Methods and data sources

Disease costs included in this report are derived from the Disease Expenditure Database. The AIHW disease expenditure database contains estimates of expenditure by Australian Burden of Disease Study (ABDS) condition, age group, and sex, for public and private admitted patient, emergency department, and outpatient hospital services, out-of-hospital medical services, and prescription pharmaceuticals (AIHW 2021a). In this database, spending on a given condition is viewed as a component of total health spending and as relative to spending on all other conditions.

The ABDS has estimated the proportion of disease burden for certain conditions that is due to physical inactivity as a risk factor. Physical activity is the inverse of physical inactivity and can be treated as such in a model. This project uses the fraction of chronic disease burden that is due to physical inactivity to calculate the health cost of inactivity, and to estimate the health costs avoided due to participation in activity.

The areas of spending in this analysis include hospital services, primary health care services and referred medical services (generally provided by medical specialists). Hospital services include public and private admitted patient services, public hospital emergency departments, and public hospital outpatient clinics. Primary health care includes general practitioner services, allied health services, pharmaceuticals and dental. Referred medical services include specialist services, medical imaging, and pathology. Due to data availability in the primary health care sector, these estimates are biased towards spending in hospitals and underestimates spending outside of these settings, and estimates should be interpreted with this in mind.

Burden of disease overview 

Physical inactivity is an important risk factor for health, and is estimated to be responsible for 2.5% of disease burden in Australia (AIHW 2021b). In the Australian Burden of Disease Study 2018 (ABDS), physical inactivity is causally linked to the burden from type 2 diabetes, bowel cancer, dementia, coronary heart disease and stroke, as well as uterine and breast cancer in females. New evidence is available to further estimate the association of inactivity with several other conditions and risk factors, discussed below in Which new conditions/mediating factors are being included?

The estimated contribution of a risk factor to disease burden is calculated by comparing the observed risk factor distribution with an alternative and hypothetical distribution (the counterfactual scenario). For physical activity, the theoretical minimum risk distribution is compared with what is currently observed in the population. The relative risk of conditions due to physical inactivity is estimated on the basis of differences in disease outcomes for varying levels of physical activity compared to the minimal risk group, and combined with data on levels of activity undertaken in the population, to generate the population attributable fraction (PAF). Physical activity is measured by the Metabolic Equivalent of Tasks (METs) performed across the Australian population, and classified into eight levels. Each of the categories has a relative risk for various conditions, with the highest risk being the most sedentary group. Various studies have used different condition sets to estimate the health costs from physical inactivity, and the differences in disease inclusions may lead to large differences in the outcomes. The AIHW models 7 conditions (breast cancer, bowel cancer, coronary heart disease, dementia, type 2 diabetes, stroke, uterine cancer), while the Global Burden of Disease Study models 5 conditions (colorectal cancer, breast cancer, ischaemic heart disease, ischaemic stroke, type 2 diabetes). The analysis in this report includes several conditions in addition to those included in the Australian Burden of Disease Study (ABDS). Details of methods used to estimate levels of physical activity undertaken are available in the  Australian Burden of Disease Study: methods and supplementary material 2018 publication (AIHW 2021c).

The ABDS risk factor of physical inactivity largely adheres to the scope from the Australian physical activity guidelines to estimate activity levels in terms of activities counted as physical activity (all physical activity domains, such as leisure, transport, occupational and household chores). This is to ensure that an accurate measure of total physical activity levels undertaken in Australia are included. However, it should be noted that the level of activity considered lowest risk for this analysis (4,200 METs/week) is different to the recommended level of physical activity in the Australian guidelines, as the purpose is to calculate total disease risk due to all levels of physical activity. Further information is available in the Frequently Asked Questions for the Australian Burden of Disease Study 2018: Interactive data on risk factor burden.

Estimating the benefits of physical activity

It is not possible to directly observe the health costs avoided by participating in physical activity. Therefore, to estimate the avoided health costs we must first examine the costs to the health system if no-one was physically active. This hypothetical condition is termed the ‘counterfactual’, as it’s ‘counter-to-the-fact’ that people are active in the population. This counterfactual is estimated in a similar way to the method outlined above, using ABDS relative risks, population age and size, and proportion of the population in each grouping of physical activity to estimate the population attributable fraction (PAF). However, rather than using the group with the highest possible level of physical activity as the comparator for a population risk analysis, the lowest level of activity is compared to the level of activity actually undertaken in the Australian population.

This PAF is applied to the condition costs to calculate the risk factor cost. The cost estimates that result from this process are the estimated cost of cancer, cardiovascular diseases, diabetes, and other conditions that may have occurred without the protective effects from physical activity. The difference between the observed expenditure and the counterfactual represents the health system savings from physical activity. This method provides an average ‘snapshot’ in a year of the health effects of physical activity, and does not capture any dynamic effects on population health through changes to life expectancy, diagnosis of other conditions, or remission of disease. This will be addressed in future stages of work to develop a population simulation model.

Which new conditions/mediating factors are being included?

The AIHW undertook a literature review to find updated evidence for associations between physical inactivity and conditions it is directly linked to, and to other risk factors that cause disease themselves.

The directly linked conditions that are being included in addition to the ABDS conditions in this model are: depression, anxiety, and falls.

The intermediate risk factors that are generally considered to be improved with physical activity are body mass index, blood pressure, blood sugar, cholesterol, and bone mineral density. This study is able to include the impact of physical activity on the ABDS risk factor ‘high blood pressure’, ‘high fasting plasma glucose’, and ‘low bone mineral density’.

While physical activity can improve BMI through decreased fat mass and maintained or increased resting metabolic rate, the evidence suggests this does not occur independently of the overall energy balance achieved through food consumption. Therefore, this model does not include physical activity reducing high body mass index, as this may overestimate the impact activity has on health and healthcare spending.

Evidence does suggest that physical activity reduces low density lipoprotein cholesterol, but the estimates available are not sufficient to incorporate this risk factor association into the model at this stage.

Methods to update risk factor model

Direct associations

For new conditions that are directly associated with physical inactivity, a PAF for each condition by sex and age group was calculated and applied to expenditure estimates for the condition.


A dose response meta-analysis of cohort studies by Pearce et al (Pearce et al., 2022) with almost 200,000 participants (from predominantly high income countries) and 2.1 million person-years found the relative risk (RR) for 17.5 marginal MET-min per week for depression was 0.72. This study found a strong dose response relationship between activity levels and depression incidence.

This study was used to estimate the PAF of burden due to varying levels of physical activity undertaken, and was applied to expenditure estimates for depression by age and sex.


A meta-analysis by McDowell involving over 80,000 participants (from predominantly high income countries) reported results separately for anxiety symptoms, anxiety disorder, and generalised anxiety disorder (McDowell et al., 2019). This analysis found an odds ratio (OR) of 0.87 for anxiety symptoms, 0.66 for any anxiety disorder, and 0.54 for generalised anxiety disorder, for people who were active compared to inactive.

The odds ratio for anxiety symptoms was used in this analysis as a conservative estimate to model spending due to inactivity, and the risk difference was applied equally to those who did >600 MET-min/week compared to no activity (0-600 MET-min/week).


A systematic review and meta-analysis by Sherrington in 2020 (Sherrington et al., 2020) included approximately 25,000 participants (from predominantly high income countries) and found that physical activity reduces the rate of falls by up to 23% across intervention analyses. The analyses included were not targeted at frail people, and were not treatment interventions. A meta-regression analysis by the authors indicated there may be a dose-response relationship as interventions with greater frequency of exercise per week had a larger reduction in falls risk (although these findings did not reach traditional levels of statistical significance testing).

This analysis used the RR of 0.77 and applied the risk reduction to those who undertook any activity (>600MET-min/week) compared to no activity (0-600 MET-min/week) to estimate spending due to inactivity.

Mediation pathways

Estimating the attributable spending due to physical inactivity through other risk factors was undertaken in a two-step process. First, the fraction of each risk factor that is related to inactivity was estimated by age and sex. Second, this fraction was applied to the PAF estimated for each of the risk factor-disease combinations. For example, if 20% of high blood pressure is due to inactivity, then 20% of disease burden attributed to high blood pressure is then estimated to be due to inactivity.

These estimates do take into account mediation between these risk factors themselves, using the mediation factors from the ABDS, and a joint effect calculation. For further details, refer to Technical notes and the Australian Burden of Disease Study: methods and supplementary material 2018.

Blood pressure

A dose-response meta-analysis by Liu et al, 2017, estimated the risk of hypertension by level of total physical activity. This meta-analysis included 330,222 participants (from predominantly high income countries) from 22 prospective cohort studies, spanning 2-20 years of follow up. The average RR per 600 MET-min/week increase was 0.94.

The RR was calculated for each activity level and the proportion of high blood pressure due to inactivity was estimated. This proportion was then applied to the PAF estimates for each condition caused by high blood pressure.

Fasting plasma glucose

There has not been any systematic reviews that have evaluated the effect of physical inactivity on fasting plasma glucose itself, but several have included diabetes as the outcome of interest. The most robust of these studies was used to estimate the RR of inactivity to diabetes in the GBD and ABDS (Kyu 2016).

This analysis uses the current physical inactivity PAF for diabetes to estimate the cost of conditions linked to fasting plasma glucose that are due to physical inactivity.

Note that estimates for fasting plasma glucose in this analysis do not include diabetes, as the ABDS assumes 100% of diabetes burden is due to high fasting plasma glucose. The PAF used to directly estimate type 2 diabetes due to inactivity is unadjusted for mediation through this pathway, and includes the component that is caused by high fasting plasma glucose.

Osteoporosis (Low bone mineral density)

A study by the World Health Organisation by Pinheiro, 2020, estimated the risk of osteoporosis for those with low physical activity using meta-analysis of randomised controlled trials (RCTs) involving approximately 2000 participants (from predominantly high income countries). These results showed a small but statistically significant effect of 0.15 SEM (Standard Effect Measure, equal to OR =1.313).  This study also provided a summary of 12 cohort studies that also showed an effect, though this was not included in the meta-analysis due to differences in study designs.

This analysis uses the risk difference applied to those who undertook any activity (>600MET-min/week) compared to no activity (0-600 MET-min/week) to estimate the fraction of low bone mineral density due to inactivity, and applies it to the low bone mineral density PAF to estimate spending due to inactivity.


There are several key limitations to the estimation methods described above. The reduction in risk due to physical activity is applied equally to all levels of physical activity above 600 MET-min/week for anxiety, falls, and bone mineral density. This is due to the estimates available being generally presented as ‘active’ and ‘inactive’. The level of activity undertaken in the ‘active’ group is generally around a minimum of 600 MET-min/week, though can be above this level. For most conditions, there is a dose-response relationship between activity level and risk reduction, which is reflected in the risk estimates. For these conditions, this dose response is not able to be estimated from the available data, and the attributable fraction is likely to be an underestimate of the impact of inactivity, as the greater reduction in risk from higher levels of activity is not captured.

Which new conditions/mediating factors are being included?In addition, data quality of the underlying studies for these risk differences is not as robust as traditionally required for a burden of disease study. However, these studies and their inputs were used in our analyses as they provide more conservative estimates of risk differences than other studies for the same condition, and are consistent with other research in the field. These estimates are intended to provide an indication of the impact of these new disease associations being explored, and should be interpreted with this in mind.