Spatial variations and social determinants of life expectancy in China, 2005–2020: A population-based spatial panel modeling study

Spatial distribution of LE in China

During 2005–2020, nationwide, LE increased from 73.1 (95%CI: 71.3, 74.4) years to 77.7 (95%CI: 76.5, 78.7) years. In 2020, for provincial disparities, LE in Shanghai (83.1, (95%CI:83.1, 83.2) years), Beijing (82.3 (95%CI:80.3, 84.0) years) and Jiangsu (80.5 (95%CI:80.2, 80.8) years) have exceeded 80 years, and most of the provinces reached 75 years, expect for Tibet (70.9 (95%CI:67.0, 74.0) years). During 2005–2020, even though LE arosen steadily in 31 provinces, change of LE varied among regions, with Guizhou (8.0 (95%CI:7.4, 9.2) years with 11.8 (95%CI:11.5, 12.2) % of relative change ) and Yunnan (6.3 (95%CI:5.7, 6.8) years with 9.0 (95%CI:8.6, 9.4)% of relative change) increased the most and have exceeded 6 years, while Shanghai (2.8 (95%CI:2.8 , 2.8) years with 3.5 (95%CI:3.5, 3.5) % of relative change) and Beijing (2.9 (95%CI:2.3, 3.5) years with 3.7 (95%CI:3.2, 4.1) % of relative change) increased the less (Figure 2). The Moran scatterplot was used to describe spatial uncertainty and detect spatial clustering of mean LE during 2005–2020, which divided the 31 spatial units into four quadrants by standardized value of observed LE and spatially lagged standardized LE (Figure 3). The first quadrant denoted High-High (HH) positive spatial correlation, which meant provinces with higher LE were encircled by similar higher units, including southeast coastal regions like Shanghai, Beijing, Jiangsu and Zhejiang. Afterwards, the second quadrant denoted Low-High (LH) negative spatial correlation, which meant lower LE provinces were encircled by higher units, like Hebei, Jiangxi and Guangxi. The third quadrant denoted Low-Low (LL) positive spatial correlation, which meant lower LE provinces were encircled by lower units, most of them were from west, southwest or midland regions such as Tibet, Xinjiang, Qinghai and Guizhou. The last, the fourth quadrant denotes High-Low (HL) spatial correlation, which meant higher LE provinces were encircled by lower ones, including Chongqing and Shanxi. Accordingly, spatial clustering of LE was thus detected statistically significant since most of provinces (23 provinces) belonged to HH and LL regions.

Figure 2Choropleth map of LE in China in 2005, 2020 and its relative change during 2005–2020

Show full caption

a. LE (years) in 2005; b. LE (years) in 2020; c. relative change (%) of LE between 2005 and 2020.

figure 3

Figure 3Moran scatterplot of mean LE in China during 2005–2020.

Association between SDOH and LE in China

Description of SDOH proxies for 31 provinces was presented by mean (standard deviation, SD) and median (Q1, Q3) and was calculated on average for period 2005–2010, 2011–2015 and 2016–2020 separately (Table 1). With VIF of less than 10 (VIF = 6.82), the linearity was not violated and all SDOH proxies were included in the analysis (Supplementary Material Part 11). Then, we performed model diagnostic tests (including LM lag test, LM error test, Robust LM lag test and Robust LM error test) and selected models according to Wald test, LR test and professional knowledge from research practice. It concluded that SPDM could not be simplified to SPAR (Wald test statistics = 96.47, P P P P P Table 2, in this study, we were interested in identifying time-lagged LE within the provinces rather than time and space-time lagged LE interactions between spatial units which lack of theoretical supports and overfitting, thus we selected SPDM with time-lagged LE in the following interpretation.

table 1Descriptive analysis of SDOH proxies on average of 31 provinces in China, 2005-2020.

Table 2The association between SDOH proxies and LE in China, NMSS 2005-2020: estimated from spatial panel data models.

For the first SDOH component socioeconomic development and equity, locally, population in provinces with higher GDP (0.02, 95%CI: 0.00, 0.03), Gini index (2.35, 95%CI: 1.82, 2.88) was associated with an increase in LE . For spatially-lagged effects, namely, the neighbor effects, GDP (0.14, 95%CI: 0.12, 0.16), Engel’s Coefficient (0.01, 95%CI: 0.00, 0.02), urbanization rate (-0.03, 95%CI: – 0.04, -0.02), unemployment rate in urban area (-0.13, 95%CI: -0.20, -0.06), average years of education attainment (-0.09, 95%CI: -0.16, -0.02) and Gini index (10.52 , 95%CI: 9.07, 11.97) exerted impacts on LE among adjacent proximity. As for healthcare resource, number of beds in health care institutions (0.02, 95%CI: 0.00, 0.05) was associated with LE decrease locally. The rest, for the population characteristics, natural growth rate of resident population was statistically significant predicted local and neighborhood LE, for its every 1‰ increase in a focused province was observed to produce a positive local impact of 0.02 years (95%CI: 0.01 , 0.02) and a similar neighbor impact of 0.06 years (95%CI: 0.04, 0.08) of LE. Besides, a statistically significant spatially lagged effect caused by sex ratio (-0.01, 95%CI: -0.01, 0.00) and gross dependency ratio (0.01, 95%CI: 0.00, 0.02) were estimated. In addition to SDOH components, time-lagged LE (1.05, 95%CI: 1.02, 1.08) and spatially lagged LE (0.29, 95%CI: 0.25, 0.32) were positively estimated to affect local LE in a specific year.

On the basis of SPDMLAG1 estimation, we further conducted direct/indirect and long/short-term effects decomposition to interpret how SDOH proxies influence LE disparities (Table 3). For long-term effects, GDP (0.36, 95%CI: 0.28, 0.44), urbanization rate (0.10, 95%CI: 0.03, 0.17), unemployment rate in urban area (0.34, 95%CI: 0.12, 0.56), average years of education attainment (0.36, 95%CI: 0.16, 0.56), Gini index (-38.37, 95%CI: -44.63, -32.10) and natural growth rate of resident population (-0.23, 95%CI: -0.33 , -0.13) statistically significantly influenced LE disparities in total. For short-term effects, it estimated that there were much more SDOH proxies presented significant influence on LE disparities (Table 3). In contrast with Table 2, it was inferred that the short-term direct effects occupied the dominant impacts of association between SDOH proxies and LE disparities. It was observed that the average years of education attainment (-0.04, 95%CI: -0.07, 0.00) and Gini index (3.18, 95%CI: 2.60, 3.76) were much different from that estimation shown in Table 2, which attributed to spatial spillover effects, and accounted for 33.33% and 35.32% of main effects, respectively.

Table 3Direct and indirect effects decomposition during long-term and short-term of LE associated SDOH proxies in China, NMSS 2005-2020.


SDPM with time lagged LE were specified individual (spatial) and time fixed effects, and adjusted for annual average temperature, temperature variability and annual average relative humidity.