Drivers of change in risk factor attributable burden

This chapter presents new analyses undertaken using data from the Australian Burden of Disease Study 2015 to explore the different drivers of change over time in burden of disease attributable to 5 selected risk factors: tobacco use, alcohol use, overweight and obesity, high blood pressure and dietary risk factors.

The disease burden attributed to a selected risk factor is referred to as ‘attributable burden’ and is measured in terms of Disability Adjusted Life Years (or DALY). It reflects the reduction in fatal burden (measured by years of life lost (YLL)) and non-fatal burden (measured by years lived with disability (YLD)) that would have occurred if exposure to the risk factor had been avoided or reduced to its lowest level. For more information on how attributable burden is calculated, see Australian Burden of Disease Study: methods and supplementary material 2015.

Between 2003 and 2015, there was an 11% increase in the total number of DALY attributed to all risk factors included in the Australian Burden of Disease Study (for those that were measured in both 2003 and 2015). Note this is different to changes in the age-standardised rate of attributable DALY per 1,000 population (which was 76.7 DALY per 1,000 in 2003 and 65.3 DALY per 1,000 in 2015).

The main factors contributing to the increase in the total number of attributable DALY were population growth (the Australian population increased by 21% between 2003 and 2015), population ageing, changes to exposure to the risk factor in the population, and changes in the amount of burden for diseases linked to each risk factor. These 4 factors and the method used to estimate the contribution of each to changes in attributable burden, are explained further in the box below.

Key results from these analyses for the 5 selected risk factors can be found in the data visualisations below and in Table S2.

How to interpret the drivers of change over time analyses and charts

The figure below is provided to help readers interpret the analyses and charts presented in this web release. Each factor included in the drivers of change over time analyses (population growth, population ageing, risk exposure and linked disease burden) may cause the attributable burden from a risk factor to rise (indicated by a positive percent change) or fall (a negative percentage change) over time. The sum of the effect of all factors represents the overall actual change in attributable burden between 2003 and 2015.

Formula for change in attributable burden

Put simply, these analyses show that if the overall attributable burden due to a risk factor is increasing (i.e. getting worse), we can see which factors are most responsible for this increase and target policy and program responses accordingly. Secondly, it also gives us additional information on those risk factors for which burden is decreasing (i.e. getting better) and whether there are still factors (e.g. increasing exposure or linked disease burden) which could be targeted and result in further improvements in the attributable burden for that risk factor.

How are the estimates of drivers of change over time calculated?

The Das Gupta method was used to decompose the changes in burden attributable to each risk factor into 4 additive components (Das Gupta 1993). Using a series of scenarios this method calculates the effect of each factor on the changes over time by assuming that all other factors, except the factor under consideration, remain the same at both time points.

The change in overall attributable burden is decomposed into changes due to:

  • population growth—in Australia population size is increasing over time
  • population ageing—in Australia the proportion of older people is increasing over time
  • risk factor exposure—changes in the prevalence of exposure to the risk factor in Australia.
  • Changes in linked disease burden— changes in the overall burden for those diseases or injuries that are linked to the selected risk factor. This may be influenced by changes in diagnosis, treatment or health intervention (resulting in changes in disease prevalence or severity), as well as changes in other risk factors. For example, increases in overweight and obesity may have some impact on coronary heart disease burden which is also linked to tobacco use.

Attributable burden is estimated as the product of these 4 factors using the formula when examining burden by type of exposure to the risk factor:

formula

where

Bt         is the amount of burden (DALY, YLL or YLD) attributable to a particular risk factor at time point t.

i           is a type of exposure to the risk factor such as current tobacco use 

n          is all types of exposure included in the estimate for the risk factor

j           is an age and sex group

m         is all age and sex groups included (males and females aged 0 to 100+)

t           is a time point

Pt         is the total population size at time t

Sijt        is the share of the population in age and sex group i at the time t

Rijt           is the rate burden of diseases linked to exposure i in the age and sex group j at the time t.

Fijt        is the population attributable fraction of diseases linked to exposure i in age and sex group j at the time t.

∑          is the sum of all of the types of exposures i and all of the age and sex groups j

Attributable burden is estimated as the product of these 4 factors using the formula when examining burden by linked disease group:

formula

where

Bt         is the amount of burden (DALY, YLL or YLD) attributable to a particular risk factor at time point t.

k           is a disease group of the burden linked to the risk factor

o          is all disease groups of diseases linked to the risk factor

j           is an age and sex group

m         is a age and sex groups included (males and females aged 0 to 100+)

t           is a time point

Pt         is the total population size at time t

Sijt        is the share of the population in age and sex group i at the time t

Rijt           is the rate burden of disease group k linked to the risk factor in the age and sex group j at the time t.

Fijt        is the population attributable fraction for disease group k in age and sex group j at the time t.

∑          is the sum of all of the disease groups k and all of the age and sex groups j

The effect of each of the 4 factors—population size, population ageing, linked disease burden and risk factor exposure—using this method on the change in attributable burden between 2003 and 2015 is calculated as:

formula

formula

formula

where

EA         is the effect of factor A (population size, population ageing, linked disease burden and risk factor exposure)

B               is the amount of burden (DALY) attributable to the risk factor in 2003 (B03) in 2015 (B15)

P          is the population size in 2003 (P03) or in 2015 (P15)

S          is the population age structure in 2003 (S03) or in 2015 (S15)

R          is the rate burden of diseases linked to risk factor in 2003 (R03) or in 2015 (R15)

F          is the population attributable fraction of diseases linked to exposure in 2003 (F03) or in 2015 (F15)

The estimates were calculated using a statistical program developed by Dr Jinjing Li from the University of Canberra (Li 2017).

What are the benefits of the methods used in this analysis and how does this compare to a stepwise methodology?

The Das Gupta method for decomposing changes over time provides an indication of the proportionate impact of the specified factors (assuming any other unspecified factors are small and independent of the specific factors). The method distributes the interaction effects (such as the relationship between an ageing population and disease burden) between the factors in proportion to the strength of the main effects (Zhai et al. 2017).

It is also possible to decompose the changes over time in attributable burden using a stepwise methodology. This methodology was used in the Australian Burden of Disease Study: impact and causes of illness and death in Australia 2015 report (Tables D12 & D13). The stepwise approach requires a logical order to be chosen for the factors to be included in the analyses and give a different result if the factors are included in a different order.

A comparison of results using the Das Gupta method and the stepwise method suggested no difference in the factors contributing the most to the change over time for each risk factor or in the direction of the change (increase or decrease) with the exception of the proportionate contribution due to risk factor exposure for alcohol use which was 1.2% using the Das Gupta method and –7.6% using the stepwise approach. There were also notable differences between the 2 methods in the proportionate contribution of changes due to risk factor exposure for overweight and obesity, and changes in linked disease burden for alcohol use and overweight and obesity (see Table S3). A comparison between the 2 methods could not be made for the individual dietary risk factors as the stepwise approach used in the Australian Burden of Disease Study: impact and causes of illness and death in Australia 2015 report only included this analysis for all risk factors combined.

What are the limitations of the methods used in this analysis?

Only factors that could be easily measured (population ageing, population growth, changes in disease/injury and changes in risk factor exposure) were included in these analyses. However, these are considered to be among the most important drivers of change in attributable burden over time. It is not possible to include other factors in the analyses such as socioeconomic status that may also have an impact on changes in attributable burden over time as they are not able to be quantified.

How do these estimates of drivers of change compare to age-standardised rates?

Both age-standardised rates (which use a 'standard' population to produce rates that can be compared independent of the age structure of the study population(s)) and the drivers of change estimates presented here are methods used to compare rates over time, while taking into account the differing age structures (population ageing) of the population over time.

The percent change in age-standardised rates of attributable burden over time is somewhat comparable to the measure of percent change due to the amount of linked disease burden in the drivers of change estimates. However, the advantage of the drivers of change estimates is that they provide an indication of the proportionate impact of each of the specified factors, not just the change in age standardised population rates. A disadvantage of age-standardised rates is that they are only useful for the purposes of comparison with other standardised rates which have used the same reference population. Once standardised, the rates no longer reflect the actual rate observed in the population.

References

Das Gupta P 1993. Standardization and decomposition of rates: a user's manual. U.S. Bureau of the Census, Current Population Reports, Series P23-186. Washington, DC: U.S. Government Printing Office.

Li J 2017. Rate decomposition for aggregate data using Das Gupta’s method. The Stata Journal 17(2): 490–502.

Zhai T, Goss J, Li J 2017. Main drivers of health expenditure growth in China: a decomposition analysis. BMC Health Services Research 17(1): 185. doi:10.1186/s12913-017-2119-1.

Acknowledgments

The authors would like to acknowledge John Goss and Jinjing Li from the University of Canberra for providing us with an analytical tool and advice to be able to break down results over time, using the method developed by Prithwis Das Gupta.