Genetic Risk, a Healthy Lifestyle, and Type 2 Diabetes: the Dongfeng-Tongji Cohort Study Free

Abstract

Objective

The objective of this study is to examine whether healthy lifestyle could reduce diabetes risk among individuals with different genetic profiles.

Design

A prospective cohort study with a median follow-up of 4.6 years from the Dongfeng-Tongji cohort was performed.

Participants

A total of 19 005 individuals without diabetes at baseline participated in the study.

Main Variable Measure

A healthy lifestyle was determined based on 6 factors: nonsmoker, nondrinker, healthy diet, body mass index of 18.5 to 23.9 kg/m², waist circumference less than 85 cm for men and less than 80 cm for women, and higher level of physical activity. Associations of combined lifestyle factors and incident diabetes were estimated using Cox proportional hazard regression. A polygenic risk score of 88 single-nucleotide polymorphisms previously associated with diabetes was constructed to test for association with diabetes risk among 7344 individuals, using logistic regression.

Results

A total of 1555 incident diabetes were ascertained. Per SD increment of simple and weighted genetic risk score was associated with a 1.39- and 1.34-fold higher diabetes risk, respectively. Compared with poor lifestyle, intermediate and ideal lifestyle were reduced to a 23% and 46% risk of incident diabetes, respectively. Association of lifestyle with diabetes risk was independent of genetic risk. Even among individuals with high genetic risk, intermediate and ideal lifestyle were separately associated with a 29% and 49% lower risk of diabetes.

Conclusion

Genetic and combined lifestyle factors were independently associated with diabetes risk. A healthy lifestyle could lower diabetes risk across different genetic risk categories, emphasizing the benefit of entire populations adhering to a healthy lifestyle.

Diabetes remains a critical public health concern all around the world. The overall diabetes prevalence is estimated to be greater than 11.6% and affects more than 100 million Chinese adults (1). Genetic and lifestyle factors are both important drivers of this complex disease. During the past 10 years, genome-wide association studies (GWAS) have identified more than 100 independent diabetes-related loci (2-13). These single-nucleotide polymorphisms (SNPs) can provide quantitative measure of genetic susceptibility and be predictive of incident diabetes events (14).

Previous interventional studies have demonstrated beneficial effects of lifestyle modification on diabetes risk. The Da Qing IGT and Diabetes Study showed that diet and/or exercise intervention decreased diabetes incidence (15). The Finnish Diabetes Prevention Study found that lifestyle change was related to decreased diabetes incidence (16). The U.S. Diabetes Prevention Program and the Indian Diabetes Prevention Program also observed that weight loss, exercise intervention, and/or metformin treatment could reduce diabetes incidence among a high-risk population (17, 18). In addition, whether genetic risk modifies the effects conferred by lifestyle modification has also been tested. The Diabetes Prevention Program Research Group observed that lifestyle intervention and/or pharmacological treatment could eliminate the increased risk of a single genetic locus such as ENPP K121Q and TCF7L2 as well as genetic risk score (GRS) on diabetes (19-22). However, owing to the different genetic susceptibility and lifestyle among different ethnicity populations, whether such associations persist or are modified by genetic factors in nonwhite populations like Asian populations remains unknown.

This study aimed to investigate the associations of a polygenic score, a healthy lifestyle, and type 2 diabetes risk; as well as to examine whether association of a healthy lifestyle with diabetes persists across different genetic profiles in a large prospective cohort of an Asian population.

Materials and Methods

Study population

The design, methods, and other details of the Dongfeng-Tongji cohort have been described elsewhere (23). Selection of the study population can be found in an online repository (24). Briefly, a total of 27 009 retired employees were recruited to the cohort and completed baseline questionnaires, underwent medical examinations, and provided baseline blood samples between September 2008 and June 2010. Among 25 978 individuals (96.2%) who completed the follow-up until October 2013, we excluded individuals with diabetes (n = 4970) and cancer (n = 1183) at baseline, as well as those with missing information related to lifestyle factors and other covariates (n = 1851), resulting in a final study sample of 19 005 individuals (8567 men and 10 438 women with a mean age of 63.2 years) to examine effects of lifestyle on incident diabetes risk. Meanwhile, among the baseline population GWAS data were available for 7532 participants. After excluding those with cancer and missing information among the 7532 individuals with GWAS data, 7344 participants (2567 diabetic cases and 4777 nondiabetic controls) remained for investigation into the associations of lifestyle with diabetes across different genetic risk groups. Characteristics between those with and without genetic data were compared and can be found in an online repository (24).

The present study has been approved by the Ethics and Human Subject Committee of the School of Public Health, Tongji Medical College, Huazhong University of Science and Technology, and Dongfeng General Hospital, the Dongfeng Motor Corporation. All study participants provided written informed consent.

Assessment of lifestyle factors and covariates

Baseline lifestyle information was collected by trained interviewers via questionnaires during face-to-face interviews. Questions about tobacco smoking included beginning or quitting age, and amount of tobacco smoked per day for current and ever smokers. Information about alcohol consumption included beginning or quitting age, drinking frequency, type of alcoholic beverage drunk habitually, and volume of alcohol drunk per time. Physical activity was assessed by asking the frequency, the usual type and duration of activities per week in the past 6 months. Exercise types included walking, biking, dancing, tai chi, jogging, swimming, climbing stairs/mountains, and playing basketball/volleyball/soccer. Habitual intake of 13 conventional food groups in the past 12 months was obtained via a simplified semiquantitative food frequency questionnaire.

Body weight, standing height, and waist circumference were measured with participants in light indoor clothing and without shoes by trained staff using calibrated instruments. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Information on sociodemographic factors such as age, sex, education (primary school/middle school/high school/college or more), marital status (married/single), medications, and health status were also included in the questionnaires.

Definition of lifestyle factors and score

Six dietary and lifestyle factors were considered to define lifestyle score, namely smoking, alcohol consumption, physical activity, diet, BMI, and waist circumference, consistent with earlier studies (25-27). These factors have also been highlighted in recent guidelines (28, 29) for the prevention of type 2 diabetes. Correlations between lifestyle variables components are detailed in an online repository (24).

Smokers were defined as those who had smoked for more than half a year, and smoked at least one cigarette per day; the low-risk group was defined as nonsmokers or those who had stopped smoking for more than half a year. Alcohol drinkers were defined as those who had drunk alcohol for more than half a year, and drunk at least once per week. Several previous studies showed that low to moderate level of alcohol drinking can have a protective effect on diabetes (30, 31) but results were not conclusive. A meta-analysis including more than 1.9 million individuals from 38 studies showed that reductions in risk among moderate alcohol drinkers were confined to women and non-Asian populations (32). A recent systemic analysis published in the Lancet (33) including 650 000 individuals across 195 countries and territories did not find significant J-shaped curves for diabetes, but found that the weighted attributable risk decreased for diabetes among women only but not among men. In China few women drink, and in the present study only 6.6% of women reported current alcohol drinking. Therefore, we defined participants who reported never drinking alcohol as being at low risk for alcohol consumption status.

Physical activity was assessed as exercise regularly for more than 20 minutes per time. Total duration per week was calculated as duration (hours per time) × frequency (times per week). A median of 7 hours per week was set as the cutoff. Participants who engaged in a higher physical activity level (≥7 h/wk) were defined as being in the low-risk group. Notably, the amount of physical activity in the study population exceeds that in the U.S. and European populations. A previous study of more than 700 000 individuals from 111 countries showed that average daily steps among Chinese were greater than 6000, which was far more than that among US and Europe populations (4000-5000 steps) (34). Participants in the Dongfeng-Tongji cohort are retired employees with an average age older than 60 years. Nearly 90% of participants reported daily physical activity for more than half an hour, and 36.5% of participants engaged in physical activity for more than 7 hours per week.

For dietary factors, we included 3 food items that were particularly emphasized in the 2013 guideline of the American Heart Association and the American College of Cardiology on lifestyle management to reduce cardiovascular risk (35). Because cardiovascular disease and diabetes share some common risk factors, we referred diet variables in the present study to that in the China Kadoorie Biobank study for cardiovascular disease and diabetes, a large cohort consisting of 0.5 million adult Chinese (36, 37). Accordingly, the low-risk group was defined as those who ate vegetables and fruits every day but meat less than daily.

For general adiposity measured by BMI, the low-risk group was defined as those who had a BMI of 18.5 to 23.9 kg/m², based on the standard classification specific for the Chinese population (38). For central adiposity measured by waist circumference, the low-risk group was defined as those who had a waist circumference of less than 85 cm for men and less than 80 cm for women (39). Overall lifestyle was subsequently categorized as poor (having no more than 2 healthy lifestyle factors), intermediate (having 3 or 4 healthy lifestyle factors), and ideal (having 5 or 6 healthy lifestyle factors) groups.

Single-nucleotide polymorphism genotyping and imputation

In the present study, SNPs were selected if they achieved genome-wide significance (P < 5 × 10^–8) in large published GWAS meta-analyses (2-13). Preliminarily, 114 previously reported type 2 diabetes-related SNPs were selected. Because the top-ranked SNPs on the X chromosome were not available in the present study, we did not include these variants in the GRS (n = 2). We also excluded the SNPs in high linkage disequilibrium (r² > 0.8) with others (n = 16), and those with a minor allele frequency of less than 0.01 (n = 8). Ultimately 88 SNPs were included. Details of the SNPs are presented in an online repository (24).

SNP imputation was performed using Minimac3 (v2.0.1) software with 1000 Genomes Project data (ALL phase 3 integrated release version 5 haplotypes, May 2, 2013) as reference panel. Finally, there were 2 SNPs imputed among the selected 88 SNPs.

The genotypes of the selected 88 SNPs were derived from the GWAS scan with Affymetrix Genome-Wide Human SNP Array 6.0 chips and Illumima Infinium OmniZhongHua-8 Chips. The effect size of each SNP on diabetes risk was derived from the meta-analysis of 2-chip data. The genotyping procedure and quality control have been described in more detail in previous studies (40, 41). All the selected SNPs were in Hardy-Weinberg equilibrium (P > .001). The β values of individual SNPs on diabetes risk were derived from the additive model via logistic regression, adjusting for age, sex, education, marital status, hypertension, and hyperlipidemia.

Genetic risk score calculation

The effect sizes (odds ratios; ORs) of the selected 88 SNPs on diabetes ranged from 1.03 (β = 0.03) for rs4275659 in MPHOSPH9 to 1.57 (β = 0.45) for rs17584499 in PTPRD. Each individual’s simple GRS was calculated by directly adding up the number of risk alleles (0, 1, or 2) at each SNP. For weighted GRS, the number of risk alleles for each individual was summed up and then multiplied by the effect size (multivariable-adjusted β derived from the published studies) of the SNP on diabetes risk (14). If multiple effect sizes were reported in multiple studies, those in the study with the largest sample size were selected. The range of simple GRS and weighted GRS is from 67 to 114 (mean, 90.0; SD, 6.30), and 5.34 to 11.55 (mean, 8.39; SD, 0.85), respectively.

We also performed sensitivity analysis to calculate the weighted GRS based on the effect sizes (β) derived from the present cohort. The effect sizes (ORs) of the selected 88 SNPs on diabetes ranged from 1.001 (β = 0.001) for rs896854 in TP53INP1 to 1.32 (β = 0.28) for rs1153188 in DCD. The range of weighted GRS is from 4.63 to 9.04 (mean, 6.84; SD, 0.64).

Ascertainment of baseline and incident diabetes

The diagnosis of diabetes was on the basis of American Diabetes Association criteria (42) as meeting any of the following criteria in the follow-up interviews or laboratory examinations: 1) self-report of a physician’s diagnosis of diabetes, 2) fasting blood glucose level of 7.0 mmol/L or greater, 3) usage of diabetes medication (insulin or oral hypoglycemic agent), 4) hemoglobin A_1c level of 6.5% or greater, and 5) 2-h 75-g oral glucose tolerance test value of 11.1 mmol/L or greater. The incident diabetic cases were those that occurred after baseline survey but before the end of October 2013. Because the oral glucose tolerance test was not conducted in this study and hemoglobin A_1c levels were assayed only during follow-up in 2013, baseline diabetic cases were thereby ascertained based on 1 to 3 criteria; incident diabetic cases were defined based on criteria 1 to 4 at the first follow-up. A total of 1555 incident diabetic cases were diagnosed during the follow-up period in the present study. When assessing the association between GRS and diabetes risk, diabetes patients were defined as those who had diabetes at baseline or during the follow-up period.

Statistical analysis

All statistical analysis was performed using SPSS 13.0 software. Categorical variables were presented in percentages and compared by chi-square analysis. Continuous variables were expressed in means (SD) and compared by Student t test or analysis of variation unless otherwise specified.

A Cox proportional regression model was used to evaluate the associations of lifestyle score with incident diabetes risk. Lifestyle score was categorized into poor (0 to 2 healthy lifestyle factors), intermediate (3 or 4 healthy lifestyle factors), and ideal (5 or 6 healthy lifestyle factors) groups; the reference group was poor lifestyle category. Hazard ratios (HRs) with 95% CIs were calculated. Multivariable models were adjusted for age, sex, education, marital status, family history of diabetes, hypertension, and hyperlipidemia. Association of GRS with diabetes risk was assessed by logistic regression, adjusting for age, sex, education, marital status, hypertension, and hyperlipidemia. To evaluate whether the association of a healthy lifestyle with diabetes persists across different genetic profiles, GRS was divided into tertiles as low, intermediate, and high genetic risk. ORs with 95% CIs were calculated between different lifestyle categories and diabetes risk (the reference group was poor lifestyle in each genetic risk group). P values and P for trend were calculated when lifestyle score or GRS entered analysis models as grouped or continuous variables, respectively. The interaction between lifestyle categories and genetic risk groups was also tested. A 2-sided P value of less than .05 was considered to be statistically significant.

Results

Baseline characteristics of the study population according to lifestyle categories are summarized in Table 1. Overall, 3257 (17.1%) individuals had an ideal lifestyle, 10 021 (52.7%) individuals had an intermediate lifestyle, and 5727 (30.1%) individuals had a poor lifestyle. Participants who had an ideal lifestyle were younger, less likely to be hypertensive or hyperlipidemic, and had lower levels of blood pressure, waist circumference, BMI, fasting blood glucose, and triglycerides as well as a higher level of high-density lipoprotein cholesterol. Participants with available genetic data were older and more likely to have higher waist circumference and BMI, and be current smokers and alcohol drinkers. More details can be found in an online repository (24).

Association of lifestyle score with incident diabetes risk

When these lifestyle factors were considered jointly, the risk of developing diabetes decreased dramatically with an increasing number of the low-risk factors in all male and female participants (all P for trend < .001) (Table 2). After adjustment for age, sex, education, marital status, and family history of diabetes, compared with participants with a poor lifestyle, those with an intermediate lifestyle had a 27% lower risk of developing diabetes (HR, 0.73; 95% CI: 0.65-0.81); and those with an ideal lifestyle showed a 51% reduction in the risk of incident diabetes (HR, 0.49; 95% CI: 0.41-0.58). Additional adjustment for hypertension and hyperlipidemia did not materially alter the association (for intermediate lifestyle: HR, 0.77; 95% CI: 0.69-0.87; for ideal lifestyle: HR, 0.54; 95% CI: 0.45-0.65).

Characteristics of participants across genetic risk categories

Simple GRS was classified into quintiles as Q1: less than 84, Q2: 84 to approximately 87, Q3: 88 to approximately 90, Q4: 91 to approximately 94, and Q5: 95 or greater. Table 3 presents the characteristics of 7344 individuals with available genetic data in the whole cohort. Compared with individuals at the lowest quintile of simple GRS, those at the upper quintiles had higher levels of fasting blood glucose and total cholesterol, and had a higher proportion of family diabetes history, hypertension, and hyperlipidemia. We also tested whether there was a difference in lifestyle score among different GRSs, and no significant difference was obtained (data not shown).

Association of genetic risk score with diabetes risk

Weighted GRS was classified into quintiles as Q1: less than 7.76, Q2: 7.76 to approximately 8.26, Q3: 8.27 to approximately 8.66, Q4: 8.67 to approximately 9.12, and Q5: 9.13 or greater. Associations of simple GRS and weighted GRS with diabetes risk are shown in Fig. 1 and in an online repository (24). In a multivariable-adjusted model, compared with participants at the lowest quintile of simple GRS, those at higher quintiles had a 35% (OR, 1.35; 95% CI: 1.13-1.62), 51% (OR, 1.51; 95% CI: 1.26-1.81), 105% (OR, 2.05; 95% CI: 1.73-2.43), and 156% (OR, 2.56; 95% CI: 2.16-3.03) higher risk of diabetes, respectively (P for trend < .001). Per SD increment of simple GRS was associated with a 39% higher risk of diabetes (OR, 1.39; 95% CI: 1.32-1.47). Compared with participants at the lowest quintile of weighted GRS, those at higher quintiles had a 19% (OR, 1.19; 95% CI: 1.01-1.39), 50% (OR, 1.50; 95% CI: 1.28-1.76), 68% (OR, 1.68; 95% CI: 1.44-1.97), and 121% (OR, 2.21; 95% CI: 1.89-2.59) higher risk of diabetes, respectively (P for trend < .001). Per SD increment of weighted GRS was associated with a 34% higher risk of diabetes (OR, 1.34; 95% CI: 1.28-1.41). In sensitivity analysis, quintiles of weighted GRS calculated based on the effect sizes of the present study were Q1: less than 6.25, Q2: 6.25 to approximately 6.66, Q3: 6.67 to approximately 7.02, Q4: 7.03 to approximately 7.43, and Q5: 7.44 or greater, and similar results were obtained. Additional adjustment for the top 3 principal components of ancestry did not materially change the results.

Associations of lifestyle score and genetic risk with diabetes risk

As shown in Fig. 2 and in an online repository (24), in each group of simple GRS and weighted GRS, increasing number of healthy lifestyle factors were associated with lower diabetes risk (all P for trend < .05). Among participants in the low simple GRS group, ideal lifestyle was associated with a 31% lower (OR, 0.69; 95% CI: 0.48-0.98) risk of diabetes. Among participants who had an intermediate genetic risk, intermediate and ideal lifestyle was reduced to a 28% (OR, 0.71; 95% CI: 0.58-0.89) and 51% (OR, 0.49; 95% CI: 0.35-0.67) risk of diabetes, respectively. Among those in the high simple GRS group, intermediate and ideal lifestyle were separately reduced to a 18% (OR, 0.82; 95% CI: 0.69-0.98) and 45% (OR, 0.55, 95% CI: 0.42-0.72) risk of diabetes. Accordingly, an ideal lifestyle was reduced to a 38% (OR, 0.62, 95% CI: 0.46-0.85), 48% (OR, 0.52, 95% CI: 0.38-0.71), and 43% (OR, 0.57, 95% CI: 0.42-0.76) risk of diabetes across low, intermediate, and high weighted GRS groups. Sensitivity analysis showed similar results. An ideal lifestyle reduced to a 35% (OR, 0.65, 95% CI: 0.46-0.92), 47% (OR, 0.53, 95% CI: 0.40-0.71), and 49% (OR, 0.51, 95% CI: 0.38-0.69) risk of diabetes respectively across low, intermediate, and high weighted GRS groups, which were calculated based on the effect size derived from the present cohort. Notably, the nonsignificant trend of greater proportional risk reduction observed here might be due to the sample size difference among these groups, which might affect power in different groups. Interaction between GRS and lifestyle score was also examined, and results can be found in an online repository (24). Considering that there might have been loss of power by defining arbitrary cutoffs for GRS, we analyzed the interaction between continuous lifestyle score, individual lifestyle components, and GRS. No significant interactions were found (all P for interaction > .05). Based on the limited population with available GWAS data at baseline, we also prospectively examined whether the effect of lifestyle on diabetes risk was modified by genetic factors and analyzed the interaction of lifestyle and GRS on incident diabetes; similar results were obtained and the results are presented in an online repository (24).

Discussion

In this large, population-based cohort study, we found that a healthy lifestyle and higher genetic risk were independently associated with diabetes risk. Within different genetic risk groups, an increasing number of healthy lifestyle factors was associated with a substantially reduced diabetes risk. However, no interaction effects were observed between genetic risk and lifestyle factors.

Observational epidemiological studies showed that lifestyle factors play an important role in diabetes risk. The Nurses’ Health Study with 16 years’ follow-up showed that 91% of diabetes risk in middle-aged women could be attributed to being overweight, having a poor diet, a lack of exercise, and smoking (25). A study conducted in 2 Finnish cohorts showed that 5 modifiable lifestyle factors defined by BMI, physical activity, smoking, alcohol consumption, and vitamin D explained 82% of diabetes risk during a 10-year period (43). Interventional studies in China, India, Europe, and the United States added a higher level of evidence that lifestyle intervention and pharmacological treatment could significantly lower diabetes risk as well as genetic effects on it (15-22). Findings in the present study provide consistent evidence for the association of a healthy lifestyle with reduced diabetes incidence in a different ethnic population.

This study provided some evidence that should be highlighted. First, inherited genetic variation and lifestyle factors independently contribute to a susceptibility to diabetes, which is consistent with previous studies (21, 44-47). However, examining multiple lifestyle factors together with equal weight in the present study might cover up a significant interaction between the effect of a single lifestyle factor and genetic risk on diabetes. Second, a healthy lifestyle was associated with a relatively lower risk of diabetes across different categories of genetic risk. Several studies conducted in Europe and the United States found that a healthy lifestyle could lower genetic risk of cardiovascular disease (48, 49), but whether the beneficial effects persist for diabetes and among other populations remains unknown. This study provided evidence for the association of multiple healthy lifestyle factors in combination with diabetes risk reduction regardless of different genetic risk profiles, supporting public health endeavors that emphasize a healthy lifestyle for everyone. However, owing to cost-conscious and limited resources, an alternative approach is to target intensive lifestyle modification to those at high genetic risk. Third, the current study showed that lifestyle did not interact with GRS on diabetes risk, in line with earlier studies that investigated a single lifestyle risk factor and genetic variants on diabetes risk (19, 20, 45, 50). Our results provide evidence that a healthy lifestyle might powerfully modify diabetes risk regardless of an individual’s genetic risk profile, suggesting that benefits of lifestyle modification apply to individuals both at low and high genetic risk. Especially from a public health perspective, it is more worthwhile to target individuals with high genetic risk for lifestyle intervention (21). In the present study, individuals with the highest quintile of GRS had the highest risk of diabetes. However, an ideal lifestyle could also modify diabetes risk even in populations with a higher genetic risk of diabetes.

Major strengths of this study are the large sample size and the prospective design of the Dongfeng-Tongji Cohort Study. Anthropometric information was measured rather than self-reported, providing more accurate estimates of BMI and waist circumference. Moreover, bias was minimized by carefully controlling potential confounding factors.

Several limitations should also be considered. On one hand, measurement errors of self-reported lifestyle factors are inevitable, and sex disparity should also be acknowledged even though we did not find interaction between sex and lifestyle factors on diabetes risk. Lifestyle factors were measured only once at baseline and might not necessarily reflect long-term exposure. Additionally, lack of more detailed dietary data limited us to include more dietary factors in the lifestyle factors, and behavioral changes before or after other illnesses occur might affect risk estimates. Moreover, although more than 80 validated genetic polymorphisms were included in the polygenic risk score, it still explained only a very small proportion of diabetes risk. Inclusion of more variants may provide more useful evidence in future analyses. On the other hand, the study population is relatively older, and results may not generalize to other population especially young population. Even though the present study provides evidence that a healthy lifestyle can reduce diabetes risk regardless of an individual’s genetic risk profile, the relatively small sample size with genetic data in the present study and observational study design limit us to draw a positive conclusion on the interaction of GRS and lifestyle on diabetes prevention; therefore, the results need to be validated in other populations with a larger sample size. Finally, examining multiple lifestyle factors together with equal weight in the present study limited us to draw conclusions about individual risk factors. In addition, lifestyle score calculated according to the actual weight of each lifestyle component might be more precise; however, population-specific complexity might be introduced.

This study showed that both combined lifestyle factors and polygenic risk score were independently associated with diabetes risk. A healthy lifestyle was associated with reduced diabetes risk across different categories of genetic risk. Behavioral lifestyle change should be encouraged for all populations through comprehensive multifactorial approaches.

Abbreviations

Abbreviations
 
• BMI

  body mass index

 
• GWAS

  genome-wide association study

 
• SNP

  single-nucleotide polymorphism

Acknowledgments

The authors would like to thank all the study participants for taking part in the present Dongfeng-Tongji Cohort Study as well as all the volunteers for assisting in collecting the samples and data.

Financial Support: This work was supported by the National Natural Science Foundation (Grants NSFC-81522040, 81473051, and 81230069); the Program for HUST Academic Frontier Youth Team; National Key R&D Program of China (2017YFC0907501); the 111 Project (No. B12004); the Program for Changjiang Scholars; the Innovative Research Team at the University of Ministry of Education of China (No. IRT1246); China Postdoctoral Science Funding (2018M630869); and the China Medical Board (No. 12-113).

Author Contributions: All authors conceived the study concept and design. X.H., Y.W., H.H., J.W., Z.L., F.W., and T.L. conducted the data analysis. All authors interpreted the data. X.H., Y.W., H.H., and J.W. wrote the first draft of the article. All authors reviewed and edited the report.

Additional Information

Disclosure Summary: The authors have nothing to disclose.

Data Availability: All data generated or analyzed during this study are included in this published article or in the data repositories listed in References.

References