Methods

Conditions associated with suicidal behaviour

Although a wide range of risk factors for suicide have been identified in the literature, this report specifically focusses on mental and behavioural disorders, substance use (alcohol or drugs) and intentional self-harm as conditions associated with suicidal behaviour. These conditions were selected for detailed analysis based on their documented significant impact on suicide risk in previous research (Causes of Death, Australia, 2021; Clapperton et al 2021). For details on how these conditions were defined and identified within the datasets analysed, please refer to the Codes and Classifications section of this report.

It is also important to highlight that suicide is a complex issue influenced by an interplay of social, environmental, and psychological factors which are not examined in this report. Furthermore, it is important to remember that the presence of one or more of these risk factors cannot predict or explain suicide or intentional self-harm as each person’s experience is unique. Experiencing any of these risk factors does not necessarily mean a person has – or ever will – attempt suicide, but establishing whether a person has any of these risk factors can help determine whether they are at increased risk. Also, some people will have suicidal thoughts without having a history of any risk factors.

Constructing the admitted patient care analysis data set 

To use the linked data for this analysis, various data processing procedures were undertaken to ensure the accuracy of findings. Apart from general data cleaning such as management of missing data and the removal of duplicates, the main procedures are described below.

Concordance with the NDI

NDI death information was linked to the patient demographic information and used to identify potential linkage errors. Data for patients was removed where any of the following discrepancies were identified where:

  • a patient had linked NDI death information and any separation with date of separation occurring more than seven days after NDI date of death (except where the care type indicated posthumous organ donation)
  • a patient had a hospital separation ending in death that preceded the NDI date by more than 7 days
  • a patient had a hospital separation ending in death and no NDI record of death.

Hospital stay

A patient’s stay in hospital for admitted patient care may include more than one episode of care. This occurs if the care type changes e.g., from acute to rehabilitation or if the patient is transferred from one hospital to another, including temporarily for a specific procedure. This analysis combined all contiguous episodes into a single stay enabling more accurate counting of hospitalisations.

Same day and overnight hospital episodes were processed together to combine all relevant episodes into a hospital stay. Contiguous, overlapping, and nested episodes were combined except when neither the separation mode of the earlier episode nor the admission mode of the later episode indicated a transfer between hospitals or a change in care type. A gap of one day was permitted between separation and admission in a single stay if the separation mode indicated a hospital transfer or care type change, to allow for overnight transfers.

Episodes were excluded where:

  • the admission or separation date was missing
  • the admission date occurred after the separation date
  • the episode could not be linked to a person

After applying exclusions and sorting by admission and separation dates, episodes were combined when:

  • the admission date for the later episode was prior to the separation date of the earlier episode (overlapping or nested episode)
  • the admission date for the later episode was the same day as the separation date of the earlier episode (contiguous episodes) except when the admission mode of the later episode did not indicate a transfer or care type change:
    • Other 

and the separation mode of the earlier episode indicated the end of the stay:

  • Left against medical advice/discharge at own risk
  • Statistical discharge from leave
  • Other (includes discharge to usual residence, own accommodation/welfare institution (includes prisons, hostels and group homes providing primarily welfare services)
  • the admission date for the later episode was one day after the separation date of the earlier episode, the separation admission mode of the later episode indicated a transfer or care type change:
  • Admitted patient transferred from another hospital
  • Statistical admission - episode type change

and the separation mode of the earlier episode indicated a transfer or care type change:

  • Discharge/transfer to (an)other acute hospital
  • Discharge/transfer to (an)other psychiatric hospital
  • Statistical discharge – type change

Resulting stays were excluded where the care type of the initiating episode was unqualified newborn days, posthumous organ donation, or hospital boarder (care type 7.3, 9, 10).

Definitions for multi-episode stays

Table 22 clarifies modifications to the definitions of key NHMD data items used for this analysis following the construction of hospital stays from multiple contiguous episodes.

Table 22: Modifications to NHMD data items for ‘stay’ based analysis

Data item 

Definition applied to multi-episode stays

Admission date

Admission date of the earliest episode in the stay.

Separation date

Latest separation date among episodes in the stay where no episode ended in death (separation mode=8). Where death was recorded in one or more episodes in the stay, after sorting by admission and separation dates the separation of the first episode ending in death was used, except where a subsequent separation due to death matched the date of death from NDI.

Admission mode

Admission mode of the earliest episode in the stay.

Separation mode

Separation mode from the episode with the latest separation date in the stay where no episode ended in death, or 8 (Died) where any episode ended in death. 

Where more than one episode shares the latest separation date, if one or more of these were same-day episodes they were treated as contiguous and the separation mode from the latest same-day record occurring on the latest separation date was used. Where neither were same day episodes, the episode with the later admission date was treated as a nested episode, and the separation mode from the earlier admission was used. 

Data set year

Reporting year for separation date as derived above.

Principal diagnosis, additional diagnoses, and external causes

Principal diagnosis, additional diagnoses, and external causes from the earliest episode in the stay to allow for identification of reason for hospitalisation.

Urgency of admission

Urgency of admission of the earliest episode in the stay

Same-day flag

Indicates whether the derived admission and separation dates for the stay are the same day.

Mental health flag

Mental health flag of the earliest episode in the stay.

Any mental health flag

Indicates whether any episode in the stay had a mental health flag regardless of flag value.

Sector

Hospital sector of the first relevant episode in the stay

Mental health, alcohol and other drug, and mechanism of self-harm diagnoses

Aggregated fields capturing whether diagnoses of interest were present in the principal diagnosis of any episode within a stay.

Intentional self-harm related stay

Captures whether any episode within a stay had a principal diagnosis and first external cause identifying an intentional self-harm admission.

Constructing the emergency department care analysis data set

Defining the index presentation

The self-harm or suicidal behaviour cohort was defined as any persons with any ED presentation throughout the study period with any diagnosis of intentional self-harm and/or suicidal ideation, as per the codes in Tables 16 and 17. The index presentation was identified as the first of these with the data ordered chronologically by presentation date and time. The index presentation, for the purposes of its definition, may or may not have also included a risk factor diagnosis. The index presentation was coded at its first documentation. For example, where contiguous ED episodes included interhospital transfer and the index diagnosis only occurred in the second ED, the date and patient demographics of the second ED were used. 

The identification of an index presentation self-harm or suicidal behaviour in this study should not be assumed to be the incident (first ever) presentation for a person, given the time-period constraint and the fact that the ED data were examined in isolation to the admitted patient data and/or primary care data which may identify earlier episodes. Some jurisdictions also have specific community mental health data collections (e.g., NSW), which would provide an additional data source to find the ‘true’ incident presentation to any healthcare practitioner for self-harm or suicidal behaviour. The index presentation can be considered a first presentation with these specific diagnoses in a period prevalence analysis. These analyses do not present data on duration of any diagnosis. 

Defining the presence of risk factors

The risk factors described in Table 20 were identified as present at any time during the study period in any of the 3 available diagnosis fields. This includes risk factor presentations that were diagnosed concomitantly with a self-harm or suicidal behaviour diagnosis. They were counted uniquely per presentation, then categorised by frequency of presentations per risk factor, per person. For consecutive ED transfers where risk factor diagnoses may be rerecorded for each presentation prior to and following transfers, risk factor diagnoses were only counted once.

ED presentation lookback analysis for risk factor presentations prior to a self-harm or suicidal behaviour presentation, only considered risk factor ED presentations that occurred separately and before the index episode for self-harm or suicidal behaviour (even on the same day). A concomitant diagnosis of any risk factors assessed with the index self-harm or suicidal behaviour presentation was not counted as ‘prior history’. 

Statistics      

Proportion and percentage

Proportion is the quotient obtained when the number of cases in a group with a characteristic of interest is divided (the numerator) by the total number in the group (the denominator). Its value is between zero and one. A percentage is a proportion multiplied by 100.

In this analysis, proportion is presented as a percentage and provides information on the number of persons affected. For example, the proportion of ex-serving ADF members who received admitted patient care in an Australian public hospital who were admitted for an intentional self-harm related stay. 

Proportion denominator

The admitted patient care data available for this analysis was incomplete. For the Australian comparator analysis, public hospital data was not available for WA and the NT, and complete private hospital data fit for this analysis was not available for WA, NT, NSW, VIC, TAS, SA, and the ACT. These data were available for eligible ex-serving ADF members receiving DVA-funded care in public or private hospitals in all states and territories.

To limit the bias introduced by differences in the availability of data, the analysis was restricted to admitted patient care provided in public hospitals (including DVA-funded care in WA and NT). Proportions were calculated as the proportion of patients for a nominated diagnostic group of all patients in participating public hospitals by reporting period.

Proportion difference

Proportion differences (PDs) also referred to as absolute differences are presented in the data tables as the absolute difference in percentage points between the two populations. They are a measure of the magnitude of the gap between populations without respect to the size of the individual rates. PDs are subject to volatility when used with small numbers and hence should be used with caution when comparing ex-serving ADF member and Australian population results. 

Relative difference

The relative difference (also referred to as the risk ratio or relative risk) is the ratio of the proportions for the ex-serving population and Australian populations and measures the scale of the difference. When the proportions are small, relative difference can be a more relevant and useful descriptive measure. A relative difference of 1 indicates that the proportion of ex-serving and Australian groups admitted for a condition/group of conditions is identical and there is no association between being admitted and being in a specific population. A relative difference greater than 1 indicates a positive association and that the risk is higher while if it is less than 1 then there is a negative association, and less risk.

Statistical significance and confidence intervals

Statistical significance is a measure that indicates how likely an observed difference would occur under the conditions of the null hypothesis i.e., the hypothesis that there is no significant difference between the specified populations, any observed difference being due to error. This report provides 95% confidence intervals (CI) to indicate a range that is likely to contain the true value with a 95% degree of confidence. For smaller populations, changes in the numerator due to random variation have a greater effect. Proportions produced for small populations will therefore have wider CIs. Wide CIs imply less certainty around a calculated value; narrow CIs imply more certainty. The result is interpreted as being a statistically significant different if the CI does not contain zero.

CIs in this report were calculated using the normal approximation method and are not reported for proportions of populations with fewer than 25 persons. It is important to note that there are other sources of uncertainty not captured by CIs, such as linkage error. Additionally, statistically significant differences between ex-serving ADF members and Australians are not necessarily explained by prior ADF service and may be explained by other socio-demographic differences between the cohorts.

Small numbers and suppression of identifiable data

Findings based on small numbers of events (such as hospitalisations for intentional self-harm) can fluctuate from year to year for reasons other than an actual change increasing risk of the event. Small groups have resulted from disaggregating the ex-serving member population by age, sex, diagnoses, and military characteristics. This has limited analysis e.g., aggregating diagnostic groups or confidentialising small cell counts. For ex-serving members, ex-serving members who died by suicide, and Australians who died by suicide, counts and proportions are not reported for fewer than 5 persons. For the Australian (all) comparator group, counts and proportions are not reported for 10 or fewer persons.

Confidentialisation was applied where states or territories were dominant contributors to the number of stay events. Data are not reported for combinations of diagnosis group, cohort, and reporting period where:

  • fewer than 3 states or territories contributed 100% of events
  • one state or territory contributed at least 85% of events
  • two states or territories combined contributed at least 90% of events

Where data are not reported due to small numbers or dominance, consequential suppression may be applied to additional data to prevent their calculation. For example, where data for females are not reported, data for total persons may be suppressed to prevent calculation by subtracting males from total persons.