Lifetime risk of developing diabetes and years of life lost among those with diabetes in Brazil

Background Given the paucity of studies for low- or middle-income countries, we aim to provide the first ever estimations of lifetime risk of diabetes, years of life spent and lost among those with diabetes for Brazilians. Estimates of Brazil´s diabetes burden consist essentially of reports of diabetes prevalence from national surveys and mortality data. However, these additional metrics are at times more meaningful ways to characterize this burden. Methods We joined data on incidence of physician-diagnosed diabetes from the Brazilian risk factor surveillance system, all-cause mortality from national statistics, and diabetes mortality rate ratios from ELSA-Brasil, an ongoing cohort study. To calculate lifetime risk of developing diabetes, we applied an illness-death state model. To calculate years of life lost for those with diabetes and years lived with the disease, we additionally calculated the mortality rates for those with diabetes. Results A 35-year-old white adult had a 23.4% (95% CI = 22.5%-25.5%) lifetime risk of developing diabetes by age 80 while a same-aged black/brown adult had a 30.8% risk (95% confidence interval (CI) = 29.6%-33.2%). Men diagnosed with diabetes at age 35 would live 32.9 (95% CI = 32.4-33.2) years with diabetes and lose 5.5 (95% CI = 5.1-6.1) years of life. Similarly-aged women would live 38.8 (95% CI = 38.3-38.9) years with diabetes and lose 2.1 (95% CI = 1.9-2.6) years of life. Conclusions Assuming maintenance of current rates, one-quarter of young Brazilians will develop diabetes over their lifetimes, with this number reaching almost one-third among young, black/brown women. Those developing diabetes will suffer a decrease in life expectancy and will generate a considerable cost in terms of medical care.

Then, using a similar approach we estimate in each self-reported ethnicity group the mortality of those with and without diabetes: 3. We estimate the mortality rate ratio (MRR), comparing those with vs. without diabetes, applying Cox regression to ELSA-Brasil cohort data adjusting for body mass index (BMI), waist circumference, schooling, hypertension, income and smoking status, and including interactions terms for age, diabetes, sex and ethnicity.
For each ethnicity, sex and age: 4. We combine the overall mortality rates previously calculated (either or ) with the corresponding diabetes prevalence (P) and MRR, to estimate the mortality rate of those with (M1) and without (M0) diabetes using similar formulas as described before.

Lifetime risk and Years of Life Lost
Cumulative lifetime risk of diabetes, years of life lost and the expected years lived with diabetes were calculated with the Illness-death model [3,4]: Whiting each ethnicity group, for each sex s and age a: Lifetime risk of developing diabetes, before age a, and conditional to surviving until a, is obtained through the following expression: Years of life lost is obtained comparing the life expectancy between individuals of same age a, that either develop diabetes (S1) or remain disease free (S0).
The expected years lived with diabetes (S1), from a certain age a, is estimated based on the diabetes mortality rate (M1): The survival of individuals without diabetes (S0), from a certain age a, is based not only on the non-diabetes mortality rate (M0), but also on the probability of acquiring diabetes in the remaining life time, at a certain age x, and therefore decrease the life expectancy from that moment on: It is important to note that our analyses, being derived from cross-sectional survey data on prevalence and incidence, and mortality information centered in the period of these surveys, and being based on the assumption of constant rates into the future, are calculated from a period, and not a cohort, perspective. Therefore, the years of life lost reported in this work is appropriately interpreted as the period expected years of life lost [5].

Uncertainty
We estimated uncertainty by randomly sampling from the distributions of the age-, sexand ethnicity-specific input parameters incidence, prevalence, and mortality rate ratio. More specifically, we randomly generated a database of 1000 values of diabetes incidence, prevalence, and mortality rate ratio for each ethnicity, sex and age between 35 and 80 assuming independence of all variables. For this purpose, the linear predictors from the logistic regressions used to estimate prevalence or incidence, as well as the log mortality rate ratio, which we estimated as the log hazard ratio through Cox regression, were assumed to follow a normal distribution. The national prevalence of ethnicity groups and all-cause mortality rates are national statistics, and thus were used as constant values. Utilizing each of the 1000 databases we performed all the calculations described to estimate lifetime risk of developing diabetes, years of life lost among those with diabetes, and years lived with diabetes, and from these results we identified the 2.5 th and 97.5 th percentiles as the confidence intervals.
All the statistical codes used to generate our estimates can be provided upon request to the corresponding author (paula.abracco@gmail.com).