Hospital and emergency department discharge against medical advice in Western Australian Aboriginal children aged 0–4 years from 2002 to 2018: A cohort study

Abstract Background Discharge against medical advice (DAMA) is a priority issue for the health system. Little is known about the factors associated with DAMA for Aboriginal and/or Torres Strait Islander (Aboriginal) children in Australia. Objectives Investigate the associations between DAMA for hospital admissions and emergency department (ED) presentations and: (i) child, family and episode of service characteristics and (ii) 30‐day readmission/ re‐presentation. Methods We conducted a cohort study of Aboriginal children born in Western Australia (2002–2013) who had ≥1 hospital admissions (n = 16,931) or ED presentations (n = 26,546) within the first 5 years of life. The outcome of interest was hospital and ED DAMA and adjusted odds ratio were derived using multilevel mixed‐effects logistic regression. Results In the Hospital Cohort, there were 43,149 hospitalisations for 16,931 children, with 684 hospitalisations (1.6%) recorded as DAMA. In the ED Cohort, there were 232,082 ED presentations in 26,546 children, with 10,918 ED presentations (4.7%) recorded as DAMA. DAMA occurring in hospitals between 2014 and 2018, the adjusted odds decreased by 75% compared to the period between 2002 and 2005. The adjusted odds of ED DAMA increased by 46% over the same period. Hospital admissions in regional and remote hospitals were almost seven times the adjusted odds of DAMA compared with hospital admissions in Perth metropolitan hospitals. The adjusted odds of ED DAMA decreased by 12% for ED presentations in regional and remote hospitals compared to those in Perth metropolitan hospitals. There was no evidence of hospital DAMA being associated with hospital readmission within 30 days and limited evidence of ED DAMA being associated with re‐presenting to an ED within 30 days. Conclusions The study identified several important determinants of DAMA, including admission status, triage status, location and calendar year. These findings could inform targeted measures to decrease DAMA, particularly in regional and remote communities.


| BACKG ROU N D
Discharge against medical advice (DAMA) refers to when a patient leaves hospital before discharge is recommended by the treating clinical team. 1,2This includes patients admitted to hospital who discharge themselves against medical advice, and patients who leave the emergency department (ED) at their own risk or without waiting to be attended to by a medical officer.6][7][8] Patients with a history of DAMA have a higher likelihood of subsequent DAMA episodes. 5,9In the Australian context, Aboriginal and Torres Strait Islander (hereafter respectfully referred to as Aboriginal) people are overrepresented in DAMA episodes. 10This discrepancy is more pronounced in the Northern Territory, South Australia and Western Australia and increases with remoteness of residence. 7,10 a review of the causes of DAMA, Shaw 3 highlighted the contribution of health systems with insufficient support for families, cultural safety and culturally appropriate care.As a result, DAMA is an ongoing issue of equity for health systems and is of both national and jurisdictional importance. 10,11e focus on DAMA in the scientific literature tends to be on adult populations.However, in Australia, between 2015 and 2017, 1.0% of hospitalisations for Aboriginal children aged <5 years reported a discharge code of left against medical advice/discharged at own risk compared with 0.3% of hospitalisations for non-Indigenous children. 10Likewise, in a study of 124,757 patients, Aboriginal children were 1.6 times more likely to have a DAMA in a tertiary paediatric hospital compared to non-Aboriginal children. 2 Predictors of DAMA among children were hospital site, mental health/behavioural diagnosis and emergency rather than elective admission. 2r ED presentations between 2015 and 2017, 2.3% of presentations for Indigenous children aged <5 years left at their own risk or did not wait, compared with 1.6% for non-Indigenous children. 10By following a cohort across time, this present study adds to this work and provides a richer use of contextual factors by using linked administrative data.
We investigated the effects of child, family, community and episode of service characteristics on DAMA in Aboriginal children aged <5 years who had one or more hospitalisation or ED presentation.
In addition, we investigated whether hospital and ED DAMA was associated with an increased likelihood of 30-day readmission to hospital or re-presentation to EDs, respectively.

| Cohort selection
The population for this study was based on the 'Defying the Odds' cohort and included all Aboriginal children born in Western Australia between 2002 and 2013 (n = 29,319). 12A child was included in the cohort if they, their parents or grandparents identified as Aboriginal using an algorithm applied to the Aboriginal status indicators within the multiple data sets by the Western Australia Data Linkage Branch. 13We excluded children if their full siblings were not identified as Aboriginal.The study cohort's relatives were identified by Western Australia's records of family links, the Family Connections System. 14This study used two cohorts derived from inform targeted measures to decrease DAMA, particularly in regional and remote communities.

K E Y W O R D S
Aboriginal, children, DAMA, data linkage, Western Australia

Study question
This study aimed to examine the associations between DAMA for hospital admissions and ED presentations, and:

What is already known
DAMA is a priority for the health system as it can negatively impact patient well-being.

| Data source
Linked data from the Midwives Notification System (MNS), the Emergency Department Data Collection (EDDC), Death Registrations and the Hospital Morbidity Data Collection (HMDC) were used.Probabilistic linkage of all records, by matching identifiers (e.g.name, address, date of birth, etc) across sets of records, was undertaken by the Western Australia Data Linkage Branch from the Western Australia Department of Health to produce deidentified records for analysis. 15Audits of data linkage quality have shown a high degree of accuracy. 16,17e MNS includes clinical (infant weight, gestational age, parity) and socio-demographic (mother's age, socio-economic status, remoteness) data on all Western Australian live births and stillbirths of more than 20 weeks gestation or birthweight >400 g, which are reported by trained midwives within 48 h of delivery.The HMDC and EDDC include data on all completed hospital admissions and ED presentations to all public and private hospitals in Western Australia, respectively. 18,19These data are entered by trained medical records staff following the episode of service.

| Community characteristics
Area level Index of Relative Socio-Economic Disadvantage and the Accessibility/Remoteness Index of Australia (ARIA) remoteness index, both from the 2016 Australian Bureau of Statistics' Census of Population and Housing, were based on the mother's address at the birth of the child.Area level disadvantage was categorised as state quintiles from most disadvantaged (1) to least disadvantaged (5).
Remoteness classifies geographic location on the basis of isolation and distance from service centres and healthcare facilities.These data were split into four categories from least remote (major cities) to most remote (very remote areas).

| Episode of service characteristics
The child's age at each episode of service was recorded, and we classified children as infants (<1 year old) or older (≥1 to <5).The calendar year of the episode of service was also recorded.
Any previous DAMA was defined based on whether the child had any history of DAMA, prior to the current episode of service, using the definitions of DAMA given in the Outcomes section.Any previous DAMA was defined separately for hospitalisation admissions and ED presentations.Potentially preventable hospitalisations for children were defined based on the ICD-10-AM code from the patient's principal diagnosis, using the scheme defined in Falster et al. 20 The hospitalisations data contained five classifications of admission status: emergency admissions, elective -waitlist, electivenot waitlist, emergency -emergency department admission and emergency -direct admission. 21We combined the three emergency admission classifications into a single category to address changes in coding in older records.The triage code for ED presentations is graded into five categories, based on the Australasian Triage Scale, 22 from most (resuscitation: immediate) to least urgent (non-urgent: within 120 min).Supplementary triage codes (dead on arrival, direct admission, inpatient) were excluded due to small cell sizes.Hospital location was classified as either metropolitan or regional and remote based on the WA Department of Health districts; all hospitals outside the Greater Perth metropolitan area were classified as regional and remote.

Hospital admission was defined as any admission to a Western
Australian hospital ward for care, including all neonatal nurseries.It excluded the normal postnatal hospital stay for healthy babies.
Hospital admissions that were serial transfers (patient moved between hospitals successively without returning home), nested transfers (patient moves to another hospital during an admission) or overlapping transfers (admission date prior to separation date on previous record) were considered a single event. 23An ED presentation was defined as any presentation to the ED regardless of whether the child was admitted to hospital.Each hospital admission and ED presentation was considered a single episode of service.
A hospital separation was defined as DAMA if the mode of separation for that admission was recorded in the HMDC as 'left against medical advice/discharge at own risk'. 19An ED discharge was defined as having a DAMA if the discharge status for that presentation was recorded in the EDDC as either 'did not wait to be attended by a medical officer' or 'left at own risk'. 18These are headline indicators for the WA Department of Health, and there is a comprehensive recording system to capture these data.For example, emergency patients are given three attempts to respond, and hospital admission records can be corrected if a patient is later found. 1 30-day readmission/ re-presentation is a standard metric for hospitals. 24A child was considered to be readmitted to hospital within 30 days if the admission date was within 30 days of the separation date of their previous hospital admission.A child was considered to be re-presenting to an ED within 30 days if the presentation date was within 30 days of the discharge date of their previous ED presentation.

| Statistical analyses
For all exposures we estimated adjusted odds ratios using multilevel mixed-effects logistic regression, which took into account the nested data structure (children recording multiple episodes of service and children sharing the same mother within their respective study cohort), using a three-level analysis (observations, children, mothers).
Directed acyclic graphs (DAGs) were used to assess potentially confounding causal relationships for the characteristics of interest (Figures S3 and S4). 25,26Based on the DAGs, a minimal sufficient adjustment set (MSA) to deconfound each exposure was identified in DAGitty 3.0. 27The adjusted odds ratios presented in this paper should be interpreted as conditional upon the nested data structure in the mixed-effects models, as well as upon adjustment for the MSA for that variable.
Statistical adjustment was based on conditioning in melogit using Stata V.16.0. 28

| Missing data
No imputation was undertaken and missing data were excluded from all analyses, as missing data were negligible (<1%), with missing cell counts specified within all the tables.Data were missing due to either the non-completion of administrative data or (in the case of area disadvantage or remoteness data) address data which could not be coded.Small cell counts (n ≤ 5) have been suppressed for confidentiality.

| Episodes of DAMA hospitalisations
In the Hospital Cohort, there were a total of 43,149 hospitalisations for 16,931 children (Table 1), with 684 hospitalisations (1.6%) recorded as DAMA (Table 2).Children with a birthweight of <2500 g had an adjusted odds ratio for DAMA 30% greater than children with a birthweight of ≥2500 g (Table 2).The adjusted odds of hospitalisation DAMA increased with remoteness from the Perth metropolitan area: compared with children born in major cities of Australia, children born in regional Australia had an aOR of 2.08 (95% CI 1.53,There was no evidence of an effect for the remaining exposures on DAMA hospitalisations (Table 2).

| Episodes of DAMA emergency department presentations
In the Emergency Department Cohort, there were a total of 232,082 ED presentations in 26,546 children (Table 1), with 10,918 ED presentations (4.7%) recorded as DAMA (Table 3).ED DAMA comprised the discharge codes of 'did not wait' and 'left at own risk', with 'did not wait' contributing 96% of all ED DAMAs (Figure S5).
Children who were born in a plural birth had adjusted odds of ED DAMA 20% less than singletons (Table 3).Children with mothers with ≥3 previous births had a 8% decrease in adjusted odds of having a ED DAMA compared with children of mothers with <3 previous births.Younger mothers (aged <20 years old) had a 13% increase in the adjusted odds of having a child who had an ED DAMA compared to older mothers (aged ≥20 years old at the birth of the study child).There was not a consistent pattern between remoteness at birth and the adjusted odds of ED DAMA: compared with children born in major cities of Australia, children born in regional Australia had an aOR of 1.03 (95% CI 0.96, 1.10), children born in remote Australia had an aOR of 1.03 (95% CI 0.96,

TA B L E 3 (Continued)
DAMA than on hospital DAMA.Studies of hospital and ED DAMA also tend to focus on specific sub-populations, such as the cohort using a specific hospital or patients with a particular disease. 7,8,29As such, our population-based study of children in both hospital and ED is an important addition to the literature.Our study also brings an explicit causal model, which provides a basis for future researchers to challenge and refine.

| Limitations of the data
Our study had some limitations.While we have described a causal model, we have made some simplifications, such as excluding maternal DAMA.We were limited by the available data.For example, we did not have measures of constructs like cultural security, which has been shown to have a substantial impact on Aboriginal people's engagement and trust in health services.Nor do we have recorded waiting times which could impact the effect of triage status on DAMA.Finally, despite important differences in DAMA between metropolitan and regional and remote hospitals, small cell counts limited our ability to stratify our analyses of hospital DAMA by location.

| Interpretation
For DAMA occurring in hospital between 2014 and 2018, the adjusted odds decreased by 75% compared to the period between 2002 and 2005.This reduction in hospital DAMA is potentially driven by improvements in regional and remote hospitals.During 2002-2018, hospitalisations in regional and remote hospitals were almost seven times more likely to DAMA than metropolitan hospitalisations.However, by 2018 the gap between metropolitan and regional and remote hospitals had substantially diminished.There has been a number of initiatives across WA Health that have been aimed at improving the care of Aboriginal people in hospital to reduce the event of a DAMA. 1,11,30It is likely that a culmination of these activities has reduced hospital DAMA in children.while it is a favourable finding that children presenting to regional and remote EDs are less likely to DAMA than presentations in Perth metropolitan EDs, we also found a higher frequency of less urgent presentations compared with metropolitan Perth.Combined, these findings suggest that more primary healthcare services are required in these locations.It is well-established that general practice (GP) access decreases with increasing remoteness in Australia, which is often reflected in increased emergency presentations for children in regional and remote and remote areas. 31For instance, it has been estimated that at least 20% of ED presentations in Western Australia could be managed in general practice. 32mission status is an important predictor of whether or not a given hospitalisation will end with DAMA.For hospitalisations, we found a substantial association between admission status and DAMA, with an aOR for non-elective admissions of more than six times greater than elective admissions from the waitlist.This finding is consistent with the extant literature.For example, in a study of children admitted to the hospitals in the Sydney Children's Hospitals Network, planned admissions were 31% less likely to DAMA than emergency admissions. 2 Further, in a study of inpatient admission for ischaemic heart disease Katzenellenbogen et al. 7 reported an aOR of 5.93 for emergency admissions compared with planned admissions.
We also found that ED presentations with a non-urgent triage code resulted in 11.5 times greater adjusted odds of DAMA compared with presentations with an emergency triage code. 33This finding is consistent with descriptive studies of adult ED presentation DAMA in Italy, Saudi Arabia and Canada, with longer waiting times associated with an increased likelihood of DAMA. 29,346][37] Another factor which drives the association between waiting times and DAMA is that some parents may choose to take their children home if symptoms improve.For example, in a study in a Canadian children's hospital, 37% of premature departures from Eds occurred after the child's symptoms resolved.
(i) child, family and episode of service characteristics, as well as (ii) the odds of 30-day readmission/re-presentation for Aboriginal children aged <5 years born in Western Australia between 2002 and 2013.
DAMA is associated with a range of characteristics, including hospital location, mode of admission/ triage code and year of admission/presentation.There was no evidence of hospital DAMA being associated with an increased likelihood of hospital readmission within 30 days and limited evidence of ED DAMA being associated with re-presenting to an ED within 30 days.the Defying the Odds Cohort: (i) the Hospital Cohort (Figure S1), which included children aged <5 years with at least one hospitalisation (57.7%; 16,931/29,319) and (ii) the Emergency Department Cohort (Figure S2), which included children aged <5 years with at least one ED presentation (90.5%; 26,546/29,319).These cohorts were defined and analysed separately.The follow-up period for each child in the study was from birth up until their fifth birthday.Children who died prior to 5 years of age during the study period (excluding stillbirths) were retained in the analysis cohorts (Hospital Cohort, n = 77; Emergency Department Cohort, n = 144), as they were eligible for hospital admissions and ED presentations up to the time of their death.
2.82), children born in remote Australia had an aOR of 3.45 (95% CI 2.55, 4.66), and children born in very remote Australia had an aOR of 4.69 (95% CI 3.64, 6.06).Children with a prior history of hospitalisation DAMA had an adjusted odds ratio for DAMA 25% less than children without any previous history of hospitalisation DAMA.Children aged <1 at hospital admission had an adjusted odds ratio for DAMA 24% greater than children aged ≥1 to <5 years at admission.The adjusted odds of hospitalisation DAMA decreased over time: compared with admissions in the period 2002-2005, the aOR for DAMA was 0.52 (95% CI 0.42, 0.64) in 2006-2009, 0.46 (95% CI 0.37, 0.57) in 2010-2013 and 0.25 (95% CI 0.18, 0.34) in 2014-2018.There were substantial differences in the adjusted odds of DAMA, based on admission status: children with admission statuses of 'elective not from waitlist' (aOR 3.42, 95% CI 1.56, 7.47), and 'emergency admission' (aOR 6.18, 95% CI 3.03, 12.59) had increased adjusted odds of DAMA compared with children who were admitted as 'elective from waitlist'.Children who attended a hospital outside the Perth metropolitan area had nearly 7 times the adjusted odds of DAMA compared with children who attended in the Perth metropolitan area.
1.11), and children born in very remote Australia had an aOR of 0.61 (95% CI 0.57, 0.65).Compared with children without any previous history of ED DAMA, children with a prior history of ED DAMA were at 24% increased adjusted odds of ED DAMA.Compared with children aged ≥1 to <5 years at ED presentation, children aged <1 at ED presentation were at 26% decreased adjusted odds of ED DAMA.There was a general trend for ED DAMA to increase over time: compared with ED presentations in the period 2002-2005, the aOR for DAMA was 1.39 (95% CI 1.30, 1.49) in 2006-2009, 1.34 (95%CI 1.25, 1.44) in 2010-2013 and 1.46 (95% CI 1.35, 1.58) in 2014-2018.There was a strong pattern of increased adjusted odds of ED DAMA with decreased urgency in triage code: compared with the triage code of emergency, the triage codes of urgent (aOR 3.72, 95% CI 2.76, 5.00), semi-urgent (aOR 10.36 95% CI 7.73, 13.90) and non-urgent (aOR 11.49, 95% CI 8.54, 15.45) were all associated with increased adjusted odds of DAMA.The triage code of resuscitation was associated with decreased adjusted odds of TA B L E 1 Socio-demographic characteristics of the study population, 2002-2013.
Although hospital DAMA decreased, the adjusted odds of ED DAMA increased by 46% over the same period.The increase in ED DAMA over time is an unexpected finding, and although it appears there is peak in ED DAMA during 2007, there is still an increase in ED DAMA in 2014-2018 compared to 2002-2005.And 34

5 |
CON CLUS IONSIn conclusion, our study highlights several factors associated with DAMA in hospitals and EDs, including hospital location, admission status, triage code and year of admission/ presentation.These findings have a number of practical implications.In particular, our results highlight the importance of admission status and triage code and suggest that further work is needed to reduce DAMA for admissions and presentations associated with less urgent entry to hospital and ED.AUTH O R CO NTR I B UTI O N SNS and DMA conceived of the paper.BM obtained the data.DC and NS undertook the data analysis.All authors contributed to writing the paper.ACK N O WLE D G E M ENTSThe authors thank staff at the Western Australian Data Linkage Branch and the data custodians of the Midwives Notification System, the Emergency Department Data Collection, Death Registrations and the Hospital Morbidity Data Collection for access to, and linkage of, the data.Aboriginal and Torres Strait Islander communities are formally acknowledged for their contribution to this research project reporting for Publications, Reports and Presentations.Georgia Whisson proof-read and edited a version of the manuscript.Open access publishing facilitated by Edith Cowan University, as part of the Wiley -Edith Cowan University agreement via the Council of Australian University Librarians.

aOR (95% CI) Minimal sufficient adjustment set used in adjusted models (see Figure S3)
TA B L E 2 Associations between characteristics and hospitalisation DAMA, 2002-2018.

aOR (95% CI) Minimal sufficient adjustment set used in adjusted models (see Figure S3)
Area socio-economic index, Birthweight, Gestational age at birth, Maternal age at birth, Potentially preventable hospitalisation, Parity, Remoteness Area, Sex, Year of Associations between characteristics and emergency department DAMA, 2002-2018.
TA B L E 2 (Continued)TA B L E 3

aOR (95% CI) Minimal sufficient adjustment set used in adjusted models (see Figure S4)
a Cells suppressed due to small cell size.