Different Associations between High Density Lipoprotein Cholesterol and Cardiovascular Diseases in Diabetic and Non-diabetic People: A Prospective Community-based Study

Background: Experimental studies found that the functionality of high density lipoprotein cholesterol (HDL-C) may be lost in the presence of diabetes mellitus (DM). We prospectively tested whether DM modied the association between HDL-C concentrations and cardiovascular disease (CVD). Methods: Included were 91,354 Chinese adults (8,244 participants with DM and 83,110 participants without DM) without use of lipid-lowering drugs and free of CVD or cancer at baseline (2006). The primary endpoint of interest was a composite of CVD (myocardial infarction, ischemic stroke and hemorrhagic stroke). Cumulative average HDL-C concentrations were calculated from all available HDL-C measures at baseline (2006) and during the follow-up period (2008, 2010, 2012 and 2014). Results: During a mean of 10.4 year of follow-up, there were 5,076 CVD events identied. Presence of DM signicantly modied the association between HDL-C concentrations and CVD risk (P interaction =0.003). The association between HDL-C concentrations and CVD followed a U-shaped curve in individuals without DM (P nonlinearity <0.001). The adjusted hazard ratio (HR) of CVD was 1.25 (95% condence interval (CI): 1.06-1.48) for HDL-C concentrations <1.04 mmol/L and 1.80 (95% CI: 1.56-2.07) for HDL-C concentrations >2.07 mmol/L, relative to the lowest risk group (HDL-C concentrations of 1.30-1.42mmol/L). In participants with DM, higher HDL-C concentrations were associated with a higher risk of CVD, in a dose-response manner (P nonlinearity =0.44; P trend <0.001). The adjusted HR of CVD was 1.63 (95%CI: 1.20-2.20) for HDL-C concentrations >2.07 mmol/L, relative to HDL-C concentrations of 1.30-1.42mmol/L. Conclusion: High cut-offs deciles. HDL-C 1.04 to 1.30, 1.30 to 1.42, 1.43 to 1.55, 1.55 to 1.81, 1.81 to 2.07 and >2.07 mmol/L[30, 31]. Reference groups were selected as the ones with the lowest hazard ratios (HRs) of CVD outcomes in the spline Cox regression model. All analyses were adjusted for age, sex, body mass index, waist circumference, cigarette smoking, alcohol consumption, monthly income, education, occupation, physical activity, baseline blood pressure status, systolic blood pressure and diastolic blood pressure during follow-up, family history of myocardial infarction, stroke, hypertension or DM, estimated glomerular ltration rate, circulating concentrations of LDL-C, triglyceride, glucose and high sensitivity C-reactive protein. Joint effects of HDL-C concentrations (5 groups) and DM status (yes/no) were further examined in a secondary analysis. Interactions of HDL-C/DM, with age, sex, cigarette smoking, hypertension and triglyceride, in relation to CVD were assessed by likelihood ratio testing, adjusting for aforementioned covariates. Several sensitivity analyses were conducted to test the robustness of results.

Interestingly, experimental studies have found that the functionality of HDL may be lost in the diabetes mellitus (DM) state [18], suggesting that DM may diminish the predictive capability of HDL-C, which has not been examined in a large-scale epidemiologic studies [19]. We thus prospectively compared the association between HDL-C concentrations and CVD risk in the presence or absence of DM in a community-based cohort consisting of 91,354 Chinese adults. These participants were followed for over 10 years. Lipid parameters and covariates were repeatedly measured every 2 years, which enables us to capture the long-term lipid patterns in these individuals.

Methods
Study Populations The design and methods of the Kailuan study have been described previously [20]. This is a prospective, community-based study of 101,510 Chinese adults (81,110 men and 20,400 women) aged 18 years or older who were residents of the Kailuan community in Tangshan Hitachi, Tokyo, Japan). Fasting blood glucose concentrations were determined with hexokinase/glucose-6-phosphate dehydrogenase method (BioSino Bio-technology and Science Inc., Beijing, China). The coe cient of variation using blind quality control specimens was <2.0% for fasting blood glucose. Fasting HDL-C and LDL-C concentrations were measured with direct test method (Mind Bioengineering Co. Ltd, Shanghai, China) and with upper limits of detection of 12.90 and 3.88 mmol/L, respectively [21].
Triglyceride with enzymatic colorimetric method (Mind Bioengineering Co. Ltd, Shanghai, China), and creatinine with sarcosine oxidase assay method (BioSino Bio-technology and Science Inc., Beijing, China). The intra-and inter-assay variable coe cients for each measurement were <10%. Information on physician diagnosed DM and use of glucose lowering medication was collected via a questionnaire in 2006 and was updated every two years during the follow-up. Individuals who were considered DM in the current analysis is they had a fasting blood glucose concentration of ≥7mmol/L, physician-diagnosed DM or self-reported use of glucose lowering medication [22]. We also conducted a secondary analysis using updated DM information during the follow-up. Ascertainment of incident death and CVD events The primary endpoint of interest was a composite of myocardial infarction, ischemic stroke or hemorrhage stroke. All death and CVD events and vital status were identi ed by directly contacting participants' family, or reviewing medical records or death certi cates at the Municipal Social Insurance Institution and all the 11 Kailuan Hospitals' Discharge Register, which included all the Kailuan study participants. Information regarding past medical history of CVD, coded as International Classi cation of Diseases-10th Revision was collected via biennial questionnaire since 2006 [23]. Speci cally, myocardial infarction was diagnosed based on cardiac symptoms, positive cardiac biomarkers or electrocardiography [24]. Ischemic stroke and hemorrhagic stroke were de ned as neurological de cit of cerebrovascular cause that lasted more than 24 hours or a signi cant lesion detected by computed tomography or magnetic resonance imaging [25]. A blinded panel of three experienced cardiologists reviewed the medical records, discharge summaries and death certi cates from local hospitals or vital statistics o ces. Assessment of potential covariates Age, sex, lifestyle (e.g., cigarette smoking, alcohol consumption and habitual physical activity), family history of CVD and socio-demographic data (e.g., education level, occupation and monthly salary) were collected in 2006 and updated every 2 years using questionnaires, as detailed previously [26]. Cigarette smoking was divided into 4 categories: "never", "former", "occasional (current smoking <1cigarette/d),'' or ''daily (current smoking≥1 cigarette/d)". Alcohol intake was assessed by a questionnaire in 2006 and participants were categorized as never, former, occasional (current drinking <1time/day) or daily (current drinking ≥1time/day). Physical activity was captured by questionnaire according to activities associated with occupation and leisure time and was classi ed as "inactive", "moderately active" and "vigorously active" [26]. Blood pressure, waist circumference and anthropometric measurements (weight and height) were measured according to standardized procedures by trained study staff [20]. Body mass index was calculated by dividing weight in kilograms by the square of the height in meters. Blood pressure was measured with a random zero sphygmomanometer after individuals had been seated quietly for at least 5 min. Hypertension was de ned as systolic blood pressure of ≥140mmHg or diastolic blood pressure of ≥90 mmHg or a history of physician-diagnosed hypertension or if on anti-hypertensive agents; pre-hypertension was de ned as systolic blood pressure of 120-139mmHg or diastolic blood pressure of 80-90mmHg; normotension was de ned as systolic blood pressure of <120mmHg and diastolic blood pressure of <80mmHg [27]. High sensitivity C-reactive protein concentrations were measured using a high-sensitivity particle-enhanced immunonephelometry assay (Cias Latex CRP-H, Kanto Chemical Co. Inc., Japan) [20]. Estimated glomerular ltration rate was calculated according to the Chronic Kidney Disease Epidemiology Collaboration equation considering creatinine, sex, and age [28]. Statistical analysis The person-time for each participant was accumulated from the nishing date of the baseline survey to whichever came rst: CVD events, death or termination based on the 5th, 27.5th, 50th, 72.5th and 95th percentiles of HDL-C concentrations that can offer adequate t of the model. The association between HDL-C concentrations and CVD events was further examined using Cox regression analysis with corresponding 95% con dence intervals (CIs) based on prede ned groups with clinically meaningful cut-offs of HDL-C concentrations and deciles. HDL-C concentrations were categorized as <1.04, 1.04 to 1.30, 1.30 to 1.42, 1.43 to 1.55, 1.55 to 1.81, 1.81 to 2.07 and >2.07 mmol/L [30,31]. Reference groups were selected as the ones with the lowest hazard ratios (HRs) of CVD outcomes in the spline Cox regression model. All analyses were adjusted for age, sex, body mass index, waist circumference, cigarette smoking, alcohol consumption, monthly income, education, occupation, physical activity, baseline blood pressure status, systolic blood pressure and diastolic blood pressure during follow-up, family history of myocardial infarction, stroke, hypertension or DM, estimated glomerular ltration rate, circulating concentrations of LDL-C, triglyceride, glucose and high sensitivity Creactive protein. Joint effects of HDL-C concentrations (5 groups) and DM status (yes/no) were further examined in a secondary analysis. Interactions of HDL-C/DM, with age, sex, cigarette smoking, hypertension and triglyceride, in relation to CVD were assessed by likelihood ratio testing, adjusting for aforementioned covariates. Several sensitivity analyses were conducted to test the robustness of results. Because circulating HDL-C concentrations may uctuate due to impending CVD events and yield reverse causal association, we excluded incident CVD events occurring during the rst 2 years of follow-up. Given that potential pharmacologic effect might confound the results, in a separate sub-group analysis, we excluded the participants who used glucose lowering drugs (n=3,618). Statistical analyses were performed using STATA12.0 (STATA Institute) and a 2-sided p value <0.05 was considered signi cant.

Results
Baseline characteristics for individuals with or without DM separately are shown in Table 1 and further divided by HDL-C concentrations cut-points in ESM Table 1 &2. Individuals with high HDL-C concentrations were more likely to be older, women, drinker, to have hypertension, low body mass index and waist circumference, high level of systolic blood pressure and monthly salary and low concentrations of triglyceride and high sensitivity C-reactive protein. In individuals without DM, high HDL-C concentrations were associated with low concentrations of fasting blood glucose and high proportion of white collars. In contrast, high HDL-C concentrations were associated with high level of diastolic blood pressure, low level of education and physical activity and low proportion of white collars in diabetic people.
During a mean follow-up period of 10.4 years, we identi ed 5,076 new cases of CVD. Presence of DM signi cantly modi ed the association between HDL-C concentrations and CVD risk (P interaction =0.003).
As for CVD subtypes (e.g., myocardial infarction, ischemic and hemorrhage stroke), only the association between HDL-C concentrations and ischemic stroke followed a U-shaped curve while low HDL-C concentrations failed to predict myocardial infarction or hemorrhagic stroke occurrence in individuals without DM (ESM Fig. 1-4). The association between HDL-C concentrations and CVD risk did not materially change when we excluded participants who developed CVD events during the rst 2 years of follow-up (ESM Fig. 5) or using updated DM information (ESM Fig. 6) or those who used glucoselowering drugs (only in diabetic people) (ESM Fig. 7). There were no signi cant interactions between HDL-C/DM and age, sex, smoking and hypertension in relation to CVD risk (P interaction >0.1 for all).

Discussion
In our large community-based prospective study including 91,354 Chinese adults during a mean of 10.4 years of follow-up, we found that the presence and absence of DM modi ed the association between circulating HDL-C concentrations and future CVD risk, after adjusting for conventional confounders. A dose-response association between high HDL-C concentrations and high CVD risk was observed in participants with DM. In contrast, in participants without DM, high CVD risk was observed in low and high HDL-C concentrations groups.
HDL-C has long been regarded as the "good" cholesterol. An inverse association between HDL-C concentrations and CVD risk has been reported previously [32]. However, recent studies implied the association between HDL-C concentrations and CVD events might not be linear through the full range of HDL-C concentrations, but follows a U-shaped curve [12,33]. In these studies, the HDL-C concentrations with lowest mortality risk ranged from 0.8 to 2.4 mmol/L, which covered that in our study (1.33mmol/L, 95%CI: 1.30-1.36mmol/L). The potential mechanism interpreting the U-shaped association between HDL-C concentrations and CVD risk remains unclear. HDL-C may have a biphasic effect, as high HDL concentrations paradoxically enhanced senescence of endothelial progenitor cells and impaired tube formation [34]. HDL-C exhibited predictive capability in a certain range of concentrations, below or beyond which higher may result in adverse outcomes. The 2019 ESC/EAS Guidelines for the management of dyslipidaemias indicated that cardiovascular hazard of an extreme high HDL-C concentrations that are inappropriate for risk assessment [35].
Our ndings found that DM attenuated the salutary effect of HDL-C. These results may partly interpret nding from recent clinical trials that increasing circulating HDL-C concentrations did not lower the CVD risk in high-risk population [14][15][16]. The relationship of HDL-C concentrations with CVD could be modi ed by hyperglycemia and insulin resistance, making it a poorer protector in individuals with DM. Of note, recent experimental studies found that hyperglycemia may modify the HDL function [36]. Increased glycation/glycoxidation of HDL impairs its reverse cholesterol transport ability and diminishes its antiatherogenic capacity [37]. Circulating HDL-C concentrations represent the sum of multiple HDL subpopulations, each with different biological capacities, particularly in special states such as DM. This suggested that the hazard of diabetic dyslipidemia for CVD risk might be better explained by number and size of circulating HDL particles rather than HDL-C concentrations [38]. DM is related to a profound redistribution of cholesterol between HDL and LDL particles, towards less of the larger, atheroprotective HDL subpopulations, and more of the smaller, lipid poor HDL particles generated, which could promote diabetic macrovascular complications [36]. Merely increasing circulating concentrations of HDL-C by pharmacological interventions or genetic conditions without any improvement of their function (e.g., cholesterol ux) and bene cial particles may not reduce CVD risk. In the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, recruiting 5518 patients with DM, pharmaceutic raising HDL-C concentrations did not reduce the CVD risk compared to simvastatin alone [39]. The CVD bene t was observed only in the subgroup of individuals with high triglyceride and low HDL-C concentrations, which was largely ascribe to lowering hypertriglyceridemia rather than raising HDL-C concentrations.
Strengths of the present study included a large sample size and a relative long follow-up period allow us to assess the capability of the joint effect of HDL-C concentrations and DM on the risk of CVD events. Cumulative average, based on repeated measures of lipid pro les at baseline and every 2 years during the follow-up period, could reduce the within-person assay uctuations and increase power to identify association independent of other confounders. Our study likewise has several limitations. We did not measure the HDL subclass distribution and the number and size of HDL particles, which are more closely aligned with HDL function. However, there remains a lack of cost-effective method measuring the size and composition of HDL particles in large-scale population study. Second, we did not collect information regarding DM subtype and thus cannot distinguish between type 1 and type 2 DM. However, the prevalence of type 1 DM was low and excluding individuals with any hypoglycemic treatment did not change results materially. Third, the observational nature of our study makes it di cult to determine the causal relationship of HDL-C concentrations and CVD outcomes. A possibility of bias due to potential unmeasured confounders could not be completely removed. Randomized clinical trials, with greater follow-up may provide higher quality evidence for HDL-based approach feasible in individuals with DM. Although the pharmacological tools currently available to improve HDL function in individuals with DM remained limited, recently developed apolipoprotein mimetics and reconstituted HDL hold much promise for CVD therapeutic strategy in individuals with DM[40-42]. Fourth, women may be underrepresented (~20%) in the Kailuan study, which challenged in generalizing these results to women although we did not nd a signi cant interaction between sex and HDL-C on mortality and CVD risk.

Conclusion
The relationship between HDL-C concentrations and CVD appeared to be modi ed by DM status. High HDL-C concentrations were paradoxically associated with increased risk of composite CVD outcomes in individuals with or without DM. However, low HDL-C concentrations failed to predict future CVD risk in individuals with DM. Further prospective studies conducted in populations with different ethnic groups are needed to con rm replicate our results.
Abbreviations LDL-C: Low density lipoprotein cholesterol; CVD: cardiovascular diseases; HDL-C: high density lipoprotein cholesterol; DM: diabetes mellitus; CI: con dence interval; HR: hazards ratio; ACCORD: Action to Control Cardiovascular Risk in Diabetes.

Declarations
Acknowledgements None.

Funding
The study was supported by the Shanghai Municipal Education Commission-Gaofeng Clinical Medicine Grant Support (20172004).

Consent for publication
Not applicable.

Con ict of Interest
No authors reported any disclosures.
Authors' contributions ZW, ZH, AL, SW and XG designed the study; CJ, SC, and FH performed the analyses; ZW, SC, SW and XG interpreted the results and wrote the manuscript; All authors read and approved the nal manuscript.

Availability of data and materials
The data can be available from the corresponding authors on reasonable request.

Ethics approval and consent to participate
The study protocol was approved by the ethics committee of Kailuan Hospital and followed the guidelines of the World Medical Association Declaration of Helsinki. Written informed consent was obtained from all the participants.  Abbreviations: DM, diabetes mellitus; BMI, body mass index; TG, triglyceride; WC, waist circumference; hs-CRP, high sensitivity C-reactive protein; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; SBP, systolic blood pressure; DBP, diastolic blood pressure; eGFR, estimated glomerular ltration rate; MI: myocardial infarction. To convert from mmol/L to mg/dL, divide by 0.0259 for HDL-C and LDL-C and 0.0113 for triglycerides.

Figure 3
Adjusted hazard ratios (HRs) and 95% con dence intervals (CIs) for cardiovascular diseases (CVD) risk among combined subgroups of high density lipoprotein cholesterol (HDL-C) clinical cut-offs and diabetes mellitus (DM) among entire cohort. The combined value of HDL-C concentrations with the lowest HRs of CVD and non-DM was used as a reference. All models were adjusted for age (years), sex (men or women), family history of myocardial infarction, stroke, diabetes mellitus or hypertension (yes or no), smoking status (never, former, occasionally or daily), body mass index (quartiles), waist circumference (quartiles), low density lipoprotein cholesterol (quartiles), triglyceride (quartiles), high sensitivity C-reactive protein Adjusted hazard ratios (HRs) and 95% con dence intervals (CIs) for cardiovascular diseases (CVD) risk among combined subgroups of high density lipoprotein cholesterol (HDL-C) clinical cut-offs and diabetes mellitus (DM) among entire cohort. The combined value of HDL-C concentrations with the lowest HRs of CVD and non-DM was used as a reference. All models were adjusted for age (years), sex (men or women), family history of myocardial infarction, stroke, diabetes mellitus or hypertension (yes or no), smoking status (never, former, occasionally or daily), body mass index (quartiles), waist circumference (quartiles), low density lipoprotein cholesterol (quartiles), triglyceride (quartiles), high sensitivity C-reactive protein (quartiles), systolic blood pressure (quartiles), diastolic blood pressure (quartiles), glucose (quartiles), estimated glomerular ltration rate (quartiles) , alcohol consumption (never, former, occasionally or daily),