Long-Range Forecasting for Emergency Care Systems in a Highly Dynamic Setting: A Singapore Case Study

Abstract

Objectives

In aging societies such as Singapore, emergency care systems (ECSs) face the challenge of ever-increasing demand for urgent care. The uncertainties around population aging are compounded by the downstream effects of population health interventions. We forecast the demand for ECS use in Singapore, in terms of emergency department length of stay (ED-LOS) hours per person per year, until 2050, with and without the effects of a major intervention that enhances the role of primary and community care.

Methods

Using a system dynamics simulation model, we applied age-specific emergency department usage rates—derived from our analysis of 1,736,405 attendances in a large tertiary hospital and stratified by acuity—to an age-stratified population forecast. We simulated a baseline and 4 intervention scenarios based on different efficacy-time and effect size levels.

Results

In the baseline scenario, median ED-LOS increased from 580 hours per 1000 residents per year in 2010 to 644 hours per 1000 residents in 2050. Median ED-LOS increased from 276 to 372 hours per 1000 residents per year for high-acuity patients, whereas it decreased from 302 to 274 hours per 1000 residents per year for low-acuity patients. Under all intervention scenarios, low-acuity ECS use per person decreased, caused by “decanting” of this patient group to primary care. However, high-acuity ECS use per person increased because of longevity.

Conclusions

In the long term, an overall increase in ECS demand is driven by an increase in high-acuity use; population health interventions can further exacerbate the high-acuity burden. Our work casts light on the relatively less-studied dynamics of long-term ECS demand.

The Bottom Line.

We forecast emergency care system (ECS) demand in terms of emergency department length of stay (ED-LOS) hours per person per year, until 2050, under a baseline scenario, as well as under 4 scenarios relating to a major population health intervention. In the baseline scenario, total ED-LOS hours per capita will increase, with high-acuity ED-LOS demand overtaking low-acuity demand. Under the intervention scenarios, the low-acuity demand per person will decline, but the high-acuity use per person will increase. Somewhat counterintuitively, improvements in population health will mean greater demands on the ECS system.

1. Introduction

1.1. Background

It is crucial to have reserve emergency care system (ECS) capacity for resilience against surges caused, for example, by pandemics or mass casualties. An aging population is a threat to the ECS reserve capacity. ECS demand is higher for older people in Australia.^1,2 In the USA, emergency department (ED) patient acuity increases with age;³ population aging will impact ED capacities.⁴ In Singapore, physical, cognitive, and social frailty among the elderly has been associated with adverse health outcomes.5, 6, 7 Japan^8,9 and Finland^10,11 are leading exemplars of rapidly aging nations; our analysis (see Supplementary Appendix 1) shows that Singapore is on a similar path.

1.2. Importance

Assessing long-range ECS demand is essential, considering the long lead time to train health care workers. The high-stress ECS environment makes it staff retention difficult; strain and burnout are prevalent among ECS providers globally.12, 13, 14, 15 An undersupply of providers will worsen burnout.

In the last 2 decades, Singapore has experienced disproportionately high ECS demand relative to its population growth, motivating stakeholders to consider alternate care models.16, 17, 18 In keeping with previous transformations,¹⁹ in 2022, the Ministry of Health launched a national population-level strategic initiative called Healthier SG.²⁰ The key features of this initiative include linking every Singapore resident with a family doctor, providing more extensive community care, strengthening technology platforms, and changing the health care funding model. The initiative is expected to increase longevity. The combination of population aging and the potential effects of this initiative creates a highly dynamic setting.

ECSs are among the most commonly modeled health systems, mainly through simulation.²¹ However, ECS simulation studies focus on short-term aspects such as crowding, resourcing, and waiting time.^21,22 Few studies have delved into long-range ECS demand. Pallin et al.⁴ found that in the USA, the ED length of stay (ED-LOS) will grow at a rate 10% faster than population growth due to aging, until 2050.⁴ Other work includes a single-ED study in Spain to estimate capacity additions required over a 15-year time horizon and an assessment of estimated emergency physician supply and demand in USA in 2030.^23,24 Thus, our study addresses a gap in research though its subject is a vital health infrastructure.

1.3. Goals

We forecast the demand for ECS use in Singapore, in terms of ED-LOS per person per year, until 2050, with and without the effects of a major intervention that improves longevity.

2. Methods

Adopting the approach of Pallin et al,⁴ we estimated the effect of age on ECS demand and projected future ECS demand as a function of the changing population age structure. However, we extended their analysis in 4 ways. First, we analyzed 9 years’ data (rather than a single base year) from the ED of a tertiary care hospital in Singapore, henceforth known as the study hospital (SH), for estimating the age-demand relationship.²⁵ Second, we used the system dynamics (SD) simulation,^26,27 a methodology that pays special attention to stock-and-flow structures, feedback, and dynamic behavior over time, and is thus well-suited to the nuances of ECS demand forecasting. Third, we stratified ECS demand by acuity. Finally, we assessed the impact of a major population health intervention on future ECS demand.

2.1. Study Design and Scenarios

2.1.1. Study design

Together with other simulation methods such as discrete event simulation and agent-based modeling, SD modeling is traditionally used for forecasting in health and other domains.²⁸ Recently, SD was used to forecast the requirements for physicians in general,^29,30 as well as patients with or health care workers for particular disease conditions.31, 32, 33 SD is preferred when the focus is on system-wide analysis rather than detailed component simulation.^28,34 Furthermore, it facilitates precise modeling of aging populations³⁵ and is particularly suited to the dynamic tendencies of complex systems.³⁶ These factors led us to choose SD modeling as the appropriate approach.

Our model, shown in simplified form in Figure 1, was built in Vensim DSS version 8.0.9. Its time horizon is 2010-2050. Instructions to use the model are provided in Supplementary Appendix 4 and the model files are included as Supplementary Appendix 5. The model has 3 sectors or submodels: population, ECS demand, and scenarios. Figure 1, following SD conventions, differentiates between stocks (“Resident population, stratified by age” and “Effect of unmet primary care needs”) and flows (eg, “Births”). Stocks are depicted by rectangles, and flows are depicted by lines with valves. The on-off switches represent scenario settings, which can be toggled between high or low levels.

Figure 1.

Schematic representation of the ECS demand forecasting model. ECS, emergency care system; ED-LOS, emergency department length of stay; SH, study hospital.

The population submodel, shown in the top left of Figure 1, was earlier used in several population health studies.37, 38, 39 It is based on an “aging clock” mechanism. The model includes the resident population of Singapore (citizens and permanent residents) stratified by age and gender and disaggregated into 1-year age cohorts. Nonsurvivors of each age cohort exit the system through the “Deaths” outflow, determined by the age-specific mortality rate. Survivors age a year as the simulation clock advances a year. The “births” inflow is determined by the birth rate and the fecund female population. The submodel is calibrated using publicly available data from the Singapore Department of Statistics. The resident population starts at about 3.7 million residents in 2010, increases to about 4.1 million by 2030, and declines to about 3.7 million in 2050. The proportion of those aged 65 years and over increases from 14% in 2010 to 39% in 2050.

The bottom half of Figure 1 explains the ECS demand submodel. Supplementary Appendix 3 provides the relevant model equations. ECS usage comprises appropriate and inappropriate use, the latter being defined as use by low-acuity patients who can be effectively managed in primary care.⁴⁰ Inappropriate use is primarily driven by gaps in primary care.¹⁸ ECS usage, in terms of LOS-hours and stratified by acuity and age group, was computed as a function of age-specific ECS usage per person per year and the age-stratified population. Inappropriate use was assumed to arise from the low-acuity patient group.

We simulated the effect of a general population health improvement program such as Healthier SG (henceforth called “PopHealth”) that could address unmet needs in primary care, lower inappropriate ECS use, as well as reduce mortality rates. This is shown in the top right of Figure 1.

2.1.2. Scenarios

The study explored potential ECS demand trajectories under 5 scenarios: a baseline and 4 intervention scenarios. The baseline models the effect of aging, without PopHealth. PopHealth’s impact depends on 2 factors: time to efficacy (10 or 15 years) and effect size (0.5 or 0.9; PopHealth’s efficacy is 50% or 90% of its full potential in reduced relative risk of mortality and reduction of inappropriate use accordingly). Intervention scenarios involve different combinations of these factors, as outlined in Table.

Table.

Scenario construction.

Scenario	PopHealth implemented flag	PopHealth effect size	PopHealth time to full efficacy, y
Baseline	0	Does not matter	Does not matter
Larger and faster	1	0.9	10
Larger and slower	1	0.9	15
Smaller and faster	1	0.5	10
Smaller and slower	1	0.5	15

2.2. Setting

Singapore is a city-state with a total population of about 6 million people and a resident population of about 4 million.⁴¹ The nation is rapidly aging—the old age dependency ratio (number of residents aged 65 years and above per 100 residents aged 15-64) was 26.4 in 2024, up from 15.2 in 2014.⁴¹ It ranks ninth globally on the Human Development Index, with life expectancy at birth being 84 years.⁴²

Singapore’s health care infrastructure delivers universal health coverage through mixed public-private financing.⁴³ Primary care is provided by private sector general practitioner clinics and public sector polyclinics. Eighty percent of primary care volumes are managed in the private sector, with the public sector focusing on chronic disease management. In direct contrast, about 84% of ED attendances are in public sector hospitals.⁴⁴ Although ED services are generally offered 24/7, primary care services are limited outside regular hours. Gaps in primary care contribute to inappropriate ED use.^18,40 Though total ED beds increased from 287 in 2007 to 628 in 2021, EDs are stressed by waiting time for inpatient beds and do resort to ambulance diversion.⁴⁴

Study data were extracted for years 2010-2018 from the SH electronic data warehouse⁴⁵ in the first week of August 2023. The SH is a large public acute care hospital serving about 130,000 attendances each year. Attendances at the ED may be walk-in or ambulance-transported patients. Patients are triaged into 4 acuity categories, P1, P2, P3, or P4, in decreasing order of acuity.⁴⁶ P1 patients require immediate life-saving attention, whereas P2 patients are stable but require urgent attention to prevent deterioration. P3 patients can walk but may have moderate symptoms; P4 patients are nonemergencies. In this study, P1 and P2 categories, who must be treated in the ED, were considered as high acuity, and P3 and P4, who can or should be treated outside the ED, were considered as low acuity.

2.3. Measures

The outcome measure of interest was ED-LOS per 1000 residents per year, stratified by acuity. Pallin et al.⁴ found that in the USA, aging would increase ED-LOS, but not the number of visits.⁴ Several studies confirm that ED-LOS is associated with ED crowding.^47,48 We combine visits per year and ED-LOS per visit in our outcome measure, which reflects the annual ED resource consumption intensity per capita in the country.

2.4. Data Analysis

SH-level analysis was used to estimate age-specific ED-LOS per SH patient per year, stratified by acuity. To this, an SH patient-to-population scaling factor was applied to estimate age-specific ED-LOS per capita per year, stratified by acuity (see Fig. 1). The SH- and national-level analyses are explained in this subsection.

2.4.1. SH-level analysis

Figure 2 shows the data wrangling process. After deduplication, 1,736,405 records detailing ED use from 2010 to 2018 were used in our analysis (clean data beyond 2018 were not available). We conducted an exploratory analysis of ED attendances and ED-LOS by age and acuity, and estimated ED-LOS per SH patient per year, stratified by acuity and by age group.

Figure 2.

Process flow for extraction of detailed data on emergency care system usage in the study hospital. LOS, length of stay.

2.4.2. National-level analysis

There were 6 groups of inputs and guesstimates for the national-level analysis (detailed in Supplementary Appendix 2): (1) population demographics; (2) a scaling factor to convert SH-level ECS usage to population level; (3) relative risk of mortality under PopHealth; (4) baseline levels of inappropriate attendance in the ED; (5) PopHealth efficacy achieved over time; and (6) PopHealth effect size.

The estimated ED-LOS figures (Supplementary Appendix 2 item 3; Fig. 4) reflect demand from ECS users in the SH. This demand was extrapolated to the population level. As detailed in Supplementary Appendix 2, this was done by comparing SH-specific attendances with population-level attendances, the latter being based on public data on emergency attendances⁴⁹ and population data from the census of Singapore.⁴¹ The value of the scaling factor was estimated to be 0.19.

Figure 4.

Attributes of demand per age group on the emergency care system in the study hospital (SH). ED-LOS, emergency department length of stay; LOS, length of stay.

We assumed that PopHealth’s effect on mortality rates stems from 1 significant change: enhanced continuity and availability of primary care. Our input for the simulation was based on findings from a previous study on the effect of continuity of care with a primary physician on mortality rates.⁵⁰

Data on inappropriate attendances were derived from a Singapore study,⁴⁰ which showed that 9.6% of all ED attendances are inappropriate. Assuming that inappropriate attendances are low acuity, as 49.2% of all attendances were low acuity, the baseline level of inappropriate attendances was 19.5% of low-acuity attendances.

For data on the time taken by PopHealth to reach full efficacy, we constructed s-curves reaching a value of 100 in 10 and 15 years, based on the diffusion of innovations paradigm.^51,52

3. Results

In this section, we first summarize the ECS demand in the SH. Next, we present the long-range ECS demand forecast based on simulation results.

3.1. ECS Demand in SH

Figure 3 depicts the median ED-LOS per attendance (top panel) and attendances per year (bottom panel) versus age. The median LOS for high-acuity attendances was consistently higher than that of low-acuity attendances.

Figure 3.

Emergency department length of stay per attendance and attendances per year as a function of patient age, 2010-2018.

Figure 4 depicts ECS usage in the SH by age group. Each age group is represented by a dot. Annual ED-LOS in hours per patient is plotted on the horizontal axis and the percentage of high-acuity ED-LOS is plotted on the vertical axis. Both the quantum of ED-LOS per patient per year and the high-acuity proportion of this usage increase with age.

3.2. Long-Range ECS Demand in Singapore

In the baseline scenario, ECS demand per capita shifts as a result of aging. Figure 5 shows baseline forecasts based on 100 simulations, where the scaling factor for conversion of SH to population usage ranged between 0.17 and 0.21, following a uniform distribution. Overall, median ED-LOS in the population increased from 578 hours per 1000 residents per year in 2010 to 645 hours in 2050. Notably, high-acuity LOS increased with the median value going up from 276 to 372 hours per 1000 residents per year, whereas the median low-acuity use decreased, going down from 302 to 274 hours per 1000 residents per year. Per person high-acuity use overtook low-acuity use in 2017.

Figure 5.

Baseline forecast of emergency care system demand per 1000 residents. ED-LOS, emergency department length of stay.

Figure 6 shows the ECS demand trajectories under the 4 PopHealth scenarios in addition to the baseline scenario. Here, demand is plotted with ED-LOS hours per 1000 persons per year on the horizontal axis and the percentage of high-acuity ED-LOS hours per person per year on the vertical axis.

Figure 6.

Future trajectories of emergency care system demand under the different scenarios. Each point on a line plots the demand profile of a particular year and scenario, with the lines representing the demand trajectories from 2010 to 2050; the axes do not originate at 0. ED-LOS, emergency department length of stay.

4. Limitations

We have modeled annual ECS demand in Singapore until the year 2050 in terms of ED-LOS hours per capita, stratified by acuity. A limitation of our work is that there is no public data to compare against, and this makes operational validation, that is, validation of model outcome measures against real-world measures,⁵³ impossible. We do have conceptual validation⁵³ for the model, through the SH-level analysis as well as discussion of the model design and results with ECS stakeholders.

A second limitation of the present work is that it does not take a granular view of disease conditions, unlike a recent study that assessed long-term ED-LOS under different levels of burden of noncommunicable diseases in Spain.²³ Although studying particular conditions is valuable, it only provides a partial view unless a comprehensive set of diseases, including dementia and cancers, is considered. Age acts as a proxy for the total burden from all conditions in our work. We assess future ECS demand with a “roughly right” analysis based purely on aging as a logical first step.

Aging has an impact not only on the demand side (the focus of our work) but also on the supply side because it leads to a shrinking pool of people who can provide care (though in Singapore, well-established talent import mechanisms do mitigate shortages of care providers). We do not model the human resource implications of changing demand. Furthermore, our present work does not examine the dynamics between capacity and ED-LOS.

Additional limitations include that this is a single-site study and that it does not model spikes in demand. We intend to refine our model following discussions with ECS stakeholders.

5. Discussion

This study projects the ECS demand for Singapore, with its rapidly aging population, accounting for the long-term effects of a large-scale population health intervention. Our analysis of SH data confirms the intuition that patient age has a bearing on attendances per year, ED-LOS per attendance, and the proportion of low- and high-acuity uses. The bottom panel of Figure 3 shows a “hump” in the age range of 17 to 24. This hump, also observed in another study,⁵⁴ is driven by low-acuity ECS use (in civilian hospitals) by military servicemen.⁵⁵ Aside from this marked hump, attendances per patient per year generally increase with age. Up to age 49, there are more low-acuity attendances per patient per year; at age 49 years, the latter overtakes the former and the gap widens.

In the baseline, total demand goes through a steady increase (Fig. 5). Underlying this apparently unremarkable increase, there is a fundamental change in the mix of high-acuity vs low-acuity demand. Health system planners need to consider ECS demand trajectories not only in terms of volume of use but also the proportion of high-acuity use (Fig. 6). Baseline demand takes a linear path in the direction of higher LOS per capita and a higher proportion of high-acuity LOS in the ED. Under intervention scenarios, the trajectory changes substantially. As PopHealth’s benefits kick in, annual ED-LOS per capita decreases as inappropriate ECS attendances decline, giving rise to the first set of inflexion points. Over time, PopHealth results in a larger pool of aged people, leading to a higher annual ED-LOS per capita and an increased proportion of high-acuity ED-LOS. The leftward bends reflect fewer inappropriate attendances; the rightward bends are caused by high-acuity attendances increasing as better care results in more aged people to care for. The health system as a whole, not just the ECS, will need to respond flexibly to shifting demand patterns.

On the one hand, PopHealth strengthens the primary care system so that inappropriate use of the ECS diminishes. On the other hand, because PopHealth takes effect and improves longevity, more people live with frailties that increase ECS service demand, especially high-acuity workload. High-acuity attendances are marked by longer, more intensive stays. Studies conducted on Canadian, German, and US populations have shown that the time spent by emergency physicians on each patient significantly increases with acuity.56, 57, 58, 59 For ED nurses too, direct care time (time spent with the patient) varies with patient dependency level.⁶⁰ This has repercussions for long-term human and physical resource planning.

Advance planning and investment in primary care capacities, including alternate service pathways,¹⁸ is critical to plug existing capacity gaps. The low-acuity ECS demand reduction in the PopHealth scenarios observed here requires timely and adequate investments in primary care. The increased load of high-acuity patients in the ED will call for adequate emergency medicine manpower. Failure to account for the shifts in ECS demand will lead to adverse health outcomes for patients; physicians will experience burnout, potentially leading to high attrition of personnel.^61,62

Policy interventions have “delayed and distal” effects; virtual simulations provide “low-cost laboratories for learning” in this context.²⁶ One may think that an intervention such as PopHealth should reduce the future load on the ECS. Simulation modeling provides evidence to the contrary. By making our model available to researchers, we hope to stimulate further research on this critical topic.

Author Contributions

Fahad Javaid Siddiqui: Conceptualization, Methodology, Investigation, Resources, Writing – Original draft, Writing – Review & Editing, Visualization, Project administration.

Ashish Kumar: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – Original draft, Writing – Review & Editing, Visualization, Project administration.

John P. Ansah: Conceptualization, Methodology, Software, Validation, Writing – Review & Editing.

Guo Yuan: Formal analysis, Investigation, Data curation.

Zhenghong Liu: Writing – Review & Editing.

Rahul Malhotra: Writing – Review & Editing.

Marcus Eng Hock Ong: Writing – Review & Editing, Supervision.

Sean Shao Wei Lam: Conceptualization, Methodology, Validation, Resources, Data curation, Writing – Review & Editing, Visualization, Supervision, Project administration.

Funding and Support

No funding was received for this work.

Conflict of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.