Association of Serum 25-Hydroxyvitamin D With Cardiovascular Outcomes and All-Cause Mortality in Individuals With Prediabetes and Diabetes: Results From the UK Biobank Prospective Cohort Study

Diabetes has become a great threat to global public health. The International Diabetes Federation estimated that 463 million adults had diabetes and 373.9 million adults had prediabetes worldwide in 2019 (1). Individuals with prediabetes and diabetes have twofold and two-to fourfold increased risks of cardiovascular disease (CVD) and mortality compared with normal controls, respectively (2,3). Furthermore, CVD mortality accounts for approximately one-half of all deaths in diabetes (1). It has gained great importance in identifying modifiable cardiovascular risk factors in individuals with prediabetes and diabetes for the prevention of diabetes cardiovascular complications and premature death.

Vitamin D status has been linked to various nonskeletal diseases, such as diabetes (4,5), hypertension (6), CVD (7–10), and cancer (11,12). Numerous epidemiological studies showed that higher serum 25-hydroxyvitamin D (25[OH]D) levels were associated with lower risk of CVD and mortality in the general population (8,13,14). These findings were largely based on general populations; however, the evidence regarding the relationships between vitamin D status and CVD events and mortality among those with prediabetes and diabetes is limited and needs to be clarified in this group of individuals, in whom vitamin D deficiency is particularly common (15–22). For instance, one prospective observational study of 245 women with type 2 diabetes failed to find an association between vitamin D and all-cause mortality (17). In contrast, one recent prospective study of 6,329 adults with diabetes from the Third National Health and Nutrition Examination Survey (NHANES III) showed that higher serum 25(OH)D levels were significantly associated with lower CVD and all-cause mortality (20). However, there were always some discrepancies in these previous studies because of limited sample size (<5,000), inconsistent definition of vitamin D deficiency, and insufficient adjustment of confounding covariates. In particular, to our knowledge, prospective high-quality evidence focusing on the relationships between vitamin D and CVD and mortality in the population with prediabetes is scarce. Therefore, data from larger prospective studies are still needed to fill the knowledge gaps.

In this large population-based cohort study from UK Biobank, we aimed to examine the associations of serum 25(OH)D concentrations with the incidence of CVD events, specific CVD events, and all-cause mortality in individuals with prediabetes and diabetes.

The UK Biobank is a large population-based prospective cohort study that recruited >500,000 participants from the general population between 2006 and 2010 across England, Scotland, and Wales (23,24). Participants completed a touchscreen questionnaire containing health information, took various physical measurements, and reported medical conditions. Blood samples were collected for biochemistry markers assay.

In the current analysis, we excluded participants who had no available data on serum 25(OH)D concentrations (n = 54,153) and those with known cancer (n = 31,451) or CVD (n = 29,402) at baseline. In total, 67,789 participants with prediabetes and 24,311 participants with diabetes were included in the current analysis (Supplementary Fig. 1).

The UK Biobank received ethical approval from the North West Multi-center Research Ethics Committee, Manchester, U.K. (REC reference for UK Biobank 11/NW/0382), and all participants provided written informed consent.

Blood samples of participants in the UK Biobank were collected at the time of recruitment. Serum 25(OH)D concentrations (nmol/L) were measured by chemiluminescent immunoassay direct competitive analysis on a DiaSorin, Ltd, LIASON XL. Details about serum biomarker measurements and assay performances have been described in the online UK Biobank Showcase (https://www.ukbiobank.ac.uk). The assay of serum 25(OH)D was registered with an external quality assurance scheme, and assay performance was externally verified via the results returned from participation in these schemes. The average coefficients of variation of 25(OH)D derived from internal quality control samples were <10%.

The main outcomes of the current study included the incidence of CVD events (i.e., coronary heart disease [CHD], stroke, heart failure, and ischemic and hemorrhagic stroke), CVD mortality, and all-cause mortality. Data on hospital admissions were collected regularly through linkages to Health Episode Statistics, the Patient Episode Database, and the Scottish Morbidity Records. Information on death was obtained from National Health Service Digital for participants in England and Wales and from the National Health Service Central Register, part of the National Records of Scotland, for participants in Scotland. At the time of analysis, health outcome data were available up to 30 November 2020. For the analyses of cardiovascular outcomes and deaths, we used this date as the end of follow-up unless death or admission occurred first. Follow-up for incident events and mortality outcomes was defined using the following ICD-10 codes: I20–I25 for CHD, I50 for heart failure, I60–I64 for stroke, I60–I62 for hemorrhagic stroke, I63 for ischemic stroke, and I00–I99 for death resulting from CVD.

Participants with prediabetes were defined by the presence of impaired fasting glucose and/or hemoglobin A_1c (HbA_1c) 5.7–6.4% (39–47 mmol/mol) according to the 2021 diagnostic criteria from the American Diabetes Association guideline (25). The definition of diabetes included diagnosed diabetes by a physician during the touchscreen questionnaire, use of insulin and hypoglycemic medication, or presence of HbA_1c ≥6.5% (48 mmol/mol) at baseline.

Information on sociodemographic factors (age, sex, ethnicity, and household income), lifestyle habits (smoking status, drinking status, and physical activity), physical measurements (weight, waist circumference, and BMI), medications (antihypertensive drug use, cholesterol-lowering medication, aspirin use, insulin treatment, and vitamin D supplementation), and medical history (hypertension) was obtained from the baseline questionnaire. Biochemical measurements including LDL cholesterol and HbA_1c were measured at baseline. Season of vitamin D assessment was categorized by the month that a participant attended assessment: spring (March, April, or May), summer (June, July, or August), autumn (September, October, or November), and winter (December, January, or February). Additional details of covariate measurements can be found in the UK Biobank online protocol (https://www.ukbiobank.ac.uk).

Statistical analyses were performed using SAS 9.4 (SAS Institute, Inc.). Two-sided P values <0.05 were considered statistically significant. Baseline characteristics were presented as mean ± SD or median (interquartile range) for continuous variables and number (percentage) for categorical variables. Multivariable linear regression analyses were used to investigate the association between serum 25(OH)D concentrations and cardiovascular risk factors. Multiple imputation with chained equations was performed for participants with missing covariate data, assuming data were conditionally missing at random.

Participants were divided into four groups according to baseline serum 25(OH)D concentrations based on the Endocrine Society Clinical Practice Guidelines: severe deficiency (<25.0 nmol/L), moderate deficiency (25.0–49.9 nmol/L), insufficient (50.0–74.9 nmol/L), and sufficient (>75.0 nmol/L) (26). The generalized linear model was used to examine the baseline characteristics of participants among groups.

Cox proportional hazard models were used to calculate the hazard ratios (HRs) and 95% CIs for the associations between serum 25(OH)D concentrations and cardiovascular outcomes and mortality. The test for linear trend was based on the variable containing the median value in each group as a continuous variable. The multivariable model (model 1) was adjusted for age and sex; model 2 was further adjusted for ethnicity (White, Black, Asian, Chinese, mixed, or other), household income (<£18,000, £18,000–30,999, £31,000–51,999, £52,000–100,000, or >£100,000), season of vitamin D assessment (spring, summer, autumn, or winter), BMI, smoking status (never, former, or current), drinking status (never, former, or current), physical activity, antihypertensive drug use (yes or no), cholesterol-lowering medication (yes or no), aspirin use (yes or no), LDL cholesterol, vitamin D supplementation (yes or no), and prevalent hypertension (yes or no). For participants with diabetes, model 2 was additionally adjusted for HbA_1c level (<7.0% [<53 mmol/mol] or ≥7.0% [≥53 mmol/mol]) and insulin treatment (yes or no). Moreover, the dose-response associations of baseline serum 25(OH)D concentrations and cardiovascular outcomes and all-cause mortality were examined using restricted cubic spline regression with three knots, adjusted for the confounding variables mentioned above.

Stratified analyses were also performed according to age (≤60 or >60 years), sex (male or female), ethnicity (White or non-White), smoking status (never, previous, or current), drinking status (never, previous, or current), prevalent hypertension (yes or no), and BMI (<30 or ≥30 kg/m²). To investigate whether the associations between serum 25(OH)D concentrations and cardiovascular outcomes and mortality differed by these stratification variables, the potential effect modification was examined using the interaction models. Sensitivity analyses were performed by excluding participants with CVD events or death within 2 years of follow-up.

Among the 67,789 participants with prediabetes, the mean (SD) age was 58.2 (7.5) years, and 36,820 (54.3%) were women; among the 24,311 participants with diabetes, the mean (SD) age was 58.5 (7.5) years, and 10,631 (43.7%) were women. The baseline cardiometabolic biomarker characteristics of the study population according to serum 25(OH)D levels are also shown in Table 1. Overall, the mean (SD) serum 25(OH)D concentration was 47.0 (20.5) nmol/L and 43.5 (20.4) nmol/L in participants with prediabetes and diabetes, respectively. Of participants with prediabetes and diabetes, 58.1% and 65.3% had deficient vitamin D (<50 nmol/L), respectively. Participants with prediabetes and diabetes who had higher serum 25(OH)D levels were more likely to be older, women, and White and were less likely to be obese or current smokers. Compared with the participants with lower serum 25(OH)D levels, participants with prediabetes and diabetes who had higher serum 25(OH)D levels had significantly lower levels of body weight, waist circumference, BMI, HbA_1c, total cholesterol, triglycerides, and LDL cholesterol and had higher levels of HDL cholesterol. In the multivariable regression analyses, serum 25(OH)D levels were associated with multiple cardiovascular risk factors (Supplementary Table 1).

During a median of 11.5 years of follow-up, 7,421 cardiovascular events, 840 deaths resulting from cardiovascular events, and 4,534 all-cause death events were documented among the 67,789 participants with prediabetes, including 5,453 CHD events, 1,761 heart failure events, and 1,453 stroke events (ischemic stroke n = 966; hemorrhagic stroke n = 428). Table 2 shows the associations of categories of serum 25(OH)D concentrations with the outcomes. After adjustment for age and sex, serum 25(OH)D levels were also inversely associated with incident type 2 diabetes, cardiovascular events, CVD mortality, and all-cause mortality in a dose-response fashion (all P for trend < 0.05) (Table 2 and Supplementary Table 2). In the multivariable adjusted models (model 2), HRs and 95% CIs across categories of serum 25(OH)D concentrations (from lowest to highest) were 1.00 (reference), 0.86 (0.81–0.93), 0.72 (0.67–0.78), and 0.62 (0.55–0.71) for type 2 diabetes (P for trend < 0.001); 1.00, 0.85 (0.79–0.90), 0.83 (0.77–0.89), and 0.78 (0.71–0.86) for cardiovascular events (P for trend < 0.001); 1.00, 0.65 (0.54–0.78), 0.61 (0.50–0.74), and 0.43 (0.32–0.59) for CVD mortality (P for trend < 0.001); and 1.00, 0.77 (0.71–0.84), 0.65 (0.60–0.72), and 0.66 (0.58–0.75) for all-cause mortality (P for trend < 0.001) (Table 2 and Supplementary Table 2).

With regard to subtypes of cardiovascular outcomes, the highest category of 25(OH)D levels was significantly associated with lower risks of CHD (HR 0.79; 95% CI 0.71–0.89), heart failure (HR 0.66; 95% CI 0.54–0.81), and stroke (HR 0.75; 95% CI 0.61–0.93) in the multivariable adjusted models (model 2). Furthermore, the highest category of 25(OH)D levels was significantly associated with a lower risk of ischemic stroke (HR 0.75; 95% CI 0.57–0.97) and hemorrhagic stroke (HR 0.66; 95% CI 0.45–0.98).

In addition, nonlinear relationships were observed for all outcomes after multivariable adjustment (all P nonlinearity < 0.001) (Fig. 1). Elevated serum 25(OH)D concentrations were inversely associated with reduced risks of cardiovascular events, CHD, and all-cause mortality, and L-shaped associations of serum 25(OH)D concentrations with risk of cardiovascular events, CHD, and all-cause mortality were observed, leveling off up to ∼50–60 nmol/L.

Over a median of 11.4 years of follow-up, the following events and deaths were recorded among the 24,311 participants with diabetes: 4,454 cardiovascular events, 646 deaths resulting from cardiovascular events, and 2,667 all-cause deaths, including 3,420 CHD events, 1,308 heart failure events, and 835 stroke events (ischemic stroke n = 605; hemorrhagic stroke n = 185). Table 2 shows the associations of categories of serum 25(OH)D concentrations with the outcomes. After further adjustment for age and sex, serum 25(OH)D levels were inversely associated with the risk of incident cardiovascular events, CVD mortality, and all-cause mortality in a dose-response fashion (all P for trend < 0.05). In the multivariable adjusted models (model 2), HRs and 95% CIs across categories of serum 25(OH)D concentrations (from lowest to highest) were 1.00 (reference), 0.88 (0.81–0.95), 0.82 (0.75–0.90), and 0.80 (0.70–0.92) for cardiovascular events (P for trend < 0.001); 1.00, 0.64 (0.53–0.78), 0.48 (0.38–0.61), and 0.58 (0.41–0.82) for CVD mortality (P for trend < 0.001); and 1.00, 0.73 (0.66–0.80), 0.59 (0.52–0.66), and 0.60 (0.51–0.71) for all-cause mortality (P for trend < 0.001).

Likewise, with regard to subtypes of cardiovascular outcomes, the highest category of serum 25(OH)D levels was significantly associated with lower risks of CHD (HR 0.80; 95% CI 0.68–0.93) and heart failure (HR 0.70; 95% CI 0.55–0.90) after multivariable adjustment, but the associations with incident stroke (HR 0.92; 95% CI 0.69–1.23) and ischemic and hemorrhagic stroke were not significant (HR 0.96; 95% CI 0.67–1.37 and HR 0.71; 95% CI 0.39–1.30, respectively). Similarly, nonlinear relationships were observed for all outcomes after multivariable adjustment (all P nonlinearity < 0.001) (Fig. 2). The L-shaped associations of serum 25(OH)D concentrations with risks of cardiovascular events, CVD mortality, and all-cause mortality leveled off at ∼50–60 nmol/L.

Stratified analyses were conducted according to potential CVD risk factors. For participants with prediabetes, the associations between serum 25(OH)D concentrations and incident cardiovascular events were stronger among women (P for interaction = 0.03) and current smokers (P for interaction <0.001) (Supplementary Table 3). Likewise, this association was stronger in older (P for interaction = 0.04) and obese (P for interaction = 0.03) individuals in the group of participants with diabetes (Supplementary Table 3). Furthermore, the associations between serum 25(OH)D concentrations and the incidence of CVD mortality were stronger among White individuals (P for interaction = 0.02) in the group of participants with prediabetes (Supplementary Table 4). Of note, the significant interactions were also observed between serum 25(OH)D concentrations and smoking status with regard to risk of all-cause mortality in participants with prediabetes and diabetes (P for interaction = 0.03 and 0.04, respectively) (Supplementary Table 5). Furthermore, no significant interactions were observed between serum 25(OH)D concentrations and risk of progression from prediabetes to type 2 diabetes in participants with prediabetes (P for interaction > 0.05) (Supplementary Table 6). In sensitivity analyses, the results were similar after excluding participants who developed CVD or death events during the first 2 years of follow-up (Supplementary Table 7).

This large prospective cohort study found that higher serum 25(OH)D concentrations were associated with reduced risk of incident cardiovascular outcomes and all-cause mortality in patients with prediabetes and diabetes. Furthermore, these inverse associations were independent of risk factors, including age, sex, ethnicity, household income, season of vitamin D assessment, BMI, smoking status, drinking status, physical activity, antihypertensive drug use, cholesterol-lowering medication, vitamin D supplementation, LDL cholesterol, hypertension, and other confounding factors. In addition, we found that the inverse associations of higher serum 25(OH)D concentrations with incident cardiovascular events and all-cause mortality were significantly modified by smoking status.

To our knowledge, this is the first large population-based prospective analysis to investigate the associations between serum 25(OH)D concentrations and CVD outcomes and all-cause mortality among individuals with prediabetes. Numerous epidemiological and prospective cohort studies have shown that low serum 25(OH)D concentrations are associated with increased risk of CVD events and mortality in older individuals and the general population (27–30). Meta-analysis studies have also shown that low 25(OH)D levels are associated with increased risk of ischemic heart disease, myocardial infarction, CVD mortality, cancer mortality, and all-cause mortality (12,31). However, prospective high-quality evidence regarding the relationships between vitamin D status and CVD and mortality in the population with prediabetes is scarce. Although previous studies have reported that low serum 25(OH)D levels are associated with increased risk of type 2 diabetes among adults with prediabetes (5,32), a recent randomized controlled trial of 2,423 participants with prediabetes found no effect of vitamin D supplementation on diabetes prevention (33). In this large prospective study using a nationally representative sample of adults with prediabetes from the UK Biobank, we found that high serum 25(OH)D levels were nonlinearly associated with low risks of cardiovascular outcomes, CVD mortality, and all-cause mortality in the population with prediabetes. Meanwhile, the dose-response relationship analysis indicated that the decreasing risk of cardiovascular events, CHD, and mortality in prediabetes appeared to level off at serum 25(OH)D concentrations of 50–60 nmol/L. Therefore, these findings underline the importance of monitoring vitamin D status and correcting vitamin D deficiency in cardiovascular health in the population with prediabetes. Future randomized clinical trials are needed to confirm these findings in adults with prediabetes, especially with vitamin D deficiency.

Although numerous studies have explored the associations between vitamin D status and CVD and mortality, the available results regarding vitamin D status and CVD and mortality risk in individuals with diabetes, who are at high risk of CVD, are insufficient and inconclusive. For instance, one prospective observational study of 245 women with type 2 diabetes failed to find an association between vitamin D and all-cause mortality (17). In contrast, another study from the NHANES III showed that higher serum 25(OH)D levels were significantly associated with lower risk of CVD and all-cause mortality in patients with type 2 diabetes (20). In addition, another cohort of 2,607 community-dwelling patients with type 2 diabetes also found that serum 25(OH) D levels were an independent predictor of future adverse CHD events (34). The discrepancy across studies may be due to small sample size, different types of vitamin D status assessment, different methods of categorizing serum 25(OH)D levels, and lack of dose-response relationship analysis. Moreover, the adjusted covariates were varied across studies, which could result in different degrees of residual confounding. For example, most previous studies did not adjust for season of vitamin D assessment, physical activity, use of vitamin D supplements, and presence of chronic diseases at baseline, which may significantly influence vitamin D status and CVD outcomes and mortality. In the current study, we found that higher serum 25(OH)D levels in the population with diabetes were significantly associated with reduced risks of cardiovascular events and CVD and all-cause mortality. The dose-response relationship analysis indicated a nonlinear curve, with decreases in cardiovascular events and mortality among the population with diabetes risk up to 50–60 nmol/L. These findings indicate that the associations of serum 25(OH)D with CVD risk and mortality in participants with prediabetes and diabetes were consistent with those in general population (28–30,35). Consistently, a recent nonlinear Mendelian randomization analysis conducted in the UK Biobank with 267,980 participants in the general population also reported that an L-shaped association of serum 25(OH)D with CVD risk was observed, and increased risk of CVD was largely restricted to individuals with low vitamin D status (36). In addition, one randomized placebo-controlled trial in which ∼13% participants had diabetes was conducted to examine the effects of vitamin D supplementation and reported no effect of vitamin D supplementation in the prevention of CVD (37). A possible reason for the null findings for vitamin D supplementation is that only 12.7% of participants had vitamin D levels <50 nmol/L. However, whether a target of 50 nmol/L can reduce the risk of CVD events and premature death in patients with diabetes needs to be confirmed in future clinical trials.

Notably, evidence regarding vitamin D status with specific subtypes of CVD, such as heart failure and stroke, was relatively limited, especially in the population with prediabetes and diabetes. In a cross-sectional study, higher circulating vitamin D concentrations were associated with better left ventricular systolic function and smaller left ventricular end-systolic diameter at baseline (38). A cohort study of 13,131 participants from NHANES III indicated that adults with inadequate serum 25(OH)D levels had a significantly higher risk of heart failure in the general population (39). Inconsistently, another prospective study of 3,731 men age 60–79 years showed no significant association of serum 25(OH)D levels with incident heart failure (40). Our results indicate that higher serum 25(OH)D levels both in the population with prediabetes and in that with diabetes were significantly associated with lower risk of heart failure. Similarly, a significant inverse relationship between serum 25(OH)D levels and CHD was also observed both in prediabetes and in diabetes, which was consistent with the previous study (34). In addition, our data showed that 25(OH)D levels were significantly associated with the incidence of stroke (ischemic and hemorrhagic stroke) in participants with prediabetes. Although the results did not reach statistical significance because of a relatively limited sample size of patients with diabetes, a similar relationship between serum 25(OH)D levels and stroke was observed in patients with diabetes. These findings might have implications for the prevention of stroke in clinical practice, especially in the population with prediabetes.

In addition, we found that the relationships between serum 25(OH)D levels and cardiovascular events and all-cause mortality were modified by smoking status in the populations with prediabetes and diabetes. Likewise, the significant relationships between serum 25(OH)D levels and major cardiovascular events were observed in older and obese patients with diabetes. Therefore, given the global burden and role of smoking and obesity in the prevention of CVD, monitoring vitamin D status and correcting vitamin D deficiency among those individuals may be particularly noteworthy.

Several potential mechanisms could explain the observed relationships between serum 25(OH)D levels and incidence of CVD and mortality. Vitamin D status is significantly related to multiple CVD risk factors, including obesity and hypertriglyceridemia (41). Furthermore, evidence has indicated that vitamin D is a negative regulator of the renin-angiotensin-aldosterone system in vivo (42). In addition, previous studies found that vitamin D can reduce inflammation, suppress matrix metalloproteinase 2/9 formation, inhibit foam cell formation, and suppress macrophage cholesterol uptake (43,44). These data suggest a protective role of vitamin D in the development of CVD and are consistent with results from previous clinical studies. However, future studies are needed to illuminate the mechanisms of vitamin D in the improvement in cardiovascular health and mortality among individuals with prediabetes and diabetes.

Our study has several strengths. First, the UK Biobank study is a well-established population-based prospective cohort that has a large sample size, and the number of cardiovascular events and deaths over a long-term follow-up provided sufficient power for the analyses in the current study. Second, abundant information was available on socioeconomic characteristics, lifestyle habits, vitamin supplementation, comorbidity, medication, season of vitamin D status assessment, and other covariates, which may enhance the validity of the conclusions through adjustment for a multitude of potential confounding factors in this study. Finally, serum 25(OH)D concentrations in the UK Biobank were measured by a standard reliable method, which allowed us to conduct detailed dose-response analyses in this study.

Our study also has some potential limitations. First, potential reverse causality might exist in the current study. However, participants with CVD events or death at baseline or within 2 years of follow-up were excluded from the analysis, and the results did not change materially, which supports the robustness of our findings. Second, although many potential confounding factors were adjusted for in the analyses, we cannot exclude the role of residual confounding resulting from errors in the measurement of covariates and other factors (i.e., medication) that were not assessed in the cohort. Third, participants without vitamin D measurements at baseline were excluded from this study, which may affect the generality of our findings. Fourth, a small number of patients with prediabetes and diabetes would be excluded in this analysis because of lack of available data on 2-h postprandial glucose in the UK Biobank. Finally, we were unable to analyze the role of parathyroid hormone in the relationship between vitamin status and CVD and the relationship between dynamic 25(OH)D levels and cardiovascular events and mortality because of the lack of parathyroid hormone and repeated 25(OH)D measurements.

In conclusion, our results indicate that higher serum 25(OH)D levels were significantly associated with lower risks of cardiovascular events and mortality in adults with prediabetes and diabetes. Furthermore, these associations were partly modified by smoking status. These findings highlight the importance of monitoring and assessing vitamin D status and correcting vitamin D deficiency in the prevention of CVD and mortality among adults with prediabetes and diabetes.

Acknowledgments. The authors are grateful to the participants of UK Biobank.

Funding. This research was conducted using the UK Biobank resource under application 68376. This study was supported by grants from the National Key Research and Development Project (2018YFA0800404), National Natural Science Foundation of China (81970736), and Key-Area Clinical Research Program of Southern Medical University (LC2019ZD010 and 2019CR022). H.Z. was partially supported by a grant from the Outstanding Youths Development Scheme of Nanfang Hospital, Southern Medical University (2017J005).

The funders of the study had no role in study design, conduct, data collection, analysis, interpretation of data, approval of the manuscript, or decision to submit the manuscript for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. P.Z., D.G., B.X., C.H., S.Y., and D.L. acquired, analyzed, or interpreted the data. P.Z., D.G., B.X., and J.L. performed statistical analysis. P.Z., D.G., and H.Z. drafted the manuscript. P.Z. and H.Z. were responsible for study conception and design. S.Y., X.W., and H.Z. supervised the study. P.Z., W.W., W.L., Y.D., and K.L. critically reviewed the manuscript for important intellectual content. D.L., Y.H., and H.Z. provided administrative, technical, or material support. H.Z. obtained funding. H.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.