Time Trends in the Rates of ED Visits in Australia, an Age‐Period‐Cohort Approach

ABSTRACT

Objectives

Assess the effects of Age, Period and Cohort (APC) in the trends in emergency department (ED) visits in 2003–2023 in Australia.

Methods

For this retrospective observational study, we obtained data on all ED visits in Australia in 2003–2023 from the Australian Institute of Health and Welfare and applied an APC model to separate Age, Period and Cohort effects.

Results

The total number of ED visits increased from 4,306,183 in 2003 to 7,194,861 in 2013 (67% increase). A 25% increase was observed in 2013–2023 (n = 9,014,526 visits in 2023). Rates of ED visits have increased in Australia in 2003–2023 from 218 per 1000 residents to 338 per 1000. Rates of ED visits were higher in patients aged > 85 years and have increased in 2003–2023 from 525 per 1000 to 835 per 1000 (59% increase). The increase in the number of ED visits per capita was slightly lower in those aged 65–74 (from 226 per 1000 in 2003 to 338 per 1000 in 2023, 50% increase) and those aged 75–84 years (350 per 1000 to 526 per 1000, 50% increase) compared to residents aged > 85 years.

Conclusions

Our APC modelling shows a deceleration in the increase in the rate of ED visits in the last decade, despite at the same age younger generations having higher rates of ED visits than older generations. The consistent increase in ED visits per capita in those aged < 75 years has major implications for healthcare planning and policies needed to reduce ED demand.

1. Introduction

In Australia, the number of Emergency Department (ED) visits has increased to unprecedented levels (from 5.8 million ED visits—292 presentations per 1000 residents for the financial year 2003–04 to 9 million visits in 2023–24 – 333 presentations per 1000) [1, 2] and this is a source of pressure for hospitals and healthcare systems. The Australasian College of Emergency Medicine has referred to this as a national state of emergency [3], highlighting that hospitals have never been busier, and that overcrowding, ambulance ramping, and workforce issues are major problems. ED crowding is associated with delayed care [4] and adverse outcomes for patients including avoidable deaths [5, 6]. ED crowding and associated negative health outcomes for patients and reduced performance of healthcare systems are not specific to Australia but a global problem [5, 7, 8, 9].

It is important to understand longitudinal trends in the number of ED visits in general and ED visits per capita and the sustainability of these trends against the capacity of hospitals to manage ED demand. However, the rapid increase in the number of ED visits is still not well understood. The Age‐Period‐Cohort (APC) modelling approach is suitable for the analysis of longitudinal trends in the number of ED visits per capita [10]. Risks of many conditions and ED visits are associated with age (Age‐effects) [11]. Period‐effects represent changes in the number of ED visits per capita that can affect all age groups at the same time [10]. The Cohort‐effects represent changes for the same birth cohort across age groups and time periods [12]. When these effects are analysed with an APC approach, they have the potential to provide insights on identifying periods of time when determinants of the increase in ED visits have more impacts on the observed trends. This approach has also the potential to highlight changes in the number of ED visits that are related to specific birth cohorts who have been exposed to factors that made them more susceptible to experiencing more ED visits while ageing.

This study assessed longitudinal trends in the number of ED visits per capita in Australia using data from major public hospitals as reported to the Australian Institute of Health and Welfare (AIHW). These trends were analysed using the APC approach to disentangle the effects of age, period and cohort.

2. Methods

2.1. Data and Study Design

For this retrospective observational study, aggregated numbers of ED visits by age groups for each year from 2003 to 2023 were obtained from AIHW who curate and maintain a nationally‐consistent reliable source of public hospital activity [13]. ED visits data for all years from 2003 to 2023 were used for describing time trends in ED visits per capita and for investigating the effects of Age, Period and Cohort with APC models. Additional analyses where we investigated the number of ED visits per capita of different birth cohorts across age groups and the number of ED visits per capita of different birth cohorts at the same age group required the same grouping (10 years) for period and age. Therefore, the ‘0–4 years’ age group was excluded from these latter graphical representations and data for the years 2003, 2013 and 2023 were selected to assess changes across two decades. Age groups provided within the AIHW data were: 0–4 years, 5–14, 15–24, 25–34, 35–44, 45–54, 55–64, 65–74, 75–84 and ≥ 85 years. Also, because the covariates age, period and cohort are included in the model as continuous variables, the age group variable was re‐coded to reflect the centre of class for each category. For example: ‘5–14 years’ became 10 years, representing those who were aged 10 years on average at a given period. A similar approach was taken for the year of birth. To be able to compute ED visits per capita, denominator populations for each age group and calendar year were obtained from the Australian Bureau of Statistics (ABS) [14].

We are reporting our study according to The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) (see the STROBE checklist Supporting Information).

2.2. Statistical Analysis

We calculated the number of ED visits per capita by dividing the total number of ED visits by the corresponding denominator population for each year and age group, multiplied by 1000 (number of ED visits per 1000 residents). The APC approach included the graphical presentation of the number of ED visits per capita by age for each year of ED presentation; the number of ED visits per capita by age for each birth cohort category; and the number of ED visits per capita by year of birth for each age group.

We modelled the number of ED visits per capita using a binomial distribution. A logistic regression model with logit link is considered the natural modelling approach for rates [15]. The statistical analysis was designed based on a Lexis diagram. A Lexis diagram is a two‐dimensional representation with calendar time (period) on the x‐axis and age on the y axis, and individuals move diagonally upward over time while they are ageing. These properties of the Lexis diagram require the same group for Age, Period and Cohort. Therefore, we conducted a sub‐analysis which is descriptive where ED cases were enumerated for each 10 × 10 × 10‐year subset of the Lexis diagram (Figures 1A and 2A,B) where we can observe changes in the number of ED visits per capita for the same cohort at different timepoints and across age groups. This also allowed us to compare different birth cohorts at the same age [12]. Figure 1A presents an example of a Lexis diagram with incomplete data, included for illustration purposes only. The complete version of the data, which is easier to interpret, is shown in Figure 2A,B. Because the APC statistical model does not make any assumption about the length of grouping of the independent variables, we were able to have the < 5 years old age group among the 10‐year groupings in logistic regression for APC modelling. The variables age, period and cohort are interdependent: if any two are known, the third is determined. For example, a person aged 10 years in 2003 had to be born in 1993 (Cohort = Period−Age); someone born in 1913 and aged 90 years was necessarily surveyed in 2003 (Figure 1A).

FIGURE 1.

(A) Lexis diagram, ED data in Australia from 2003 to 2023. We display the number of ED visits per capita in a 10‐year grouping for age and cohort observed a decade apart. The squares in light blue correspond to a cohort of residents born in 1993 on average (1989–1998) followed over time, and their rates of ED visits are noted from 2003 to 2023. The orange squares represent those aged 90 years in 2003, 2013 and 2023. This allows us to compare rates of ED visits in those who 90 years old in 2003 versus those who had the same age in 2013 and 2023. Figure 1B presents trends in the number of ED visits per 1000 residents over time for each age group.

FIGURE 2.

(A) Number of ED visits per 1000 residents by age for each year of birth. This figure follows generations while they are ageing; their rates of ED visits are observed at different calendar years (2003, 2013, 2023). Figure 2B. Number of ED visits per 1000 residents by year of birth for each age. In the figure, we can observe (compare) rates of ED visits for different generations at the same age.

We aimed to investigate the independent effects of age, period and cohort. However, these factors cannot be analysed in the same model due to collinearities. Collinearity occurs when variables in the model are not independent and are therefore highly linearly related (correlated). In the case of this study, the linear trend known as ‘drift’ cannot be uniquely attributed to the effects of Period or to the effects of Cohort. Therefore, the fully unrestricted APC model is not identifiable without additional constraints or choices about how to allocate the drift. To overcome this, we adopted parametrisations of the model—that is a different way of setting up the model—that can attribute the linear effect (drift) to Period‐effects or the Cohort‐effects while one of these two is assessed as a non‐linear effect [16]. Age‐effects are set as rates for the reference cohort or for the reference period when one of the two has the linear effect. For the effects of Period, 2003 was set as a reference. While for the Cohort effects, 1963 was chosen as reference. The choice of the reference cohort (and reference period) is arbitrary. However, it is common practice to choose a middle cohort as reference. To investigate the goodness of fit of the full APC model, we compared different statistical models that include combinations of Age, Period, Cohort, and ‘drift’ effects to explain variation in rates ED visit. The drift is the linear effect that we cannot uniquely attribute to the effects of Cohort or the effects of Period because of the collinearity.

R version 4.3.2 was used for the analysis. R package Epi version 2.48 was used for the Age‐Period‐Cohort modelling.

3. Ethics

An ethics exemption for the study was obtained from the Metro South Human Research Ethics Committee (EX/2022/QMS/89905) on the basis that it is a quality improvement study with the intent to publish the findings.

4. Results

Overall, the total number of ED visits increased from 4,306,183 in 2003 to 7,194,861 in 2013 (67% increase). A 25% increase was observed between 2013 and 2023 (n = 9,014,526 visits in 2023). Overall, rates of ED visits per capita have increased in Australia between 2003 and 2023 from 218 per 1000 residents to 338 per 1000 residents (55% increase). The number of ED visits per capita increased for all age groups between 2003 and 2023 (Figure 1B). For Australian residents aged < 5 years, the rate increased from 410 per 1000 in 2003, 493 per 1000 in 2013 (20% increase), and to 547 per 1000 residents in 2023 (11% increase between 2013 and 2023). Among those < 75 years, rates of ED attendances appeared to be higher among residents around 0–4 years of age. Rates of ED visits were higher in patients aged > 85 years and have increased from 2003 to 2023 from 525 per 1000 to 835 per 1000 residents (59% increase). The increase in the number of ED visits per capita was slightly lower in those aged 65–74 years (from 226 per 1000 residents in 2003 to 338 per 1000 residents in 2023, 50% increase) and those aged 75–84 years (350 per 1000 to 526 per 1000 residents, a 50% increase).

Figure 2A allows us to follow cohorts of residents while they are ageing. Australians born in 1993—aged 10 years on average in 2003—experienced 161 visits per 1000 residents in 2003; this rate increased in 2013–340 per 1000 residents at 20 years of age—and a slight decrease afterwards: 300 visits per 1000 residents in 2023 at 30 years of age, which is an age bracket that appears more protective for seeking emergency care. After the age of 30, ED visit rates appeared to increase while generations were ageing.

In Figure 2B we observe different generations at the same age. At the same age, younger generations experienced higher rates of ED visits per capita than older generations. For example, at 90 years of age, the rate of ED visits for Australians born in 1913 was 525 per 1000 residents, 665 per 1000 residents for those born in 1923 and 794 per 1000 residents for those born in 1933.

The model in Figure 3 was parametrised so that the Cohort‐effects include both linear and non‐linear effects. The Period‐effects represent non‐linear effects of period only and can be interpreted as accelerations and decelerations. The Age‐effects represent rates of ED visits per 1000 residents for the reference cohort (1963). Results showed that the younger the generation, the higher the odds of ED visits per capita. The non‐linear Period‐effects show accelerations in the increase of ED visits between 2003 and 2013 and decelerations in the last decade.

FIGURE 3.

Results of the Age‐Cohort‐Period model with ‘ACP’ parametrisation. Age‐effects are rates for the reference cohort (1963), rates for those born in 1963 at different ages. Cohort‐effects (red line) are odds ratios relative to the reference cohort but this cohort effect includes a constant increase over time (linear) component that cannot be attributed to the effects of Cohort or the effects of Period. The blue line represents the Period‐effects that are set to 0 on average to represent the non‐linear effects of period. These Period‐effects show an accelerated increase in the first decade while the increase in the rate decelerated in the last decade.

For results displayed on Figure 4, the model parametrisation has been set to include non‐linear cohort effects, combined linear and non‐linear period effects, while age effects represent rates of ED for the reference period (2003). Results show that the increased hazards of ED visits observed in the period 2003–2023 were faster for those born before the mid 1980's (non‐linear cohort effects) compared to those who were born after. Odds ratios of the period effects show that Australians were 59% more likely to visit EDs in 2013 compared to 2003 and 85% more likely to visit EDs in 2023 compared to 2003.

FIGURE 4.

Results of the Age‐Period‐Cohort‐ model with ‘APC’ parametrisation. Age‐effects (black line) are rates for the reference period (year 2003), rates for all age groups in 2003. Cohort‐effects are non‐linear effects of cohort (red line). The blue line represents the Period‐effects that are odds ratios relative to 2003. All the subsequent years have higher rates of ED visits per capita than 2003.

Table 1 compares different statistical models that include combinations of age, period and cohort effects to explain variation in ED visit rates. Lower AIC and lower model deviance indicate better model fit. While adding Period or Cohort to the model resulted in an improvement compared to the model with age only, the full Age‐Period‐Cohort model is the best model.

TABLE 1.

Comparison of nested age‐period‐cohort models.

Model	AIC	Model DF	Model deviance	Pr (> Chi)
Age	7,673,992	203	7,670,885
Age‐drift*	4,888,287	202	4,885,178	< 0.001
Age‐Cohort	4,467,237	197	4,464,118	< 0.001
Age‐Period‐Cohort	4,222,933	192	4,219,804	< 0.001
Age‐Period	4,669,353	197	4,666,234	< 0.001
Age‐drift*	4,888,287	202	4,885,178	< 0.001

^*Drift is the linear trend that cannot be uniquely attributed to period or cohort.

5. Discussion

The value of applying an Age‐Period‐Cohort modelling approach when analysing longitudinal trends in the number of ED visits per capita over two decades is that it allows us to follow cohorts of residents while they are ageing. This study confirms that Australian ED attendance per capita has been increasing with time. However, the APC approach also highlights how changes in the number of ED visits per capita relate to both patient age and the decade they were born in. We found that the increase in the number of ED visits per capita was larger in the decade 2003–2013 compared to 2013–2023 and among residents aged ≥ 85 years compared to other age groups.

Although the rate of increase in per capita ED attendances has slowed, the consistent increase in the number of ED visits per capita in those aged 30–70 years between 2003 and 2023 is of concern, suggesting that an increase in the rate of attendance growth may be impending as this generation ages. Similar trends were observed in the USA, with the number of ED visits per capita increasing in the US since the 1990s [7, 8, 9, 17, 18]. Sun and colleagues (2018) reported that the increase in the number of ED visits per capita in the US between 2006 and 2015 was larger among those aged 45–64 years. The deceleration in the increase of the number of ED visits per capita in 2013–2023 is of interest. This slower increase in the number of ED visits per capita comes at a time when the Australian population is larger and older. Older patients have higher risks of ED presentation and the impact of their healthcare needs on the healthcare system and ED crowding is growing [11]. Projections up to 2050 with US data have shown that the number of ED visits per capita is unlikely to cause an increase in ED visits more than what would be expected from population growth [19]. However, the time spent in ED is expected to increase 10% faster than population growth while admissions from ED will increase 23% faster than population growth [19]. These data suggest that even if the increase in the number of ED visits per capita were to reach a plateau, difficulties currently faced by EDs across Australia may still increase. The decelerating trend observed in the last decade may reflect effects of policies that create alternative pathways to ED, but more research is needed to identify factors associated with these trends. The deceleration in the increase of the number of ED visits per capita for those born after 1970 is likely due to the fact that these generations have reached ages associated with lower ED visits during our study period. It is likely that these generations will experience higher rates of ED visits when they reach ages > 50 years.

The increasing rates in younger age groups suggest that strategies based mainly on chronic disease management may not be adequate. The finding that at the same age (e.g., 90 years), younger generations have higher rates of ED visits suggests long‐term upward pressure on ED even if demographic trends stabilise. Understanding birth cohort differences in adequate primary care access, for example differences in socioeconomic vulnerability, will be important for designing policies aimed at reducing reliance on ED for non‐urgent conditions.

6. Limitations

This study has some limitations. Although the study period is long (two decades), it was not long enough to capture the lifelong trajectories of generations. The Age‐Period‐Cohort model did not eliminate the collinearities between the three factors but by assigning the linear trend to period or cohort we can identify and interpret the non‐linear effects of these two variables. This study did not assess determinants of changes in ED visits but periods of time when major changes occurred and cohorts of patients (or age groups) that were more likely to experience these changes. Data from public hospitals only were used for this study. Although this dataset comprised a large number of hospitals (700 for the last year of the study), it is acknowledged that the private hospital sector also forms a component of the nation's emergency care.

7. Conclusion

This study shows that the increase in the number of ED visits per capita in Australia was slower in 2013–2023 compared to 2003–2013. However, the increasing trends even in younger age groups suggest that interventions based mainly on chronic disease management will not be enough to reverse the trend.

Funding

This work was supported by the National Foundation of Emergency Medicine.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: emm70220‐sup‐0001‐Supinfo.docx.

Diouf I., Boyle J., Yoon H.‐J., et al., “Time Trends in the Rates of ED Visits in Australia, an Age‐Period‐Cohort Approach,” Emergency Medicine Australasia 38, no. 1 (2026): e70220, 10.1111/1742-6723.70220.

Data Availability Statement

The data that support the findings of this study are available in AIHW National Non‐admitted Patient Emergency Department Care at https://www.aihw.gov.au/hospitals/other‐resources/about‐the‐data. These data were derived from the following resources available in the public domain: AIHW, https://meteor.aihw.gov.au/content/756107.