Limitations in ascribing injuries to weather

When trying to identify a relationship between injury and weather patterns, the available data present a range of challenges.

Data fragmentation

In Australia, several agencies produce data relevant to weather-related injuries. Each agency has a different purpose and produces reports relating to its own remit.

The Bureau of Meteorology (BOM) does not produce data on injuries but provides weather forecasts, warnings and data that can be used to identify areas at risk. There are a range of organisations focused on specific injury mechanisms that collate data on deaths and injuries, but these are not routinely linked to indicators of extreme weather involvement.  

The Australian Institute of Health and Welfare (AIHW) collects and analyses data on a wide range of health-related issues, including injury hospitalisations and deaths caused by severe weather events. It currently does not have access to consolidated data assets enabling analyses of injuries in the primary care or emergency management sectors.

NHMD data captures admitted hospital patients only. It is likely that only a small proportion of weather-related injuries lead to a stay in hospital, and these are typically more severe injuries. The data do not include information on people who sought treatment in the ED, general practices, outpatient settings or pharmacies.

The National Coronial Information System (NCIS) also provides data on deaths, including those caused by severe weather events such as heatwaves, storms, and floods.

The reports produced by these agencies are often using data that is collected by other organisations at the national and state levels, making it difficult to obtain a comprehensive picture of the problem. Additionally, some types of weather-related injuries, such as those caused by floods or landslides, may be underreported due to a lack of specific data collection efforts.

The use of data collected for other primary purposes

Administrative health datasets under-estimate the occurrence of weather-related injuries for the following reasons:

  • limitations of ICD-10 external cause codes in capturing all weather-related injury hospitalisations (for example no transport accidents appeared in the results – this is discussed below).
  • use of NMD data without comparison to NCIS data which can provide a more detailed picture of circumstances surrounding a death.
  • the likelihood of there being cases where weather contributed to the injury, but where this was not reported by the patient, and/or not recorded in clinical notes or coded in the patient record.
  • there is no standardised way to capture the level of contribution of a weather event to a diagnosis.

They may over-estimate the occurrence of some weather-related injuries because some codes include, alongside weather-related injuries, some injuries that may not have been weather-related, such as, for example, injuries caused by non-bush fires or hypothermia caused by sustained contact with a cold environment such as the ocean.

Available codes are not specific to weather-related events

The relative contributions of different types of weather to the data presented may in part reflect the available ICD-10 codes (methodological bias). For example, the fact that heat is the most frequent weather event may be due, in part, to heat being easier to record using the available codes. Available data do not allow for examination of transport injury hospitalisations with a weather-related external cause code. This is because there is no suitable ICD-10-AM code to indicate the involvement of weather in a transport injury.  

Case type category for weather-related deaths

Typically, a doctor certifies deaths due to natural causes and a coronial investigation is not required. Deaths due to natural causes (for example, chronic condition exacerbation from a heatwave) are types of deaths that may not result in coronial investigation. Therefore, the NCIS is not suitable as a standalone data source for weather-related deaths.

Separation between weather and injury event data

Weather data and injury data are stored in separate databases. Weather conditions are linked to the location of events such as bushfires, droughts, floods, heatwaves, severe thunderstorms and cyclones. However, injury data are not explicitly linked to weather events.

Further, within health datasets, diagnosis and servicing information may be separate (e.g. the NHMD and the MBS/PBS datasets).

Confounding effects of adaptation

This analysis does not account for injury prevention or risk-reduction behaviour. For example, if weather-related injury numbers are higher in 2017–18 compared to 2018–19, it does not necessarily mean the weather was less extreme in 2018–19. Instead, this could be due to:

  • the effectiveness of injury prevention initiatives.
  • behavioural responses such as forgoing or rescheduling activities that were inappropriate for the weather conditions.

Causation and correlation

The extreme weather event-based approach enables health care data sets that contain no weather-related information to be used if time and place of injury information can be matched with the time and place of an extreme weather event. It is difficult to attribute causation of the injury directly to weather events using this approach, but it is possible to establish correlation. The secondary effects of extreme weather also need to be accounted for. For example, injuries may be sustained during clean-up operations undertaken due to the effects of adverse weather events, or existing health conditions may be exacerbated by extreme weather events hampering access to health care.

Dataset size and granularity

Bureau of Meteorology data are fine-grained, with thousands of data points collected daily. The data are available on demand with no privacy restrictions, however the size of the national dataset, combined with health service data may pose issues relating to the computing power and processing time required to run such analyses.

Data quality

Statistical analysis of NHMD and NMD data is a secondary use. As such, some data are missing or limited due to data quality issues, in particular, activity and place-of-occurrence data.

In 2021–22 data, 66% of NHMD injury cases have a missing or unspecified value for activity at the time of injury, and 39% have a missing or unspecified value for place of occurrence. This may be because the information was not reported by the patient, and/or the information was not recorded in the clinical notes or the information was not coded into the hospital record. 

There are similar data quality issues with these data items in the NMD. The NMD only contains summary data on the cause of injury deaths in terms of ICD-10 cause-of-death codes. More detailed information is available on the circumstances and causes of most injury deaths from the National Coronial Information System.

The NNAPEDCD covers ED presentations at public hospitals, but diagnosis information may be incomplete where a definitive diagnosis is not possible within the ED episode of care (for example, where further diagnostic tests are needed), and it does not include any information on the external cause or circumstances of the injury event. Therefore, no weather-related information is included in the data set.