Geography and time-specific health data for environmental analysis

Interpreting data variation over time and geography

As noted in the Data considerations and limitations sections (under each of the source data sets) in this report, there are a range of factors that can cause variation in counts and rates for given data items over time and across or between jurisdictions. Causal links cannot be determined without further detailed analysis. Some of the factors that may lead to variation in the data are listed below.

Fundamental factors include:

• changes/differences in population demographic profiles
• policy shifts (for example, the introduction of the Better Access Scheme for MBS mental health services; changing PBS medication restriction levels and definitions of the eligible population) 
• changes in disease prevalence, health characteristics or treatment options
• changing markets for health service provision
• seasonality, for example:
  • changes in incidence of illness (for example, respiratory infections during winter, ‘back-to-school asthma’)
  • impact of public holidays on demand/ supply of health services
  • policy-induced changes (for example, the increase in PBS prescriptions filled towards the end of the calendar year associated with the PBS safety net policy)
• natural volatility.

Methodological factors include:

• changes/differences in diagnosis coding schemas and practices adopted by different jurisdictions or hospitals at different times
• changes in data scope (for example, the scope of the PBS collection expanded to capture under co-payment scripts on 1 April 2012)
• the application of mapping files to construct 2016 ASGS SA4 geography.

Additionally, intermittent shocks (for example, natural disasters – but also, events such as the COVID-19 outbreak) can cause effects which can vary across time and geographically, and can affect policies, practices or service use/ access.

Impact of COVID-19 on health service use

The COVID-19 outbreak led to significant shifts in many data series – and these were both fundamental and methodological in nature. For example, there were shifts in demand for different types of health services (for instance, due to lockdowns, and associated changes in the profile of disease), the delivery of health services changed considerably (for instance, with greater use of telehealth, prescription stockpiling), and relatedly in some cases, new codes were introduced to reflect the evolved health system. For further information on the impact of COVID-19 on service use, see:

• Australia's hospitals at a glance – Impact of COVID–19 on hospital care
• Admitted patient activity – What impact has COVID-19 had on admitted patient activity
• Emergency department care activity – Impact of COVID-19 on emergency department activity
• Impacts of COVID-19 on Medicare Benefits Scheme and Pharmaceutical Benefits Scheme service use.

Implications for research design

This data set, due to its innate structure, lends itself to particular research methods. 

For example, some empirical methods attempt to exploit the panel aspect of data – analysing variation across both geographical and temporal dimensions. These methods might allow researchers to attribute shifts in health service use in particular locations at particular times to environmental events, through assessing the extent to which evident trends and volatility in affected areas are also evident in unaffected areas. However, issues around the quality of the data (including its geolocation), and the comparability of the data over time and geography (as discussed throughout these notes) will need to be taken into consideration. Additionally, the aggregation of data to a weekly time period by SA4 geography will need to be considered in interpreting any results (for example, in interpreting estimates of the magnitude of changes in service use associated with environmental exposures).

In the context of bushfires, research might also benefit from a consideration of other factors (such as extreme heat), which are temporally and geographically correlated with bushfire, and which could separately impact health service use.

There are limitations to the utility of these data in assessing longer-term impacts of bushfire on health and health service use. For example, this data set would not facilitate a cohort study (following particular individuals over time). It reflects service use counts, rather than individuals. Moreover, since people tend to relocate over time, the correlation between an individual's recorded place of residence and the likelihood they were exposed to an historical environmental event in that place, declines the longer the time lag since the event occurred.