations was not considered in the model, which might
justify a gap in prediction. Moreover, since there is no information on the rate at which chronic patients have been dying in
Portugal due to causes other than chronic diseases, it was
assumed that chronic patients only die due to their chronic
condition. However, most chronic patients die due to their
chronic condition and so not considering deaths due to other
causes may raise only small deviations. With regard to the
largest deviation found for the youngest age group, this might
be explained by the lower quality of data for forecasting the
behavior of this group, given the small number of deaths and
low levels of incidence and prevalence rates for that age group.
In addition, comparing model predictions with figures
reporting the delivery of LTC in other countries shows that:
&

4.1 Validation results

According to model predictions, 56 % of the group of


individuals requiring LTC is composed of women (and
this proportion differs from the proportion of women
living in the county of Lisbon, which is around 52 %).
This finding is in accordance with Burke et al. [63], who
argue that women typically represent the largest share of
LTC demand;
Model predictions show that 60 % of those who died due to
chronic diseases needed to receive LTC during the 2005
2009 period, which is in line with estimations presented by
McNamara et al. [64], who estimated that between 50 %
and 89 % of all the patients that die from a specific group of
diseases will require palliative care (note that palliative care
is typically provided in the last year of a patients life [48]).

Model validation is critical to ensure that the model can effectively be used for prediction [61, 62]. Therefore, the validity of
the model was carried out by performing two tests using
predictive validation techniques [61], namely by comparing:

&

i. The number of inhabitants in the county of Lisbon as


predicted by running retrospectively the model for the
20052009 period with historical demographic data for
the same period. Analysis of the demographic evolution
indicates whether prevalence, incidence and mortality
rates are well calibrated;
ii. The same model predictions with figures reported in
international studies, such as the proportion of women
within the group of people needing LTC. This comparison
helps to understand whether the simulation results follow
some expected trends, although the level of demand might
differ across countries (e.g., due to differences in the
demographic and socio-economic characteristics of populations). However, the alternative comparison with

Table 7 Maximum deviation (in percentage) between real and predicted values for the number of inhabitants in the county of Lisbon
during the 20052009 period
Age group

Total (%)

Men (%)

Women (%)

[15;29]
[30;64]
+65

3.31
0.76
0.40

3.12
0.66
0.43

3.51
0.86
0.38

Modeling the demand for long-term care services

Following these validation results, we found that the


implemented model is able to satisfactorily predict the demand for LTC services.

4.2 Individuals in need of LTC during the 20102015 period


Figure 2 shows the results obtained for the total number and
percentage of individuals in need of formal and informal
home-based care (FHC and IHC), ambulatory care (AC) and
institutional care (IC) in 2010 (point estimates [PE] and
95 % CI), as well as the corresponding average annual
growth rate (AAGR) predicted for the total number of
individuals in need until 2015. One can see that Santa Maria
dos Olivais is the Lisbon borough with the highest number
of inhabitants requiring LTC. Also, the number of individuals in need of LTC is expected to increase for all the
services and for all the boroughs during the 20102015
period, as shown by the positive AAGR. This increase is
expected to be higher for AC e.g., the total number of
individuals in need of AC in Santa Maria dos Olivais is
expected to increase, on average, by 4.53 % per year,
whereas the number of individuals in need of FHC/IHC
and IC is expected to increase annually, on average, by
1.06 % and 3.77 %, respectively (Fig. 2).
A detailed analysis of the group of individuals requiring
AC (the type of care with the highest AAGR) shows that
those older than 65 are responsible for a large share of
estimated needs. In particular, in Santa Maria dos Olivais
and Benfica this group represents more than half of the total
need (Fig. 3). The same behavior was found for the need for
FHC, IHC and IC. Figure 3 also depicts the probability

397

distribution for the number of individuals aged over 65 in


need of AC in 2010 and 2015 in Santa Maria dos Olivais.
Since all the individuals in need of AC are chronic patients
(Table 2), it is worthwhile analyzing the influence of different
types of chronic diseases within this group. In particular, if we
observe the group of individuals aged over 65 from Santa
Maria dos Olivais (the group responsible for the largest share
of AC needs in this borough), it is possible to verify that
cerebrovascular and ischemic heart diseases are responsible
for the highest and lowest numbers of individuals in need,
respectively (Fig. 4). Similar trends are found for the remaining boroughs in Lisbon. This finding has implications for the
planning of the type of care required for these individuals. The
model also produces important information regarding the
distribution of income among individuals in need of LTC in
the coming years. Figure 5 depicts that distribution among
those requiring AC in 2015.
According to Fig. 5, around half of the individuals requiring AC have VLI. Figure 5 also shows that there are
significant differences between boroughs. In some of these
boroughs, individuals with VLI and aged over 65 are responsible for the largest share of the need for AC (e.g.,
31.41 % with a 95 % CI of [29.8;33.1] in Santa Maria dos
Olivais). On the other hand, in other boroughs individuals
without very low income (NVLI) and aged under 65 are
responsible for the largest share of those needs (e.g., 38.1 %
with a 95 % CI of [35.4;41] in Lumiar). Moreover, the
groups responsible for the largest share of AC need in each
borough are mostly made up of females (Fig. 5, below).
A similar analysis was performed on the number of individuals requiring IC. Since this type of care can be provided
for both chronic and non-chronic patients, it is also relevant to

Fig. 3 Number of individuals aged over/under 65 in need of AC in 2010 and 2015 (PE and 95 % CI); and probability distribution generated by the
Monte Carlo simulation for the number of individuals aged over 65 in need of AC in 2010 and 2015 in Santa Maria dos Olivais

398

T. Cardoso et al.

Fig. 4 Number of individuals


aged over 65 with oncological
diseases/chronic obstructive
pulmonary diseases/
cerebrovascular diseases/
ischemic heart diseases in need
of AC in 2010 and 2015 in
Santa Maria dos Olivais

understand how the existence (or absence) of chronic diseases


will influence IC needs. Figure 6 depicts this influence together with the distribution of income among individuals older
than 65 requiring IC in Santa Maria dos Olivais during the
20102015 period. According to Fig. 6, the majority of the IC
need (for those older than 65) in Santa Maria dos Olivais is
due to NVLI individuals suffering from chronic diseases.
Moreover, in line with the pattern previously described, women represent the largest group in need within that group.
4.3 Volume of resources/services required to meet
LTC need predicted for the 20102015 period,
and associated costs
The results obtained concerning the volume of resources/
services for the 20102015 period include: a) the number of

informal caregivers required to provide IHC; b) the number of


domiciliary visits (provided by nurses and doctors) needed to
provide FHC; c) the number of consultations needed to provide AC; d) the number of beds required to provide IC; and e)
the costs associated with the provision of these services (and
these costs should be read as a proxy for the total costs to be
funded, assuming that current prices and costs in public delivery are observed). The proportion of LTC spending on the
GDP is also presented in this section. Scenarios 1 and 4
(scenarios in which the LOS was set to the minimum, which
is 30 days, see Table 6) are compared in Table 8, with reference to the Santa Maria dos Olivais borough.
Comparing both scenarios, it can be seen that a scenario in
which all the individuals in need of IC receive IC for a period
equal to the minimum LOS (scenario 4) is associated with a
total cost of 45 euros per capita in 2010 (corresponding to a

Fig. 5 Percentage of VLI/NVLI individuals aged over/under 65 requiring AC in 2015 (above) (PE and 95 % CI); and distribution of men and
women within the group of individuals responsible for the largest share of the AC needs in 2015 (below)

Modeling the demand for long-term care services

399

Fig. 6 Number of VLI/NVLI individuals aged over 65 with/without


chronic diseases requiring IC during the 20102015 period in Santa
Maria dos Olivais (above, left) (PE); percentage of those patients

requiring IC in 2015 (above, right) (PE); and distribution of men and


women within the group of individuals responsible for the largest share
of the IC needs in 2015 in Santa Maria dos Olivais (below)

total of 2.2 million euros, approximately, and to 0.28 % of


the GDP for the same year). This is 1 euro per capita
cheaper (which corresponds to savings of 46,000 euros) than
the total cost associated with a scenario in which a proportion of the individuals in need of IC receive home-based
care (both FHC and IHC) (scenario 1). This trend is
expected to be found during the 20102015 period, although

with total costs increasing, on average, between 2.84 % and


2.87 % per year (the total cost associated with scenario 1
and scenario 4 for 2015 is 341,000 and 330,000 euros
higher, respectively).
On the other hand, performing the same comparison
between the other two extreme scenarios (scenarios 3 and
6, Table 9; scenarios in which the LOS was set to the

Table 8 Volume of resources/services for the 20102015 period (PE;


95 % CI; and AAGR); total cost associated with the provision of all
types of LTC service (total cost [TC]; TC per capita; and AAGR); and

proportion of LTC spending on the GDP (% GDP) according to


scenarios 1 and 4 and for the Santa Maria dos Olivais borough

Scenarios

LTC
Services

Volume of resources/services

Total Cost

2010

Scenario 1

Scenario 4

IHC
FHC

AC
IC
IHC
FHC

AC
IC

Informal caregivers
Domiciliary visits
(nurses)
Domiciliary visits
(doctors)
Consultations
Beds
Informal caregivers
Domiciliary visits
(nurses)
Domiciliary visits
(doctors)
Consultations
Beds

2010-2015

2010

%
GDP
2010-2015

PE

95 % CI

AAGR
(%)

TC ()

TC per
capita ()

AAGR (%)

437
20,952

1.72
1.75

2,246,507

46

2.87

0.29

5,238

44,085
26
332
15,936

[43,272;44,952]
-

2,200,565

45

2.84

0.28

3,984

44,085
35

[43,272;44,952]
-

4.52
3.58
1.06
1.09

4.52
3.71

400

T. Cardoso et al.

Table 9 Total cost associated with the provision of all types of LTC
services (TC; TC per capita; and AAGR); and proportion of LTC spending on the GDP (% GDP) in Santa Maria dos Olivais according to the
following pairs of scenarios: scenarios 2 and 5; and scenarios 3 and 6
Scenarios

Total Cost

Table 10 Volume of resources/services required in Santa Maria dos


Olivais in 2015 (PE and 95 % CI)
LTC Services

Scenarios

% GDP

2010

2015

2010-2015

TC ()

TC per
capita ()

AAGR (%)

Scenario 2
Scenario 5
Scenario 3

2,903,695
3,076,816
3,591,593

60
63
74

3.10
3.14
3.18

0.37
0.39
0.46

Scenario 6

3,994,013

82

3.22

0.51

IHC

FHC

Informal
caregivers

Domiciliary
visits (nurses)

Domiciliary
visits (doctors)

maximum, which is 90 days) allowed it to be verified that


providing home-based care (both FHC and IHC) for 25 % of
the individuals in need of IC (scenario 3) would result in
savings of 8 euros per capita (which corresponds to savings
of 402,000 euros). The same pattern was found for the
20102015 period, but with total costs increasing, on average, between 3.18 % and 3.22 % per year. The same conclusion was reached when comparing scenarios 2 and 5:
providing beds for institutionalizations characterized by average LOS (60 days) for all the individuals in need of IC
(scenario 5) is 3 euros per capita more expensive (which
amounts to 173,000 euros) than providing home-based care
(both FHC and IHC) for a proportion of these individuals
(scenario 2). The same trends were also found for the
remaining boroughs in Lisbon.
An additional result concerns the differences found between each pair of scenarios, which were found to be significant at the 5 % level the 95 % CIs obtained as a result
for each scenario do not overlap. Similar results can be
found in Table 10 for the year 2015.
4.4 Comparison between LTC demand as predicted
by the model and current LTC supply
A comparison between predicted LTC demand and current
LTC supply was performed for the number of beds in the
Lisbon and Tagus Valley region. Results, in Fig. 7, show
that the current supply of services far from meets the need
for services.
According to the base scenario (scenario 2), 2,608 beds
were required in Lisbon and Tagus Valley in 2010, which
means that only 34 % of the required beds existed. In the most
optimistic scenario (scenario 1), only 1,304 beds were required in Lisbon in 2010, which means that only 68 % of
the beds required to attend to all the people in need of IC
existed; and in the most pessimistic scenario (scenario 6), that

Volume of resources/services

IC

Beds

Scenario
2&3
Scenario
5&6
Scenario
2&3
Scenario
5&6
Scenario
2&3
Scenario
5&6
Scenario
Scenario
Scenario
Scenario
Scenario
Scenario

PE

95 % CI

1,

476

[468;485]

4,

350

[342;359]

1,

22,852

[22,428;23,292]

4,

16,819

[16,416;17,232]

1,

5,713

[5,607;5,823]

4,

4,205

[4,104;4,308]

1
2
3
4
5
6

31
93
377
42
125
503

[31;32]
[92;95]
[369;385]
[41;43]
[122;127]
[492;513]

figure decreases to 17 %. One should bear in mind that


existing beds in Lisbon and Tagus Valley can also be used to
provide care to populations outside the Lisbon region (e.g.,
from the Centre and South regions of Portugal) [10], meaning
that the gap between predicted demand and current supply
might be even more pronounced.
In Portugal, the large discrepancy found between the
predicted demand for care and current supply is an expected
result because, as explained in the Methodology section, the
RNCCI has only recently been implemented and there has
been no extra public funding to extend the LTC network.
These budget constraints, as well as the economic context in
Portugal, have been limiting the development of LTC provision because [10, 52]: i) as noted before, the government
is a key player in directly or indirectly funding the provision
of LTC; ii) by limiting public funding in recent years,
greater financial pressure is put on the population side, with
most of it not having the financial means to pay for these
services; and iii) a reliance on private finance for LTC
investments is even more problematic in the context of the
current economic crisis. Along with these issues, the mixed
and fragmented nature that characterizes Portuguese providers of LTC might also contribute to the gap found between predicted demand and current supply. A proper
coordination between all the services, providers and entities
involved in LTC provision and funding is essential for an
adequate supply of LTC, but this coordination is still far
from being satisfactory [10, 52].

Modeling the demand for long-term care services

401

Fig. 7 Average number of beds


required to meet the demand for
IC in the region of Lisbon and
Tagus Valley as predicted by
the simulation model for 2010
(grey bars); and number of beds
that were available in that
region at the end of 2010 (black
line)

4.5 Discussion
Applying the simulation model to the Lisbon borough level
has shown that the demand for LTC services is expected to
increase during the 20102015 period. When planning
changes to the LTC network, policymakers should consider
these projections and policy responses should take into
consideration that:
&

&

&

&

&

The highest average annual growth rate is observed for


ambulatory care (Fig. 2) this is because Portuguese
people suffering from chronic diseases are typically nondependent on others (according to the 4th NHSur); thus,
most of the Lisbon inhabitants will not need to be
institutionalized or receive home-based care;
Individuals older than 65 are responsible for a large
share of the LTC needs (Fig. 3) this happens because
older people are characterized by higher prevalence,
incidence and dependency rates;
Different chronic diseases result in different levels of
LTC need (Fig. 4) e.g., cerebrovascular and ischemic
heart diseases are responsible for the highest and lowest
numbers of individuals in need, respectively; these differences are relevant for planning because patients suffering from different types of diseases might require
specialized care;
Different boroughs are characterized by different patterns of LTC need (Fig. 5) this happens because of
demographic differences in the proportion of individuals
aged over 65 (this being lower in Lumiar and higher in
Santa Maria dos Olivais); moreover, those older than 65
have approximately 50 % probability of having very low
income (according to the 4th NHSur), but the same does
not apply for those younger than 65;
Women represent the largest share of the LTC needs for
both the elder and younger groups (Figs. 5 and 6);

&
&

&

&

Around half of the population needing LTC has very low


income, and these individuals will be unable to pay for
those services (Figs. 5 and 6);
Different assumptions related to the choice of an LTC
delivery model have significant impact on model predictions e.g., the number of beds that should be made
available in institutions differs hugely when assuming
distinct LOS for institutionalizations. For instance, an
LOS of 30 days results in a 42-bed requirement in 2015
for the population of Santa Maria dos Olivais, whereas
an LOS of 90 days results in a 125-bed requirement
(assuming that all the demand for institutionalizations
is met by institutional care);
Planning an LTC network characterized by short institutionalizations (such as in scenarios 1 and 4) suggests
that providing institutional care for all the individuals in
need of institutionalization might result in significant
savings (Table 8). On the other hand, when longer
institutionalizations are observed (such as in scenarios
3 and 6), it is worth considering the option of providing
home-based care as a substitute service for institutional
care for some of the individuals in need (Table 9);
The current supply of LTC services in the Lisbon and
Tagus Valley region is not enough to meet the predicted
demand for LTC in that region (Fig. 7). This result
supports the assertion that utilization of services cannot
be used as a proxy for demand, as a low supply of
services is expected to constrain the utilization of services and correspond to unmet needs.

However, several issues must be considered when deciding


upon replacing institutional care with home-based care. Based
on the results presented in Table 9, it seems that, under the
assumption of longer institutionalizations, it is cheaper to
provide home-based care for some of the people in need of
institutional care. Nevertheless, this result might change if we

402

T. Cardoso et al.

consider that providing home-based care for individuals in


need of institutional care is more expensive than providing the
same type of service for those in need of home-based care. For
example, people in need of institutional care might not have an
informal caregiver available (see Table 2), and so providing
formal home-based care for these individuals might imply a
higher number of domiciliary visits or longer domiciliary
visits (and thus, higher costs).
One should bear in mind that the methods used in this study
have some limitations, and thus results should be analyzed
with caution, given that:
&

&

&

&

&

Transition probabilities in the long-term model were assumed to be constant over time it was assumed that the
epidemiological patterns do not change over time, which
is typically not true; however, this assumption had to be
made because there was no data available to estimate
changes on transition probabilities;
Survey data was used for the application of the model
many issues arise over the reliability and validity of
survey-based data, e.g., there might be problems such as
sampling and reporting bias [51]. Nevertheless, the 4th
Portuguese National Health Survey was a rich source of
information, providing a wide range of data with representation by health region;
It was assumed that past data (such as past mortality data
and past information collected from the above-mentioned
survey) will remain unchanged in future years but since
there is no clear understanding on how past health and
socio-economic characteristics will evolve over time [58,
59], and since we are dealing with a short planning horizon
(only 5 years), making this assumption seemed reasonable;
It was assumed that the demand for LTC services will
remain constant for a whole year this assumption was
used to simplify the model. However, it might be acceptable in the planning context because the need for
care was converted into indicators of demand for services and resources using average ratios that translate the
pattern of care for each type of user and their need for
care (e.g., while predicting the need for institutionalizations for a whole year, the model considers that institutionalizations last no more than the average LOS used as
an input to the model);
The distinction between single and non-single people was
used as a proxy for the availability of informal care
although single people are not likely to receive informal
care [15], there might be exceptions (e.g., when those in
need of LTC have family/friends outside their household
available to take care of them); similarly, non-single people might not have children or companions available at
home. Nevertheless, according to the literature, distinguishing between non-single and single people is widely
used as a proxy for the availability of informal care;

&

Several issues arise regarding the unit costs used as a


basis for the application of the model: i) these costs
represent the average unit costs that were incurred in
the first semester of 2009, as no more recent data exists;
ii) these costs represent health care provider costs and
lack other components, such as the costs associated with
the provision of informal care (e.g., social costs associated with employment and earnings of caregivers) and
with patients transportation to health care providers;
and iii) the total cost predicted using these unit costs as
a basis should not be regarded as representing the total
costs to be incurred by the government, since there
might be associated savings if acute care in hospitals
provided to long-term patients can be provided by specialized LTC services (acute care resources are more
technology dependent in Portugal, and typically more
expensive than LTC resources [65]). All these issues
should be taken into account when analyzing results
in particular, the choice of balance between institutional
and home-based care depends on the organizational
model of LTC delivery and should be informed by those
costs and additional evidence on the cost-effectiveness
of different LTC delivery models and LTC interventions.

Finally, one should note that the results from the proposed model were shown to provide important detailed
information for policymakers and health care planners.
While previous studies have considered different factors
(e.g., the influence of age on demand), to the best of our
knowledge, those studies have not comprehensively considered most socio-economic factors that are expected to influence LTC demand. Moreover, although model predictions
show that LTC demand is expected to increase in the coming
years (which is in line with previous models results), they
also show that this increase will be higher for ambulatory
services. To our knowledge, these services have not been
typically considered in the literature but should be
accounted for when planning a network of LTC services.

5 Conclusion
The development of networks of LTC services is nowadays
seen as a policy priority in many countries, mainly due to
the ageing phenomenon and to the increase in the prevalence of chronic diseases that are currently affecting populations all over the world. Nevertheless, there is still a lack
of information to inform the planning of those services. The
present work aims to contribute to filling this gap by proposing a simulation model, based on a Markov cycle tree
structure, to predict future demand for LTC services. The
model adds to the literature since it does not rely on information on the current (or past) utilization of services. This is

Modeling the demand for long-term care services

a desirable feature when predicting demand in countries


where utilization of LTC is highly constrained by a low
supply of LTC services [9]. Our results confirm that this is
the case observed for Portugal. For instance, according to
model results, 1,300 was the approximate number of beds
required in Lisbon and Tagus Valley in 2010 in the most
optimistic scenario, and this figure is well above the 900
beds that were actually available in that region. Therefore,
instead of relying on utilization data, our model departs
from an assessment of the absolute need for care that
depends on a wide range of health and socio-economic
determinants, such as the incidence and prevalence of
chronic diseases, the level of dependency, the household
composition and the level of income of populations. Information on these determinants is routinely collected in most
European countries, for instance through National Health
Surveys, which typically include socio-demographic variables, incidence and prevalence rates of diseases and disabilities [66]. Thus, the model can be easily adapted and
applied to any country. Considering these socio-economic
determinants allowed a comprehensive model to be built
that provides results with a high level of detail and with
relevance to policymakers and health care planners. However, one should note that although not relying on
utilization-based information, supply side factors need to
be considered since the potential demand for services
depends on the choice of a predefined LTC delivery model
and is influenced by supply. Therefore, the choice of delivery model should involve considering the historical supply
of services, evidence on the consequences of distinct LTC
delivery models and expected effects due to an interaction
between demand and supply. Additionally, using the model
involves employing an integrated approach for assessing the
different types of uncertainty expected to affect future predictions of LTC demand, whereas previous studies have
only used a scenario analysis for the same purpose. Finally,
this study applies the model to the small-area level (for
which there has been a lack of applications in the area of
LTC demand prediction), providing relevant information for
policymakers in the LTC sector, since health characteristics
are known to vary geographically.
This work was applied to Portugal, a country where there
is a critical lack of data to inform the development of a
network of LTC to our knowledge and to date, only one
study has presented predictions for LTC demand in Portugal
(a study developed under the scope of the FELICIE project),
but it only covered the demand for those aged over 75 and
did not consider the impact of chronic diseases on future
demand. Applying the simulation model developed in this
article at the borough level in Lisbon for the 20102015
period has shown that LTC demand in Lisbon is expected to
increase during this period, with an higher increase being
registered for ambulatory care, a type of service not usually

403

considered in studies in this area. Results also show that the


current supply of services far from meets the predicted
demand and that the model can be used to provide policymakers with relevant information for informing financing
and planning decisions. The model gives information on the
main group of individuals in need of each type of LTC
service; on the type of chronic diseases that are associated
with higher levels of demand; on the income distribution
among those needing LTC; and on which cases providing
home-based care as a substitute service for institutional care
might be financially advantageous for.
Following this article, several research topics seem to be
worth pursuing. First, it might be relevant to adapt and apply
the simulation model proposed in this study to different
contexts and health system settings. Among other things,
these applications might allow policymakers and health care
planners to compare some of the consequences of adopting
different organizational models. Additional research is also
needed to obtain more reliable estimates of the total cost
associated with meeting predicted demand according to an
LTC delivery model. In particular, one needs to consider
cost components other than health care provider costs (e.g.,
costs related to the provision of informal care such as those
associated with the employment and earnings of caregivers).
Further work is also required to create tools to inform LTC
providers on how LTC services should be geographically
organized, taking into account other features of the health
care system. For example, it might be relevant to develop
mathematical programming models to plan the supply of
LTC services, so that the supply meets the demand for
services as predicted by the proposed simulation model.
Acknowledgments The first author acknowledges financing from
the Fundao para a Cincia e Tecnologia (Portugal) (SFRH/BD/
63966/2009). The authors thank three anonymous referees for their
thorough and insightful comments on an earlier version of this paper.
The authors remain responsible for any omissions and inaccuracies.

Appendix A: Assumptions and mathematical


formulation of the model
This appendix describes the main assumptions and the complete mathematical formulation of the simulation model.
The following assumptions were used in the proposed
simulation model:
1. Non-dependent individuals (both non-chronic and
chronic without symptoms) do not need to receive any
type of LTC service [48];
2. Non-dependent and chronic patients with symptoms
(with symptoms depending on the chronic disease from
which they suffer) need to receive specialized ambulatory care. This type of patients are able to perform their

404

3.

4.

5.

6.
7.

T. Cardoso et al.

daily activities, and to go to an outpatient unit to see a


doctor whenever necessary [67];
Single people are much more likely to be institutionalized than non-single people, because non-single people
can receive care from a companion (ex. spouse) and are
much more likely to have children, another important
source of informal care [15]. Thus, dependent and single
individuals need to receive institutional care, while dependent and non-single individuals need to receive formal and informal home-based care;
States involving chronic conditions cannot evolve to
states with non-chronic conditions, because chronic diseases are long-lasting diseases for which a cure is not
expected [50];
No transitions exist between states characterized by
non-single and single individuals. This assumption
was used to simplify the model, although the model
can be easily changed to include those transitions;
The level of income is assumed not to vary within the
time horizon;
Chronic patients only die due to their chronic condition.

Additional assumptions were also required for applying


the model to Portuguese data:
1. Mortality, incidence and dependency rates do not
change over time;
2. The following chronic diseases were considered: i) oncological diseases, ii) chronic obstructive pulmonary
diseases, iii) cerebrovascular diseases and iv) ischemic
heart diseases;
3. Individuals were defined as dependent if i) are always in
bed, ii) are always sitting in a chair, or iii) are limited to
their own house;
4. Two types of household were used as a proxy for the
availability of informal care: single and non-single
people;
5. Individuals are split into a very low income group (those
living below the poverty line) or not a very low income
group (the remaining ones).
For the purpose of presenting the mathematical formulation of the model, each of the 20 states of the long-term
model was numbered with a state number (see Table 11).
One should note that chronic individuals who have
shown some symptoms recently (states 2, 4, 6, 12, 14 and
16) include two main types of individuals: individuals suffering from a chronic disease for the first time (CS1st); and
chronic individuals who have shown some symptoms during the last year but not for the first time (CSagain). One
needs to distinguish between these two groups of individuals because the Markov states from which they come from
are different. E.g., a very low income (VLI) individual with
a chronic disease and recent symptoms (CS) at t+1 (states 2,

Table 11 Long-term model states


State of the long-term State
model
number
(s)

State of the long-term


model

State
number
(s)

NDjCWS \ VLI
NDjCS \ VLI
SjD \ CWS \ VLI
SjD \ CS \ VLI
NSjD \ CWS \ VLI
NSjD \ CS \ VLI
NDjNC \ VLI
SjD \ NC \ VLI
NSjD \ NC \ VLI

NDjCWS \ NVLI
NDjCS \ NVLI
SjD \ CWS \ NVLI
SjD \ CS \ NVLI
NSjD \ CWS \ NVLI
NSjD \ CS \ NVLI
NDjNC \ NVLI
SjD \ NC \ NVLI
NSjD \ NC \ NVLI

11
12
13
14
15
16
17
18
19
20

Death due to CD

1
2
3
4
5
6
7
8
9
10

Death due to OC

4 or 6) might have two different origins: i) he/she might


have been non-chronic (NC) at t (belonging to states 7, 8 or
9 at t, which means that he/she is suffering from a chronic
disease for the first time at t+1); or ii) he/she might have
already been suffering from a chronic disease at t (belonging
to states 1, 2, 3, 4, 5 or 6 at t, which means that he/she is not
suffering from a chronic disease for the first time at t+1).
The sets, parameters and variables used within the model
are identified in Tables 12, 13 and 14, respectively.
Since the proposed simulation model includes a
short-term decision tree and a long-term model, its
mathematical formulation can be split into two main
groups of equations: a first group that defines the number of individuals in each one of the 18 branches of the
short-term decision tree (Equations 16); and a second
group for predicting how the number of individuals in
each state of the long-term model will evolve over time
(Equations 712). Each branch and Markov state is
divided into different groups of individuals, depending
on their age group i I, gender j J, borough k K
and type of chronic disease d D (if any).
The number of individuals in each of the 18 branches of
the short-term decision tree corresponds to the initial number of individuals in 18 of the 20 states of the long-term

Table 12 Sets
Set

Definition

i2I

Age group
Gender
Borough
Chronic disease
Time period
Long-term Markov model state

j2J
k2K
d2D
t2T
m, s 2 S

Modeling the demand for long-term care services

405

Table 13 Parameters
Parameter

Definition

nInd ijkt0

Number of individuals from age group i 2 I, gender j 2 J and borough k 2 K in the


first year of the planning horizon (t00).
Probability of an individual from age group i 2 I and gender j 2 J who suffers from
a chronic disease d 2 D(with or without symptoms) being in state s 2 Sin the first
year of the planning horizon.
Probability of a non-chronic individual from age group i 2 I and gender j 2 J being
in state s 2 Sin the first year of the planning horizon.
Probability of an individual from age group i 2 I and gender j 2 J who suffers from
a chronic disease d 2 D(with or without symptoms) changing from state m 2 S to
state s 2 S (state transition probability).
Probability of a non-chronic individual from age group i 2 I and gender j 2 J
changing from state m 2 S to state s 2 S (state transition probability).
Probability of an individual from age group i 2 I and gender j 2 J being nondependent on others given that he/she: suffers from a chronic disease d 2 D(but
without showing any symptom during the last year); and has very low/not very
low income.
Probability of an individual from age group i 2 I and gender j 2 J being nondependent on others given that he/she: suffers from a chronic disease d 2 D for the
first time; and has very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being nondependent on others given that he/she: suffers from a chronic disease d 2 D and
has shown some symptoms during the last year (but not for the first time); and has
very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being nondependent on others given that he/she: suffers from a chronic disease d 2 D and
has shown some symptoms during the last year; and has very low/not very low
income.
Probability of an individual from age group i 2 I and gender j 2 J being single given
that he/she: is dependent on others; suffers from a chronic disease d 2 D(but
without showing any symptom during the last year); and has very low/not very
low income.
Probability of an individual from age group i 2 I and gender j 2 J being single given
that he/she: is dependent on others; suffers from a chronic disease d 2 D for the
first time; and has very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being single given
that he/she: is dependent on others; suffers from a chronic disease d 2 D and has
shown some symptoms during the last year (but not for the first time); and has
very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being single given
that he/she: is dependent on others; suffers from a chronic disease d 2 D and has
shown some symptoms during the last year; and has very low/not very low
income.
Probability of an individual from age group i 2 I and gender j 2 J being non-single
given that he/she: is dependent on others; suffers from a chronic disease d 2 D(but
without showing any symptom during the last year); and has very low/not very
low income.
Probability of an individual from age group i 2 I and gender j 2 J being non-single
given that he/she: is dependent on others; suffers from a chronic disease d 2 D for
the first time; and has very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being non-single
given that he/she: is dependent on others; suffers from a chronic disease d 2 D and
has shown some symptoms during the last year (but not for the first time); and has
very low/not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being non-single
given that he/she: is dependent on others; suffers from a chronic disease d 2 D and
has shown some symptoms during the last year; and has very low/not very low
income.

pijds

pijs
pijdms

pijms

.
PNDjCWS \ LVI ijd PNDjCWS \ NVLI ijd
.
PNDjCS 1st \ VLI ijd PNDjCS 1st \ NVLI ijd

 . 

P NDjCS again \ VLI ijd P NDjCS again \ NVLI ijd
.
PNDjCS \ VLI ijd PNDjCS \ NVLI ijd
.
PSjD \ CWS \ VLI ijd PSjD \ CWS \ NVLI ijd
.
PSjD \ CS 1st \ VLI ijd PSjD \ CS 1st \ NVLI ijd

 . 

P SjD \ CS again \ VLI ijd P SjD \ CS again \ NVLI ijd
.
PSjD \ CS \ VLI ijd PSjD \ CS \ NVLI ijd
.
PNSjD \ CWS \ VLI ijd PNSjD \ CWS \ NVLI ijd
.
PNSjD \ CS 1st \ VLI ijd PNSjD \ CS 1st \ NVLI ijd

 . 

P NSjD \ CS again \ VLI ijd P NSjD \ CS again \ NVLI ijd
.
PNSjD \ CS \ VLI ijd PNSjD \ CS \ NVLI ijd

406

T. Cardoso et al.

Table 13 (continued)
Parameter

Definition

PNDjNC \ VLI ij PNDjNC \ NVLI ij


.
PSjD \ NC \ VLI ij PSjD \ NC \ NVLI ij
.
PNSjD \ NC \ VLI ij PNSjD \ NC \ NVLI ij

PDdsijkdt
PDndsijkt
nInfCp
nVdmp
nVnmp
nCmp
LOS
UCdp
UCc
UCb

Probability of an individual from age group i 2 I and gender j 2 J being nondependent on others given that he/she: is not a chronic patient; and has very low/
not very low income.
Probability of an individual from age group i 2 I and gender j 2 J being single given
that he/she: is dependent on others; is not a chronic patient; and has very low/not
very low income.
Probability of an individual from age group i 2 I and gender j 2 J being non-single
given that he/she: is dependent on others; is not a chronic patient; and has very
low/not very low income.
Probability of an individual from age group i 2 I, gender j 2 J and borough k 2 K
dying due to a chronic disease d 2 D during time period t 2 T.
Probability of an individual from age group i 2 I, gender j 2 J and borough k 2 K
dying (due to causes other than chronic diseases) during time period t 2 T.
Number of informal caregivers required per patient.
Number of domiciliary visits to be provided by doctors per month and per patient.
Number of domiciliary visits to be provided by nurses per month and per patient.
Number of consultations to be provided per month and per patient.
Average length of stay (LOS) that characterizes institutional stays (in years).
Unit cost per domiciliary patient.
Unit cost per consultation.
Unit cost per bed.

Table 14 Variables
Variable

Definition

NStijkts
nStijkdts

Total number of individuals from age group i 2 I, gender j 2 J and borough k 2 K who belong to state s 2 S during time period t 2 T.
Number of individuals from age group i 2 I, gender j 2 J and borough k 2 K who suffer from a chronic disease d 2 D (with or without
symptoms) and belong to state s 2 S during time period t 2 T.
Number of non-chronic individuals from age group i 2 I, gender j 2 J and borough k 2 K who belong to state s 2 S during time period t 2 T.
Total number of individuals from borough k 2 K who are in need of formal home-based care during time period t 2 T.
Number of individuals from borough k 2 K suffering from a chronic disease d 2 D (with or without symptoms) who are in need of
formal home-based care during time period t 2 T.
Number of non-chronic individuals from borough k 2 K who are in need of formal home-based care during time period t 2 T.
Total number of individuals from borough k 2 K who are in need of informal home-based care during time period t 2 T.
Number of individuals from borough k 2 K suffering from a chronic disease d 2 D (with or without symptoms) who are in need of
informal home-based care during time period t 2 T.
Number of non-chronic individuals from borough k 2 K who are in need of informal home-based care during time period t 2 T.
Total number of individuals from borough k 2 K who are in need of ambulatory care during time period t 2 T.
Number of individuals from borough k 2 K suffering from a chronic disease d 2 D (with or without symptoms) who are in need of
ambulatory care during time period t 2 T.
Total number of individuals from borough k 2 K who are in need of institutional care during time period t 2 T.
Number of individuals from borough k 2 K suffering from a chronic disease d 2 D (with or without symptoms) who are in need of
institutional care during time period t 2 T.
Number of non-chronic individuals from borough k 2 K who are in need of institutional care during time period t 2 T.
Total number of informal caregivers that is required in borough k 2 K and during time period t 2 T.
Total number of domiciliary visits that must be provided by doctors in borough k 2 K during time period t 2 T.
Total number of domiciliary visits that must be provided by nurses in borough k 2 K during time period t 2 T.
Total number of consultations that must be ensured in borough k 2 K during time period t 2 T.
Total number of beds that must be made available in borough k 2 K during time period t 2 T.
Total cost associated with meeting predicted demand in borough k 2 K during time period t 2 T.

nStijkts
NFHCkt
nFHCkdt
nFHCkt
NIHCkt
nIHCkdt
nIHCkt
NACkt
nACkdt
NICkt
nICkdt
nICkt
NInfCkt
NDVdkt
NDVnkt
NCkt
NBkt
TotalCostkt

Modeling the demand for long-term care services

407

Markov model (the remaining two states correspond to the


absorbing states of the long-term model and have no associated branches in the decision tree). This initial number of
individuals is denoted by NStijk(t00)s (individuals from age
group i I, gender j J and borough k K that belong to
state s S in the first year of the planning horizon (t00)) and
is computed according to Equation 1, in which nStijkd(t00)s
and nStijk(t00)s correspond to the initial number of individuals suffering from a chronic disease d D (with or without
symptoms) and to the initial number of individuals without

(
nSt ijkdt0s

(
nSt ijkt0s

pijds

nInd ijkt0  pijds

8i 2 I; j 2 J ; k 2 K; d 2 D; s 2 f1; 2; 3; 4; 5; 6g [ f11; 12; 13; 14; 15; 16g


0 otherwise

nInd ijkt0  pijs

8i 2 I; j 2 J ; k 2 K; s 2 f7; 8; 9g [ f17; 18; 19g

8
PNDjNC \ VLIij
>
>
> PSjD \ NC \ VLI
>
>
ij
>
< PNSjD \ NC \ VLI

ij

PNDjNC \ NVLIij
>
>
>
>
>
> PSjD \ NC \ NVLIij
:
PNSjD \ NC \ NVLIij

8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d
8i 2 I; j 2 J ; d

8i 2 I; j 2 J ;
8i 2 I; j 2 J ;
8i 2 I; j 2 J ;
8i 2 I; j 2 J ;
8i 2 I; j 2 J ;
8i 2 I; j 2 J ;

2 D; s 1
2 D; s 2
2 D; s 3
2 D; s 4
2 D; s 5
2 D; s 6
2 D; s 11
2 D; s 12
2 D; s 13
2 D ; s 14
2 D; s 15
2 D; s 16

s7
s8
s9
s 17
s 18
s 19
5

The probabilities of an individual belonging to each state of


the long-term model in the first year are computed similarly to

0 otherwise

8
PNDjCWS \ VLIijd
>
>
>
>
PNDjCS \ VLIijd
>
>
>
> PSjD \ CWS \ VLIijd
>
>
>
>
PSjD \ CS \ VLIijd
>
>
>
>
PNSjD \ CWS \ VLI
>
>
< PNSjD \ CS \ VLI ijd
ijd

PNDjCWS \ NVLI ijd


>
>
>
>
PNDjCS \ NVLIijd
>
>
>
>
>
> PSjD \ CWS \ NVLIijd
>
>
>
> PSjD \ CS \ NVLIijd
>
>
>
PNSjD \ CWS \ NVLIijd
>
:
PNSjD \ CS \ NVLIijd

pijs

any chronic disease, respectively (Equations 2 and 3). On


the other hand, nStijkd(t00)s and nStijk(t00)s depend on the
probability of an individual being in states Sin the first
year of the planning horizon (pijds and pijs, computed based
on Equations 4 and 5, respectively).
8X
nStijkdt0s
<
d
NSt ijkt0s
8i 2 I; j 2 J ; k 2 K; s 2 S
:
nSt ijkt0s

Equation 6 (which describes one particular case).


pijds1 PNDjCWS \ VLI ijd

PND \ CWS \ VLIijd

2D

PCWS \ VLIijd

8i 2 I; j 2 J ; d
6

The second group of equations predicts how the number


of individuals in each one of the 20 states of the long-term

408

T. Cardoso et al.

model evolves over time. The number of individuals from


age group i I, gender j J and borough k K that belong
to state s S during the time period t T is denoted by
NStijkts and is computed based on Equation 7, in which
nStijkdts and nStijkts correspond to the number of individuals
suffering from a chronic disease d D (with or without
symptoms) and to the number of individuals without any
chronic disease, respectively (Equations 8 and 9). On the
other hand, nStijkdts and nStijkts depend on the probability of

nSt ijkdts

nSt ijkts

pijdms

8 6 h
i


P
>
>
nSti1jkdt1m 1  PDdsi1jkdt1 pijdms
>
>
>
m1
>
>
>
i
9 h
>


P
>
>

nSt
1

PDnds
p
>
i1jkt1m
i1jkt1
ijdms
>
>
m7
>
>
i
>
16 h


> P
>
>
nSt i1jkdt1m 1  PDdsi1jkdt1 pijdms
>
>
< m11
i
19 h



P
>
nSt i1jkt1m 1  PDndsi1jkt1 pijdms

>
>
>
>
> 6 m17
i
>
Ph
>
>
>
nSti1jkdt1m PDdsi1jkdt1 pijdms
>
>
>
m1
>
>
i
>
16 h
P
>
>
>

nSt i1jkdt1m PDdsi1jkdt1 pijdms


>
>
>
: m11
0 otherwise

changing from state m S to state s S (pijdms and pijms,


called the state transition probabilities and that are computed based on Equations 10 and 11, respectively).
8X
nSt ijkdts
<
d
NSt ijkts
8i 2 I; j 2 J ; k 2 K; t 2 T ; s
:
nSt ijkts
2S

8i 2 I; j 2 J ; k 2 K ; d 2 D; t  1; s 2 f1; 2; 3; 4; 5; 6g

8i 2 I; j 2 J ; k 2 K ; d 2 D; t  1; s 2 f11; 12; 13; 14; 15; 16g

1st

ijd

8i 2 I; j 2 J ; k 2 K ; d 2 D; t  1 ; s 10

8 9 h
i


P
>
>
8i 2 I; j 2 J ; k 2 K ; t  1; s 2 f7; 8; 9g
nSt i1jkt1m 1  PDndsi1jkt1 pijms
>
>
>
m7
>
>
h
i
>
19


>
P
>
>
8i 2 I; j 2 J ; k 2 K ; t  1; s 2 f17; 18; 19g
nSt i1jkt1m 1  PDndsi1jkt1 pijms
>
>
<
m17
h
i
9

P
>
nSt i1jkt1m PDndsi1jkt1 pijms
>
>
>
m7
>
>
h
i
>
19
>
> P nSt
>
PDnds
p
8i 2 I; j 2 J ; k 2 K; t  1 ; s 20
>
i1jkt1m
i1jkt1
ijms
>
>
: m17
0 otherwise

8
PNDjCWS \ VLI ijd
>
>


>
>
>
P
NDjCS again \ VLI ijd
>
>
>
>
>
PNDjCS 1st \ VLI ijd
>
>
>
> PSjD \ CWS \ VLI ijd
>
>


>
>
> P SjD \ CS again \ VLI ijd
>
>
>
>
> PSjD \ CS 1st \ VLI ijd
>
>
>
PNSjD \ CWS \ VLI ijd
>
>


>
>
>
P NSjD \ CS again \ VLI ijd
>
>
>
>
< PNSjD \ CS 1st \ VLI ijd
1
>
>
>
PNDjCWS \ NVLI ijd
>
>
>
>
PNDjCWS \ NVLI ijd
>
>
>
>
>
> PNDjCS 1st \ NVLI ijd
>
>
PSjD \ CWS \ NVLI ijd
>
>

> 
>
>
P SjD \ CS again \ NVLI ijd
>
>
>
>
PSjD \ CS 1st \ NVLI ijd
>
>
>
>
>
PNSjD \ CWS \ NVLI ijd
>
>


>
>
P NSjD \ CS again \ NVLI ijd
>
>
>
: PNSjD \ CS \ NVLI

8i 2 I; j 2 J ; d 2 D ; m 2 f1; 2; 3; 4; 5; 6g; s 1
8i 2 I; j 2 J ; d 2 D; m 2 f1; 2; 3; 4; 5; 6g; s 2
8i 2 I; j 2 J ; d 2 D; m 2 f7; 8; 9g; s 2
8i 2 I; j 2 J ; d 2 D; m 2 f1; 2; 3; 4g; s 3
8i 2 I; j 2 J ; d 2 D; m 2 f1; 2; 3; 4g; s 4
8i 2 I; j 2 J ; d 2 D; m 2 f7; 8g; s 4
8i 2 I; j 2 J ; d 2 D; m 2 f1; 2; 5; 6g; s 5
8i 2 I; j 2 J ; d 2 D; m 2 f1; 2; 5; 6g; s 6
8i 2 I; j 2 J ; d 2 D; m 2 f7; 9g; s 6
8i 2 I; j 2 J ; d 2 D ; m 2 f1; 2; 3; 4; 5; 6g [ f11; 12; 13; 14; 15; 16g; s 10
8i 2 I; j 2 J ; d 2 D; m 2 f11; 12; 13; 14; 15; 16g; s 11
8i 2 I; j 2 J ; d 2 D; m 2 f11; 12; 13; 14; 15; 16g; s 12
8i 2 I; j 2 J ; d 2 D; m 2 f17; 18; 19g; s 12
8i 2 I; j 2 J ; d 2 D; m 2 f11; 12; 13; 14g; s 13
8i 2 I; j 2 J ; d 2 D ; m 2 f11; 12; 13; 14g; s 14
8i 2 I; j 2 J ; d 2 D ; m 2 f17; 18g; s 14
8i 2 I; j 2 J ; d 2 D ; m 2 f11; 12; 15; 16g; s 15
8i 2 I; j 2 J ; d 2 D ; m 2 f11; 12; 15; 16g; s 16
8i 2 I; j 2 J ; d 2 D ; m 2 f17; 19g; s 16

10

Modeling the demand for long-term care services

pijms

8
PNDjNC \ VLI ij
>
>
>
> PSjD \ NC \ VLI ij
>
>
>
> PNSjD \ NC \ VLI
<
ij
PNDjNC \ NVLI ij
>
>
PSjD \ NC \ NVLI ij
>
>
>
>
>
PNSjD \ NC \ NVLI ij
>
:
1

409

8i 2 I; j 2 J ; m 2 f7; 8; 9g; s 7
8i 2 I; j 2 J ; m 2 f7; 8g; s 8
8i 2 I; j 2 J ; m 2 f7; 9g; s 9
8i 2 I; j 2 J ; m 2 f17; 18; 19g; s 17
8i 2 I; j 2 J ; m 2 f17; 18g; s 18
8i 2 I; j 2 J ; m 2 f17; 19g; s 19
8i 2 I; j 2 J ; m 2 7; 9 [ 17; 19; s 20

Transition probabilities are computed similarly to


Equation 12 (which describes one particular case).
pijdms1 PNDjCWS \ VLI ijd

PND \ CWS \ VLIijd


PCWS \ VLIijd

8i 2 I; j

2 J ; d 2 D ; m 2 f1; 2; 3; 4; 5; 6g

12

Based on the number of individuals in each state and


per time period (nStijkdts and nStijkts), it is possible to
compute:
1. The total number of individuals from borough k K who
are in need of formal home-based care during time period
t T (NFHCkt, Equation 13), in which nFHCkdt and
nFHCkt correspond to the number of individuals suffering
from a chronic disease d D (with or without symptoms)
and without any chronic disease, respectively, who are in
need of this service (Equations 14 and 15);
2. The total number of individuals from borough k K who
are in need of informal home-based care during time
periodt T(NIHCkt, Equation 16), in which nIHCkdt and

nFHC kdt

XX

NFHC kt

nSt ijkdts5 nSt ijkdts6 nSt ijkdts15 nSt ijkdts16

XX

nSt ijkts9 nSt ijkts19 8k 2 K ; t 2 T
i

nIHCkt correspond to the number of individuals suffering


from a chronic disease d D (with or without symptoms)
and without any chronic disease, respectively, who are in
need of this service (Equations 17 and 18);
3. The total number of individuals from borough k K
who are in need of ambulatory care during time period t
T (NACkt, Equation 19), in which nACkdt corresponds
to the number of individuals suffering from a chronic
disease d D (with or without symptoms) who are in
need of this service (Equation 20);
4. The total number of individuals from borough k K
who are in need of institutional care during time
period t T (NICkt, Equation 21), in which nICkdt
and nICkt correspond to the number of individuals
suffering from a chronic disease d D (with or
without symptoms) and without any chronic disease,
respectively, who are in need of this service (Equations 22 and 23).
8X
nFHC kdt
<
:

8k 2 K ; t 2 T

13

nFHC kt

8k 2 K ; d 2 D; t 2 T

14

nFHC kt

11

15

nIHC kdt nFHC kdt

17

16

nIHC kt nFHC kt

18

NIHC kt NFHC kt

410

T. Cardoso et al.

NAC kt

nAC kdt 8k 2 K ; t 2 T

19

nAC kdt

XX

8X
nIC kdt
<

NIC kt

nIC kdt

8k 2 K ; d 2 D; t 2 T

20

8k 2 K ; t 2 T

21

nIC kt

XX

nSt ijkdts3 nSt ijkdts4 nSt ijkdts13 nSt ijkdts14

8k 2 K ; d 2 D; t 2 T

22

XX
i

nIC kt

nSt ijkdts2 nSt ijkdts12

nSt ijkts8 nSt ijkts18

8k 2 K ; t 2 T

TotalCost kt NFHC kt  UCdp NC kt  UCc

NBkt  UCb

23
The volume of resources/services required to meet
the needs predicted for each borough k K and time
period t T can thus be computed using as a basis the
total number of individuals in need of each type of
service the number of informal caregivers is denoted
by NInfCkt, and is computed based on Equation 24; the
number of domiciliary visits to be provided by doctors
and by nurses are denoted by NDVdkt and NDVnkt,
respectively, and are computed based on Equations 25
and 26; the number of consultations is denoted by NCkt,
and is computed based on Equation 27; and the number
of beds is denoted by NBkt, and is computed based on
Equation 28.
NInfC kt NIHC kt  nInfCp

24

NDVd kt NFHC kt  nVdmp

25

NDVnkt NFHC kt  nVnmp

26

NC kt NAC kt  nCmp

27

NBkt NIC kt  LOS

28

Finally, the total cost associated with meeting predicted


demand in borough k K during time period t T is
computed using Equation 29.

29

References
1. World Health Organization (2000) Home-based long-term care:
report of a WHO study group Technical Report Series, vol 898.
World Health Organization, Geneva
2. Brodsky J, Clarfield AM (2009) Organization of health services: long
term care in health services. In: Carrin G, Buse K, Heggenhougen K,
Quah SR (eds) Health systems policy, finance, and organization. 1st
edn. Academic Press, pp 351357
3. Ministry of Health (2006) Decreto-lei n 101/2006: Cria a Rede
Nacional de Cuidados Continuados Integrados [Create the National Network of Long-Term Care]. Dirio da Repblica: I Srie-A, n
109 de 6 de Junho
4. Leichsenring K (2004) Developing integrated health and social
care services for older persons in Europe. Int J Integr Care 4:e10
5. Sundstrm G, Malmberg B, Castiello M, Barrio , Castejon P,
Tortosa M, Johansson L (2007) Family care for elders in Europe:
policies and practices. In: Szinovacz M, Davey A (eds) Caregiving
contexts: cultural, familiar and societal implications. Springer,
New York, pp 235267
6. Comas-Herrera A, Costa-Font J, Gori C, di Maio A, Patxot C,
Pickard L, Pozzi A, Rothgang H, Wittenberg R (2003) European
study of long-term care expenditure. PSSRU discussion paper
1840. Personal Social Services Research Unit, London
7. Comas-Herrera A, Wittenberg R, Costa-Font J, Gori C, di Maio A,
Patxot C, Pickard L, Pozzi A, Rothang H (2006) Future long-term
care expenditure in Germany, Spain, Italy and the United Kingdom. Ageing Soc 26(2):285302
8. Inter-American Conference on Social Security (2008) The Americas Social Security Report 2008: organization and management of
long-term care programs. CISS, Mexico
9. Roberfroid D, Leonard C, Stordeur S (2009) Physician supply
forecast: better than peering in a crystal ball? Hum Resour Health
7(10)
10. Barros PP, Machado SR, Simes JA (2011) Portugal: Health
system review. Health Systems in Transition 13(4):1156

Modeling the demand for long-term care services


11. Chung RY, Tin KYK, Cowling BJ, Chan KP, Chan WM, Lo SV,
Leung G (2009) Long-term care cost drivers and expenditure
projection to 2036 in Hong Kong. BMC Health Serv Res 9:172
12. Lane D, Uyeno D, Stark A, Kliewer E, Gutman G (1985) Forecasting demand for long-term care services. Health Serv Res 20
(4):435460
13. Rivlin AM, Wiener JM (1988) Caring for the disabled elderly: who
will pay? 1st edn. Brookings Institution, Washington, D.C
14. Robinson J (1996) A long-term care status transition model. The oldage crisis - Actuarial opportunities: The 1996 Bowles Symposium
15. Wittenberg R, Pickard L, Comas-Herrera A, Davies B, Darton R
(1998) Demand for long-term care: projections of long-term care
finance for elderly people. University of Kent, Personal Social
Services Research Unit
16. Wittenberg R, Pickard L, Comas-Herrera A, Davies B, Darton R
(2001) Demand for long term care for older people in England to
2031. Health Stat Q 12:517
17. Comas-Herrera A, Pickard L, Wittenberg R, Davies B, Darton R
(2003) Future demand for long-term care, 2001 to 2031: projections of demand for older people in England. PSSRU Discussion
Paper 1980. Personal Social Services Research Unit, London
18. Wittenberg R, Comas-Herrera A, King D, Malley J, Pickard L,
Darton R (2006) Future demand for long-term care, 2002 to 2041:
projections of demand for long-term care for older people in
England. PSSRU Discussion Paper 2330. Personal Social Services
Research Unit, London
19. Pickard L, Wittenberg R, Comas-Herrera A, Davies B, Darton R
(2000) Relying on informal care in the new century? Informal care
for elderly people in England in 2031. Ageing Soc 20(6):745772
20. King D, Malley J, Wittenberg R, Darton R, Comas-Herrera A
(2010) Projections of demand for residential care for older people
in England - Report for BUPA. PSSRU Discussion Paper 2624.
Personal Social Services Research Unit, London
21. Comas-Herrera A, Wittenberg R, Pickard L, Knapp M (2007)
Cognitive impairment in older people: future demand for longterm care services and the associated costs. Int J Geriatr Psychiatry
22(10):10371045
22. Wittenberg R, Pickard L, Malley J, King D, Comas-Herrera A,
Darton R (2008) Future demand for social care, 2005 to 2041:
projections of demand for social care for older people in england Report to the Strategy Unit (Cabinet Office) and the Department of
Health. PSSRU discussion paper 2514. Personal Social Services
Research Unit, London
23. Comas-Herrera A, Casado D, Wittenberg R, King D, Pickard L
(2005) Projections of demand for long-term care for older people
in Wales to 2030, by local authority. PSSRU discussion paper
2253. Personal Social Services Research Unit, London
24. Comas-Herrera A, Wittenberg R, King D, Pickard L (2005) Projections of demand for long-term care in Norfolk to 2016. PSSRU
discussion paper 2146. Personal Social Services Research Unit,
London
25. King D, Wittenberg R, Comas-Herrera A, Pickard L (2005) Projections of demand for long-term care in Worcestershire to 2011.
PSSRU discussion paper 2144. Personal Social Services Research
Unit, London
26. Xie H, Chaussalet TJ, Millard PH (2005) A continuous time
Markov model for the length of stay of elderly people in institutional long-term care. J R Stat Soc A 168(1):5161
27. Xie H, Chaussalet TJ, Millard PH (2006) A model-based approach
to the analysis of patterns of length of stay in institutional longterm care. IEEE Trans Inf Technol Biomed 10(3):512518
28. Chahed S, Demir E, Chaussalet TJ, Millard PH, Toffa S (2011)
Measuring and modeling occupancy time in NHS continuing
healthcare. BMC Health Serv Res 11:155
29. Costa-Font J, Wittenberg R, Patxot C, Comas-Herrera A, Gori C, di
Maio A, Pickard L, Pozzi A, Rothgang H (2008) Projecting long-term

411

30.

31.

32.

33.

34.

35.

36.

37.

38.
39.

40.
41.
42.

43.
44.
45.

46.
47.

48.
49.
50.

51.

52.

care expenditure in four European Union member states: the influence


of demographic scenarios. Soc Indic Res 86(2):303321
Hostetler S (2011) Evaluating long-term care demand, availability
and needs: a 50-year outlook. In: Doolen T, Aken EV (eds)
Industrial Engineering Research Conference (IERC) 2011. Reno,
Nevada
Karlsson M, Mayhew L, Plumb R, Rickayzen B (2006) Future
costs for long-term care: Cost projections for long-term care for
older people in the United Kingdom. Health Policy 75(2):187213
Zuttion R, Bidino RD, Tubaro E, Garbo M, Ponta S, Cecotti E
(2008) Projection of demand for long-term care in an Italian
region: a macrosimulation approach. In: European Population
Conference (EPC), Barcelona, Spain, July 2008
Leung E (2003) Projecting the needs and costs of long term care in
Australia Research Paper Number 110. Economics and Commerce:
Centre for Actuarial Studies, Victoria Australia
Economic Policy Committee and European Commission (2005)
The 2005 EPC projection of age-related expenditure: agreed underlying assumptions and projection methodologies. ECFIN/
CEFCPE, Brussels
Lagergren M (2005) Whither care of older persons in Sweden? - a
prospective analysis based upon simulation model calculations,
20002030. Health Policy 74(3):325334
Lagergren M (2007) A simulation model concerning future needs
for long-term care for the elderly persons in Sweden. Int Symp
Econ Theory & Econom 16:281296
CORDIS (2001) Europa : CORDIS. Available via http://cordis.
europa.eu/search/index.cfm?fuseaction0proj.document&PJ_
RCN06059809. Accessed September 2011
INTERLINKS (2011) INTERLINKS. Available via http://inter
links.euro.centre.org/. Accessed September 2011
ANCIEN (2011) Assessing Needs of Care in European Nations,
ANCIEN Available via http://www.ancien-longtermcare.eu/.
Accessed September 2011
Parsons S (2011) Qualitative methods for reasoning under uncertainty. The MIT Press, Massachusetts
Papadopoulos CE, Yeung H (2001) Uncertainty estimation and Monte Carlo simulation method. Flow Meas Instrum 12(4):291298
Coyte PC, Goodwin N, Laporte A (2008) How can the settings
used to provide care to older people be balanced? Health Systems
and Policy Analysis, World Health Organization, Copenhagen
Evashwick CJ (2005) The continuum of long-term care, 3rd edn.
Delmar Publishing, New York
Sonnenberg FA, Beck JR (1993) Markov models in medical decision making: a practical guide. Med Decis Making 13(4):322338
Karnon J, Brown J (1998) Selecting a decision model for economic
evaluation: a case study and review. Health Care Manag Sci 1
(2):133140
Briggs A, Sculpher M, Claxton K (2006) Decision modelling for
health economic evaluation. Oxford University, USA
Petrou S, Gray A (2011) Economic evaluation using decision
analytical modelling: design, conduct, analysis, and reporting.
BMJ 342:d1766
World Health Organization (2002) Current and future long-term
care needs. World Health Organization, Geneva
Ibe OC (2009) Markov processes for stochastic modeling. Elsevier
Academic, Massachusetts
Busse R, Blmel M, Scheller-Kreinsen D, Zentner A (2010)
Tackling chronic disease in Europe: strategies, interventions and
challenges. World Health Organization, Observatory Studies
Series, 20
Asthana S, Gibson A (2008) Health care equity, health equity and
resource allocation: towards a normative approach to achieving the
core principles of the NHS. Radic Stat 96:626
Lopes M, Mendes F, Escoval A, Agostinho M, Vieira C, Vieira I,
Sousa C, Cardozo S, Fonseca A, Novas VC, Eliseu G, Serra I,

412

53.

54.

55.

56.

57.

58.

59.

60.

T. Cardoso et al.
Morais C (2010) Plano Nacional de Sade 20112016: Cuidados
Continuados Integrados em Portugal analisando o presente, perspectivando o futuro [Health National Plan 20112016: Long-term
care in Portugal - analyzing the present, predicting the future]. Alto
Comissariado da Sade, vora
Ministries of Finance and Public Administration, Labour and Social Security and Health (2006) Define os preos dos cuidados de
sade e de apoio social prestados no mbito das experincias piloto
da Rede Nacional de Cuidados Continuados Integrados [Define the
prices for health and social care provided under the pilot experiments of the National Network of Long-Term Care]. Dirio da
Repblica: I Srie, n181 de 19 de Setembro
AARP Public Policy Institute (2006) European experiences with
long-term care: France, the Netherlands, Norway, and the United
Kingdom. Washington, DC
Portuguese National Institute of Statistics (2011) Portal do Instituto
Nacional de Estatstica [Statistics Portugal]. Available via www.
ine.pt. Accessed October 2011
Mission Unit for Integrated Continuing Care (2011) Unidade de
Misso para os Cuidados Continuados Integrados [Mission Unit
for Integrated Continuing Care]. Available via URL: www.rncci.
min-saude.pt. Accessed 2011 December
Hughes BB, Kuhn R, Peterson CM, Rothman DS, Solrzano JR
(2011) Improving global health: forecasting the next 50 years Patterns of potential human progress, vol 3. Paradigm, USA
Merlis M (1999) Financing long-term care in the twenty-first
century: the public and private roles. Institute for Health Policy
Solutions, The Commonwealth Fund
Cairns AJG, Blake D, Dowd K (2006) A two-factor model for
stochastic mortality with parameter uncertainty: theory and calibration. J Risk Ins 73(4):687718
Mission Unit for Integrated Continuing Care (2010) Relatrio
de monitorizao do desenvolvimento e da actividade da Rede

61.

62.

63.

64.

65.

66.

67.

68.

Nacional de Cuidados Continuados Integrados (RNCCI), 1


semestre 2010 [Report monitoring the development and the
activity of the National Network of Long-Term Care
(RNCCI), 1st semester 2010]. Cuidados Continuados - Sade
e Apoio Social
Petty MD (2009) Verification and validation. In: Sokolowski JA,
Banks CM (eds) Principles of modeling and simulation - A multidisciplinary approach. Wiley, New Jersey, pp 121149
Gallivan S (2008) Challenging the role of calibration, validation and
sensitivity analysis in relation to models of health care processes.
Health Care Manag Sci 11(2):208213
Burke SP, Feder J, Water PNVD (2005) Developing a better longterm care policy: a vision and strategy for Americas future.
National Academy of Social Insurance, Washington
McNamara B, Rosenwax LK, Holman DAJ (2006) A method for
defining and estimating the palliative care population. J Pain
Symptom Manage 32(1):512
Mission Unit for Integrated Continuing Care (2010) Trs anos de
Rede Nacional de Cuidados Continuados Integrados [3 years of
National Network of Long-Term Care]. Cuidados Continuados Sade e Apoio Social
Bullinger M (2003) International comparability of health interview
surveys: An overview of methods and approaches. In: Nosikov A,
Gudex C (eds) EUROHIS: Developing common instruments for
health surveys. Ios Press, Amsterdam, pp 111
Mission Unit for Integrated Continuing Care (2008) Critrios
gerais de refernciao de doentes para unidades de internamento,
de ambulatrio e equipas da RNCCI [General criteria for the
referral of patients to inpatient and outpatient units and RNCCI
teams]. Directiva Tcnica 1. UMCCI, Lisboa
National Institute of Health Dr. Ricardo Jorge (2006) Base de
dados do Inqurito Nacional de Sade 2005/2006 [National Health
Survey Database 2005/2006]. INSA, Lisboa

Copyright of Health Care Management Science is the property of Springer Science & Business Media B.V. and
its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.

S-ar putea să vă placă și