Background

The latest World Health Organization (WHO) report shows that more than one in eight adults and adolescents suffers from a mental health disorder [1]. The associated public health burden has steadily risen since 1990 in line with population growth [2]. This large and increasing disease burden requires a strong and efficient mental health care system to provide treatment, limit personal suffering, and avoid the significant productivity and societal costs induced by mental illnesses. As increasing demand leads to mental health care systems exceeding their capacity, emergency departments (ED) often become the first point of entry for distressed individuals. However, busy EDs exert strain on hospital resources and worsen distress for people in the midst of a mental health crisis. Individuals with mental health issues are likely to wait longer than others, leave before completing their treatment, and experience acute distress while waiting for care [3]. In Australia, more than 280,000 people presented to EDs for mental and behavioural disorders in 2021–22, including 25% for psychoactive substance use-related disorder, 27% for neurotic and stress-related disorders, and 12% for schizophrenia and schizotypal disorders [4].

Young people are of particular concern, as 75% of serious mental health disorders and alcohol and substance misuse first appear before the age of 25 [5]. Drivers of poor mental health are complex, and many pathways can lead to mental health crises [6]. Young people often find mental health care systems complex and difficult to access and navigate, and are discouraged by waiting times and inappropriate levels of care. This leads to inadequate treatment and increases the rates of presentation at EDs [4].

The mental health care system is often managed as a supply and demand-driven phenomenon, so policies to address services capacity constraints tend to focus on simply increasing the supply of services. Alternatives can however involve improving the coordination and effectiveness of services as well as implementing policies to reduce demand. System dynamics is a powerful way to model complex systems. It consists in creating a map of the components of a system and defining the interactions, or flows, between these components, including feedback loops. The description of the flows between the components, backed by mathematical equations, enables the modelling of the complex nonlinear dynamic behaviour of large systems. Systems dynamics models are particularly well suited to the evaluation of proposed changes to complex systems such as health systems, and may be used to model and optimise movements of people through acute care services. This allows expert knowledge and empirical data to be combined in an interactive decision support tool that decision makers and other stakeholders can use to test the impacts of interventions before their implementation in the real world.

We built a system dynamics model to evaluate strategies to decrease mental health-related ED presentations in the Australian Capital Territory (ACT), using a participatory modelling approach involving local stakeholders, including health planners, health providers and young people with lived experience of mental health disorders. The ACT is an Australian territory of 454 500 inhabitants that contains the federal capital, Canberra [7]. The ACT population grew by more than 27% over the years 2011–2021, with a third of the growth during the last five years coming from people born in India and Nepal [7]. A lagging investment in mental health care capacity has led the ACT to report the worst percentage of mental health-related ED presentations seen within an acceptable time (38.2% in 2021, compared to 62.2% nationally) [4]. The ACT has a young and educated population, with 37.1% holding a bachelor’s degree or above (compared with 22% in Australia), a median age of 35 years, and nearly 40% of the population in the 20–44 age range [7]. We modelled the effect of interventions for the 15–24 age group and for the general population.

Methods

Model development

A system dynamics model was developed in partnership with the ACT Office for Mental Health, which provides strategic oversight and coordination of the delivery of mental health services in ACT. The model was built using a participatory modelling approach that involved 33 local stakeholders, including 10 health agency representatives (Office of Mental Health and Wellbeing, ACT Health, ACT Primary Health Network, Children and Young People Commissioner’s Office, and Mental Health Justice Health Alcohol and Drug Services), five mental health professionals and allied health service providers (Child and Adolescent Mental Health Services, community mental health centres and non-government agencies), two representatives of the Australian Federal Police, two primary care practitioners, two patient advocates (Office of the Public Advocate), one representative of education (ACT Education Directorate), two representatives of carers (ACT Together and Carers ACT)), one counsellor from the crisis telephone support line (Lifeline), one youth peak body representative (Youth Coalition), six young people with lived experience of mental health conditions and one representative of both Gender Agenda and Mental Illness Education ACT [8, 9]. Stakeholders participated in three workshops (attendance: 28 to 33 stakeholders at each workshop) that were conducted between March and October 2022. During these workshops, participants mapped the causal pathways of the ACT mental health system, and identified and thoroughly defined new interventions to be included in the model. This input informed the design of the model, which was further expanded and refined through an iterative process of stakeholder review and model revision and analysis; this process was conducted through bi-weekly meetings with a subset of seven stakeholders called the model development group, which included five health agency representatives and two young persons with lived experience. After identification of the interventions by the stakeholders during the workshops, a shortlist was created through consensus by the model development group. The final set of interventions was further narrowed based on the availability of high-quality evidence identified in the literature (usually published systematic reviews with meta-analyses or controlled trials). During the last workshop, participants provided feedback after interacting with the decision support interface of a prototype of the model, which allowed further refinements of the model and interface to meet the ACT requirements. The participatory modelling approach was evaluated via analysis of ACT stakeholder survey responses and interviews using mixed methods [10]. The cost effectiveness of selected interventions of the model was assessed [11]. Model construction and analysis were performed using Stella Architect version 3.1.1 (isee systems [12]).

Model structure

The model represents the interactions between population, education, employment status, homelessness, family and domestic violence, psychological distress, suicidal behaviour, and mental health services (Fig. 1). The model is stratified by age group (ages 0–4, 5–11, 12–14, 15–17, 18–24, and 25 years and older) and by males and females. These age groups reflect the focus of the model on youth mental health, feedback from the participatory workshops, demarcation of schooling stages, and data availability from the Australian Bureau of Statistics (ABS).

Fig. 1
figure 1

Overview of the components and causal structure of the system dynamics model. Arrows represent how components of the model affect other parts of the model. The population component influences all components of the model. NEET: Not in Education, Employment or Training

Full size image

The model is divided into:

  1. 1.

    A population component that models changes in the size and age structure of the ACT resident population due to births, ageing, mortality, and migration;

  2. 2.

    An education component that captures transitions between primary, secondary, and post-secondary, and the highest level of qualification for people aged 15 and older. Children turning five years old flow into the primary education group, and then flow to the secondary education group, and optionally to the post-secondary education group. Students in secondary or post-secondary education may discontinue studies at rates dependent on sex and levels of psychological distress;

  3. 3.

    An employment component that models flows between employment, unemployment, and not participating in the labour force, for people aged 15–24 and 25 and older. A sub-component captures the proportion of young people aged 15–24 who are not in employment, education or training (NEET). The rates of transition between employed and unemployed, and between not participating in the labour force and unemployed, are dependent on age, sex, levels of psychological distress, and highest level of qualification;

  4. 4.

    A strength and difficulties component that represents the psychopathological vulnerability of children aged under 12 as measured by the strengths and difficulties questionnaire (SDQ) and captures their increased risk of developing mental health disorders as they age into adulthood. Children can flow between three possible levels of SDQ with rates dependent on age, sex, rates of family and domestic violence, rates of disengagement with the mental health services system, and rates of treatment;

  5. 5.

    A psychological distress component that models the onset of and recovery from moderate to very high psychological distress as measured by the Kessler Psychological Distress Scale (K10) (score 16–50) in the population [13]. People aged 12 and over can belong to three levels of distress: “Low”, “Moderate” and “High”, the last capturing people at high or very high levels of psychological distress (K10 score 22 or above). People can flow between low to moderate levels, and between moderate to high levels of psychological distress. These transition rates are dependent on age, sex, homelessness rate, rate of family and domestic violence, unemployment rate, rate of disengagement with the mental health services system, and rate of treatment;

  6. 6.

    A mental health services component that represents the flow of people with moderate to very high psychological distress or elevated SDQ as they engage with the mental health care system. Modelled services pathways include primary care practitioners, psychologists, psychiatrists, allied health professionals, online services, EDs, psychiatric and non-specialised hospital services, and community mental health centres (CMHCs) services. Waiting times are modelled to reflect service capacity constraints prior to commencing treatment. People may perceive a need for service and engage with the mental health system either by presenting directly to an ED or by seeking help from a primary care practitioner or online mental health services. After engaging with services, people may either a) recover through treatment and return to the general population with a lower level of psychological distress, b) be treated and not recover, c) disengage from the service due to either excessive waiting times or unsatisfactory care, or d) be referred on to different services;

  7. 7.

    A suicidal behaviour component that models self-harm hospitalisations and suicide deaths. The rates of suicide attempts are dependent on age, sex, and levels of psychological distress. Due to data availability constraints, self-harm hospitalisations are treated as suicide attempts, and suicidal behaviours are only modelled for people aged 15 and above;

  8. 8.

    A homelessness component that models transitions into and out of homelessness, where homelessness is defined according to the criteria developed by the ABS. Children aged under 15 enter homelessness at rates dependent on age, sex, and family and domestic violence rates. Homeless rates for people 15 years and over are further influenced by levels of psychological distress and unemployment;

  9. 9.

    A family and domestic violence component that captures the variation in family and domestic violence rates as a function of age, sex, secondary school completion rates, and unemployment rates.

For all modelled groups of people, mortality, which includes suicides, is modelled with a mortality outflow, and immigration and emigration with a net migration bidirectional flow..

Data sources for calibration

Calibration is a process that seeks to adjust the parameters of the model based on what is known, namely observed real-world historical data, thereby enabling robust projections into the future and the testing of interventions. We calibrated the model using historical time-series data extracted from the ABS (population sizes, birth and mortality rates, overseas and internal migration rates, employment rates and flows, education data, homelessness rates, family and domestic violence incidence, psychological distress prevalence, and suicide deaths), the Australian Institute of Health and Welfare (mental-health-related ED presentations and mental health services statistics), the Longitudinal Study of Australian Children (SDQ levels among children), the ACT Education Directorate (additional education data), the Young Minds Matter Survey (psychological distress for 12–17-year-olds), and the ACT Government’s Office of Mental Health and Wellbeing (intentional self-harm hospitalisations). Detailed references are available in Additional file 1. Parameters for which no data were available were estimated by constrained optimisation using Powell’s method, whereby the optimal values were determined by minimising the mean absolute percent error of the model’s outputs compared to the historical data [14].

Interventions

Used as a decision-support tool, the model allows exploration of the impact of different strategies for decreasing mental health-related ED presentations. We simulated a series of interventions (Table 1) that target early prevention of mental ill-health, alternative responses to emergency calls, prevention of re-presentations to EDs, increased support for frequent users of ED for mental health reasons, and improvement of the mental health care system itself. All interventions by default started in January 2023.

Table 1 Description of interventions
Full size table

Projected outcomes

The model was used to forecast mental health-related ED presentations, self-harm hospitalisations, and deaths due to suicide. The model was stratified by age and by sex. We present the results for the 15–24 age group and the whole ACT population over 10 years (beginning of 2023 to end of 2032).

Sensitivity analysis

Sensitivity analyses were conducted to assess the impact of uncertainty in estimates of the effect sizes of each intervention on the projected outcomes. Parameters were varied according to the 95% confidence intervals of the original research papers from which these effects were taken. Latin hypercube sampling was used to draw 100 sets of values for these model parameters, using lognormal distributions for odds and hazard ratios, and beta distributions for percentage parameters. The increases in capacity growth rates were sampled from a uniform distribution within ± 20% of the default value. Projected outcomes were calculated for each of the 100 runs and summarised using simple descriptive statistic. The resulting uncertainty intervals provide a measure of the uncertainty in the projected outcomes.

Results

Figure 2 shows the impact of interventions on preventing cumulative mental health-related ED presentations over the period January 2023 to December 2032. The crisis response program, which involves reducing ED transfers from police cars and ambulances, is the most effective intervention. It is forecast to decrease mental health-related ED presentations by 10.2% (95% interval 9.5–11.1%) for both youth (15–24 age group) and the general population. The ED suicide prevention program, which limits re-presentations to the ED, is the second most effective intervention, and is projected to reduce mental health-related ED presentations by 6.4% (95% interval 0–10.9%) for youth and similar numbers for the general population. Family psychosocial education for people with schizophrenia and their carers, doubling the annual capacity growth rate of mental health services (primary care practitioners, psychologists, psychiatrists, allied mental health professionals and CMHCs), and implementing technology-enabled care are each projected to decrease mental health-related ED presentations by 4.3 to 5.1% for youth and by 5.2 to 5.9% for all ages.

Fig. 2
figure 2

Effect of interventions on mental health-related ED presentations for youth (aged 15–24) and all ages: Projected effect of interventions on cumulative mental health-related emergency department presentations between January 2023 and December 2032, compared to baseline (no interventions activated). The first column displays the projected absolute number of prevented cumulative presentations. The second column and Forest plot show the projected mean percentage reduction of cumulative presentations versus baseline, and uncertainty intervals (50% in thick line and 95% in fine line) derived from the sensitivity analysis. ED: emergency department

Full size image

Combination of all the above interventions delivers the greatest benefits by reducing cumulative presentations by 26.6% (95% interval 19.7–33.5%) for youth and by 28.6% (95% interval 21.2–36.4%) for all ages. The larger uncertainty intervals of the ED suicide prevention and technology-enabled care interventions reflect the wider 95% confidence intervals of the original research papers from which the parameter estimates were taken (see Additional file 1). The school-based suicide prevention program has no effect on mental health-related ED presentations, while the youth connectedness program is projected to reduce presentations by 2.4% (95% interval -1.7–5.7%) for youth. Figure 3 displays the effect across time of the interventions on mental health-related ED presentations for youth. Interventions by default started in January 2023 and take two years to reach full effect. The crisis response program is the fastest intervention to take effect. The combination of the four interventions (crisis response, ED suicide prevention, family psychosocial education and technology-enabled care) has a rapid and increasing effect over the years, and demonstrates how mental health-related ED presentations can be substantially reduced without increasing mental health care capacity.

Fig. 3
figure 3

Effect of interventions on mental health-related ED presentations for youth (aged 15–24) across time: Projected effect of interventions on mental health-related ED presentations across time, compared to baseline. Interventions started in January 2023. ED: emergency department. The combination of all 7 interventions includes the crisis response program, ED suicide prevention, technology-enabled care, family psychosocial education, doubling of the yearly capacity growth rate of mental health services (primary care practitioners, psychologists, allied mental health professionals, psychiatrists and community mental health centres), school suicide prevention program, and youth connectedness program

Full size image

Figure 4 depicts how interventions impact self-harm hospitalisations. Only interventions that specifically target suicide and suicidal ideation have a noticeable impact on self-harm hospitalisations. The ED suicide prevention intervention, which is an outreach program for people who presented with self-harm or suicidal ideation, is forecast to decrease self-harm hospitalisations by 11.9% (95% interval 0.1–19.2%) for youth and by 8.9% (95% interval 0.1–12.2%) for all ages. The school-based suicide prevention program is projected to reduce self-harm hospitalisations by 9.4% (95% interval 7.1–11.5%) for youth. Technology-enabled care and doubling of the annual capacity growth rate of mental health services are projected to decrease self-harm hospitalisations by 1.2% to 1.5% depending on the age group, while other interventions have an effect of less than 1%. Similar trends are forecast for deaths by suicide (Additional File 1: Fig. S1).

Fig. 4
figure 4

Effect of interventions on self-harm hospitalisations for youth (aged 15–24) and all ages: Projected effect of interventions on cumulative self-harm hospitalisations between January 2023 and December 2032, compared to baseline (no interventions activated). The first column displays the projected absolute number of prevented cumulative self-harm hospitalisations. The second column and Forest plot show the projected mean percentage reduction of cumulative self-harm hospitalisations versus baseline, and uncertainty intervals (50% in thick line and 95% in fine line) derived from the sensitivity analysis. ED: emergency department

Full size image

Discussion

This study used systems dynamics modelling and simulation to investigate five complementary strategies (Fig. 5) to decrease mental health-related ED presentations for youth (15–24 age group) and for the general population in the ACT region. The first strategy consists of sending mental health clinicians into the field to directly meet people in mental health crisis and attempt to de-escalate the situation or divert people to CMHCs if appropriate. This strategy, simulated as the crisis response program, is the most effective single intervention and is projected to reduce mental health-related ED presentations by 10.2% for youth and all ages. The mental health clinician, usually a mental health nurse, travels with ambulance services or police, which is appropriate as mental health patients as more likely to arrive at ED via ambulance and police than other people [3]. This intervention has been piloted in the Australian state of Victoria from 2007 to 2011, evaluated in 2012, and further implemented in Victoria [15]. Similar interventions deployed internationally involve mobile crisis teams that send mental health clinicians to people in crisis in unmarked or ambulance vehicles [24], with the same goal of meeting distressed people on site and providing care, referring, or transferring people to community services; these similar interventions also markedly decrease mental health-related ED presentations.

Fig. 5
figure 5

Diagram of strategies to decrease mental health-related emergency department presentations. ED: emergency department. MH-ED: mental health-related emergency department. CMHC: community mental health care centre

Full size image

The second strategy aims to limit the re-presentation to EDs of individuals who have previously presented with self-harm or suicidal ideation. Perera et al. showed that these individuals, on average, re-present 3.8 times to ED in the year after the first presentation [25]. This strategy was simulated as the ED suicide prevention program; it involves an initial universal suicide risk screening and an active outreach program via phone calls after discharge. It is the second most effective intervention and is projected to decrease mental health-related ED presentations by an average of 6.5%. In addition, this intervention is forecast to decrease self-harm hospitalisations and suicides of youth by 11.9% and 12.8% respectively.

The third strategy focuses on individuals with psychotic disorders, whose distressing symptoms and behaviour result in a significant use of the mental health care system. While the prevalence of schizophrenia in Australia is 0.47%, people with schizophrenia represent 12% (15% in the ACT) of mental health-related ED presentations [4], 23% of overnight admitted mental health hospitalisations, and 26% of CMHCs service contacts [26]. Psychosocial education for families and carers of people with schizophrenia has been extensively evaluated and significantly reduces the rate of relapse [18]. It involves much more than providing people affected and families with technical information about the condition and its treatment; it includes understanding each person and their family’s explanatory model for their experiences and helping them develop insights into their condition, thereby promoting a person’s autonomy, self-management, restoration of self-esteem and reduction of relapses [27]. The family psychosocial education program is projected to reduce mental health-related ED presentations by an average of 5.2%. This intervention has been conservatively restricted to persons with schizophrenia and families, as this is where psychosocial education has been the most extensively studied. This intervention could be extended to a wider group of people with psychotic or severe mood disorders and carers, who can also benefit from family psychosocial education [28].

The fourth strategy focuses on early prevention, with a school-based suicide prevention and youth connectedness program. School-based suicide prevention programs, which consist of education and role play around suicidal ideation to help young people develop skills to help peers, have been extensively trialled in the United States and in Europe [29]. Our modelling forecasts for the 15–24 age group a decrease of self-harm harm hospitalisations and suicides by 9.4% and 8.8% respectively, but no effect on mental health-related ED presentations. The youth connectedness program is more abstract than the other interventions and allows stakeholders to quantify the effect of a broad intervention that can be fulfilled by different approaches. It draws on the fact that young people with good levels of social connectedness have lower risks of anxiety and depressive symptoms, as well as psychological distress [29]. This intervention is projected to reduce mental health-related ED presentations by 2.4% for youth.

The final strategy involves improving the mental health care system by increasing its capacity and implementing a new model of care enabled by digital technologies. Doubling the annual capacity growth rate of primary care practitioners, specialised care (psychologists, psychiatrists, and allied mental professionals), and CMHCs services could decrease youth mental health-related ED presentations by 4.3%, while technology-enabled care is projected to achieve a 4.8% reduction. Disengagement from an overwhelmed mental health care system can cause psychological distress, worsened mental health symptoms, and mental health crises that lead to increased presentations to ED. Technology-enabled care involves the use of digital technologies to improve the coordination of health care by multi-disciplinary teams and the measurement of real-time individual outcomes. Hence, it has the capacity to increase the efficiency of mental health services. There are several real-world constraints on expanding mental health services capacity, including workforce shortages, an aging workforce (in 2021, 27% of psychologists and 42% of psychiatrists in Australia were aged 55 or over [30]), and stress and poor job satisfaction leading to high staff turnover [31]. Reforms can improve the efficiency of the mental health care system and further expand on the interventions presented herein. The triage and referral pathway to psychologists and mental health allied professionals can be broadened to facilitate access to care [32]. Analyses of longitudinal data from primary care-based youth services indicate that the majority of young people with emerging mental health disorders did not gain meaningful improvement in social and occupational functioning [33]. While many people with mild mental health symptoms could be directed to self-care strategies, and ongoing monitoring or psychosocial support (such as financial, employment, social, and education support), people with more complex disorders require team-based and multidisciplinary care, and close monitoring of their mental health symptoms and trajectory.

System dynamics, conducted in collaboration with local stakeholders, can provide a powerful decision support tool for local health planners to identify the best approaches to the allocation of limited resources.

This analysis has limitations. Most data inputs for our model were derived from population health survey data, published health, demographic and economic data that may vary in quality. Triangulation of multiple data sources, parameter estimation via constrained optimisation and local verification were performed to minimise potential measurement biases.

Conclusions

This study, conducted for the ACT region in Australia with the active participation of local stakeholders, showed that in a mental health care system already operating at capacity, a decrease in mental health-related ED presentations is best achieved by implementing new policies to address upstream and recurrent flows to the ED, from early prevention to mobile crisis teams, active prevention of representations, use of digital technologies to coordinate care, and other structural reforms, rather than simply relying on increasing mental health care capacity.