Several in vivo and in vitro studies have demonstrated the effect of glucose variability (GV) on adverse cardiovascular events (Fig. 1). The primary mechanism is suggested to be the activation of oxidative stress and production of inflammatory cytokines that induce endothelial damage and dysfunction. Research on human coronary artery endothelial cells under fluctuating glucose levels has shown an increase in tumor necrosis factor-α, intercellular adhesion molecule-1, and interleukin-6 [8,9]. Oxidative stress produced by GV leads to apoptosis through the nuclear factor-erythroid 2-related factor 2 (Nrf2)/heme oxygenase-1 (HO-1) pathway [10]. Overproduction of reactive oxygen species (ROS) in mitochondria by GV induces vascular endothelial cell growth factor (VEGF) production and triggers undesirable angiogenesis [11]. GV also induces dysfunction of endothelial nitric oxide (NO) synthase, leading to vasoconstriction from decreased NO levels [12].

Costantino et al. [12] reported that epigenetic modification of p66Shc, the adaptor protein, causes inadequate angiogenesis and vascular damage in type 2 diabetes mellitus (T2DM). Some studies suggested microRNA (miRNA) produced in glucose fluctuating condition influence endothelial dysfunction. Guo et al. [13] indicated miRNA-1273-3p is related to GV-induced autophagy, attenuation of cell proliferation, endothelial dysfunction. La Sala et al. [14] revealed upregulated miRNA-185 in high glucose fluctuation inhibit elevation of glutathione peroxidases-1 which normally suppress ROS damage. Furthermore, glucose fluctuation contributes to activated monocyte and macrophage recruitment, resulting in vascular smooth muscle cell proliferation and migration [15]. In addition, high GV may cause platelet activation, aggregation, and autonomic dysfunction in endothelial cells [16].

Since the time Monnier et al. [7] demonstrated that not only sustained hyperglycemia but also glucose swing cause oxidative stress and accompanying microvascular and macrovascular complications, numerous studies have published reinforcing evidence (Table 1) [3,17-31] and attempted to find an appropriate index reflecting inter- and intra-day GV. In addition to the traditionally used standard deviation (SD) and coefficient of variation (CV) showing dispersion of data, various metrics have been reported and applied [32]. The mean of daily difference calculated using the glucose value of 2 consecutive days simultaneously reflects between-day glycemic variation, and Continuous Overlapping Net Glycemic Action (CONGA), measured by the SD of glucose value for the defined period of time, reveals within-day glycemic fluctuation. In addition, indices indicating risk and patient prognosis (lability index) and assessing long-term diabetic complications (time in range) have been used in several studies [16,33].

Early studies were conducted on patients who underwent percutaneous coronary intervention (PCI) or coronary artery bypass graft and measured the periprocedural glucose level of participants [23-25,34-36]. Fluctuating glucose levels are associated with major adverse cardiovascular events (MACEs), atrial fibrillation, and post-procedural complications. However, most of the studies were conducted in an in-hospital setting and used short-term glycemic variability. Subsequent studies performed secondary analysis of randomized clinical trials including T2DM patients, and reported an association between long-term GV and CVD risk [3,26-31]. Hirakawa et al. [3] reported that increased glycosylated hemoglobin (HbA1c) variability was associated with vascular events (hazard ratio, 1.64; 95% confidence interval [CI], 1.05 to 2.55); Zhou et al. [18] reported that high fasting blood glucose (FBG) variability increased CVD risk; and Wan et al. [37] demonstrated this relationship is applied to all age groups regardless of DM duration. The EMpagliflozin and daPAgliflozin in patients hospiTalized for acute decompensated Heart failure (EMPATHY) trial recently conducted in Japanese T2DM patients identified that adverse effects of intermittent hyperglycemia were enhanced in patients with HbA1c levels below 7% [19]. These findings indicate the harmfulness of glycemic variation not only in metabolically high-risk patients but also in relatively well-controlled T2DM patients or people whose metabolic risk is low.

Accumulating evidence suggests that glucose fluctuation increases CVD risk in T2DM or patients who undergo coronary artery procedures; however, the possibility of a reverse causality effect must be addressed. Two studies [20,38] demonstrated the relationship between intermittent hyperglycemia and high CVD using data from people without DM and previous CVD from the Korean National Health Insurance. Analysis of data from a Denmark private clinic cohort with non-diabetics also showed that HbA1c variability is a predictor of MACE [21]. In addition, Kim et al. [39] illustrated that increased FBG variability is predictive of gestational DM.

In selecting anti-diabetic agents, their effects on GV as well as improving hyperglycemia should be considered. Dipeptidyl peptidase-4 (DPP4) enzyme inhibitors, glucagon-like peptide 1 (GLP1) receptor agonists, and sodium-glucose co-transporter 2 (SGLT2) inhibitors might be desirable oral hypoglycemic agents (OHAs) that reduce GV. Kim et al. [40] reported that sitagliptin reduced GV compared to glimepiride in T2DM patients who used metformin. Bae et al. [41] demonstrated a significant reduction in MACE and SD of glucose in patients treated with teneligliptin compared with placebo. Although several clinical trials have also reported that DPP4 inhibitors blunted glucose fluctuation [42-44], other studies reported that DPP4 inhibitors were unsuccessful in reducing ASCVD risk [45,46]. The Variability of Glucose in Patients with Type 2 Diabetes Treated with Four Different Insulin Combination Regimens (VARIATION) study illustrated that GLP1 receptor agonists and basal insulin decreased GV and hypoglycemic events, compared to basal and bolus insulin or basal insulin plus OHAs [47]. The FLuctuATion reduction with inSUlin and GLP-1 Added together (FLAT SUGAR) trial also demonstrated that adding GLP1 receptor agonist to basal insulin and metformin is better than metformin plus basal/bolus regimen for GV improvement [48]. The superiority of DPP4 inhibitor and GLP1 receptor agonist on GV might be attributable to their intrinsic mechanism of action. While GV aggravation is induced by decline of endogenous insulin and insufficient glucagon suppression, DPP4 inhibitor and GLP1 receptor agonist, which keep incretin level, inhibit postprandial glucagon secretion as well as improve pancreatic-βcell function [49-51]. As for SGLT2 inhibitors, despite limited evidence of the superiority of GV to other OHA, Nomoto et al. [52] showed similar GV between dapagliflozin plus glargine and DPP4 inhibitor plus glargine. In addition, adding dapagliflozin to long-acting insulin and other OHA ameliorated GV in a study including 36 subjects from a Chinese hospital [53]. However, a recently published placebo-controlled, randomized study with 84 Korean patients reported that adding dapagliflozin to basal insulin plus OHA did not show superiority to placebo in terms of GV [54]. Future large-scale studies are needed to demonstrate the effects of SGLT2 inhibitors on GV.

Several theories explain the detrimental effects of lipid variability on CV risk. One hypothesis is that cholesterol variability causes transient hypercholesterolemia and fluctuation of cholesterol plaque composite and structure, inducing endothelial dysfunction and cholesterol plaque instability [55]. Clark et al. [56] demonstrated that high variability of low-density lipoprotein (LDL), non-high-density lipoprotein (HDL) cholesterol, and total cholesterol/HDL (measured by SD) was correlated with percent atheroma volume progression by analyzing nine clinical studies that used intravascular ultrasonography to assess coronary atheroma burden. Second, the existence of cholesterol variability implies general frailty and epiphenomenon of systemic conditions [57]. Another theory is that the adverse effects of lipid variability are linked to statin non-adherence [58,59]. Mann et al. [60] reported that visit-to-visit lipid variability was associated with statin non-adherence after adjusting for mean LDL, age, and sex using pharmacy data. Statin discontinuation induces a rebound phenomenon, and the favorable effect of statin is acutely lost when it is discontinued [58]. Patients who discontinued statin therapy after acute coronary syndrome showed higher mortality than those who did not.

Since the importance of lipid variability in coronary disease was reported in an observational study on air force officers [5], it took >30 years to shed light on lipid variability. The Framingham study showed that high LDL variability in a population without cardiovascular risk was associated with future CVD risk [61]. In addition, the Treating New Target (TNT) study reported that LDL variability in people with known CVD risk increases coronary and cardiovascular events by 16% and 11%, respectively, regardless of absolute LDL levels [62].

Factors that affect lipid variability have been investigated. The TNT study illustrated that 80 mg of atorvastatin lowered LDL variability than 10 mg of atorvastatin. In contrast, doses and types of statins showed no effect on lipid variability in several other studies [63,64]. Boey et al. [63] reported no significant differences in LDL variability among patients treated with simvastatin (10, 20, and 40 mg). The VOYAGER (an indiVidual patient data meta-analysis Of statin therapY in At risk Groups: Effects of Rosuvastatin, atorvastatin and simvastatin) meta-analysis [64] consists of 37 clinical studies using atorvastatin, simvastatin, and rosuvastatin and reported low relevance between lipid variability and types of statins. Discordant results were shown regarding the influence of previous statin use on the harmful effects of lipid variability. Most studies focused on patients with coronary artery disease (CAD) or underwent PCI, and Kim et al. [22] and Zhu et al. [65] excluded people with previous CV events to leave out possible reverse causality. The outcomes were in line with those of previous studies showing the association between lipid variability and CV risk. Furthermore, Kim et al. [58] illustrated that the adverse effect of cholesterol variability was magnified in patients without conventional risk factors and in patients who did not take lipid-lowering medication. However, a Korean nationwide population-based cohort study conducted with statin-naïve healthy young people [66] revealed no significant association between lipid variability and CV risk. Moreover, total cholesterol variability augments the risk of atrial fibrillation [67] and dementia [68] which increases morbidity and mortality in the elderly population.

Future well-designed studies addressing improvements on several points are required (Fig. 2). The intervals and numbers of blood lipid measurements must be standardized. Smith et al. [69] reported that lipid variability tends to increase with longer daily, weekly, and monthly measurement interval. The appropriate interval and frequency of cholesterol measurement for optimal variability estimation that reflects CVD risk should be investigated [55]. Controversies on proper statistical methods and variability indices must also be addressed. A large number of previous studies compared the highest quartile (Q4) to the lowest quartile (Q1) or high quartile (Q4) to low quartiles (Q1–3); such approaches might exaggerate the association [70]. In addition, Cox regression is adequate to compare between-group variability but not within-individual variability or correlation across repeated measurements. Multilevel or mixed linear regression models and joint modelling are suggested as proper tools for this type of analysis [70]. Blood lipid level variability was calculated using the following formulas: CV, average successive variability [ASV=∑(Xi–Xi+1)/n], SD, or variation independent of mean [VIM=100%×(SD/meanβ), where β is the regression coefficient]. Although VIM is less relevant to the mean than SD and CV, controversy over association with mean value and adequacy for variability estimation still exists [70]. The frequency of deviation outside the designated range or the greatest absolute change is suggested as an alternative index [55]. Future studies are also required to define the normal range of cholesterol variability so that clinicians can pursue an optimal approach to reduce lipid variability. Proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitor which has a long half-life, is expected to maintain a stable cholesterol level. Lipid-lowering agents and treatment strategies that lower not only mean cholesterol levels but also variability are required [37].

Blood pressure (BP) fluctuation is necessary to maintain homeostasis and to provide adequate organ perfusion, but consistent blood pressure variability (BPV) reflects dysfunction of regulatory function, leading to pathological problems. Baroreflex dysfunction [71], prolonged sympathetic activation, and increased arterial stiffness cause short-term BP variations [72,73]. Seasonal changes and behavioral factors, such as sleep disturbance or work-related stress, can cause long-term BPV [74,75]. Periodic high BP induced by these factors induces shear stress on the vessel wall and cardiovascular system, which consequently increases intima-media thickness, vascular dysfunction, and atherosclerosis progression [76]. Furthermore, patients’ poor adherence to drugs and inappropriate titration of BP medication lead to visit-to-visit BPV [75]. In addition, Kim et al. [22] suggested that general frailty caused by systemic conditions might play a role in intermittent high BP.

Parati et al. [75] suggested the classification of BPV according to time frame as very-short-term (beat-to-beat), short-term (day-to-night, hour-to-hour), mid-term (over various days), and long-term (over months, over seasons, over clinic visits). Short-term BP fluctuations from ambulatory BP monitoring is considered to cause target organ damage, particularly left ventricular hypertrophy [77], and predict cardiovascular events and death [78]. Studies on home-monitored BP fluctuation (mid-term) were few, but they proved that BPV was associated with cardiovascular events and mortality [79-82]. Research over 19 years discovered the predictive value of home BPV on cardiovascular events over clinic-based BPV [83]. Several clinical trials and cohort studies identified that visit-to-visit BP fluctuation was predictive of CADs, myocardial infarction (MI), and related mortality, and this association was confirmed in several meta-analyses [22,75,84-87]. Parati et al. [75] indicated that long-term BPV is more strongly associated with CVD than short-term BPV, whereas Meng et al. [84] illustrated that within-visit systolic BPV was inappropriate for CVD prediction. Kim et al. [22] and de Havenon et al. [85] reported that adverse effects of BPV on cardiovascular risk and all-cause mortality were maintained in a general healthy population. However, the predictive value of BPV on CV events was not significant in patients with end-stage renal disease [86].

Several clinical trials have revealed that calcium-channel blockers (CCBs) and non-loop diuretics cause lower BPV than beta-blockers (BBs), angiotensin receptor blockers (ARBs), and angiotensin-converting-enzyme inhibitors (ACEi) [88-90]. A meta-analysis including 398 trials [91] reported that the ratio of variance (VR) increased in the order of CCB, thiazide, ACEi, ARB, and BB (VR=0.81, 0.87, 1.08, 1.16, and 1.17, respectively). The combination of antihypertensive medications has been researched in subsequent studies. Sato et al. [92] reported that adding amlodipine was better than adding thiazide in patients with inadequate BP control with ARB monotherapy in terms of low BP fluctuation. Rakugi et al. [93] demonstrated that the combination of ARB and CCB decreased BPV compared to a combination of ARB and diuretics. Nagai et al. [94] suggested that CCB plus diuretics are more favorable than ARB plus CCB, which is superior to ARB and diuretics in terms of lowering BPV. Although the mechanism of different effect on BPV among medications is uncertain, decrease of arterial stiffness via vasodilation by CCB and non-loop diuretics is suggested as a possible explanation [88,91,95].