By Thomas Horvath, Thomas Leoni, Peter Reschenhofer & Martin Spielauer 

 

Background
Population ageing is associated with rising healthcare expenditure. To inform policy and adapt health systems accordingly, a detailed quantitative analysis of the different components of ageing and other factors that influence cost dynamics is needed.

Methods
We use dynamic microsimulation to project healthcare expenditure in Austria and disentangle the effects of changes in longevity, population age-structure, healthy life years and socio-economic health disparities. By combining price weights for healthcare services with information on healthcare consumption from the Austrian Health Interview Survey, we construct average cost profiles by gender, age, and education. These profiles, aligned with the System of Health Accounts, are integrated into the microDEMS model, along with official population projections, to estimate expenditure scenarios until 2060. We examine the relationship between rising life expectancy and changes in healthy life years and assess the potential impact of closing the gap in costs currently observed between education groups. Total and per-capita cost trajectories are derived and evaluated against two indicators for the size of the labor force to assess economic implications.

Results
In all scenarios, demographic ageing increases the financial burden on the economically active population, even with morbidity compression. Nearly two-thirds of the projected cost increase stems from declining mortality, while one-third results from age-structure changes. Per-capita costs rise by 26% under a morbidity expansion scenario but could decrease by 5% if lower mortality is accompanied by an extension of healthy life years and a reduction in socio-economic health disparities. In economic terms, costs per working-age person increase by 12% to 48%, depending on the scenario. When adjusting for labor force expansion and the associated economic benefits, the increase ranges between 5% and 39%.

Conclusions
Rising healthcare expenditure poses a major challenge in an ageing society. However, policies that extend healthy life years and reduce socio-economic disparities offer viable strategies to significantly mitigate the economic impact of ageing.

Introduction
In recent decades, healthcare expenditure (HCE) has outpaced economic growth in most advanced economies, and long-term projections typically forecast that future health expenditures will continue to rise as a share of Gross Domestic Product (GDP), [27, 49]. These developments, which cause concern about the sustainability of public health spending among policy makers, have led to intensive research into the drivers of HCE. Assessing the role and relative importance of different drivers is crucial for designing and prioritizing adequate policies. This paper investigates the long-term effects of aging on HCE, providing a detailed analysis of different impact channels and including in the analysis the role played by socio-economic differences as driver of healthcare costs.

Broadly speaking, the literature distinguishes between demographic and non-demographic factors affecting HCE, with changes in national income, in the age and morbidity distribution of the population, and in medical technology as primary drivers of aggregate expenditure [22, 53, 56]. To capture the effect of income on HCE, typically GDP-related measures are used, whereas the effect of technological advancement is estimated by treating it as a residual in HCE regression models or using proxy indicators for medical innovation such as patents, the number of clinical trials or R&D expenditure [14, 30, 52].

There is some consensus that, in the past, rising incomes and innovations in health technology have been the main causes behind the observed expansion in healthcare costs [24, 47, 68]. However, demographic factors can be expected to take full effect in the coming decades, at least in those countries and regions of the world where ageing is already well advanced [13]. In the European Union, for example, the combination of low fertility rates, continuing gains in life expectancy and the ageing of the large cohorts born in the 1950 s and 1960 s will result in a significant increase in median population age and population growth will be concentrated in the group of people aged 70 years and older [26, 28].

Demographic change impacts HCE through various channels, which clearly include the different components of population age-structure, but also (changes in) the health and disability status of the ageing population as well as indirect effects related to the supply and utilization of healthcare services [21]. In recent decades, an intensive debate has developed in the literature about which aspects of ageing are decisive for healthcare costs and must therefore be considered when projecting future developments. According to the “red herring” hypothesis, the commonly observed link between ageing and increasing HCE may be deceptive [69]. Since healthcare costs tend to rise significantly in the final years of life and older age groups include a higher proportion of individuals nearing death, it can falsely appear that age alone is driving the increase in healthcare spending. In contrast, the hypothesis posits that the time remaining until death is a more accurate predictor of healthcare expenditure than age.

Although individual studies use different methodological approaches and reach sometimes different conclusions, it has been shown that the inclusion of proximity to death increases the explanatory power of HCE models and that projections based solely on age tend to overestimate the future growth of healthcare costs [29, 34, 58, 59, 62, 70]. However, time to death loses most of its significance as an explanatory factor for healthcare expenditure when the health status of the population is sufficiently taken into account. Research, particularly more recent studies, clearly shows that both age and time to death are proxy indicators for the decisive cost driver, namely morbidity [16, 38, 40, 45, 66]. The key question is therefore whether the projections of healthcare costs take sufficient account of the development of the population’s health status. This is all the more significant as chronic diseases and multimorbidity account for a large and growing proportion of healthcare costs in affluent, ageing societies [23, 51, 63]. Other studies have pointed out that the influence of age, morbidity and time-to-death varies depending on the type of health services, with age having a greater impact on out-of-hospital expenditures and hospital expenditures being primarily determined by morbidity and time-to-death [5, 65].

This complexity explains why estimates of the relationship between ageing and healthcare costs vary considerably and why the debate on the correct methodology for measuring the impact of population ageing is still wide open [13, 34]. Moreover, recent findings highlight the role of social inequalities as determinants of HCE [3, 20] and the potential savings resulting from policies that close the gap in morbidity and mortality between socio-economic groups. Lower socio-economic status, which is usually measured using income, education, labour market status or a combination of these characteristics, is associated with increased healthcare utilization and expenditure [20, 42]. The positive link between education and health is a particularly well-established empirical fact. A recent meta-analysis estimated that an additional year of schooling is associated with an average reduction in all-cause adult mortality risk of 1.9% [7]. Including the socio-economic dimension can therefore add an important facet to the projection of HCE and the associated policy implications.

Using a dynamic microsimulation model and a combination of micro and macro data for Austria, this paper contributes to clarifying and quantifying the relative importance of different effects associated with demographic change. We project HCE and investigate different cost drivers up to 2060, including also differences in healthcare costs across educational groups. There is a well-documented educational gradient in health and robust evidence that the attained level of education is associated with differences in life expectancy [12, 25, 35]. We thus use educational level as a proxy to highlight the role of socio-economic status as a determinant of life-time healthcare costs. We calculate total and per-capita healthcare costs and place these projections in an economic context by contrasting them with projections of the number of working-age persons and the number of economically active persons.

Different scenarios are applied to disentangle and quantify the impacts of changes in population age-structure, life expectancy, healthy lifeyears and socio-economic differences. We align our model with official population projections and healthcare cost accounting and apply stylized assumptions that allow us to highlight the key determinants of HCE and their relative importance. The primary objective is to show the (relative) magnitude of the different cost drivers and thus also the scope for different policy strategies to cushion the impact of rising healthcare costs. Our results provide a benchmark for the development of healthcare costs in a highly industrialised, ageing country with a well-developed public healthcare system. They also contribute to the literature by improving the understanding of cost projections and increasing the transparency of the associated assumptions, which are often made only implicitly.

Methods
We build on previous work [31] and use the dynamic microsimulation model microDEMS (Demography, Employment and Social Security) to project healthcare costs for Austria up to 2060. In contrast to Horvath et al. [31], who analyze gender and educational disparities in the projected lifetime healthcare costs of a single birth cohort by integrating life expectancy differentials with education-specific cost profiles, the present study examines the evolution of overall health cost, highlighting the long-term impact of different factors and assessing the sensitivity of different assumptions. This analysis specifically evaluates the influence of demographic shifts, the potential ameliorative effects of reducing socio-economic health gradients, and the impact of increased healthy life expectancy. Furthermore, we contextualize these projected costs within the broader framework of anticipated changes in labour force size.

MicroDEMS is designed as a modular, multi-purpose platform using MODGEN programming technology. It is a more detailed national implementation of the comparative microWELT (https://www.microWELT.eu/) model, which has been utilized in various contexts [11, 31, 60, 61). In contrast to microWELT, which relies on comparative survey data, microDEMS is based on detailed administrative records and implements more detailed institutionalised settings, such as Austrian pension regulations. A detailed model description is available in Horvath et.al. [33].

The model is based on cross-sectional data derived from Austrian Microcensus data (2018), which are representative of the population in the base year. It simulates the further individual life courses over time, whereby the various processes (such as partnerships, fertility, educational pathways, labour force participation, changes in health status or death) are informed by empirical estimates from various data sources. Microsimulation allows the analysis and testing of different “what-if” scenarios that can provide valuable insights that go beyond what is available from retrospective population studies [4].

Estimation of healthcare cost profiles by gender, age, and education
In the first step, we combine survey data on healthcare use and administrative information on HCE to calculate average cost profiles by gender, age, and education consistent with aggregate expenditure according to the System of Health Accounts (SHA). These cost profiles are combined with official population projections in the microsimulation model microDEMS to project different HCE scenarios isolating the effects of changes in crucial cost parameters.

The use of healthcare services is estimated using microdata from the representative Austrian Health Survey (ATHIS) for 2019. The data contain information on the number of inpatient hospital stays (excluding stays related to childbirth), daycare stays, and visits to general practitioners (GPs) and specialist doctors (including hospital outpatient visits). Until a few years ago, primary care in Austria has taken place in the practices of general practitioners and in hospitals [10]. As part of ongoing efforts to reform primary care, Austria is establishing primary health care units (PCUs) to provide multi-professional and interdisciplinary primary care. This is however a new and slow development. Until 2023 only 40 PCUs had been established, against plans to reach 75 operational PCUs by 2021 [55].

As our estimates of healthcare system utilization are based on earlier survey data, primary care in the form of visits to GPs and specialists is included. However, we have not included preventive medical exams in our analysis, as although these were surveyed in ATHIS, their costs are not covered in the healthcare expenditure profile by gender and age-group provided by Statistics Austria that we used (see below). The consumption of pharmaceuticals, on the other hand, is only reported very roughly in the ATHIS survey; we lacked the means to quantify use and unit costs and could for this reason not include it in our study.

The distribution of healthcare service consumption was calculated by gender, age, and education. Using the International Standard Classification of Education (ISCED), we distinguish between low educational attainment (at most compulsory schooling, ISCED 0–2), medium educational attainment (lower and upper secondary education and apprenticeship, ISCED 3–4), and high educational attainment (tertiary education, ISCED 5 +).

In the second step, aggregate healthcare spending by gender and age-group is derived from official statistics provided by Statistics Austria. Following the SHA methodology, current expenditure on health is defined as spending on the core functions of healthcare (HC.1-HC.9). This approach distinguishes current healthcare expenditure from long-term care (LTC) expenditure [27]. The data used in this analysis cover healthcare functions HC.1 to HC.5, as defined by the SHA methodology. It represents the total cost of healthcare services and goods, excluding investments and prevention. Our analysis focused on the abovementioned inpatient, outpatient, and daycare services, which account for over 90% of personal healthcare service costs and 71% of total expenditure, according to SHA healthcare functions HC.1 to HC.5.

Using information provided by the Austrian Ministry for Work, Social Affairs, Health, and Consumer Protection and the Austrian National Public Health Institute, we determine price weights for different healthcare service categories under scrutiny. Inpatient hospitalisations have the highest average unit cost (856 Euro per day) and are the most significant factor in cost estimation. The cost of a GP visit was assigned a price weight of 57 Euro, whereas a specialist doctor visit was assigned a price weight of 76 Euro. For daycare, which is not frequently used and for which no price reference was available, we assumed a unit cost of 600 Euro.

The resulting cost profiles by gender, age, and education are shown in Figure 5 in the Appendix. With increasing age, the cost profiles tend to rise notably, yet individuals with higher levels of education typically incur lower costs compared to individuals in other educational groups across all age groups. Although women demonstrate more pronounced variations based on education after the age of 40, men exhibit greater variation at younger ages. Projections based on these cost profiles have shown that, even after accounting for the social gradient in mortality and thus the higher life expectancy of better-educated groups, the lifetime healthcare costs of men and women with higher education are respectively around 40% and 10% lower than for men and women with lower education [31].

Microsimulation of total healthcare expenditure
In the next step, we use the healthcare cost profiles as inputs in microDEMS to calculate the future evolution of HCE. Based on a representative cross-sectional database of the Austrian population, microDEMS simulates individual life courses over time. The model is fully consistent with official population projections by directly implementing their demographic assumptions, including age and gender specific fertility and mortality rates as well as international migration flows by place of birth.

The population’s educational composition is subject to changes over time, due to the differing levels of education attained between birth cohorts and younger cohorts (typically characterized by higher levels of education) replacing older (less educated) cohorts as time progresses. Our projections assume that all future educational improvements stem from intergenerational transmission. The distribution of children’s educational outcomes by parental education remains constant over time, which constitutes a conservative scenario. We take into account the different educational characteristics of immigrants by assigning them an educational status according to their age at arrival, gender, and country of birth, under the assumption that these patterns remain stable over time. All corresponding model parameters were estimated based on labour force survey data and complementary schooling statistics provided by Statistics Austria. For a detailed model description see Horvath et al. [33]. In addition, our model also takes into account education-specific differences in mortality by linking the actuarial mortality tables provided by Statistics Austria, which are incorporated in the model, with OECD data on remaining life expectancy by education for 25- and 65-year-olds [46]. microDEMS therefore reproduces changing age- and education-specific mortality rates and accounts for the overall increase in life expectancy according to official population projections.

Applying the average healthcare costs by gender, age and education, to each individual in the population, microDEMS allows us to simulate how future HCE evolves over time.

Scenario description
To quantify and disentangle how different channels impact future HCE dynamics, we run a set of scenarios highlighting how different assumptions with respect to mortality, healthy life years and socio-economic differences in healthcare cost affect total HCE (HCEtot) and per-capita HCE (HCEpc) over time. Table 1 provides an overview of the seven scenarios that we apply in our analyses, indicating the assumptions in the relevant dimensions of change.

Table 1 Scenario definitions
Full size table
In our first scenario (S0), we assume that mortality rates, the distribution of health status and healthcare cost profiles remain constant by age and gender (discarding differences by education), reflecting the levels observed in 2020. This scenario is suitable for showing the effects of ageing on HCE that result solely from demographic shifts in the population structure by gender and age. In other words, this scenario reflects past changes in line with official population projections, but without assuming further changes in mortality rates or one of the other drivers of healthcare costs investigated in this study.

In our next scenario (S1), we introduce differences in healthcare costs by education. We assume that age-, gender- and education-specific health status, mortality and healthcare costs remain constant. Healthcare cost dynamics are influenced not only by demographic changes, but also by the educational expansion in the population. This educational expansion is driven by the extrapolation of existing trends in combination with modelling intergenerational transmission of education, whereby higher education of parents further increases the probability of children to attain higher education [11]. Although the transmission channels between education and health are far from being fully understood, existing research has emphasised numerous pathways for positive health effects of education, such as higher incomes, higher social status, stronger psychosocial resources and lower behavioural risks [15, 17, 19, 48],

Scenario S2 relies on the same assumptions as scenario S1, while also incorporating expected increases in life expectancy in line with official population projections. By keeping health status constant by age, gender and education, this scenario implicitly assumes that the Austrian population attains a higher life expectancy, without increasing the proportion of healthy people at a given age. While this scenario is pessimistic, it is consistent with the “expansion of morbidity” hypothesis, according to which longer life expectancy does not lead to an equal increase in the number of years spent in good health but rather to a constant or even increasing proportion of years spent in ill health. Some of the empirical evidence in the literature does indeed highlight how increasing longevity is accompanied by an increase in the number of years with morbidity [8, 37, 64, 67]. Consequently, recent scholarship has called for the inclusion of scenarios with a steepening expenditure profile by age group in all HCE forecasts [38].

However, the evidence for the development of morbidity is by no means uniform and numerous studies show a decline in morbidity or more nuanced findings, depending on the health indicator, the country and the population (sub)groups studied [36, 41, 54]. Following OECD [50], where the impact of population ageing on the demand for long-term care is estimated using a pessimistic, an optimistic and an average scenario for the extent of health ageing, we complement our “pessimistic” scenario S2 with two additional scenarios where increasing life expectancy is accompanied by a compression of morbidity.

Scenario S3 assumes that individuals not only live longer, but also in better health. In this scenario, we define a person’s “health age” such that from 50 onward, a person ages only 8 years in 9 calendar years for HCE-related processes (having a birthday only every 1.12 years). A person aged 88 will then have a”health age”of 84, thus adjusting age to increasing life expectancy. In this way, we implicitly also address the issue of time-to-death as a relevant cost parameter. HCE projections covering future cohorts with greater longevity may be biased upwards if they fail to account for the fact that a significant share of healthcare costs accrues at the end of life, irrespective of age [62]. By extending the number of healthy life years in conjunction with increasing life expectancy, we can take into account that certain costs in the healthcare system will be incurred later due to the ageing of the population, while the costs in earlier phases of life may be lower than at present. In our modelling approach, we limit this process to the over-50 age group. The choice to concentrate the gains in healthy life years on those aged over 50 is motivated by the fact that the healthcare cost profiles become steeper above that age (particularly for women, cfr. Figure 5 in Appendix) and that a large share of the projected gains in life expectancy at birth will be due to lower mortality at later stages in life [39]. As in the OECD study, the “pessimistic” and “optimistic” scenarios are complemented with a scenario where healthy ageing develops along a path averaging the other two scenarios.

In our “average” scenarios S4, we assume that only half of the gain in longevity (i.e. approximately 2 years) from population projections translates into a slower progression of “health age”.

Finally, scenarios S5 and S6 complement scenarios S3 and S4 by projecting how, in addition to a compression in morbidity, removing social inequality in health could affect HCE by closing the gap in the healthcare cost profiles between education groups over time. In both scenarios, all education groups converge to the healthcare cost profile of the high education group. Scenario S5 includes the “optimistic” assumption that healthy life years keep pace with increasing life expectancy, while scenario S6 includes the “average” assumption that half of the gains in longevity translate into healthy life years.

Assessing the economic relevance of HCE developments
The future development of absolute HCE is a relevant policy parameter, but it does not in itself say much about the resulting impact on public finances and the sustainability of the social protection system. Austria’s public health system provides nearly universal coverage and mandatory social security contributions linked to employment are its main financing source [6]. By comparing the development of healthcare costs with the development of labour supply, we can provide additional context to our HCE projections. The labour force is the most important pillar for financing the healthcare system, not only because of compulsory health insurance contributions but also because the number of economically active persons influences the overall tax base as well as economic growth [18, 44]. Unlike a purely demographic view based on old-age dependency ratios, comparing healthcare costs with the labour force can provide a more accurate outlook on the fiscal sustainability of healthcare.

For this reason, we contrast our cost projections with two different indicators for changes in the size of the labour force. Of course, a full assessment of the impact of HCE trends on public finances would require a number of additional modelling steps and assumptions, including changes in productivity and GDP, which are beyond the scope of this paper. Our aim is to show, on a more modest scale, how demographic change may affect not only public spending but also public revenues, using changes in the labour force as a proxy indicator. This allows us to highlight the role of changes in labour supply, and hence also labour force participation rates, as a highly relevant policy area to address the rising HCE associated with demographic ageing. It also implicitly relates cost increases to average incomes. If we assume that unit prices for healthcare services rise in line with increases in average wages, the development of the ratio between HCE and the economically active share of the population is a good approximation for how HCE evolves as a share of national income.

MicroDEMS distinguishes several labour market states and individuals can change their states throughout the simulation. While some of these transitions are based on decision models (such as retirement, which is automatically triggered when a person is eligible for retirement), most transitions are modelled based on a series of piecewise constant hazard rate regressions that account for individual characteristics such as age, education, gender and country of birth, as well as the duration of the current state. These estimations are performed on administrative social security data that cover the entire labour market history of almost all workers in Austria. A detailed description and corresponding estimation results are reported in Horvath et al. [33].

First, we use a ratio (DEPpop) defined as total costs (HCEtot) divided by the number of working age people (i.e. persons aged 20 to 64 years). Although the size of the working age population is widely used to calculate dependency ratios, it is a purely demographic indicator that does not necessarily accurately reflect the number of economically active persons in a society. In this respect, it is important to go beyond age and consider also economic characteristics of the population, such as length of schooling, retirement age, and labour supply behaviour [43, 57]. Looking into the future, we can expect the continuing educational expansion to reduce labour force participation at younger ages, while later retirement will extend working careers and a combination of higher education and shifting gender roles will continue to increase the labour force participation rates of women. In addition, participation rates will also be affected by the health status of the population. To capture these factors and changes, we use a second indicator (DEPlfs), defined as HCEtot divided by the number of economically active people (i.e. labour force participants, irrespective of age).

While the size of the working age population is pre-determined by demographic projections and thus exogenous to our model, we model the Austrian labour supply up to the year 2060 accounting for the impact of personal, family and job characteristics on labour force participation as well as for cohort-specific retirement regulations. Labour force participation and changes between different labour market states are determined by estimations based on Austrian Microcensus data as well as longitudinal administrative data [32]. The projections of future changes in the labour supply in Austria are thus consistent with external demographic forecasts but also account for the effect of compositional changes (such as education expansion or increasing labour force attachment of women) on labour supply. Detailed pension modelling in microDEMS also allows us to account for the impact of the ongoing harmonization of retirement age in Austria, increasing regular retirement age for women by 5 years over the next decade. For a more detailed description of the underlying methodology, please refer to Bittschi et al. [9].

To allow for a consistent comparison between scenarios, we project the working age population and the labour supply using the population characteristics underlying our scenarios S2 to S6, which correspond to official population projections, and use them together with HCEtot from scenarios S0 to S6 to calculate the ratios DEPpop and DEPlfs for each scenario.

Results
Figures 1 and 2 show the evolution of healthcare costs, expressed by total and per-capita costs (HCEtot and HCEpc). The left-hand panels highlight the cost trajectories resulting from the purely demographic scenarios S2, S3 and S4, whereas the right-hand panels focus on scenarios S5 and S6, which include the assumption of closing the socio-economic gap in healthcare profiles. Scenarios S0 and S1, which are the benchmarks for quantifying the impact that changes in specific cost-drivers and assumptions are expected to have on HCE, are included in both panels.

Fig. 1
figure 1
Total Health care expenditure (HCEtot) Notes/Source: Projections of future healthcare cost expenditure by scenario (Table 1) based on microDEMS. Relative change to 2020. Left panel shows scenarios S0 to S4, right panel shows scenarios S5 and S6 in comparison to S0 and S1

Full size image
Fig. 2
figure 2
Health care expenditure per capita (HCEpc) Notes/Source: Projections of future per capita healthcare cost expenditure by scenario (Table 1) based on microDEMS. Relative change to 2020. Left panel shows scenarios S0 to S4, right panel shows scenarios S5 and S6 in comparison to S0 and S1

Full size image
According to official projections, the Austrian population is expected to grow by approximately 13.5% between 2020 and 2060. For this reason, total healthcare costs as shown in Fig. 1 experience a stronger dynamic than per-capita costs shown in Fig. 2. Scenarios S0 and S1 must be considered separately in this respect, however, because the assumption that life expectancy will not increase any further means that population growth would only amount to 4.9% instead of 13.5% between 2020 and 2060. This also explains why the percentage changes in HCEtot and HCEpc are more similar in these scenarios than in the other scenarios.

As the projections for scenario S0 in Fig. 1 show, without further changes in mortality rates, and keeping gender-, age- and education-specific healthcare expenditures at their respective 2020 levels, total HCE would be approximately 20% higher in the late 2040 s than in 2020. In the following years the costs would decrease slightly, leading to a difference of 18% between 2060 and 2020. The ageing of the baby boomers is the main driver behind this pattern. Accounting for the educational expansion and for the related health gains (scenario S1) leads to a slightly more favourable development, with a cost increase of 15% in 2060 compared to 2020. From a per-capita perspective (Fig. 2), the cost curves for scenarios S0 and S1 show the same pattern, with a slightly lower increase until the end of the 2040 s and a flatter development thereafter. The HCEpc values are 13% (S0) and 10% (S1) higher in 2060 than in 2020. Together these scenarios highlight that, without increases in life expectancy, population ageing would have a comparatively modest impact on long-term cost dynamics, especially when factoring in positive health effects associated with the educational expansion.

Scenario S2, which incorporates expected increases in life-expectancy while again keeping health-care cost profiles constant, leads to a much stronger increase in HCEtot, exceeding 2020 levels by more than 41%. Per capita, the scenario results in a 26% increase in costs. In other words, incorporating increasing life expectancy in the projection more than doubles the per-capita cost dynamics that we can expect for the coming decades as a result of demographic change. This scenario, however, implicitly assumes that increases in longevity will be associated with an equivalent expansion of morbidity.

Scenarios S3 and S4 show how improvements in healthy life years could cushion the cost pressure resulting from demographic aging. While scenario S3 basically assumes that increasing life expectancy fully translates into a slowing process of ageing by proportionally increasing the time spent in each phase of the HCE age profile, scenario S4 attributes only half of overall life expectancy gains towards a slowed down ageing process. As Fig. 1 shows, the assumptions about changes in morbidity strongly affect HCE over time. Scenario S4 results in an increase in HCEtot of 33% compared to 2020, whereas in scenario S3 the increase amounts to 26%. From a per-capita perspective, healthcare costs would increase by 19% in scenario S4 and by only approximately 13% in scenario S3.

Scenarios S5 and S6 finally show how, in addition to a compression in morbidity, removing social differences in health could affect costs by closing the gap in the healthcare cost profiles between education groups over time. Assuming that all education groups converge to the high education group strongly reduces HCE over time. Assuming strong morbidity compression in scenario S5 (the “optimistic” assumption used also in scenario S3) results in HCEpc even lower than in 2020, by −5%. Owing to population growth, HCEtot would still increase over the projection period, by approximately 7%. In the intermediate scenario S6 (including the “average” morbidity assumption used also in scenario S4), overall HCEtot would increase by 15% compared to 2020 and HCEpc would remain roughly constant over the projection scenario.

Figures 6 and 7 in the Appendix provide some insights on how projected inpatient, outpatient and daycare services will develop over time, as well as the breakdown of costs by type of health service. To highlight the effect of changes in longevity, which represent the primary demographic cost driver, we focus on scenarios S1 and S2. Hospital nights account for the largest proportion of costs, accounting for 54.7% of the total in 2020. Without increasing life expectancy, this share would peak at 55.3% around 2050 and then decline slightly, to 54.4% in 2060. Due to changes in longevity, this cost component is increasing disproportionately: In scenario S2, which includes projected improvements in mortality, it increases by 47% up to 2060, whereas total costs increase by 41%. Therefore, potential efficiency gains in hospital stays would be particularly effective in reducing costs. For example, a 10% reduction in the length of stay by 2060 would lead to a 5.7% reduction in total costs.

Figures 3 and 4 present the results for the dependency indicators DEPpop and DEPlfs, helping to assess the impact that changes in costs will have on the financing base of the healthcare system. Projections for the working age population and also those for the labour supply lead us to expect a decrease in the number of economically active people in Austria in the coming decades. Although the total population will increase by 13.5% between 2020 and 2060, the number of people of working age (20 to 64 years) will decrease by 5%. Our projections for the number of economically active people (15+ years) are more favourable, with a modest increase by 2% over the period 2020 to 2060.

Both the DEPpop and the DEPlfs indicators highlight that rising HCE will represent a challenge for public finances. However, as expected, the assessment varies depending not only on the scenario but also on the dependency indicator chosen, with the indicator based on the projection of labour supply (DEPlfs) resulting in more favourable developments than the indicator based on the projection of the working age population (DEPpop).

In the most challenging scenario (S2), where we achieve greater life expectancy but no gains in health life years, the ratio of healthcare costs per working-age person (20 to 64 years) increases by close to 50% over the next decades. With respect to the economically active population, the picture is slightly more favourable, with an increase of approximately 40%.

According to scenarios in which population health improves and the number of life years with high healthcare use increases by less than life expectancy (S4) or even remains constant (S3), costs per working-age person rise by 33% (S3) to 40% (S4), while those per economically active person rise by slightly over 24% and over 31%, respectively. In the most ambitious scenarios, in which positive health developments are coupled with reducing socio-economic health inequalities, we can still expect HCE to grow more dynamically than the labour force. The trajectories of DEPpop and DEPlfs would however be much flatter. The former would increase between 2020 and 2060 by approximately 20% according to S6 and by approximately 12% according to S5, while the latter would increase by approximately 12% (S6) and 5% (S5).

Table 2 synthesizes the main results of the projections, showing how healthcare cost levels change according to the different scenarios and indicators between 2020 and 2060.

Discussion
In this study we used dynamic microsimulation to disentangle the impacts of changes in longevity, population age-structure, morbidity and socio-economic health differences on healthcare expenditure (HCE) in Austria. If current age-specific HCE were to remain constant, demographic changes would lead to a 41% increase in total HCE and a 26% increase in per-capita HCE by 2060, almost two-thirds attributable to decreasing mortality and one-third to the changing age composition of the population. Disregarding price changes in healthcare services, which represent a further uncertainty factor, this scenario is pessimistic, as it implicitly assumes an expansion of morbidity. We run two types of alternative scenarios in which morbidity is affected by two mechanisms: the first proportionally translating increases in life expectancy into increases in healthy life years, and the other closing the considerable gap in HCE currently observed between education groups.

Both mechanisms significantly mitigate– and together even offset – the impact of demographic change. While the evidence on the impact of increasing longevity on morbidity is inconclusive in the literature, leading to very high uncertainty in cost projections, the observed socio-economic gradient in health and health expenditure suggests considerable policy scope to influence health outcomes. The first mechanism addresses changes in morbidity that are entirely due to changes in longevity and simulates the effects on HCE of full, partial or no compensation for increasing life expectancy in terms of healthy life years. In this way, we also address the potential bias, highlighted by the “red herring” literature, resulting from projecting future healthcare costs based only on the changing age structure of the population. Closing the education gap additionally addresses changes in morbidity because of changes in population health that are unrelated to longevity and driven, for example, by the increasing diffusion of healthy lifestyles or the reduction of work-related health risks.

Of course, there are limits to the extent to which policy changes can affect the individual determinants of health care demand. This applies to demographic factors but also to the educational composition of the population, which is slow to change. However, our results are intended to show how much scope there is for mitigating the effects of demographic change on HCE, provided that the health of the population can be improved. The literature has identified numerous ways in which education may improve population health, but in many instances this link is not strictly causal and exclusive. Accordingly, it is not always necessary to address the educational level, but rather directly health-related determinants such as health literacy and risk behaviour. For example, anti-tobacco campaigns, tobacco control policies such as smoking bans, health warnings and tax increases have proven to be very effective for curbing smoking [1], and there is strong evidence that taxation on sugar-sweetened beverages is associated with significant reductions in sales [2].

Our projections do not directly address the price dimension of healthcare services, such as shifts in relative prices resulting from technological innovation combined with shifts in supply and demand. To facilitate the economic interpretation of changes in HCE, we relate projected total costs to changes in the projected labour force. It is noteworthy that the anticipated rise in total HCE until 2060, spanning between + 7% and + 41% depending on the scenario, closely aligns with the projected increments in costs per economically active person. In contrast, increases in costs per capita are lower (as the population size increases), while increases in costs per working age person (20–64) are higher (as it is the dependent-age population which increases over-proportionally). Combining HCE with labour force projections on one hand enables the quantification of the beneficial impact of increases in labour force participation (mitigating projected cost escalations by approximately 7 to 9 percentage points comparing active age to economically active persons, see Table 2). Moreover, cost increases can be related to average wages, i.e. the increase in projected costs provides a robust measure from the perspective of an average worker, assuming increases of unit-prices of health services in the range of increases in average wages.

All our scenarios indicate that demographic ageing is likely to increase future healthcare costs, even after taking into account a marked compression of morbidity over time. In addition to the high uncertainty about future cost dynamics, reflected in a wide range of outcomes across the scenarios, the observed socio-economic gradient in health and health expenditure suggests considerable policy scope for influencing outcomes. In our scenarios, we focus on factors affecting population health and thus the demand for healthcare services, but efficiency gains and improvements in the supply of services can make important contributions to adapt the healthcare system to the needs of an ageing society. According to our projections, demographic change will lead to an over proportional increase in inpatient costs. Therefore, potential efficiency gains in hospital stays would be particularly effective in reducing costs. For example, a 10% reduction in the length of stay by 2060 would lead to a 5.7% reduction in total costs.

To check the validity of our results, we can compare our HCE projections with the projections in the latest Ageing Report of the European Commission [28], which to our knowledge is the only current study that projects healthcare costs for Austria over a similar time horizon (2022 to 2070). The Ageing Report projects healthcare costs as part of a macroeconomic model and sets the costs in relation to GDP, whereby the underlying assumptions and scenarios differ in part from ours. When the results are converted to per capita expenditure at fixed prices and limited to the more comparable scenarios, the projections nevertheless show comparable results.

In the Ageing report’s baseline scenario, which assumes that half of the additional years of life gained through higher life expectancy will be spent in good health, Austria’s healthcare expenditure increases from 7.8% to 8.8% of GDP between 2022 and 2060. If we account for the Ageing Report’s macroeconomic assumptions that GDP will expand by 1.3 percent and unit costs in the health sector will develop by 1.1 percent per year over the projection period, this implies that HCE will increase by 15% per capita. In the “pure demographic” scenario, which deviates from the baseline by assuming that the elasticity of demand for healthcare with respect to income is 1 rather than 1.1, HCE will amount to 8.6% of GDP in 2060. In absolute values, this corresponds to a per capita increase by 13%. Based on the underlying assumptions, these figures can be compared with the results for our scenario S4, where per capita HCE increases by 19% between 2020 and 2060.

Conclusions
Our results help to shed light on the relevance of different cost determinants and can provide guidance to policy-makers when seeking to adapt healthcare systems to demographic change. While specific to Austria, our findings can be of interest for other advanced economies with a comprehensive public healthcare system. They also underscore the advantages of using dynamic microsimulation in combination with official demographic projections and health systems accounts to provide consistent what-if scenarios and long-term projections.

At the same time, our study has limitations that will have to be addressed in future work and provide scope for further research. In the absence of better data, we used cross-sectional data and had to make the simplifying assumption that age profiles of healthcare costs can be used to project life-course cost trajectories. The validity of this assumption depends on the extent to which the positive relationship between HCE and age is determined by time-to-death, a question that must be answered empirically [13]. While providing different what-if scenarios to estimate outcomes depending on the relationship between changes in mortality and in morbidity, in this study we did not explicitly model the relationship between healthcare consumption and time-to-death. The availability of longitudinal data will be crucial to enable future research to investigate in greater detail and with greater accuracy the relationship between ageing and healthcare expenditure, particularly toward the end of life. This research should include long-term care (LTC) and the resulting costs in the analysis, as population ageing is likely to have a different impact on LTC than on other healthcare services [38] and we can expect ageing to have a larger impact on LTC than on acute care [13]. Also, it would be interesting to include costs for preventive care and to explore the interaction between these costs and the costs for curative care.

 

 

 

Read more @bmchealthservres biomedcentral