Modelling Diabetes: A multi-state life table model

Transcription

1 Modelling Diabetes: A multi-state life table model Public Health Intelligence Occasional Bulletin No 9

2 Published in March 2002 by the Ministry of Health PO Box 5013, Wellington, New Zealand ISBN ISBN HP 3490 This document is available on the Ministry of Health s Website:

3 Contents List of Figures List of Tables Foreword Acknowledgements and Disclaimer iv iv v vi Introduction 1 Data sources and methods 2 Life table construction 2 Diagnosed or total diabetes burden 3 Analysis 4 Results 5 I Descriptive epidemiology of diabetes 5 II The burden of diabetes 18 Discussion 25 References 28 Contents iii

4 List of Figures Figure 1: Principle of the multi-state life table 3 Figure 2: Figure 3: Figure 4: Figure 5: Modelled incidence of diagnosed diabetes (onset years), by age, gender and ethnicity, Modelled prevalence of diagnosed diabetes (onset years), by age, gender and ethnicity, Population pyramids of diagnosed diabetic (onset years) and nondiabetic populations at ages 25 and above, by gender, Modelled mortality rates of diagnosed diabetes (onset years), by age, gender and ethnicity, Figure 6: Modelled 12-month case fatality rates of diagnosed diabetes (onset years), by age, gender and ethnicity, Figure 7: Figure 8: Modelled relative risk of all-cause mortality conditional on diagnosed diabetes (onset years), by age, gender and ethnicity, YLL rates (remaining life expectancy method) attributable to diagnosed diabetes (onset years) by gender, ethnicity and age, List of Tables Table 1: Table 2: Modelled incidence of diagnosed diabetes (onset years), by age, gender and ethnicity, Modelled average age at diagnosis of diabetes (onset years), by gender and ethnicity, Table 3: Modelled prevalence rates and counts of diagnosed diabetes (onset years), by age, gender and ethnicity, Table 4: Table 5: Table 6: Table 7: Table 8: Table 9: Table 10: Table 11: Table 12: Modelled average duration of diagnosed diabetes (onset years), by gender and ethnicity, Modelled mortality attributable to diagnosed diabetes (onset years), by age, gender and ethnicity, Life table risk of diagnosed diabetes (onset years), by gender and ethnicity, Multi-state life table life expectancy at birth estimates, by gender and ethnicity, Modelled diabetic and non-diabetic life expectancies at birth, by gender and ethnicity, YLL rates and counts of diagnosed diabetes (onset years), by gender and ethnicity, Ethnic group comparisons selected diabetes-related variables, by gender, Mäori and Pacific peoples observed versus expected counts of selected diabetes-related variables, by gender, Summary statistics of modelled diagnosed diabetes (onset years), by gender and ethnicity, iv Modelling Diabetes: A multi-state life table model

5 Foreword New Zealand is currently experiencing a diabesity epidemic an epidemic of obesity closely shadowed by a consequential epidemic of type 2 diabetes. These epidemics are already impacting severely on the health of New Zealanders and are contributing to the alarming health inequality between the major ethnic groups. This has been recognised by the inclusion of both diabetes and obesity, as well as nutrition and physical activity, as priority health objectives within the New Zealand Health Strategy, launched by the Minister of Health on 14 December Improving population health and reducing inequalities in health between ethnic groups will require greater effort to prevent diabetes and to provide effective care for people with diabetes, in order to delay the onset or progression of micro- and macrovascular complications. Such efforts need to be guided by analyses of the current and future burden of this disease. Another publication in the Occasional Bulletin series, Modelling Diabetes: The mortality burden, provides estimates of the current burden imposed on society by this disease with particular reference to its contribution to premature mortality. This work has now been extended through the construction of a multi-state life table model which yields estimates of diabetes incidence in addition to mortality and prevalence. Estimates of incidence are needed to plan diabetes services and allocate resources. The model also provides the necessary foundation for preparing forecasts of future diabetes burden under different policy options. Such forecasts are reported in another bulletin in this series, Modelling Diabetes: Forecasts to The model is also available to researchers, policy advisors, community groups, funders and providers to investigate scenarios of interest. Comments on this report are welcomed, and should be sent to Public Health Intelligence, Ministry of Health, PO Box 5013, Wellington or martin_tobias@moh.govt.nz. Don Matheson Deputy Director-General Public Health Directorate Foreword v

6 Acknowledgements The model was built and the report prepared by Martin Tobias and Jit Cheung (Public Health Intelligence, Ministry of Health). The authors acknowledge the assistance of Mick Roberts (AgResearch) with an earlier version of the model. The report was peer reviewed both within the Ministry of Health and by experts outside the Ministry in New Zealand, Australia and the Netherlands. Disclaimer Opinions expressed in the report are those of the authors and do not necessarily reflect the views of peer reviewers or those of the Ministry of Health. The Ministry of Health accepts no liability for decisions or actions based on the content of this report. Note on terminology This report uses the term diabetics rather than the more cumbersome people living with diabetes solely for ease of description. In general, the latter term is preferred to avoid any suggestion of labelling people. vi Modelling Diabetes: A multi-state life table model

7 Introduction The prevalence of self-reported diabetes among adult New Zealanders, most of which is believed to represent type 2 disease (insulin resistance leading to relative rather than absolute insulin deficiency), has been rising steadily since the 1980s (Simmons 1996). The main albeit not exclusive modifiable driver of this epidemic is the rising trend in body fat mass, especially abdominal fat, that has characterised the New Zealand population since the 1970s (Wilson et al 2001), itself the result of changing dietary and physical activity patterns. The potential impact of these social trends in diet and physical activity on the health of the affected populations is severe, and a major pathway linking these risk factors to health outcomes is through overweight and obesity leading to insulin resistance, which, given genetic susceptibility, may in turn lead to type 2 diabetes with its serious micro- and macrovascular complications (Prentice 2001). Another report in this series (Ministry of Health 2002a) provides estimates of the current mortality burden of diabetes, based on an observed prevalence life table model. We now extend this to a multi-state life table model, in order to obtain estimates of diabetes incidence. This type of model is based on the mathematical relationships between incidence, prevalence, duration and mortality, and so ensures internal consistency between the modelled estimates of these epidemiological variables. The result is a comprehensive and internally consistent description of the epidemiology of diabetes. This description includes estimates of current incidence, life table risk, prevalence, mortality (including premature mortality or years of life lost, and life table functions such as life expectancy at different ages), case fatality, duration and age of onset all indexed by age, gender and ethnicity. In this way a picture of the demography of the diabetic population is built up including its size, age structure, gender ratio and ethnic composition. The model also enables inequalities in the burden of diabetes experienced by the major ethnic groups in New Zealand to be quantified. Yet the model is limited in that: only diagnosed diabetes with onset in the age range years is included; the burden of undiagnosed diabetes and of prediabetic states (impaired glucose tolerance (IGT) and impaired fasting glucose (IFG)) is excluded while (diagnosed) diabetes with onset in this age range is mainly type 2, a small proportion is type 1 the increasing incidence of type 2 diabetes in childhood and adolescence is not captured by the model. Provided these model restrictions are understood, the model can be applied to help answer a variety of policy questions. Policy applications of the model include forecasting future diabetes burden and its distribution, given projected demographic, epidemiological and health care trends. Such forecasts are reported in a separate bulletin in this series (Ministry of Health 2002b). Introduction 1

8 Data Sources and Methods Life table construction Six multi-state life tables were constructed (Mäori, Pacific and European/Other ethnic groups; male and female) using the following data: number of deaths by five-year age group averaged for (European ethnic group) or for (Mäori and Pacific ethnic group) (source: Statistics New Zealand) number of people by five-year age group estimated for mid-1996 (source: Statistics New Zealand) prevalence rates of self-reported clinically diagnosed diabetes by five-year age group for ages 25 years and above, averaged for (sources: the 1996/97 New Zealand Health Survey [Ministry of Health 1999] and (for the Pacific ethnic group only) the 1992/95 South Auckland Diabetes Household Survey [Simmons et al 1999]) estimates of the relative risk of all cause mortality for diabetics compared with nondiabetics (RR) by five-year age group (source: systematic review of 10 published cohort studies, mainly European and North American see Ministry of Health 2002a for references). Zero remission from the state of diagnosed diabetes was assumed. The model was restricted to the age range years. Ages under 25 years were excluded because of the difficulty in distinguishing type 2 from type 1 diabetes at younger ages. Ages over 90 years were excluded because of lack of reliable prevalence or RR data at older ages. The age threshold of 25 years is intended to provide a rough proxy for type 2 diabetes, although this fails to capture the small but increasing proportion of type 2 diabetes with onset in childhood or adolescence (American Diabetes Association 2000). Full details of the data sources used, and references to published studies, are provided elsewhere (Ministry of Health 2002a). The mathematics underlying multi-state life table models is also described elsewhere (Roberts and Tobias 2001), but the principle involved may be simply illustrated, as shown in Figure 1. 2 Modelling Diabetes: A multi-state life table model

9 Figure 1: Principle of the multi-state life table Susceptible i m r m Dead p Diabetic a Key: i = incidence rate r = remission rate (zero for diabetes) m = other cause mortality rate a = diabetes-attributable mortality rate p = prevalence Note: age, gender and ethnicity indexes suppressed. The life table was constructed by first running the model in the backward calculation mode, to estimate incidence from the empirical prevalence and RR data. The initial incidence rates were then smoothed by cubic spline interpolation, and the model run in the forward calculation mode to estimate amended prevalence and mortality rates. This process was then iterated until the model output stabilised. Diagnosed or total diabetes burden Rather than modelling only diagnosed diabetes (as is done here), assumptions could be made about the prevalence of undiagnosed diabetes (for example, that the ratio of diagnosed to undiagnosed diabetics is 1:1), and about the relative risk of mortality in prediabetic states such as IGT (for example, that this is half that of diabetics), so allowing the total burden of diabetes to be modelled. Uncertainty introduced by these assumptions could be quantified by sensitivity analysis. However, we prefer empirical to assumption-driven modelling, and so choose to model only diagnosed diabetes (for which we have empirical data). This leaves users free to apply whatever multipliers they wish to the modelled estimates for diagnosed diabetes, in order to obtain an indication of the total diabetes burden (including prediabetic states if desired). It needs to be emphasised that the estimates in this report refer only to diagnosed diabetes (and, furthermore, only to diabetes with onset in the age range, as a proxy for type 2 diabetes). Data Sources and Methods 3

10 Analysis Direct inputs or outputs of the model include incidence, prevalence, diabetes-attributable and other-cause mortality, case fatality, relative risk of mortality conditional on diabetes, age of onset of diabetes and duration of diabetes, all by five-year age group, gender and ethnicity. The multi-state life table was also used to calculate the life table risk of diabetes; this is considered to be a more useful statistic than the more commonly estimated cumulative (lifetime) incidence, as it takes mortality into account. Separate life tables for diabetics and non-diabetics by gender and ethnic group were also constructed (using the model output), complementing the set of multi-state life tables. These separate life tables provide estimates of the burden of diabetes on the diabetic subpopulation, rather than on the whole population (comprising both diabetics and non-diabetics). The burden of diabetes is also estimated in units of YLL (years of life lost) to provide an estimate of premature mortality. YLL were estimated in two ways: by the arbitrary upper age limit method, using 75 years as the cut-off by the remaining life expectancy method, using model life table West (level 26 for females and level 25 for males) as the standard for consistency with international practice (Murray and Lopez 1996). YLL have been discounted at a rate of 3% per year (Gold et al 1996). YLL rates have been age standardised by the direct method, using the WHO standard world population as the reference (World Health Organization 2000). DALYs (disability-adjusted life years) for diabetes could also be calculated by combining YLL and YLD (years of life lost to disability, adjusted for severity). YLD could be calculated by applying average disability weights to the incidence and duration estimates from the model. However, in the absence of robust estimates for disability weights, DALY calculations are not presented here. Ethnic differentials are analysed mainly by comparing incidence and case fatality rates. This enables the influence of ethnicity to be seen on both the risk of developing diabetes in the first place, as well as the risk among diabetics (of differing ethnicity) of dying from their disease. Excess Mäori and Pacific incidence and mortality from diabetes are quantified by applying the European rates for these variables to the respective ethnic populations. The differences between these expected counts and the observed counts output directly by the model are a measure of excess burden. 4 Modelling Diabetes: A multi-state life table model

11 Results The results are reported in two sections. The first section provides a complete epidemiological description of (diagnosed, mainly type 2) diabetes, representing the direct output of the model. The second section analyses the population-level health impact of diabetes and, particularly, the excess burden borne by ethnic groups. I Descriptive epidemiology of diabetes Incidence Estimated incidence rates and number of new diagnoses (for 1996) are summarised in Table 1. Incidence rates by five-year age group for each gender and ethnic population are illustrated graphically in Figure 2. Table 1: Modelled incidence of diagnosed diabetes (onset years), by age, gender and ethnicity, 1996 Incidence rates (per 100,000 person years) Ethnicity Age std RR Mäori Pacific European Total Mäori Pacific European Total Notes: Age standardisation is to the WHO World population age 25+ years. RR is the ratio of the ethnic group rate to the European rate. Results 5

12 Counts (number of new diagnoses) Ethnicity Mäori Pacific European Total Mäori Pacific European Total Figure 2: Modelled incidence of diagnosed diabetes (onset years), by age, gender and ethnicity, Incidence rate (per 100,000) Mäori Pacific European Modelling Diabetes: A multi-state life table model

13 Incidence rate (per 100,000) Mäori Pacific European The incidence of diagnosed (mainly type 2) diabetes across all adult ages for the whole population in 1996 is estimated to be approximately 220 per 100,000 person-years for males and 196 per 100,000 for females. Standardising for age, the incidence rate among Mäori is 3.5 (males) and 5.2 (females) times that among Europeans. The corresponding rate ratios for Pacific people are 2.9 and 4.1 respectively. Incidence rates of diagnosed diabetes (with onset in the age range years) start to accelerate exponentially from young adulthood through to middle age, and peak earlier for Pacific peoples at ages than Mäori (around age 60) and Europeans (around age 65). At their respective peaks, the Mäori incidence rate of over 1100 per 100,000 person-years (genders pooled) is more than three times that of Europeans. At older ages, incidence rates tail off, and by ages the rates have returned to the levels attained at about years; this apparent decline may represent an age-cohort effect, a selection effect, data artefacts, or a combination of these. Among Mäori and Pacific peoples, females exhibit higher incidence rates than their male counterparts in nearly all age groups, particularly at the peak, where female rates are 7 9% higher than the corresponding rates for males. In contrast, the opposite gender pattern is demonstrated among Europeans, with European female incidence rates being consistently lower than those of European males at all ages. Translating incidence rates into absolute counts shows that there were an estimated 4742 new diagnoses of diabetes among people aged in Despite the higher risk experienced by Mäori and Pacific peoples, most new cases (65%) are European. The European share of new cases increases with age to 82% at ages 65 and above, reflecting this ethnic group s large population size and older age structure. Although true underlying incidence rates tend to increase with age, diabetes is most often recognised initially in middle age, as illustrated in Figure 2 and Table 1. In the mid- to late 1990s (ethnic groups pooled), 49% of new diagnoses involved middle-aged adults, 29% young adults and only 22% older people. Just over half (52%) of new diagnoses were male. Results 7

14 Age at diagnosis Estimated age at diagnosis of diabetes (with onset in the age range years) within any fiveyear age group is an output directly extractable from the model. The average age at diagnosis for each gender and ethnic group, averaged across all age groups, is shown in Table 2. This statistic reflects the interaction within each group of its age-specific incidence rates and population age structure. Mäori and Pacific peoples have almost identical average ages at diagnosis approximately 48 years for males and 47 years for females. The average age at diagnosis is approximately six years older for Europeans. Despite this, as will be shown below, European longer life expectancy (and lower case fatality) means that European diabetics live longer after diagnosis on average than do Mäori or Pacific diabetics. Table 2: Modelled average age at diagnosis of diabetes (onset years), by gender and ethnicity, 1996 Average age at diagnosis Mäori Pacific European Prevalence Prevalence rates of (diagnosed, mainly type 2) diabetes are both an input 1 and an output of the model. The output prevalence estimates differ from the input data used to initiate the modelling process because the values are changed by running the model iteratively until all outputs have stabilised. Modelled prevalence estimates are plotted in Figure 3 and summarised in Table 3. 1 Input prevalence data are provided in Ministry of Health 2002a, (Appendix 1). 8 Modelling Diabetes: A multi-state life table model

15 Figure 3: Modelled prevalence of diagnosed diabetes (onset years), by age, gender and ethnicity, Prevalence rate (per 100) Mäori Pacific European Prevalence rate (per 100) Mäori Pacific European Results 9

16 Table 3: Modelled prevalence rates and counts of diagnosed diabetes (onset years), by age, gender and ethnicity, 1996 Prevalence rates (per 100) Ethnicity Age std RR Mäori Pacific European Total Mäori Pacific European Total Notes: Age standardisation is to the WHO World population. RR is the ratio of the ethnic group rate to the European rate. Counts (number of existing cases) Ethnicity Mäori Pacific European ,774 13,306 31,790 Total ,644 14,858 40,807 Mäori Pacific European ,280 13,701 28,707 Total ,390 16,096 40,684 The age trajectory of (diagnosed) diabetes prevalence mirrors that of incidence (see Figure 2), accelerating through the young adult and middle ages and peaking at older ages. For Mäori and Pacific peoples, prevalence peaks at ages 65 69, some five to 10 years after incidence rates have peaked. For Europeans, the lag between incidence and prevalence peaks is even longer, at around 15 years, with prevalence peaking at ages The decreasing trend of prevalence at older ages is a direct reflection of the empirical data, and may be exaggerated (a decline of this extent was not seen in a recent major Australian study [Dunstan et al 2001]). This downward movement in prevalence also drives the decreasing trend in the modelled incidence rates at older ages. 10 Modelling Diabetes: A multi-state life table model

17 At their respective peaks, Mäori prevalence of around 19% (genders pooled) is over 2.5 times that of Europeans. Prevalence rates for Pacific peoples are slightly lower than those for Mäori, given the available survey data. However, the rates for Pacific peoples are uncertain because of small numbers in the surveys, and recent data (Bell et al 2001) suggests that prevalence may have been under-estimated for this ethnic group. Gender differentials in prevalence follow the same pattern as those in incidence, with Mäori and Pacific males having lower prevalence rates than their female counterparts, but European males having higher rates than European females. Demography of the diabetic population The size of the diagnosed diabetic population in 1996, as estimated by the model, is also shown in Table 3. In 1996 the model estimates that there were 81,491 people aged with diagnosed (mainly type 2) diabetes, with the number split almost evenly between the two genders slightly less than the estimate of approximately 87,000 obtained previously using an observed prevalence life table (Ministry of Health 2002a). Mäori were considerably over-represented in the diabetic population, at 20% (genders pooled) compared to less than 14% in the overall population aged 25 years and over. In contrast, Europeans were under-represented in the diabetic population, at 74% compared to 81% in the overall population aged 25 and over. For Pacific peoples the proportions were 6% and 5% respectively. The diabetic population has a much older age structure than the non-diabetic population, resulting from the highly age-dependent incidence rates, as noted earlier. The contrasting age structures of the diabetic and non-diabetic populations at age 25 and above are illustrated using population pyramids in Figure 4. The structure of the (diagnosed) diabetic population resembles an inverse pyramid with proportionately more people at older ages than younger ages. Around two-thirds (64%) of the diabetic population are aged between 45 and 75, whereas the corresponding estimate for the nondiabetic population is 42%. For diabetic males, the age distribution is further concentrated at ages 60 69, and for diabetic females at ages These age ranges capture roughly a quarter of the diagnosed diabetic population, double the corresponding level of 12% for the non-diabetic population. Results 11

18 Figure 4: Diabetic population Population pyramids of diagnosed diabetic (onset years) and non-diabetic populations at ages 25 and above, by gender, 1996 Non-diabetic population % 10% 5% 0% -5% -10% -15% Percentage of population 15% 10% 5% 0% -5% -10% -15% Percentage of population Duration The average duration of diagnosed diabetes (onset years) is measured as the average time from diagnosis to death from all causes. Estimated duration for each five-year age group is an output directly extractable from the model. The average duration for each gender and ethnic group, averaged across all age groups, is shown in Table 4. The average duration of diabetes is determined by the average age at diagnosis (which in turn reflects the interaction of age-specific incidence rates and population age structure as noted earlier), and the age-dependent mortality of diabetics (that is, age-specific diabetes-attributable mortality and other-cause mortality rates). For both genders, the average duration of diabetes is least for Mäori (at approximately 18 years), intermediate for Pacific peoples (approximately 20 years) and longest for Europeans (approximately 23 years). The higher mortality of Mäori and Pacific diabetics more than offsets their earlier onset ages (compared to Europeans), resulting in shorter disease durations for these ethnic groups. Table 4: Modelled average duration of diagnosed diabetes (onset years), by gender and ethnicity, 1996 Average duration (years) Mäori Pacific European Modelling Diabetes: A multi-state life table model

19 Diabetes-attributable mortality Mortality rates and counts estimated from the model are summarised in Table 5 and illustrated graphically in Figure 5. These are estimates of deaths attributable to diagnosed diabetes (with onset in the age range) regardless of their ICD codes. The denominator for the rates is the whole population, including both diabetics and non-diabetics. Table 5: Modelled mortality attributable to diagnosed diabetes (onset years), by age, gender and ethnicity, 1996 Diabetes-attributable mortality rates (per 100,000 person years) Ethnicity Age std. RR (mortality) RR (case fatality) Mäori Pacific European Total Mäori Pacific European Total Notes: Age standardisation is to the WHO World population. RR is the ratio of the ethnic group rate to the European rate. Counts (number of diabetes-related deaths) Ethnicity Mäori Pacific European Total Mäori Pacific European Total Results 13

20 Figure 5: Modelled mortality rates of diagnosed diabetes (onset years), by age, gender and ethnicity, Diabetes-related mortality rate (per 100,000) Mäori Pacific European Diabetes-related mortality rate (per 100,000) Mäori Pacific European In total, 1494 deaths occurring in 1996 are estimated to be attributable to diagnosed diabetes, accounting for 5.3% of all deaths aged registered in that year. However, for Mäori the proportion of deaths attributable to diagnosed diabetes is 19.5% (512 out of 2629 recorded deaths) and for Pacific peoples it is 15.9% (117 out of 735), compared with only 3.5% for Europeans (865 out of 25,013). Only 59% of the diabetes-attributable deaths occur in older people, aggregating across all ethnic groups. As with incidence, 53% of these deaths are male. 14 Modelling Diabetes: A multi-state life table model

21 Comparing age-standardised rates, while Mäori males are 3.5 times more likely than European males to develop diabetes (see Table 1), they are 7.4 times more likely to die from this condition (that is, the former group experiences a diabetes case fatality rate approximately 2.1 times (7.4/3.5) that of the latter group). For Mäori females the relative risks compared to European females are 5.2 and 13.0 for incidence and mortality respectively. Similarly, Pacific males and females have higher relative risks for mortality (5.3 and 8.3 respectively) than for incidence (2.9 and 4.1 respectively). The age trajectory of diabetes-attributable mortality follows that for case fatality (see below), increasing exponentially at middle adult ages and peaking at around ages The diabetesattributable mortality rate is itself a function of the age-specific prevalence rate, case fatality rate (see below) and mortality rate (all cause) for the whole population (comprising both diabetics and non-diabetics). For example, the declining diabetes-attributable mortality rate at advanced ages is likely to be at least in part the result of competing causes of mortality at ages where the overall risk of mortality is very high (other explanations include a cohort effect and bias in the underlying data used to run the model). Case fatality Case fatality is defined here as the probability of death from diabetes-attributable causes in the 12 months between ages x and x+1 years among those with diagnosed diabetes at exact age x years (that is, we model only the 12-month case fatality rate). The age trajectory of case fatality (Figure 6) shares many characteristics with other epidemiological variables described previously. The rate increases rapidly through the middle ages and peaks at older ages. For both genders, case fatality is highest for Mäori, intermediate for Pacific peoples and lowest for Europeans. Among diabetics, the risk of dying from diabetes is highest between ages 70 and 74 for Mäori and Pacific peoples, but peaks at the slightly older age of years for Europeans. The peak case fatality for Mäori of around 0.30 (genders pooled) is over 20% higher than that for Pacific peoples and 2.5 times that for Europeans. The decline in case fatality at extreme old ages is likely to be the result of competing causes of mortality. In contrast to the gender patterns in incidence and prevalence, for all three ethnic groups males have higher case fatality rates than females at all ages. At its peak the male excess is 11% for Mäori, 16% for Pacific peoples and 24% for Europeans. That is, while Mäori and Pacific females have a greater risk of being diagnosed with diabetes than their male counterparts, their chances of surviving are also higher. European males, however, not only have higher risks of developing diabetes but also higher risks of dying from diabetes compared to their female counterparts. Results 15

22 Figure 6: Modelled 12-month case fatality rates of diagnosed diabetes (onset years), by age, gender and ethnicity, Case fatality Mäori Pacific European Case fatality Mäori Pacific European Modelling Diabetes: A multi-state life table model

23 Relative risk of mortality Relative risks of dying conditional on being diagnosed with diabetes are both inputs 2 and outputs of the model, as is the case for prevalence. For the same reason as applies to prevalence, the final relative risk (RR) estimates differ from the empirical (input) data used for initiating the iterative modelling process. The higher relative risks experienced by Mäori and Pacific peoples, and the differences in age dependency of these risks compared with those experienced by Europeans, are illustrated in Figure 7. For Europeans, RR peaks at ages at a level of approximately 2.5 for both males and females, whereas for Mäori and Pacific peoples the peak occurs earlier at around age 40 or younger and at a higher level of 3.7 to 4.1. At older ages RR returns to approximately 1.5 or below in all gender and ethnic groups. Figure 7: Modelled relative risk of all-cause mortality conditional on diagnosed diabetes (onset years), by age, gender and ethnicity, Relative risk Mäori Pacific European Input RR and prevalence data are provided in Ministry of Health 2002a (Appendix 1). Results 17

24 Relative risk Mäori Pacific European II The burden of diabetes Life table risk of diabetes The life table risk of diabetes is a summary measure defined here as the probability of being diagnosed with diabetes (with onset in the age range years) during the typical individual s lifetime. This statistic takes into account the lifetime risks of mortality, both from diabetesattributable and other causes; that is, it is the complement of the probability that a person is never diagnosed with diabetes before his or her death. This feature makes the life table risk a more realistic measure of risk than the conventional cumulative incidence of diabetes, which ignores mortality (and so overestimates the true risk). Which is to say, cumulative incidence measures what the probability of developing diabetes over the specified interval would be in the absence of death an unrealistic scenario except over short intervals at young ages. The life table risk from birth as derived from the model output is shown in Table 6. The probability in 1996 that a person of European ethnicity will develop diabetes during their lifetime is estimated to be approximately 9% (genders pooled). The corresponding lifetime risk for Mäori and Pacific peoples is approximately 30% and 25% respectively (genders pooled), more than 2.5 times the risk of Europeans. 18 Modelling Diabetes: A multi-state life table model

25 Table 6: Life table risk of diagnosed diabetes (onset years), by gender and ethnicity, 1996 Mäori Pacific European While Mäori and Pacific females have an approximately 20% greater risk of developing diabetes during their lifetimes than their male counterparts, European females have about a 20% lower risk than European males. This contrasting ethnic gender pattern is consistent with those observed in the modelled incidence and prevalence rates. Life expectancy Life expectancy estimates at birth derived from the multi-state life tables are summarised in Table 7. Despite the different methods used, these results agree reasonably closely with those reported previously using an observed prevalence instead of a multi-state life table approach (Ministry of Health 2002a). Table 7: Multi-state life table life expectancy at birth estimates, by gender and ethnicity, Mäori Pacific European Mäori Pacific European LE LE 0 without diabetes LE 0 with diabetes (LE 0 d ) LE 0 d / LE 0 (%) 6.4% 5.7% 2.9% 8.4% 7.5% 2.6% Note: estimates relate only to diagnosed diabetes with onset in the age range years. LE 0 = life expectancy at birth. = expectation at birth of years lived with diabetes. LE 0 d In 1996 European males could expect to live on average 75.3 years, of which 73.1 years are expected to be free of diabetes and 2.2 years (or 2.9% of the lifetime) are years spent living with (diagnosed) diabetes. Mäori and Pacific males can expect to live shorter lives than their European counterparts, but spend longer times (4.3 years and 4.0 years respectively) and a greater proportion of their shorter lives (6.4% and 5.7% respectively) living with (diagnosed) diabetes. In terms of both the absolute number of years lived with diagnosed diabetes and as a proportion of their respective life expectancies, Mäori and Pacific males experience double the burden of their European counterparts. Of all gender and ethnic groups, European females in 1996 could expect to live the longest life of 78.5 years free of diabetes and the shortest duration of 2.1 years with diagnosed diabetes. The corresponding estimates for Mäori and Pacific females are 71.6 years and 75.6 years free of (diagnosed) diabetes and 6.0 years and 5.7 years with diabetes respectively. As a percentage of their respective life expectancies at birth, time spent with diagnosed diabetes for Mäori females (8.4%) is over three times that for European females (2.6%). Results 19

26 Life expectancy of diabetics and non-diabetics While the above analysis spreads the impact of diabetes across the whole population (comprising both diabetics and non-diabetics), this section reports the results derived from constructing separate life tables one for the diabetic population (the diabetic life table) and one for the non-diabetic population (the non-diabetic life table) (Table 8). Table 8: Modelled diabetic and non-diabetic life expectancies at birth, by gender and ethnicity, Mäori Pacific European Mäori Pacific European LE LE 0 dia LE 0 non-dia LE 0 non-dia LE 0 dia LE 0 non-dia LE Notes: LE 0 = life expectancy at birth of the whole population; LE non-dia 0 = life expectancy at birth of the non-diabetic population; LE dia 0 = life expectancy at birth of diabetic population. In 1996 European male non-diabetics could expect to live on average 75.7 years, about 6.5 years and 4.2 years longer than their Mäori and Pacific counterparts respectively. By contrast, European male diabetics can expect to live for only 69.0 years, yet this is 11.6 years and 9.3 years longer than their Mäori and Pacific counterparts. Similar ethnic patterns are observed among female diabetics and non-diabetics. European female non-diabetics can expect to live 80.9 years, 6.7 years and 3.2 years longer than their Mäori and Pacific counterparts. However, European Mäori differentials are noticeably larger among female diabetics: European female diabetics life expectancy at birth of 74.1 years is 12.5 years longer than that for Mäori female diabetics, and is almost the same as the non-diabetic life expectancy for Mäori females. The impact of diabetes on diabetics can be quantified by the difference between the non-diabetic and diabetic life expectancies. The model estimates that diabetes reduces the life expectancy at birth of European diabetics by almost seven years, compared to approximately 12 years for Mäori or Pacific diabetics. Similarly, the impact of diabetes on the whole population (both diabetics and non-diabetics included) can be summarised by the difference between non-diabetic and ordinary life expectancies. If diabetes were eliminated, European life expectancy at birth would increase by approximately 0.4 years (genders pooled), whereas the gain for Mäori and Pacific peoples would be 2.3 years and 1.9 years respectively. Thus eliminating diabetes would close the gap between European and Mäori or Pacific life expectancies at birth by approximately two years and 1.5 years respectively. This would represent a narrowing of the life expectancy gap (relative to Europeans) by about 25% for both ethnic groups. The estimate would be even greater if undiagnosed type 2 diabetes and prediabetic states (IGT and IFG) had been included in the model. 20 Modelling Diabetes: A multi-state life table model

27 Years of life lost (YLL) The burden of fatal outcomes of diabetes may be denominated in years of life lost (YLL) to capture the prematurity of deaths attributable to diabetes. YLL can be calculated using an arbitrary upper age limit or the life expectancy remaining at each age. Both methods are reported here, with age 75 set as the upper age limit for the first method, and model life table West level 26 (females) and level 25 (males) used as the standard life table for the second method (Murray and Lopez 1996). YLL have also been discounted to net present value at a rate of 3% per year, in keeping with conventional practice (Gold et al 1996). YLL counts and rates are summarised in Table 9. YLL rates by five-year age group using model life tables are illustrated graphically in Figure 8. YLL rates with age 75 as the cut-off point follow similar age patterns to those calculated by the remaining life expectancy method except for being more compressed into the age range, and also show the same gender and ethnic patterns (data not shown). Table 9: YLL rates and counts of diagnosed diabetes (onset years), by gender and ethnicity, 1996 Mäori Pacific European Mäori Pacific European Number of YLL Age-std. YLL 75 rate (per 1000) Number of YLL west Age-std. YLL west rate (per 1000) Notes: Age standardisation is to the WHO World population age 25+. YLL are discounted at a rate of 3% per year. Results 21

28 Figure 8: YLL rates (remaining life expectancy method) attributable to diagnosed diabetes (onset years) by gender, ethnicity and age, YLL rate (per 1000) Mäori Pacific European Notes: Reference life table is West level 25; YLL are discounted at a rate of 3% per year YLL rate (per 1000) Mäori Pacific European Notes: Reference life table is West level 26; YLL are discounted at a rate of 3% per year. The markedly greater burden of premature mortality imposed by diabetes on Mäori and Pacific peoples, compared to Europeans, reflects the interaction of the former groups higher incidence and case fatality of diabetes, the shift in incidence and fatality to younger age groups, and the younger age structure of these populations. 22 Modelling Diabetes: A multi-state life table model

29 Gap analysis The differential burden of diabetes (diagnosed, mainly type 2) on Mäori and Pacific peoples is summarised in Table 10. This table shows both absolute (rate difference) and relative (rate ratio) comparisons between these ethnic groups and the European/Other ethnic group. Table 10: Ethnic group comparisons selected diabetes-related variables, by gender, 1996 Mäori European Pacific European Mäori European Pacific European Rate diff. Rate ratio Rate diff. Rate ratio Rate diff. Rate ratio Rate diff. Rate ratio Incidence rate, (per 100,000) Prevalence rate, (per 100) Diabetes-attributable mortality, (per 100,000) Life table risk of diabetes % of LE 0 spent with diabetes LE 0 non-dia LE 0 dia (years) LE 0 non-dia LE 0 (years) YLL west rate (per 1000) Notes: All rates are age standardised to the WHO World population; comparisons are all based on the European ethnic group as the reference. The excess burden borne by Mäori and Pacific peoples appears to vary depending on which variable is examined. Much of this variation results from there being two types of statistics included in the table: those relating to the whole population (for example, the YLL rate ratio), and those relating directly to the diabetic subpopulation (for example, the life table risk). The former have higher values for excess burden because they conflate the variable of interest with the higher prevalence of diabetes among Mäori or Pacific ethnic groups. The information summarised in Table 10 can be used to calculate excess counts for Mäori and Pacific ethnic groups, by subtracting from the modelled ( observed ) count (for example, of new diagnoses or deaths) the expected count (obtained by applying the European rate for the variable of interest to the Mäori or Pacific population) (Table 11). Results 23

30 Table 11: Mäori and Pacific peoples observed versus expected counts of selected diabetesrelated variables, by gender, 1996 Mäori Pacific Observed (1) Expected (2) Excess (1) (2) Observed (1) Expected (2) Excess (1) (2) New diagnoses Existing diagnoses Deaths YLL west New diagnoses Existing diagnoses Deaths YLL west Notes: The number of new diagnoses, existing diagnoses and deaths are for the population aged years. Expected values are calculated by applying the appropriate European rate to the respective Mäori or Pacific ethnic group populations. Observed values are those output directly by the model for each ethnic group. 24 Modelling Diabetes: A multi-state life table model

31 Discussion To accurately measure the incidence, prevalence, and mortality of type 2 diabetes in New Zealand, a large cohort study would be necessary. Such a study would be prohibitively expensive, take decades to yield useful data, and is unlikely ever to be done. Instead, epidemiological modelling can be used to derive reasonably robust estimates of these key variables, using only the cross-sectional diabetes prevalence data available from surveys such as the New Zealand Health Survey, data about the excess mortality experienced by diabetics (irrespective of coded cause), and theory about the fundamental relationship between incidence, prevalence and mortality (the multi-state life table model). The model uses total rather than cause-specific mortality, so avoiding data quality problems in cause of death coding for diabetes-related deaths. However, this method requires knowledge of the excess mortality (all-cause) risk of diabetics, disaggregated by age, gender and ethnicity. Yet such information is not available from New Zealand studies, and the assumption had to be made that relative risks estimated from the international literature could be used, although such estimates are derived from different populations and even from different periods. This assumption is clearly a limitation of the method (especially for Mäori and Pacific ethnic groups, for whom higher relative risks were assumed to apply), although sensitivity analysis could be done to gauge the extent of uncertainty in the epidemiological estimates so obtained. Given these assumptions, the model has provided us with a consistent set of estimates for the incidence, prevalence and mortality of diabetes (diagnosed, mainly type 2) by age, gender and ethnicity for New Zealand in The incidence rates can be compared with the few empirical reports for similar populations available in the literature, and are reasonably consistent with these (Blanchard et al 1996; Garancini et al 1996). The estimates of incidence in old age are strongly influenced by the empirical prevalence data, which at least in New Zealand shows a decline at advanced ages. Such a decline was not found to a similar extent in a recent Australian health examination survey (Dunstan et al 2001). The New Zealand data may reflect a strong cohort effect, a lower proportion of diagnosed to undiagnosed diabetes in older cohorts, or a lesser propensity to report diabetes by older respondents (perhaps because of comorbidities, or lesser familiarity with questionnaires). Also, New Zealand survey sample frames typically exclude the institutionalised population (who are older on average). The model as currently configured may therefore underestimate incidence (and prevalence) at older ages. The model output is derived from a period (multi-state) life table and so will not reflect cohort effects. If, for example, incidence has recently increased rapidly, it will take several years for actual prevalence to reach the new higher equilibrium value, but this happens instantly in the context of the model life table population. Some divergence between modelled and observed rates is therefore to be expected. In this respect, the observed prevalence life table method (Ministry of Health 2002a) can be expected to yield better estimates of current burden. Nevertheless, this freedom from distortion by historical cohort effects can be seen as an advantage of the multi-state life table approach, in that the approach is forward looking and hence more useful for examining future scenarios. Another strength of the modelled epidemiological estimates reported here is that they are internally consistent with one another; it is this property that makes these estimates suitable as a basis for forecasting. Discussion 25