• Open Access

Projecting the burden of diabetes in Australia – what is the size of the matter?


Correspondence to:
Dianna J. Magliano, Baker IDI Heart and Diabetes Institute, 250 Kooyong Road, Caulfield, Victoria, 3162. Fax: (03) 9258 5090; e-mail: dianna.magliano@bakeridi.edu.au


Objective: To analyse the implications of using different methods to predict diabetes prevalence for the future.

Approach: Different methods used to predict diabetes were compared and recommendations are made.

Conclusion: We recommend that all projections take a conservative approach to diabetes prevalence prediction and present a ‘base case’ using the most robust, contemporary data available. We also recommend that uncertainty analyses be included in all analyses.

Implications: Despite variation in assumptions and methodology used, all the published predictions demonstrate that diabetes is an escalating problem for Australia. We can safely assume that unless trends in diabetes incidence are reversed there will be at least 2 million Australian adults with diabetes by 2025. If obesity and diabetes incidence trends, continue upwards, and mortality continues to decline, up to 3 million people will have diabetes by 2025, with the figure closer to 3.5 million by 2033. The impact of this for Australia has not been measured.

We are amid a global diabetes epidemic.1,2 In acknowledgement of this crisis, in 2006 the United Nations General Assembly passed a resolution on diabetes recognising it as a chronic, debilitating and costly disease.3 Diabetes prevalence has increased beyond what would have been expected due solely to our ageing population.2,4 But what does this mean for the future of Australians and what can we truly expect to be the burden of diabetes in coming decades? The impact of this on future national budgets and infrastructure needs will be significant but it is impossible to accurately assess the real damage without informed projections. A number of recent predictions about the potential future burden of diabetes for Australia are based on different assumptions.5–8 For future planning, it is critical to understand the differences between these predictions. Here we examine the different methods and recommend some basic principles to increase confidence in these analyses.

Quantifying the dimensions of a public health problem such as diabetes is critical to assist governments in formulating a proportionate policy response. Having agreed measures of current and projected impact, in terms of disease burden and costs, helps to shape the resources a government is prepared to commit to address the problem. Other information that will inform the policy response includes evidence for effectiveness of action, public concern and strength of counter-arguments from key interested parties, including industry. For example, the release of the Foresight project in the UK,9 a study modelling obesity projections and policy measures provided the necessary data, and was the stimulus for the British government to commit considerable resources and policy measures to tackle the obesity problem in the UK.

Four drivers have led to increasing diabetes prevalence: the ageing population, which has resulted from increases in life expectancy due to improvements in risk factors and advances in medical treatment, and the ageing of the ‘baby boom’ generation; increases in diabetes incidence (driven primarily through increases in obesity),4 decreases in mortality in those with diabetes10,11 and immigration from countries with higher diabetes prevalence rates that may also increase diabetes prevalence.

Ideally, each of these factors should be considered when projecting the future burden of diabetes, although supporting evidence is often unavailable or severely limited. Consequently, projections of future prevalence necessarily involve assumptions. We argue that these assumptions must be as transparent and as parsimonious as possible. Both the quality of the data used and the precise question of interest must be considered when deciding on the appropriate set of assumptions. With respect to the latter point, we recommend that authors consider presenting both sensitivity and uncertainty analyses that present a primary analysis using the most accurate information that is supported by alternative scenarios where elements with greater uncertainty are varied. Uncertainty analysis provides readers with an idea of the range within which the most likely result lies.

In the past two years there have been three attempts at projecting diabetes prevalence for future Australians; each giving a different estimate of the number of adult Australians who will have diabetes by the year 2025: 1.3 million,6 2.0 million5 and 2.5 million (2023)7,8 (Table 1). A summary of the methods used for each of the approaches is given in Table 2. The methods vary according to whether trends in population figures, diabetes incidence rates and mortality rates are accounted for, and how future trends in these rates are extrapolated.

Table 1.  Summary of the different published projections of diabetes prevalence in Australia.
 YearAge-group (years)Numbers of people with diabetes (millions)Predicted prevalence of diabetes (%)
  1. Notes:

  2. a) Static diabetes incidence and mortality rates for 2005 applied until 2025.

  3. b) Incidence of diabetes rises by 4% every year5 and mortality (for those with and without diabetes) falls by 2.2% each year.13

  4. c) The prevalence of diabetes across all ages would be approximately 12.%.

  5. d) Increasing BMI trends are applied until the year 2023.

Sicree et al.6202520–791.37.7
Magliano et al.5    
 Static incidence and mortality ratesa2025≥252.011.4
 Dynamic incidence and mortality ratesb2025≥253.017.0c
Begg et al.8,d2023All ages2.510.4
Table 2.  Summary table of methods used in the prediction of diabetes prevalence. Thumbnail image of

The first approach takes the most recent diabetes prevalence data for Australia (from 1999-2000) and applies the prevalence observed for each age-group to the expected population distribution for the year 2025. Thus, as we expect there to be more elderly people in the population by 2025, and the elderly have a much higher prevalence of diabetes, the prevalence of diabetes (and the total numbers with diabetes) for the whole population rises. It assumes that the prevalence of diabetes within each age-group will not change in the next 20 years. We know that this has certainly not been the case over recent decades, with the age-adjusted prevalence of diabetes more than doubling between 1980 and 2000,2 and this is therefore a conservative estimate.

The second approach starts with the most recent prevalence data for Australia2 to provide the total number of Australians with diabetes. It then applies Australian diabetes incidence and mortality data for diabetes5,12 to keep adding to and subtracting from this number each year until 2025. In the base, and most conservative, case, current incidence and mortality data are used throughout the period. As current prevalence rates are based on historical incidence and mortality rates, this approach is a more realistic prediction of future prevalence than the first one. However, this approach is also conservative in nature as it does not apply potentially increasing diabetes incidence rates or decreasing mortality rates into the future. Consequently, the estimate of 2 million cases of diabetes by 2025 could be seen as the minimum prevalence of diabetes cases that we would expect to see, assuming no improvements in diabetes incidence rates occur in the interim. In this approach, scenario analyses were performed by applying temporal trends in mortality and diabetes incidence. By varying the incidence (based on overseas data), showing a rise in incidence over the past 20 years and mortality13 (which has consistently fallen over recent decades) between now and 2025, a less conservative, and perhaps more realistic, estimate can be derived. Changing the incidence and mortality in this way increases the estimated number of people with diabetes for 2025 from 2 to 3 million.

The third approach is more complex. It uses recent data on the prevalence and on the mortality from diabetes, it calculates the current incidence of diabetes. It then extrapolates historical trends in body mass index (BMI) and uses the established relationship between BMI and diabetes risk to modify (i.e. increase) the calculated current incidence of diabetes. This new incidence (which increases over the projected time, with the projected increase in BMI), together with decreasing mortality rates over time are applied to the prevalence of diabetes at the starting point and the prevalence for subsequent years can be calculated. This study estimated 2.5 million cases of diabetes by the year 2023.8

The third approach uses a slightly different, but not very dissimilar, population series than the first two methods. Both population series used have an identical age distribution and use the population from the 2001 Census as the initial input to derive the projected population numbers. However, the population series used by Begg et al.8 contains about 3% more people overall in the year 2023 due to slightly different assumptions. It is unlikely that this difference would have contributed to the disparity between the diabetes projections.

Each of these approaches is accompanied by substantial uncertainty in each set of critical data and without uncertainty intervals to accompany each point estimate it is not possible to know whether the projected estimates of 2.5 million7,8 and 3 million5 are indeed different. Thus there is no easy way of proving that one method is preferable to the others, short of waiting for time to pass and then measuring the prevalence in a nationally representative cross-sectional survey.

The future is inherently uncertain, with significant debate about the extent to which diabetes incidence, obesity and mortality will change in the coming years. In this setting, a conservative approach is often adopted, so as not to be accused of overstating the problem. While such a conservative approach may be adequate for raising public awareness, it may not be ideal for health planning. Thus, it may be best to be clear about the reason why we require projection of diabetes data and then we can critically evaluate the assumptions (either explicit or otherwise) of each of the approaches and comment on the likelihood that these will be borne out. It is also important to note that while these analyses and the accompanying assumptions can appear complex, in the absence of an ongoing commitment to measuring health in Australia such as occurs in the US in the NHANES surveys, such extrapolations and manipulations are unavoidable.

It is clear that any analyses aiming to indicate the general magnitude of the problem must take care not to overestimate. In contrast, if we are trying to estimate more precise numbers for service health planning purposes, it will be important to model all likely scenarios. In all these analyses, we encourage the use of scenario analysis and uncertainty analyses to provide the decision-makers with a better understanding of the accuracy and comparability of the results and how they may change under different scenarios.

In summary, all the published predictions demonstrate that diabetes is an escalating problem for Australia (as it is in most parts of the world).1 We recommend that all projections are presented as a ‘primary analysis’ using the most robust, contemporary data available and that uncertainty analyses be included. We feel we can safely assume that unless trends in diabetes incidence are reversed there will be at least 2 million Australian adults with diabetes by 2025. If, instead, obesity trends, and consequently diabetes incidence trends, continue upwards, and mortality continues to decline, the projections predict around 2.5-3 million people will have diabetes by 2025,5,7,8 with the figure closer to 3.5 million by 2033.7,8