Can real time suicide data support suicide cluster detection?

Feasibility Study for Identifying Suicide Clusters Using Real-time Coronial Data

by Leo Roberts, Angela Clapperton and Matthew Spitta

Published 28 September 2023

What's the issue?

A suicide cluster is a group of suicides that occur within a short time frame in a population group or place at a rate higher than previously recorded or expected.

Suicide clusters are more common in young or indigenous people. It is important to identify suicide clusters early to use prevention measures to prevent further suicide deaths.

Access to real-time data may support the early identification of suicide clusters. In Australia, each State and Territory have independent data collection systems. However, a time lag of up to 18 months between the reporting of suicide deaths and the data becoming available is a problem for researchers and service providers. An additional problem impacting the ability to identify a suicide cluster is that available suicide data records the location of a suicide death as the deceased person’s place of usual residency (their home) rather than the location where the death occurred. Suicide deaths are also categorised by population density areas (Statistical Areas). Previous collection systems and analysis measures have hindered the ability to identify a suicide cluster if it has occurred on the border of or within two different geographical areas. In Victoria, the Coroner’s Court of Victoria (CCV) has developed a real-time suicide register called the Victorian Suicide Register (VSR). This study aimed to test the feasibility of using the Victorian Suicide Register to undertake regular monitoring of suicide clusters in Victoria using software and modelling.

What was done?

The authors of this report aimed to determine whether modern cluster detection methods can be used on real-time data with precise geocoordinates to monitor the emergence of suicide clusters.

The researchers sought ethical approval, developed a secure platform for storing sensitive data relating to suicide deaths, and obtained data on population sizes at various geographical levels and Statistical Areas.

The CCV provided researchers with data extracts from the VSR, which included date of suicide, age, sex, incident location latitude, incident location longitude, incident suburb, incident location description (e.g., usual residence or non-residential location), a residential location latitude, residential location longitude, residential suburb and the method of suicide.

The data covered the period from January 1, 2008, to June 30, 2022.

Cluster detection

Researchers used a software called SaTScan, which detects large groups of observations using a method called the scan statistic. The scan statistic investigates the presence of clusters by looking at different variables and geographical locations.

Researchers also used modelling to determine what the expected counts of suicide deaths would be for people under the age of 25 years across places of different population densities. Suicide deaths were classified as clusters if their p-value was ≤ 0.01.

To determine if the software and modelling would be able to detect clusters in real-time, the researchers also analysed the data in 2-month intervals to determine if the same identification of clusters occurred across the comparable time period. This helped the researchers to analyse if a cluster could be identified as it occurred, or would only be identified months after, or as part of a bigger cluster.

What was found?

The researchers found that the software used was able to detect significant suicide clusters in Victoria and, if applied to the VSR data, could potentially support the detection of suicide clusters in real-time.

SaTScan software may also have the ability to detect clusters from only four weeks or two weeks of data, aiding in the ability to prevent clusters in real-time.

Why are findings important?

The real-time identification of clusters creates new opportunities for community-level interventions to prevent future suicide in a particular geographical area or population group.

SaTScan-based modelling techniques could replace current cluster identification methods, with the potential to use this software to detect clusters without aggregating the suicide deaths by geographical or statistical area.

There is potential to apply SaTScan and modelling to other Australian States and Territories to identify suicide clusters in real-time and respond in appropriate ways to prevent further suicide deaths.