Paediatric Emergency Department Mental Health and Behavioural Presentations in Australia Before and After the Onset of the COVID-19 Pandemic: Retrospective Observational Study

2.1 Study Design and Setting

A multi-centre retrospective cohort study of school-aged children and adolescents (6–17-years old) attending Australian EDs with a primarily mental health-related presentation was conducted. Data were obtained from 10 hospitals in three states: eight metropolitan Victorian hospitals (two paediatric tertiary referral, one mixed adult/paediatric tertiary referral, and five mixed non-tertiary), one regional mixed adult/paediatric tertiary hospital in Tasmania, and one metropolitan mixed non-tertiary hospital in Queensland. Hospitals were recruited via an invitation to participate, distributed to 30 sites within the Paediatric Research in Emergency Departments International Collaborative (PREDICT) Network [15], a collective of paediatric EDs across Australia and New Zealand. We examined presentations between 1 January and 31 December 2019 (pre-COVID-19 group) and between 1 January and 31 December 2021 (COVID-19 group). Ethical approval was granted by the Monash Health Human Research Ethics Committee (RES-19-0000-896A).

2.2 Participants

Eligible participants were identified by searching the local ED Information System (EDIS) for all patients aged 6–17 years who attended the ED within the pre-COVID-19 or COVID-19 period with a primarily mental health-related presentation. Due to the different drivers, symptoms, and diagnostic methodologies of mental health in children under 6 years old [11, 14, 16], presentations by this age group were excluded from the current analysis. Presentations were eligible if the patient was: (a) referred to the mental health team from the ED; and/or (b) discharged with a mental health diagnosis (defined according to the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification [ICD-10-AM] codes relating to mental and behavioural disorders [F01–F99], emotional state, appearance and behaviour [R45.0–R46.8]); and/or (c) a diagnosis consistent with intentional self-harm (X71–X83). Presentations were excluded if the only presenting complaint was intoxication with drugs or alcohol (Y91.0–Y91.9; excludes poisoning: T519). Table S1 contains a full list of included diagnoses.

2.3 Diagnostic Classification

Within Australian EDs, patients who present for mental health reasons are reviewed by ED clinicians and/or mental health teams, who categorise the presenting complaint using ICD-10-AM diagnostic codes to determine the most appropriate treatment/referral options [17]. These codes are numerous, highly detailed, and precise, allowing for rapid categorisation of patients within the busy ED setting. Codes used in the ED typically capture the presenting behaviours (e.g., destructive behaviours, drug ingestion) and new or existing mental health diagnoses (e.g., anorexia nervosa, borderline personality disorder). Discharge diagnoses were identified in discharge notes or, where these were incomplete, within mental health assessment notes. Given the precise detail of ICD-10-AM codes, diagnoses were grouped within the major categories of the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-5) (Table S2). Diagnostic impressions documented in the ED may not result from an extensive assessment as would typically occur in an inpatient, outpatient, or community setting [15]. Therefore, diagnostic categories described in this study may reflect preliminary impressions and should be interpreted as an indication of the behaviours and personality structures patients presented with and the likely treatment and referral pathways indicated for them within current practice guidelines, rather than a conclusive diagnosis.

2.4 Data Collection

At each hospital site, all eligible mental health presentations from 2019 and 2021 were exported from the EDIS into two excel spreadsheets (one for each year). All presentations, including repeat presentations, were included in the list, which was randomised using the ‘RAND’ function in excel. At each site, researchers were instructed to select up to 100 unique patient presentations per year (200 across both years) sequentially from their randomised list. It was decided a priori that 100 presentations from each period would be feasible and adequately representative of overall ED mental health visits. Where a patient presented multiple times, only the presentation closest to the top of the randomised list was selected, with any additional presentations lower down the list excluded (regardless of which presentation occurred first in the calendar year). For one site, less than 100 eligible presentations occurred, so randomisation of the list was not necessary (Table S3).

Medical record data were extracted for each randomly selected presentation from the computerised EDIS by trained research staff (abstractors), in line with recommended chart review methods [18]. Abstractors were given explanatory notes and met J.N.D. and M.J.W. regularly to resolve uncertainty. Data were de-identified locally and entered into a password-protected, central web-based database (REDCap; Research Electronic Data Capture, hosted at Monash University [19, 20]). Abstracted data included routinely collected administrative and clinical (e.g., diagnosis/es, precipitants, history of mental illness, and social background) information. Although not routinely collected, gender and/or sexual diversity (GSD) was extracted from medical notes where available. Gender was defined as girl, boy, or non-binary, according to the recorded gender/sex of patients in the medical notes. Transgender young people were categorised according to their preferred gender. If patient sex was described as female or male using the binary coding available in the computerised EDIS, with no mention of gender identity in other notes, this was assumed to align with preferred gender. The collection instrument included pre-defined multiple-choice options, with some opportunities for free text description. Free text data were coded using directed content analysis, first attempting to reintegrate them into the pre-defined categories, with new categories created only if they could not be reintegrated. Coding was conducted by J.N.D., with 20% dual coded by E.A. for consistency. Discrepancies were resolved through discussion with R.B., S.C., K.K., and G.M.

2.5 Urgency

The urgency of ED presentations was defined using the five-level Australasian Triage Scale (ATS; see Table S4). Each presentation was classified into a high urgency (ATS category l or 2; i.e., needing to be seen immediately or within 10 min, respectively) or lower urgency episode (ATS categories 3–5; i.e., needing to be seen within 30–120 min).

2.6 Psychosocial Factors

Psychosocial data were extracted from patient assessment records and grouped into broad categories: family/personal history of mental illness; social issues/vulnerabilities; housing and economic factors; trauma, abuse and neglect; relational problems (with parents, partners, and peers); substance use; medication changes; living situation; and other life stressors (including school-related stressors).

2.7 Statistical Analysis

Data were analysed in Stata IC (V16.1, StataCorp, College Station, Texas, USA). Outliers were identified and either corrected or left untouched, based on site feedback and review of available data.

Descriptive statistics were generated to describe the participant profile. Categorical data were described as counts and percentages, while continuous data were described with medians and interquartile ranges. To account for clustering within hospitals, comparisons for binary categorical variables were made using the Mantel–Haenszel Χ ² test. For variables with three or more groups, categories were compared to a single reference group [21]. Due to the skewed nature of hospital time and date data [22], a log transformation of continuous variables was conducted. Continuous data comparisons were made using a linear mixed model with a fixed effect of year and a random effect of hospital, following the restricted maximum likelihood (REML) method to account for clustering within a multilevel exploratory design [21]. Coefficients were exponentiated to report geometric means and confidence intervals (CIs).

3.1 Demographic and Clinical Characteristics

Figure 1 provides a flow diagram of patient recruitment. Across 10 sites, 996 ED records were included from 2019 and 968 were included from 2021.

One site had fewer than 100 eligible presentations, and some larger sites inadvertently collected over 100 from their randomised list of eligible presentations (20 additional presentations across five sites, mean of 3.33 per site [range of 1–7]; see also, Table S3). Since the analysis does not require an equal number of cases from each site [21] and removal of these cases would reduce power without decreasing risk of bias, they were retained. Following sample selection from randomised lists at each site, 19 patients selected at a single hospital network (n = 3 sites) from 2021 (2% of 2021 total across sites) were identified as acute alcohol intoxication presentations with no mental health-related concerns—review of records confirmed ineligibility and they were removed prior to analysis. As data collection had concluded, it was not possible to replace removed patients. As such, site totals ranged from 82 to 107 in 2019 and 68 to 106 in 2021. In both 2019 and 2021, the majority of those presenting were girls (63.4% and 68.0%, respectively), were aged 12–17 years (89.7% vs. 92.8%, respectively), and were born in Australia (Table 1). Due to the low number of non-binary young people (N = 16), gender is only reported for boys and girls. There was a 157% increase in reporting of gender and/or sexual diversity amongst young people presenting in 2021, compared to 2019 (95% CI [62%, 401%]).

In both years, almost half of all presentations occurred during daytime hours, with a 9% higher likelihood of out-of-hours presentations in 2021 (Table 1). Over half arrived on their own or with friends and/or family members in both years, with a higher likelihood of this arrival mode and a lower likelihood of arrival via ambulance in 2021.

Compared to 2019, Median LOS in the ED increased in 2021, and the likelihood of admission to the ED short stay unit was 37% higher. Patients waited longer on average to be seen by ED clinicians in 2021 and time to be seen by a mental health clinician increased (Table 1).

In 2021 participants were 1.24 times more likely to have presented to the same ED in the previous 12 months, and the number of previous presentations per patient was 16% higher (Table 1). We observed no changes in the likelihood of patients being discharged or leaving without being seen/against medical advice. Similarly, there was no change in length of inpatient stay.

3.2 Psychosocial Factors

Figure 2 shows the psychosocial factors associated with presentations in 2019 and 2021. In 2021, young people presenting to the ED were more likely to be using substances; to present with relational issues (e.g., bullying, romantic relationship breakdown, and family discord); to have documented past or present trauma, neglect, and abuse; and to present with other life stressors (e.g., school pressures, change in environment, appearance or health-related stress, and sleeping difficulties). Medication changes were more likely to be a precipitant for presentations in 2021.

Young people presenting to the ED in 2021 were slightly less likely to be living in single-parent households or with other relatives than to be living with both parents compared to those presenting in 2019. The proportion living in either foster care, residential care, temporary/crisis accommodation, or no fixed address (other/precarious living arrangements) remained stable.

Notably, there was an overall increase in documentation of psychosocial factors associated with presentations in 2021 (RR = 1.13; 95% CI [1.09–1.18]).

3.3 Diagnoses and Diagnostic Impressions

Figure 3 provides an overview of the relative risk of documentation of one or more diagnoses/diagnostic impressions in 2021 compared to 2019. We observed significant increases in the likelihood of presentations with documented self-harm and suicidal ideation and behaviour, neurodevelopmental and neurocognitive disorders, and other mental health concerns/disorders (including obsessive compulsive disorder, bipolar disorder, undefined aggression and behavioural issues, and other mental health concerns not further specified). We found a decreased likelihood of presentations with documented disruptive, impulse control, and conduct disorders in 2021.

4.1 Strengths and Limitations

By sampling across diverse hospital sites, our methodology allows examination of broader trends. Our study utilised a more in-depth data collection approach than previous studies, enabling a richer profile of presenting young people to emerge.

There are also limitations. Use of a single year pre- and post-pandemic means we cannot determine whether these changes were expected based on prior years. However, our findings are broadly similar to those of studies which modelled changes over time [11, 13]. Hospitals were recruited via convenience sampling and may not be representative at the national/state level. Almost all sites were metropolitan, and most were based in Victoria, with only two interstate hospitals participating. This may skew results towards the Victorian experience, particularly that of metropolitan Melbourne. However, given the stringency of restrictions in Victoria [3], this may have increased the likelihood of detecting the impact of the COVID-19 pandemic. A sensitivity analysis was conducted excluding non-Victorian sites (Supporting Information A), which found a slight increase in the strength of some differences (except arrival by ambulance and admittance to ED short stay, which were no longer significant, Table S5), but minimal changes overall (Table S5, Figures S1 and S2). The reduced likelihood of missing data in 2021 indicates a possible bias, either in data collection or in ED recording practices between years. Diagnostic and psychosocial categories were created by combining original questionnaire items with those that emerged from the open-ended ‘other’ option provided to abstractors, the latter of which has a lower likelihood of being systematically collected. These limitations are common within medical record cohort studies and are balanced by the reduced risk of bias associated with data collected for non-research purposes [18]. Finally, it remains possible that the use of mental health diagnoses to describe child and adolescent behavioural and emotional problems may have unintended consequences, including stigma and risk of pathologisation and medicalisation of transient behaviours associated with social and environmental issues [31]. Controversy surrounds the use of mental health diagnostic labels for adolescent populations, due to the ongoing development of personality structures throughout adolescence and early adulthood and the heterogeneity of diagnoses across the life course [31]. In this study, we included psychosocial factors where available for a more holistic view of the patients and the elements contributing to their presentation. However, future research could examine psychosocial and biological determinants of mental health presentations and the supports available in addressing these, with less emphasis on diagnoses.