Higher Long-term Visit-to-Visit Variability in Fasting Plasma Glucose Predicts New-Onset Heart Failure in the General Population

Background: Previous studies suggested an adverse association between higher fasting blood glucose (FBG) and heart failure. However, FBG values uctuate continuously over time, the association between FBG variability and the risk of heart failure is uncertain. Aims: We investigated the relationship between visit-to-visit variability in FBG and the risk of new-onset heart failure. Methods and Results: This was a population-based cohort study using the Kailuan dataset, which comprises of medical claims and a biennially health checkup information from a Chinese cohort. A total of 98 554 individuals (mean age: 53.63 years) who had at least two health checkups with FBG measurement between 2006 and 2012 without preexisting heart failure were included. FBG variability was calculated using the variability independent of the mean, coecient of variation, standard deviation, and average successive variability (ARV). Participants were divided into quartiles of ARV. Cox regression was used to identify heart failure. Over a mean follow-up of 6.27 years, 1218 individuals developed heart failure. The incidence of heart failure was 1.97 per 1000 person-years. After adjusting for baseline FBG and other potential confounders, individuals in the highest quartile of the ARV of FBG had 32.6% higher risk of developing heart failure compared to those in the lowest quartile (hazard ratio, 1.326; 95% condence interval, 1.120-1.570). This association remained signicant in patients with or without prevalent hypertension. In subgroup analyses, individuals who were younger (<65 years), without diabetes mellitus or chronic kidney disease, and with a body mass index<25 kg/m 2 experienced a higher risk of heart failure. Conclusions: Our data demonstrated that high FBG variability is independently associated with the development of new-onset heart failure. Future studies should explore whether measures to reduce variability can lead to improve clinical outcomes. eGFR, glomerular ltration heart cholesterol; TG, Hs-CRP, coecient of VIM, variability independent of SD, standard deviation; ARV: average


Introduction
Diabetes mellitus is becoming increasingly prevalent worldwide, placing signi cant burdens on many healthcare systems. Glycemic impairment, including in the nondiabetic range, is an independent risk factor for heart failure. [1] Hitherto, previous studies investigating the hyperglycemia-related complications have mainly relied on punctual assessment of blood glucose, which may not capture the true levels over time. Sustained hyperglycemia is the predominant condition in patients with diabetes, especially at its advanced stages. However, recent studies have revealed that not only constant hyperglycemia, but also large glucose uctuations, may in uence cardiovascular risk in diabetic patients with even more deleterious effects than sustained hyperglycemia. [2,3] Glycemic variability, an emerging marker of glycemic control, is a measurement that determines the uctuations in glucose over a de ned time interval. High glycemic variability appears to have more deleterious effects than sustained hyperglycemia in the pathogenesis of diabetic cardiovascular complications [4,5], with potential implications for cardiovascular disease. [6] Previously published studies have found that visit-to-visit variability in fasting blood glucose (FBG) is an independent predictor of incidence of left ventricular adverse remodeling in type 2 diabetes mellitus patients with ST-segment elevation myocardial infarction. [7] However, the association of visit-to-visit FBG variability with new-onset heart failure in non-diabetic patients is unknown. Therefore, the purpose of our study was to prospectively investigate the association between visit-to-visit FBG variability and the risk of new-onset heart failure in the Kailuan study cohort.

Methods
Population Page 3/16 The participants were recruited from Kailuan study (registration number: ChiCTR-TNC-11001489), a prospective cohort study established at the Kailuan General Hospital and ten other a liated hospitals, as described in detail elsewhere.
[8] Among these subjects, who had at least two health examinations biennially, were included for the evaluation of FBG variability (n=100 305). Those subjects with pre-existing heart failure were excluded (n=477). Finally, the remaining 98 554 participants were included in the nal analysis ( Figure 1). All participants received interviews with pre-de ned questionnaires, clinical examinations, and laboratory assessment. The study was performed according to the guidelines of the Declaration of Helsinki and was approved by the Ethics Committee of the Kailuan General Hospital (Approval Number: 2006-05). Written informed consent was obtained from all participants.

Data collection
Information on the demographic variables of age, sex, smoking status, alcohol drinking status, and use of antidiabetic medications was collected using standard questionnaires. Body weight and height were measured by trained nurses according to the standard protocols, allowing calculation of the body mass index (BMI) (weight in kilograms divided by height in meters squared). Blood samples for the measurement of FBG were drawn after an overnight fast. All blood measurements were performed using an automatic analyzer following standard operating procedures and were subjected to regular quality control. The baseline estimated glomerular ltration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation. [9] chronic kidney disease (CKD) was de ned as an eGFR (calculated using the Modi cation of Diet in Renal Disease formula) under 60 mL/min at the baseline of health examination. [10] Measures of fasting plasma glucose variability FBG variability was determined as visit-to-visit intra-individual FBG variability during at least 4 years before the baseline. Visit-to-visit FBG variability was de ned as variability in FBG recorded in at least 2 checkups. The alternative visit-to-visit of FBG metrics include: (1) the coe cient of variation (CV)=(SD/x×100%), (3) the variability independent of the mean (VIM), which is calculated as 100×SD/mean β , where β is the regression coe cient based on natural logarithm of SD on natural logarithm of mean, and All the aforementioned measures of variability have been previously described. [11,12] Given that there are no internationally agreed upon gold standard measures of FBG variability in general and of visit-to-visit in particular, we opted to include a wide range of measures, which would potentially capture different aspects of FBG variability. [13] Follow-up and primary outcomes The participants were followed from the baseline examination until December 31, 2016, or until the diagnosis of heart failure. The method for diagnosis of heart failure was described previously. [14] Incident heart failure was de ned in accordance with the criteria of the European Society of Cardiology [15] on the basis of clinical symptoms, echocardiography, chest X-ray, and electrocardiography. The diagnosis of heart failure was also con rmed based on an annual review of each patient's medical records by a team of experienced cardiologists.

Statistical analysis
Statistical analyses were performed using SAS software (SAS Institute, Inc., Cary, NC; version 9.4). Continuous variables with normal and skewed distributions were described as the mean ± standard deviation (SD) and the medians with interquartile ranges, respectively, whereas categorical variables were described as frequencies and percentages. Missing data for baseline covariates were dealt with using multiple imputation. χ 2 test for linear trend was used for categorical variables, and one-way analysis of variance for linear trend and Kruskal-Wallis test was performed for continuous variables with normal and skewed distributions, respectively. Visit-to-visit FBG variability was assessed by both quartiles and continuous variables. Univariate survival analysis was performed by the Kaplan-Meier method and the log-rank test.
Person-years were calculated from the date that the baseline interview was conducted to the date when the rst occurrence of heart failure was detected, date of death, or date of participating in the last interview in this analysis, whichever came rst.
The association of FBG variability with incidence of heart failure was determined by calculation of hazard ratios (HR) and 95% con dence intervals (CI) from Cox regression, after veri cation of the proportional hazard assumption with Schoenfeld residuals. All participants were divided into quartiles using ARV. Model 1 was adjusted for age and sex. Model 2 was further adjusted for history of diabetes mellitus, hypertension, whether to take hypotensive drugs and hypoglycemic drugs, baseline FBG, resting heart rate, LDL-C and HDL-C. Model 3 was further adjusted for SBP (systolic blood pressures), DBP (diastolic blood pressures), BMI, high-sensitivity C-reactive protein (Hs-CRP), smoking, alcohol abuse and physical activity.
Sensitivity analysis was conducted to test the robustness of our results, using three exclusion criteria. To assess the possibility that these inverse associations were driven by other risk factors of heart failure, we excluded participants who had a diagnosis of myocardial infarction and arrhythmia at baseline. In addition, we excluded individuals use of β-adrenergic blocking, angiotensin-converting enzyme inhibitors (ACEI), angiotensin receptor blockers (ARB), aldosterone antagonist to assess whether the main analysis could be affected by these agents which improve heart failure outcomes.

Baseline characteristics
A total of 98 554 subjects (mean age, 53.63±12.34 years; 78.83% male) were included in this study. Over a mean of 6.27±0.91 years of follow-up, 1218 subjects were newly diagnosed with heart failure (1.97 per 1000 person-years).
Baseline characteristics of subjects according to quartiles of the ARV are presented in Table 1, and those according to VIM, CV, SD are presented in Table S2-S4. Subjects with the highest variability of ARV (ARV-Q4) were older, had higher proportion of male subjects, and had a higher number of comorbidities, higher proportion of current smokers, more subjects with low monthly income, higher body weight, BMI, SBP and DBP.
The baseline characteristics strati ed by the outcome status of new onset heart failure are summarized in Table S1. Approximately 11.00% of participants with incident heart failure took hypoglycemic medications. Participants with incident heart failure were older, mostly male (85.39%), more obese, had higher rates of cardiovascular risk factors such as diabetes, smoking, and worse renal function compared to those without incident heart failure (p<0.01, respectively). The resting heart rate was signi cantly higher (76.75±12.56 vs. 73.60±10.46, p<0.01), and regular exercise was more frequently observed in participants with incident heart failure compared with those without it (73.56% vs. 70.08%, p<0.01).
FBG variability and the risk of heart failure As FBG variability increased from quartiles 2, 3 and 4, the incidence of heart failure increased from 1.41 to 1.55, 1.94, and 2.97 per 1000 person-years, respectively. The incidence rate of heart failure signi cantly increased with a higher variability of FBG. The associations between FBG variability and new-onset heart failure are shown in Table 2 Figure 2. Individuals in the top quartile of variability of FBG experienced a higher risk than participants of the other quartiles over the 6.27 follow-up years for heart failure (log-rank test, P<0.001).

Sensitivity analyses
The following sensitivity analyses were performed. In the rst sensitivity analysis, we excluded all individuals with prevalent MI (n= 471). Among the remaining patients, Table 2  Second, we excluded individuals diagnosed with arrhythmias (eg. atrial brillation, atrial utter, atrial premature beat, ventricular ectopic beats, and/or atrioventricular block) at baseline (n= 1331). In second sensitivity analysis using a higher ARV and SD of FBG measures to estimate variability, the association between visit-to-visit variability in FBG with the risk of incident heart failure was consistent with that observed in the primary analysis. Higher CV and VIM was not signi cantly associated with incident heart failure.
The last sensitivity analysis demonstrated similar to those of the main analysis when individuals use of β-adrenergic blockers, ACEI, ARB and aldosterone antagonists were removed (n= 1340).

Subgroup analyses
Subgroup analysis for age demonstrated that the association of FBG ARV with heart failure was stronger among those <65 years compared with those ≥65 years [age<65 years: HR 1.356 (95% CI 1.074-1.712); age ≥65 years: HR 1.276 (95% CI 0.998-1.633), interaction p-value 0.0471] (Table S5). There was no interaction by sex, hypertension, diabetes, BMI or eGFR on the association of FBG ARV with incident heart failure ( Figure 3). Moreover, ARV of FBG was associated with the risk of heart failure across both patients with or without prevalent hypertension. In subgroup analysis, the impact of high FBG variability on heart failure development was stronger in high-risk subjects, who were younger (<65 years), without diabetes mellitus or chronic kidney disease, and with a BMI<25 kg/m 2 . Strati ed analysis of subjects according to VIM, CV, SD are presented in Table S6-S8, the results were similar those using ARV.

Discussion
In this study, we examined the in uence of FBG variability on new onset heart failure. The primary ndings from this large, prospective, population-based cohort study showed that higher visit-to-visit variability of FBG was associated with an increased risk for heart failure events over 6.27 years of follow-up. The positive association of FBG variability and heart failure was signi cant and greater in magnitude among people without diabetes compared with those with diabetes. The direction and the magnitude of these associations were roughly consistent across measures of variability. These ndings were obtained after adjustment for traditional risk factors for heart failure, including a prior history of diabetes mellitus, hypertension, high BMI, resting heart rate and FBG levels. Our ndings add to the growing body of evidence on the risk of cardiovascular disease with higher visit-to-visit variability in FBG and highlight the importance of glycemic uctuations. FBG variability has emerged as a measure that can accurately capture the pathological processes that mediate different complications. However, there remains an extensive debate about FBG variability as a risk factor for cardiovascular disease. Our study expands on the ndings from previous studies which have also shown an association of FBG variability with new onset of cardiovascular disease. [3,6,16,17] However, these previous studies have been limited by their small size, [16,17] the restriction to individuals with diabetes only[17], main outcomes not including heart failure, [3,6] a variable interval between visits at which glycemia was assessed (ranging from days to months),[18] and the methodologies used to estimate variability.
[19] All these factors not only would in uence variability in FBG but also could impact the strength of the association with outcomes.
The exact mechanisms linking increased FBG variability to an increased risk of adverse outcomes are unknown, but there are several hypotheses[20-23]. Glucose oscillation enhanced human tubule-interstitial cell growth and collagen synthesis and accelerated apoptosis in human endothelial cells more than exposure to a constantly high glucose concentration.
[24] Oxidative stress was shown to be the key player in mediating endothelial damage and dysfunction.
[25] Several other studies have since con rmed that oscillating glucose concentrations, via increased oxidative stress, can adversely affect the cells of different organs.
[26] These pathophysiological changes may trigger cardiovascular disease.
We did not identify any association between FBG variability and increased risk of heart failure in women. A possible explanation for the sex-speci c observations may be the higher estrogen level in this subgroup.[27, 28] Studies have suggested that higher estrogen levels can up-regulate the expression of anti-oxidant genes [29, 30] and inhibit the expression of pro-in ammatory genes.
[31] Therefore, because of estrogen protection, women may be more adaptable to FBG variability than men, which may reduce the risk of cardiovascular disease mediated by increased FBG variability.
The lack of association of FBG variability with outcomes among those with diabetes may be due to a number of factors, including the use of diabetes medications in this subgroup that may have blunted glycemic variability, as well as the possible underestimation of the number of diabetes cases, as we did not have data on HbA1c or 2-h post load glucose, especially as 2-h post load glucose variability may be strongly associated with outcomes.
[32] It is also possible that long-term glycemic variability matters more among those without diabetes, and among those with diabetes short-term variability is a predictor of outcomes. Among those with diabetes, the lack of association of FBG variability with heart failure is apparently consistent with results of previous studies. [16] Strengths and limitations The strengths of the present study include its prospective design and large sample size, and that the association between visit-to-visit FBG variability and risk of heart failure was determined for the rst time. However, several limitations should be noted. Firstly, the observational study design limited the ability to establish causal pathways. Secondly, we did not have data on glycemic markers other than FBG, such as HbA1c or 2-hour post-prandial glucose, and long-term variability in HbA1c or in 2-hour post-prandial glucose may provide additional value for risk strati cation. Thirdly, although we adjusted antidiabetic drugs, we could not fully eliminate some unmeasured confounding factors, which may blunt the FBG variability, such as diet management. Fourthly, FBG values were only measured for two or three times and the median follow-up time was approximately 6 years, which is relatively short. Finally, all the participants in our study were Chinese, which may limit the generalizability of the results of our investigation to other ethnic groups.

Conclusions
Our ndings demonstrate that higher visit-to-visit variability in FBG is associated with an increased risk of new onset heart failure, especially among people without diabetes. Future studies are needed to further elucidate the mechanisms underlying a high level of FBG variability and whether measures that reduce this variability can lead to improve clinical outcomes.

Acknowledgments
We thank all study participants, their relatives, the members of the survey teams at the 11 regional hospitals of the Kailuan Medical Group for contributing to the dataset.
Authors'contributions TL, SLW, YYW and WWQ conceived and designed the study. YYW, WWQ and NZ analyzed, interpreted the data and drafted the rst version of the manuscript. All authors have interpreted the data, critically revised, provided intellectual contributions and approved the nal version of the manuscript. TL is the guarantor. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Funding
This work was supported by the grants from Tianjin Natural Science Foundation (20JCZDJC00340, 20JCZXJC00130 to TL).

Availability of data and materials
The data that support the ndings of this study are available from Kailuan General Hospital, but restrictions apply to the availability of these data, and so are not publicly available. Data can be accessed via application to Kailuan General Hospital.  Tables   Table 1 Variables Abbreviations: CKD, chronic kidney disease; eGFR, estimated glomerular ltration rate; Scr, serum creatinine levels; BMI, body mass index; DBP, diastolic blood pressure; SBP, systolic blood pressure; FBG, fasting blood glucose; HDL-C, highdensity lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; RHR, resting heart rate; TC, total cholesterol; TG, triglyceride; Hs-CRP, High-sensitivity C-reactive protein; CV, coe cient of variance; VIM, variability independent of mean; SD, the standard deviation.

Ethics approval and consent to participate
*Triglyceride levels, Hs-CRP levels, the SD of fasting plasma glucose levels, the variability independent of the mean and the average real variability are presented as median (interquartile range). Model 3, Model 2 + SBP, DBP, BMI, Hs-CRP, smoking status, drinking status, and physical exercise a Sensitivity analysis was excluded the participants with myocardial infarction event at baseline, and adjusted for variables in model 3 b Sensitivity analysis was excluded the participants with arrhythmia (eg, atrial brillation, atrial utter, atrial premature beat, ventricular ectopic beats, and/or atrioventricular block) at baseline, and adjusted for variables in model 3 c Sensitivity analysis was excluded the participants use of β-adrenergic blocking, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers and diuretics at baseline, and adjusted for variables in model 3 Abbreviations: CI, con dence interval; HR, hazard ratio; RHR, resting heart rate; DBP, diastolic blood pressure; SBP, systolic blood pressure; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; BMI, body mass index; Hs-CRP, High-sensitivity C-reactive protein; FPG, fasting plasma glucose. Figure 1 Selection of the study population.  Hazard ratio for incident heart failure in different subgroups in overall population. Strati ed analysis by age, gender, hypertension, diabetes mellitus, chronic kidney disease and body mass index was performed. HR signi cantly differed between older (≥65 years) and younger (<65 years) participants (interaction p-value <0.05). In this analysis, HR was adjusted for age, gender, systolic and diastolic blood pressure, resting heart rate, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, baseline fasting plasma glucose levels, high-sensitivity C-reactive protein, body mass index, history of diabetes mellitus, history of hypertension, antihypertensive medications use, antidiabetic medications use, current smoking, alcohol intake, and physical exercise. Boxes indicate the hazard ratio, limit lines indicate the 95% con dence interval, and the vertical line (at hazard ratio 1) indicates no difference in the hazard ratios between heart failure and no heart failure. Note: CI, con dence interval; HR, hazard ratio; Bold values were the P values for interaction.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.