Introduction
Over the past decade, the prevalence of type 2 diabetes in China has more than doubled from approximately 3% in 1994 to 7–10% in 2008 [
1,
2], which is of concern as China is home to more than 1.3 billion people and comprises one-fifth of the world's population. Rapid urbanisation, economic growth and economic and healthcare reforms have been implicated in the increased diabetes prevalence [
3‐
6]. Urbanisation may influence diabetes prevalence through increased time engaged in sedentary lifestyle behaviours and greater consumption of animal products, high-fat foods and highly processed foods [
7,
8]. Yet, little is known about which specific elements of urbanisation are most associated with diabetes.
Most of the published literature relies on dichotomous contrasts between rural and urban areas. However, in China there is complexity and heterogeneity between and within urban and rural communities [
9]. For example, in China's uneven economic transition, some rural areas became larger than small cities, while other urban areas lost services and physical infrastructure with development [
9]. Provincial and regional differences in services and infrastructure due to decentralised healthcare and economic systems are exacerbated because of the urbanisation process [
5,
6]. Research in China suggests that a dichotomous classification of urban and rural areas inadequately captures this heterogeneity, whereas a multidimensional measure that encompasses healthcare, economic systems, services and infrastructure adequately characterises variation across the spectrum of urbanicity [
9]. While it is important to understand the association between overall urbanisation and diabetes prevalence, understanding the elements of urbanisation most associated with diabetes can inform community-level diabetes prevention and treatment efforts.
In this paper, we capitalise upon unique population-based, community- and individual-level data from the China Health and Nutrition Survey (CHNS) [
10] to examine the association between urbanisation and diabetes using a validated, multidimensional urbanisation index [
9]. This urbanisation index includes a total score as well as 12 component measures (reflecting a diverse range of factors such as healthcare, economic systems, services and infrastructure) that improve with urbanisation. In addition, we use multilevel modelling to account for the inherent interdependence between community-level factors and individual health outcomes among members of the same community or province [
11]. We hypothesise that: (1) total urbanisation is positively associated with diabetes prevalence and (2) each of the 12 components of urbanisation is separately and differentially associated with diabetes prevalence.
Methods
Statistical analysis
All analyses were conducted in Stata 12 (Stata, College Station, TX, USA). In descriptive analyses, we examined individual-level characteristics, total urbanisation level and each of the 12 urbanisation components across the nine CHNS provinces. We tested sex-specific differences in individual-level characteristics across low, medium and high urbanisation and across the nine CHNS provinces by ANOVA for continuous outcomes and χ2 tests for categorical outcomes. In addition, we tested differences in medians (due to the skewed distributions of the urbanisation index components) using the K-sample equality of medians test. The significance level for all descriptive statistics was set at p < 0.05.
Given the high prevalence of undiagnosed diabetes in China [
2], we examined the prevalence of undiagnosed and diagnosed diabetes by province and urbanisation levels. Using the question ‘Has a doctor ever told you that you suffer from diabetes?’, a ‘yes’ response was coded as diagnosed diabetes and a ‘no’ response plus an FBG ≥7.0 mmol/l was coded as undiagnosed diabetes. The two mutually exclusive categories together represent total diabetes used for the central analyses.
To examine diabetes prevalence across urbanisation levels, we used multilevel logistic regression models, with individual-level age and household income as fixed effects and with the community and province of residence as random effects. Given differences across regions of China in lifestyle, climate and economic factors, we assessed whether northern (Lianoning, Heilongjiang, Shandong and Henan) and southern (Jiangso, Jubei, Hunan, Guangxi and Guizho) region confounded the association between urbanisation and diabetes prevalence. Geographic region did not meet the 10% change-in-estimate criterion of confounding, so it was not included in the final models. However, variation due to geographic region was captured by the province random effect, which accounted for clustering in lifestyle, climate and economic factors across China.
We did not include BMI in the final models on the basis that it is a causal intermediate on the pathway from urbanisation to diabetes and, as such, could potentially bias results [
16]. However, in a secondary analysis we assessed the confounding by BMI of the association between urbanisation and diabetes. Findings indicated that BMI did not meet the 10% change-in-estimate criterion of confounding, thus BMI was not included in the final models (see electronic supplementary material [ESM] Table
1).
Likelihood ratio tests showed a statistically significant sex difference in the association between the education and income diversity urbanisation component with diabetes prevalence (p < 0.05), thus all models were stratified by sex. When age was categorised as <50 or ≥50 years, there was no evidence for effect modification by age in the association between urbanisation and diabetes prevalence (men χ
2 = 5.11, p = 0.08; women χ
2 = 0.64, p = 0.73). Tests for the interaction between age and the 12 separate urbanisation index components yielded similar results (p > 0.05), showing no empirical basis for effect modification by age, thus analyses were not age-stratified.
First, we examined prevalence of diabetes across low, medium and high urbanisation to assess whether levels of the overall multicomponent urbanisation measure was significantly associated with diabetes (statistical significance set at p < 0.05). Second, using the individual component measures as mean-centred continuous variables, we examined whether each of the 12 components of urbanisation was separately associated with diabetes. Given colinearity between the 12 urbanisation index components (correlation ranging from 0.2 to 0.7) and interest in assessing the associations across urbanisation components, we constructed 12 separate models per sex for each of the 12 urbanisation index components, resulting in a total of 24 models. We used the Bonferroni correction for multiple comparisons with a type 1 error rate of 0.05 and 24 models (p value set at 0.002). Half of the components were linearly related to diabetes prevalence on the logit scale; thus, all components were retained as linear variables for consistency and comparability across models. The components are scored independently and vary in their distribution across provinces and in the total sample. To facilitate comparison across the components and for comparability with the total urbanisation measure, we present ORs with 95% CIs comparing the 75th (high) vs the 25th (low) percentile for each of the 12 urbanisation components. Variance and SE of the random effects are presented to characterise community- and province-level variation in diabetes prevalence. We calculated the intraclass correlation coefficients (ICCs) at the community and province levels to examine the relative contribution of province, community and individual levels to diabetes prevalence.
Results
Diabetes prevalence differed across low, medium and high areas of urbanisation, with highest prevalence in highly urbanised areas (Table
2). In addition, diabetes prevalence differed by sex (
p = 0.001, χ
2 test), with a higher mean prevalence across all levels of urbanisation in men (8.7% [SE 0.5]) vs women (6.7% [0.4]).
Table 2
Individual-level characteristics by low, medium and high urbanisation level
Age, mean (SE) | Male | 50.5 (0.4) | 50.3 (0.4) | 52.1 (0.4) | 0.005 |
Female | 50.4 (0.4) | 50.3 (0.4) | 52.4 (0.4) | <0.001 |
Diabetesc,d, % (SE) | Male | 5.8 (0.7) | 8.8 (0.8) | 11.7 (0.9) | <0.001 |
Female | 4.5 (0.6) | 6.5 (0.7) | 9.1 (0.8) | 0.001 |
Overweight (WHO criterion [BMI 25–30 kg/m 2]), % (SE) [ 15] | Male | 21.0 (1.2) | 27.8 (1.3) | 28.6 (1.3) | <0.001 |
Female | 23.9 (1.2) | 26.7 (1.2) | 25.9 (1.2) | 0.21 |
Obese (WHO criterion [BMI ≥30 kg/m 2]), % (SE) [ 15] | Male | 2.8 (0.5) | 3.8 (0.6) | 4.9 (0.6) | 0.03 |
Female | 4.2 (0.5) | 4.7 (0.6) | 4.0 (0.5) | 0.64 |
Overweight (Chinese criterion [BMI 24–28 kg/m 2]), % (SE) [ 14] | Male | 27.1 (1.3) | 32.9 (1.4) | 36.0 (1.4) | <0.001 |
Female | 28.9 (1.2) | 29.5 (1.2) | 29.2 (1.2) | 0.93 |
Obese (Chinese criterion [BMI ≥28 kg/m 2]), % (SE) [ 14] | Male | 5.9 (0.7) | 9.5 (0.8) | 10.8 (0.9) | <0.001 |
Female | 10.0 (0.8) | 11.0 (0.8) | 10.4 (0.8) | 0.69 |
Across provinces, diabetes prevalence ranged from 4.8% to 12.1% (Table
3). Total urbanisation and the 12 urbanisation components varied across provinces, indicating a differential degree of urbanisation across the provinces.
Table 3
Individual-level and community-level urbanisation characteristics, by province
Individual-level factors |
Sample, no. (%) | 847 (10.9) | 657 (8.5) | 856 (11.1) | 723 (9.3) | 879 (11.4) | 808 (10.4) | 968 (12.5) | 956 (12.4) | 1,047 (13.5) | 7,741 (100.0) |
Diabetes, % (SE)a,b
| 6.3 (0.8) | 5.6 (0.9) | 8.6 (1.0) | 11.6 (1.2) | 12.1 (1.1) | 5.8 (0.8) | 4.8 (0.7) | 6.8 (0.8) | 7.7 (0.8) | 7.7 (0.3) |
Household income (Yuan), mean (SE)b
| 34,017 (1,183) | 32,345 (1,514) | 31,284 (1,428) | 37,685 (1,164) | 39,420 (1,503) | 38,696 (1,693) | 32,842 (1,229) | 39,759 (1,709) | 56,511 (1,474) | 38,622 (496) |
Community-level total urbanisation level)c
|
Low, no. (%)d
| 560 (66.1) | 351 (53.4) | 493 (57.6) | 108 (14.9) | 233 (26.5) | 213 (26.4) | 361 (37.3) | 109 (11.4) | 143 (13.7) | 2,571 (33.2) |
Medium, no. (%) | 118 (13.9) | 235 (35.8) | 168 (19.6) | 277 (38.3) | 384 (43.7) | 340 (42.1) | 211 (21.8) | 424 (44.4) | 431 (41.2) | 2,588 (33.4) |
High, no. (%) | 169 (20.0) | 71 (10.8) | 195 (22.8) | 338 (46.8) | 262 (29.8) | 255 (31.6) | 396 (40.9) | 423 (44.3) | 473 (45.2) | 2,582 (33.4) |
Median (IQR)b
| 49.7 (43.1, 75.6) | 52.8 (48.0, 68.3) | 54.1 (40.6, 81.2) | 65.6 (56.1, 89.1) | 67.5 (54.7, 82.7) | 68.9 (52.8, 85.3) | 71.0 (47.5, 85.8) | 73.9 (61.0, 92.0) | 82.1 (56.3, 90.6) | 65.0 (50.9, 85.4) |
Community-level 12 urbanisation index componentsd
|
Population densityb, median (IQR) | 5.0 (3.5, 6.5) | 5.5 (5.0, 6.5) | 5.5 (5.0, 6.0) | 6.5 (5.5, 8.0) | 6.0 (5.0, 6.5) | 6.5 (5.0, 8.0) | 6.0 (5.0, 6.5) | 6.0 (5.0, 6.5) | 6.5 (5.5, 7.0) | 6.0 (5.0, 7.0) |
Social servicesb, median (IQR) | 1.3 (1.3, 2.5) | 1.3 (1.3, 2.5) | 2.5 (1.3, 5.0) | 2.5 (1.3, 5.0) | 1.3 (1.3, 7.5) | 2.5 (1.3, 6.3) | 2.5 (1.3, 5.0) | 5.0 (2.5, 8.8) | 1.3 (1.3, 8.8) | 2.5 (1.3, 5.0) |
Health infrastructureb, median (IQR) | 6.4 (1.3, 8.8) | 6.3 (2.5, 7.5) | 5.0 (3.8, 7.5) | 7.5 (6.3, 8.8) | 7.5 (5.0, 8.8) | 5.0 (3.0, 7.5) | 5.0 (2.5, 8.0) | 7.5 (5.0, 8.0) | 7.5 (4.0, 8.8) | 7.5 (3.8, 8.8) |
Modern marketsb, median (IQR) | 1.5 (0.5, 5.0) | 2.0 (1.0, 4.5) | 3.5 (1.5, 5.0) | 5.0 (3.5, 7.0) | 2.5 (2.0, 6.0) | 5.5 (4.0, 8.0) | 6.0 (1.5, 8.5) | 6.3 (4.5, 8.0) | 4.5 (2.0, 6.5) | 4.5 (1.5, 7.0) |
Traditional marketsb, median (IQR) | 5.0 (0.0, 5.0) | 5.0 (0.9, 5.1) | 2.9 (0.0, 5.0) | 2.9 (1.4, 5.0) | 5.0 (1.4, 7.5) | 5.0 (3.8, 8.8) | 5.0 (4.4, 8.8) | 5.0 (5.0, 10.0) | 5.0 (0.0, 10.0) | 5.0 (1.4, 7.9) |
Transportation infrastructureb, median (IQR) | 6.7 (5.0, 6.7) | 5.0 (3.3, 6.7) | 5.0 (3.3, 6.7) | 6.7 (6.7, 6.7) | 5.0 (3.3, 6.7) | 6.7 (5.0, 6.7) | 6.7 (5.0, 6.7) | 6.7 (6.7, 6.7) | 6.7 (3.3, 6.7) | 6.7 (5.0, 6.7) |
Communicationsb, median (IQR) | 6.5 (5.8, 7.2) | 4.9 (4.0, 6.8) | 6.9 (5.3, 8.1) | 6.5 (6.0, 7.2) | 7.0 (6.1, 8.1) | 7.2 (6.5, 7.8) | 6.6 (5.5, 8.4) | 7.0 (5.4, 8.6) | 7.9 (6.8, 8.5) | 6.8 (5.8, 8.0) |
Housing infrastructureb, median (IQR) | 7.0 (4.9, 9.3) | 6.7 (6.1, 8.8) | 4.6 (3.8, 8.3) | 9.4 (6.9, 9.8) | 7.1 (5.1, 9.3) | 7.8 (6.3, 9.2) | 8.5 (5.8, 9.6) | 8.5 (7.6, 9.9) | 9.3 (7.3, 10.0) | 7.6 (5.9, 9.6) |
Sanitationb, median (IQR) | 4.5 (3.0, 6.0) | 6.5 (3.2, 9.0) | 4.8 (3.3, 9.5) | 8.8 (5.0, 10.0) | 7.3 (5.8, 9.2) | 7.6 (4.3, 9.8) | 7.3 (2.5, 9.5) | 8.6 (6.3, 10.0) | 9.8 (8.5, 10.0) | 7.5 (4.3, 9.8) |
Economic activityb, median (IQR) | 5.0 (3.0, 10.0) | 5.5 (3.0, 7.0) | 6.0 (2.0, 10.0) | 9.9 (3.1, 10.0) | 8.0 (3.0, 10.0) | 5.8 (3.0, 10.0) | 7.0 (5.0, 9.6) | 9.0 (5.5, 10.0) | 9.5 (6.0, 10.0) | 7.0 (4.0, 10.0) |
Educationb, median (IQR) | 2.8 (2.4, 3.7) | 2.2 (1.6, 3.0) | 2.9 (2.4, 4.0) | 4.2 (3.0, 4.8) | 3.2 (2.6, 4.0) | 2.9 (2.4, 3.9) | 3.2 (2.6, 3.7) | 3.3 (2.8, 4.0) | 3.3 (2.5, 4.5) | 3.0 (2.5, 4.0) |
Education and income diversityb, median (IQR) | 4.5 (4.5, 5.0) | 5.3 (4.5, 6.0) | 5.5 (4.5, 5.8) | 5.5 (4.5, 6.0) | 5.5 (5.0, 6.5) | 5.5 (4.5, 6.0) | 5.0 (4.5, 5.5) | 5.5 (5.0, 6.0) | 6.0 (5.0, 6.5) | 5.5 (4.5, 6.0) |
Over half of the individuals with diabetes in this study were not previously diagnosed by a doctor (ESM Table
4). The proportion of the sample with undiagnosed and diagnosed diabetes differed across urbanisation level and province.
High vs low urbanisation was associated with an approximate twofold higher diabetes prevalence (men OR 2.02, 95% CI 1.47, 2.78; women OR 1.94, 95% CI 1.35, 2.79) after controlling for individual-level factors and accounting for clustering at the community and province levels (Table
4). In general, higher urbanisation across the 12 components was positively associated with diabetes prevalence, albeit with variation in statistical significance (ESM Figs
1 and
2). Comparatively stronger associations were observed for men.
Table 4
ORs and province- and community-level variance for diabetes prevalence across total urbanisation and its componentsa,b
Total urbanisation levelc
|
Low | 1 | | 0.13 (0.08) | 0.04 | 0.00 (0.00) | 0.00 | 1 | | 0.07 (0.05) | 0.02 | 0.12 (0.10) | 0.03 |
Medium | 1.50 (1.08, 2.08) | 0.02 | | | | | 1.52 (1.04, 2.21) | 0.03 | | | | |
High | 2.02 (1.47, 2.78) | <0.001 | | | | | 1.94 (1.35, 2.79) | <0.001 | | | | |
12 urbanisation index componentsd
|
Population density | 1.27 (1.08, 1.50) | 0.005 | 0.12 (0.07) | 0.04 | 0.01 (0.08) | 0.00 | 1.18 (0.97, 1.43) | 0.09 | 0.07 (0.06) | 0.02 | 0.17 (0.10) | 0.05 |
Social services | 1.11 (0.96, 1.28) | 0.16 | 0.14 (0.08) | 0.04 | 0.04 (0.08) | 0.01 | 1.22 (1.05, 1.43) | 0.01 | 0.06 (0.05) | 0.02 | 0.15 (0.10) | 0.04 |
Health infrastructure | 1.39 (1.07, 1.81) | 0.01 | 0.11 (0.07) | 0.03 | 0.04 (0.08) | 0.01 | 1.19 (0.88, 1.60) | 0.25 | 0.06 (0.06) | 0.02 | 0.18 (0.10) | 0.05 |
Modern markets | 1.49 (1.17, 1.91)e
| 0.001 | 0.15 (0.09) | 0.04 | 0.02 (0.07) | 0.01 | 1.46 (1.10, 1.94) | 0.008 | 0.08 (0.06) | 0.02 | 0.15 (0.10) | 0.04 |
Traditional markets | 1.30 (1.03, 1.64) | 0.03 | 0.15 (0.09) | 0.04 | 0.03 (0.08) | 0.01 | 1.33 (1.01, 1.74) | 0.04 | 0.09 (0.07) | 0.02 | 0.16 (0.10) | 0.04 |
Transportation infrastructure | 1.11 (1.01, 1.22) | 0.04 | 0.14 (0.08) | 0.04 | 0.02 (0.08) | 0.01 | 1.18 (1.06, 1.32) | 0.004 | 0.08 (0.06) | 0.02 | 0.14 (0.10) | 0.04 |
Communications | 1.48 (1.22, 1.79)e
| <0.001 | 0.13 (0.07) | 0.04 | 0.00 (0.00) | 0.00 | 1.36 (1.10, 1.69) | 0.005 | 0.06 (0.05) | 0.02 | 0.15 (0.10) | 0.04 |
Housing infrastructure | 1.61 (1.27, 2.05)e
| <0.001 | 0.15 (0.09) | 0.04 | 0.00 (0.00) | 0.00 | 1.32 (1.00, 1.73) | 0.05 | 0.08 (0.06) | 0.02 | 0.16 (0.10) | 0.05 |
Sanitation | 1.59 (1.23, 2.06)e
| <0.001 | 0.12 (0.07) | 0.04 | 0.01 (0.07) | 0.00 | 1.59 (1.19, 2.12)e
| 0.002 | 0.07 (0.06) | 0.02 | 0.13 (0.10) | 0.04 |
Economic activity | 1.51 (1.19, 1.91)e
| 0.001 | 0.13 (0.08) | 0.04 | 0.01 (0.08) | 0.00 | 1.25 (0.95, 1.63) | 0.11 | 0.07 (0.06) | 0.02 | 0.17 (0.10) | 0.05 |
Education | 1.29 (1.15, 1.46)e
| <0.001 | 0.12 (0.07) | 0.03 | 0.00 (0.07) | 0.00 | 1.10 (0.94, 1.29) | 0.22 | 0.06 (0.05) | 0.02 | 0.18 (0.10) | 0.05 |
Education and income diversity | 1.42 (1.20, 1.66)e
| <0.001 | 0.10 (0.07) | 0.03 | 0.00 (0.08) | 0.00 | 1.14 (0.94, 1.39) | 0.18 | 0.07 (0.06) | 0.02 | 0.18 (0.10) | 0.05 |
Low variance and ICC at the community level for men suggests low variation in diabetes across communities (Table
4). In contrast, community-level variance was larger in women than men, and women had a higher ICC at the community level than at the province level. Thus, for women, variation in diabetes prevalence across communities remained even after urbanisation components were taken into account.
Although BMI was not an empirical confounder of the association between urbanisation and diabetes prevalence, findings from models with adjustment for BMI are shown in ESM Table
2. With BMI in the model, ORs were somewhat attenuated, but in general showed the same pattern of direction and statistical significance.
While models for total diabetes prevalence are presented in Table
4, we also considered associations with urbanisation for diagnosed and undiagnosed diabetes (ESM Table
5). We observed a similar pattern of association for diagnosed diabetes at high vs low urbanisation (men OR 5.86, 95% CI 3.08, 11.15; women OR 4.03, 95% CI 2.09, 7.78) after controlling for individual-level factors and accounting for clustering at the community and province levels. In contrast, the prevalence of undiagnosed diabetes did not differ by high vs low urbanisation level (men OR 1.25, 95% CI 0.84, 1.85; women OR 1.38, 95% CI 0.88, 2.17).
Sensitivity analyses related to diagnostic criteria for diabetes suggest that the use of HbA
1c alone (ESM Table
4), a combination of high HbA
1c and/or FBG (ESM Table
5), and FBG alone (Table
4) yielded findings similar in direction and magnitude.
In sensitivity analyses, we also compared an administratively defined classification of urban vs rural communities to the multidimensional urbanisation measure used in the central analysis. ORs for diabetes prevalence suggest a similar direction and magnitude for high vs low urbanisation using the traditional (men OR 1.54, 95% CI 1.21, 1.97; women OR 1.38, 95% CI 1.04, 1.84) vs multidimensional (men OR 2.02, 95% CI 1.47, 2.78; women OR 1.94; 95% CI 1.35, 2.79) measure.
Discussion
Our findings suggest that diabetes prevalence is elevated across areas of low and high urbanisation in China. However, using a detailed and multidimensional measure of urbanisation, we found that individuals living in more-urbanised areas had a twofold higher diabetes prevalence than individuals living in less-urbanised areas, even after community and province of residence, age, sex and household income were accounted for. These findings are concerning given the vast number of affected individuals. China is home to the world's largest population and approximately 80% of diabetes deaths occur in low- and middle-income countries [
17]. Although we observed higher diabetes prevalence in more-urbanised areas, the relatively high level of diabetes in less-urbanised areas suggests that these areas are still in need of attention. We also observed variation in diabetes prevalence across specific components of urbanisation. Our findings suggest that community economic factors, modern markets, communications and transportation infrastructure might be areas deserving of further research and could potentially be the focus of community-level diabetes prevention efforts.
Although others have examined disparities in diabetes [
4,
18,
19] and other have examined noncommunicable diseases [
2,
20,
21] across regions of China using traditional measures of urbanisation, our analysis improves on these studies in three key ways. First, the CHNS intentionally sampled the capital city and rural areas within each province to span the breadth of economic development within provinces and thus provides substantial variation across more- and less-urbanised areas. Second, our multidimensional measure of urbanisation, which has been tested for reliability and validity [
9], captures heterogeneity in a variety of services and infrastructure within and between high- and low-urbanised areas. This multidimensional measure is particularly relevant in China or other countries undergoing rapid and asymmetric urbanisation, which is inadequately captured by dichotomous classification of urbanicity [
22‐
24]. Having observed the separate associations between the 12 urbanisation index components and diabetes prevalence, we suggest that some potentially modifiable factors, namely community economic factors, modern markets, communications and transportation infrastructure, might be foci for further research and community-level policy.
Third, we used multilevel modelling to account for the clustering of individuals at the community and province levels. This technique is critical given our research question, as well as the nature of the sampling method, and the fact that the main exposures of interest are measured at the community level [
11]. Indeed, descriptive statistics suggested differences in the distribution of urbanisation components and diabetes prevalence across provinces. Accounting for these dependencies via random-effects modelling did not attenuate associations between urbanisation and diabetes. Therefore, although community and province of residence were associated with differential diabetes prevalence in China, unmeasured factors remain that might influence diabetes.
We observed variation in the level of urbanisation and prevalence of diabetes across provinces of China. The total measure of urbanisation captures synergies among its components across low-, medium- and high-urbanisation levels, and we observed higher prevalence of diabetes in high- vs low-urbanised areas of China. Our findings across the 12 urbanisation components provide additional information not captured in the overall measure. For example, we observed substantial variation in economy-related factors across provinces and increased diabetes prevalence in communities with high vs low economic activity, education, and education and income diversity. Together, these findings might be attributed to limited/regional free trade agreements, economic development that has favoured eastern China and port cities [
25], and a decentralised healthcare system that determines health regulation and infrastructure at community and province levels [
5,
6]. Modern markets (which in our study included supermarkets and fast-food restaurants) have also increased in number with economic development [
26] and were positively associated with diabetes prevalence. Our findings, namely that multiple economy-related factors are positively associated with diabetes prevalence, suggest that the largest disease burden may be found in wealthier and more modernised areas of China.
Given inter-correlation across the 12 urbanisation components, combining them into a single model was not possible. Furthermore, we were explicitly interested in the net effect of each individual component. Looking at the associations between high vs low levels of the 12 urbanisation components with diabetes prevalence highlights modifiable factors with potential for community-level efforts to prevent and treat diabetes. While our findings are suggestive, further research on these community-level factors is needed before incorporating such factors into policy and programmes. Nonetheless, community-level factors, such as transportation and communication, which were positively associated with diabetes, have been useful for addressing chronic diseases in other developing-country contexts. For example, in Brazil, changes in transportation infrastructure have had an impact on population-level physical activity [
27,
28]. In Pakistan, a communications-oriented health education campaign has positively affected nutrition-related chronic disease awareness [
29]. Increased alcohol taxes in Thailand have been associated with reductions in alcohol-related mortality [
30]. In China, community-level tobacco cessation efforts have been associated with reductions in smoking [
31]. Thus, community-level strategies offer promise in deterring lifestyle-related diseases such as diabetes.
In general, associations between the 12 urbanisation index components and diabetes prevalence appeared to be stronger in men. The extent to which this difference represents a differential relationship between community-level factors and diabetes by sex or whether other factors (e.g. educational and economic opportunities for men vs women) drive this difference is unknown. In the CHNS, a comparatively stronger association between urbanisation and BMI for men has been shown [
32]; therefore, it is possible that lifestyle in urban vs rural areas differs by sex.
In contrast, we did not find differences in the association between urbanisation and diabetes prevalence by age. Cumulative exposure to the urban environment has been shown to be related to increased prevalence of overweight and noncommunicable diseases [
33]; however, because urbanisation is a relatively recent and ongoing phenomenon in China, cumulative exposure to the urban environment may not be greater in older people, so differential associations between urbanisation and diabetes by age may not be apparent.
There are limitations to the current study that should be noted: (1) the main exposure variables are based on community-level data and while our urbanisation index measure includes diverse and detailed components of urbanisation, elements such as local food prices and individual-level exposures are not captured; (2) our cross-sectional analysis does not take into account changes in urbanisation over time; (3) diabetes prevalence is influenced by a number of factors in addition to incidence; thus, differential mortality among individuals with diagnosed or undiagnosed diabetes may influence prevalence of diabetes (e.g. diabetes prevalence may be artificially lower in areas with higher mortality rates [
34]). However, our findings do provide insight related to province- and community-level factors that might be targets for policy in China and other recently urbanised countries. Although we used just one FBG measure to classify individuals with diabetes, this is standard epidemiological practice for population-based studies. While FBG and HbA
1c capture different populations of people with diabetes [
35,
36], we found consistency in results using HbA
1c and FBG in terms of the effect size and direction of association with overall urbanisation and its components, albeit with some differences by sex. Researchers in Korea found a greater proportion of women with diabetes when using HbA
1c vs FBG as the diagnosis criterion [
37]; however, the reasons why HbA
1c would capture more women with diabetes are unknown. We are not able to distinguish type 1 from type 2 diabetes in our sample, but the incidence of type 1 diabetes in China is among the lowest in the world, with an estimate of 0.1 per 100,000 each year [
38]. Due to differences between the WHO and Chinese BMI cut-off points for overweight and obesity, both were included in Table
2. While the internationally recognised WHO cut-off points facilitate comparison across studies and nations, the lower BMI Chinese cut-off points [
15] reflect the higher noncommunicable disease risk at lower BMI in Chinese populations [
14].
Conclusion
Diabetes is becoming a major public health concern across China. Using a multidimensional measure of urbanisation, which captures healthcare, economic systems, services and infrastructure, we observed a twofold higher prevalence of diabetes in more- vs less-urban areas, even after community and province of residence, age, sex and household income were accounted for. We also observed that high (i.e. more urbanised) values for some components of urbanisation, particularly community economic factors, modern markets, communications and transportation infrastructure, were separately and positively associated with diabetes. As such, these factors might present opportunities for community-level diabetes prevention efforts.
Acknowledgements
The authors are thankful to J. Jones-Smith (University of California San Francisco) for guidance regarding the urbanisation index and S. Albrecht (Carolina Population Center) for helpful comments on an earlier draft of the manuscript.