We obtained China DKD mortality data from the Global Burden of Disease (GBD) Study 2019, which is a multinational collaborative study that estimates disease burden in 204 countries and territories worldwide (17, 18). The methods used in GBD 2019 have been reported in detail elsewhere (17–19). In brief, GBD 2019 used vital registration and verbal autopsy data to model mortality due to chronic kidney disease (CKD) (17). The Bayesian geospatial regression model was used to increase the comparability of mortality data sources that used location-specific covariates to create smoothed time trends. Data from the ESRD registry were used to estimate five causes of CKD: type 1 diabetes, type 2 diabetes, glomerulonephritis, hypertension, and a residual category of other and unspecified causes. The DKD data analyzed in this paper refer to the data on CKD due to type 2 diabetes. An epidemiologic state-transition disease modeling tool was used to produce consistent estimates by location, year, age, and sex. These adjusted proportions were applied to the parent CKD regression model to obtain type-specific estimates of CKD. As the data were publicly available and data were aggregated and de-identifiable, institutional review board approval and informed consent were not needed.

We used the age-period-cohort framework to estimate the following parameters: (1) net drift, representing the overall log-linear trend by period and birth cohort, indicating the overall annual percentage change of the expected age-adjusted rate; (2) local drifts, representing the log-linear trends for each age group by period and birth cohort, indicating the annual percentage change of the expected age-specific rate over time; (3) longitudinal age curve, showing the expected age-specific rates adjusted for period effects in reference cohort; (4) period (or cohort) rate ratios (RR), representing the ratio of age-specific rates in each period (or cohort) relative to the reference one.

For age-period-cohort analyses, we arranged the DKD mortality and population data into consecutive 5-year periods from 1990 to 2019, and successive 5-year age intervals from 15-19 years to 95 plus. The birth cohort was defined using the difference between the medium value of the age interval and the period interval. We obtained the estimated parameters by the age-period-cohort Web Tool provided by the National Cancer Institute (20). For relative rate measurements, the reference period interval was from 2000 to 2004, and the reference birth cohort interval was from 1945 to 1949. We used the Wald chi-square test to test the significance of the estimable parameters and functions. All statistical tests were two-sided.

We used the Bayesian age-period-cohort analysis with integrated nested Laplace approximations to project the future age-specific number of death cases from DKD from 2020 to 2030 (21), which shows better coverage and precision than other prediction methods (22). Based on the assumption that the effects of age, period and cohort adjacent in time are similar, the Bayesian inference in age-period-cohort model applies the second-order random walk for smoothing priors of age, period, and cohort effects and to project posterior mortality rates. The integrated nested Laplace approximations are used with this Bayesian age-period-cohort model to approximate the marginal posterior distributions avoiding any mixing and convergence issues introduced by Markov chain Monte Carlo sampling techniques traditionally used in the Bayesian approach. We conducted the Bayesian age-period-cohort analysis using R-package BAPC (version 0.0.34). We provided additional details in Text S1. The population predictions for China were taken from the 2019 revision of the United Nations (UN) World Population Prospects and were used to estimate China’s population in 2020 and beyond (23).

To analyze the drivers of the changes in the number of DKD deaths from 1990 to 2030, we used a newly developed decomposition method to attribute changes in the total number of DKD deaths to population growth, population aging, and age-specific changes in DKD mortality between 1990 and each subsequent year from 1991 to 2030 (16, 24). Briefly, this decomposition method has considered the 2-way and 3-way interactions of the three components and is robust to the choice of the decomposition order of the three factors, and the selection of the reference year compared to previous decomposition methods (25, 26). Details about the decomposition method were described elsewhere (16, 24) and in the Text S2. This method has been used to quantify the impact of population aging on mortality for 195 countries or territories and 169 causes of deaths (24), and to quantify the demographic and epidemiologic drivers for the impacts of air pollution and high sodium intake (27–29). We calculated the absolute and relative contributions of the three drivers to the change in the number of DKD deaths. The absolute contribution was the number of attributed DKD deaths, while the relative contribution was estimated as the number of attributed DKD deaths divided by the total DKD deaths in 1990×100%. A positive contribution indicates an increase in total DKD deaths, while a negative contribution indicates a decrease in total DKD deaths. The age-specific changes in DKD deaths refer to epidemiologic changes, which include all differences in mortality that cannot be explained by population growth and population aging (30), such as new treatments or medications for DKD. The net changes in these three components are equal to the difference in the total number of observed deaths. We performed statistical analyses with R software (Version 3.6.3, R core team).

In 2019, there were 63,354 (95% UI: 49,787 to 77,280) DKD deaths in China, and the age-standardized mortality rate of DKD was 3.6 (95% UI: 2.8 to 4.3) per 100,000. Between 1990 and 2019, the total number of DKD deaths increased dramatically from 13,269 (95% UI: 9,998 to 16,930) in 1990 to 32,296 (95% UI: 24,275 to 41,393) in 2019 for males and from 14,144 (95% UI: 10,909 to 17,470) in 1990 to 31,058 (95% UI: 23,720 to 38,907) in 2019 for females (Figure 1A). On the contrary, the age-standardized mortality rate of DKD was relatively stable for both sexes (Figure 1B).

Net drift represents the overall annual percentage change across the study period (Figure 2). We found marked sex differences in net drift with -0.75% (95% confidence interval, CI: -0.93% to -0.57%) for males and -1.90% (95% CI: -2.19% to -1.62%) for females, reflecting less improvement in reduction of DKD mortality for males than for females from 1990 to 2019. Local drift reflects additional age-specific variations in DKD mortality trends (Figure 2). Values lie predominantly below 0 for both sexes for most age groups, indicating improvements in reducing DKD mortality. The exceptions were males aged above 65 to 69 and females aged above 70 to 74, indicating increased mortality from DKD.

For both sexes, in the same birth cohort, the risk of death from DKD showed an accelerated increase with age. We performed a curve estimation for the longitudinal age curves and found that both sexes showed an exponential distribution (Figure 3). The relationship between age and mortality rate can be expressed as mortality rate=0.023×e^0.097×age for males (R-squared=0.997) and mortality rate=0.056×e^0.081×age for females (R-squared=0.997), where age is the median age of each age interval. These indicated that the mortality risk of DKD was 128-fold higher for males and 57-fold higher for females aged 75 to 79 years compared to the corresponding males and females aged 20 to 24, respectively.

The period (cohort) relative risks are the ratio of age-specific rates in each period (cohort) relative to the reference period (cohort). We found decreased period relative risks for both sexes, with a more quickly decreasing trend for females than for males during the whole study period after adjusting for age and birth cohort (Figure 4). Cohort relative risks were also found in similar patterns for both sexes, starting to decline after 1935 for females and after 1945 for males and then declining more rapidly for females (Figure 5). In addition, using the specific results of Wald tests, we found cohort and period effects for both sexes, and the net drifts and local drifts were all statistically significant (p<0.05) (Table S1).

We next conducted a Bayesian age-period-cohort analysis to project future mortality trends for DKD in China. Our results showed that the total number of deaths from DKD in China would continue to increase, with 88,803 deaths from DKD in China by 2030 (Table S2). However, there were significant differences in the distribution of DKD deaths across age groups, with more occurring in the older age groups and a continued increase in the older age groups (above 60 years), but a decreasing trend in the younger age groups (under 50 years).

Finally, we conducted a demographic decomposition analysis to identify DKD mortality drivers from 1990 to 2030. Our results showed that demographic factors drove the increasing trend in the number of DKD deaths in China, with population aging playing a dominant role, especially after 2010 (Table 1, Figure 6).

There were 35,942 additional DKD deaths in China in 2019 from 1990, an increase of 131.2%. The increase was driven by changes in the number of DKD deaths due to population aging (90.2% increase from 1990) and population growth (53.5% increase from 1990). Our projection suggests that the increasing trend in the number of DKD deaths will continue. By 2030, China will have 224.2% more DKD deaths than in 1990, with a contribution of 173.9% increase in deaths due to population aging and a 74.9% increase due to population growth, despite a 24.7% decrease in age-specific death rates (Table 1, Figure 6).