Diabetes mellitus is now among the largest contributors to global cardiovascular morbidity. The International Diabetes Federation estimated that 536.6 million adults aged 20–79 years were living with diabetes in 2021, or 10.5% of the world’s adult population, and projected 783.2 million by 2045, with the steepest relative growth in middle-income countries [1]. Cardiovascular disease remains the dominant cause of death in this population, and the management of type 2 diabetes mellitus (T2DM) has correspondingly shifted from glucose lowering as an end in itself toward the reduction of cardiovascular risk [2].

For four decades, glycated haemoglobin (HbA1c) has been the single metric against which that management is judged. Its limitations are, however, structural rather than incidental. HbA1c reports an integrated average of glycaemia over roughly three months; it is silent about how that average was achieved [3]. Two patients with identical HbA1c values of 8.0% may differ entirely in their day-to-day glucose profiles: one stable, the other oscillating between hypoglycaemia and marked postprandial excursions. Population analyses of T2DM confirm that HbA1c is by far the most frequently used stratification variable, yet it captures only part of the risk that glycaemia conveys [4]. Trials of intensive glucose lowering have made this concrete. In the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, targeting near-normal HbA1c did not reduce major cardiovascular events and increased mortality, with hypoglycaemia requiring assistance substantially more frequent in the intensive arm [5]. A lower average, in other words, is not automatically a safer one.

Glycaemic variability (GV) was proposed to fill this gap. It describes the amplitude and frequency of glucose oscillations around the mean and is operationalised either as long-term visit-to-visit variability in HbA1c or fasting glucose, or as short-term within- and between-day variability derived from continuous glucose monitoring (CGM) [6,7]. The mechanistic case is strong. In a case-control study using CGM and urinary 8-iso prostaglandin F2α, the mean amplitude of glycaemic excursions correlated with oxidative stress far more tightly than HbA1c, fasting glucose or 24-hour mean glucose did, and the relationship survived adjustment for every other marker of glycaemic control [8]. Oscillating glucose appears to activate reactive oxygen species generation, endothelial dysfunction, inflammation and advanced glycation end-product formation more potently than stable hyperglycaemia of equivalent average [9,10].

The epidemiological evidence, however, is weighted toward long-term variability and long follow-up. In 101,533 people with T2DM free of cardiovascular disease, HbA1c variability predicted myocardial infarction, stroke and all-cause mortality over five years, and did so even among those whose mean HbA1c sat within target [11]. A systematic review and meta-analysis found that the highest quantile of every HbA1c variability metric — standard deviation, coefficient of variation and variability score — predicted mortality, cardiovascular events and progression of kidney disease [12]. Post hoc analyses of ACCORD and the EMPATHY cohort report similar associations, the latter strongest in patients whose mean HbA1c was below 7% [13,14]. Short-term, CGM-derived variability has been studied mainly in acute settings, where glucose variability during hospitalisation predicted mid-term major adverse cardiac events in patients with diabetes and acute coronary syndrome [15,16].

What remains unclear is whether short-term glycaemic variability, measured once with ambulatory CGM in stable outpatients, carries prognostic information over a clinically immediate horizon — the kind of horizon over which a clinician actually adjusts therapy. Three questions follow. Does CGM-derived %CV predict cardiovascular events within twelve months? Is any such association independent of mean glycaemia, HbA1c and conventional risk factors? And how much of it is explained by hypoglycaemia, which travels with high variability and is itself an established risk marker [17]? This prospective cohort 

Study design and setting

This was a prospective, observational cohort study conducted at three centres — one tertiary referral hospital and two affiliated secondary-care diabetes clinics — between January 2023 and September 2025. The protocol was approved by the institutional ethics committee ([reference number]), registered prospectively ([registry, identifier]), and conducted according to the Declaration of Helsinki. All participants gave written informed consent. Reporting follows the STROBE statement for cohort studies.

Participants

Consecutive adults attending the outpatient diabetes clinics were screened. Eligible participants were aged 30–80 years with T2DM of at least one year’s duration, diagnosed by standard criteria, and able to wear a CGM sensor for 14 days. We excluded type 1 and secondary diabetes; pregnancy; an acute coronary syndrome, stroke or coronary revascularisation within the preceding three months; decompensated heart failure; estimated glomerular filtration rate (eGFR) below 15 mL/min/1.73 m² or dialysis; active malignancy; haemoglobinopathy or recent transfusion, either of which invalidates HbA1c; and a life expectancy under 12 months. Recruitment and exclusions are shown in Figure 1.

Exposure assessment

Each participant wore a blinded, factory-calibrated CGM sensor for 14 consecutive days at enrolment. Blinding was deliberate: an unblinded sensor is itself an intervention that changes glycaemic behaviour and would have made the exposure a moving target. Recordings with less than 70% of expected sensor readings over 14 days were excluded, consistent with international consensus on the data sufficiency required for reliable CGM metrics [18].

The pre-specified primary exposure was the coefficient of variation of sensor glucose (%CV = 100 × SD ÷ mean glucose). %CV was chosen over the raw standard deviation because it is independent of mean glucose and is the variability metric endorsed for standardised CGM reporting [18,19]. Secondary glycaemic metrics, all computed over the same 14-day window, were mean sensor glucose; glucose SD; the mean amplitude of glycaemic excursions (MAGE); time in range 70–180 mg/dL (TIR); time below range <70 mg/dL (TBR); and time above range >180 mg/dL. %CV was analysed three ways: continuously per 1-SD increment, in quartiles with the lowest as reference, and dichotomised at 36%, the threshold above which a glucose profile is considered unstable by consensus [18].

Outcomes

The primary outcome was a composite of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke, hospitalisation for heart failure and urgent coronary revascularisation, occurring within 12 months of the baseline CGM assessment. Secondary outcomes were the individual components and all-cause mortality. Participants were followed at 3, 6, 9 and 12 months by clinic visit, with telephone contact and review of regional hospital records for those who did not attend. All events were adjudicated by a two-member committee of cardiologists using standard definitions, blinded to CGM data; disagreements went to a third adjudicator. Follow-up was censored at death from a non-cardiovascular cause, loss to follow-up or 365 days, whichever came first.

Covariates

Covariates were specified in advance on clinical grounds rather than selected by statistical screening. They comprised age, sex, body-mass index, duration of diabetes, current smoking, hypertension, established atherosclerotic cardiovascular disease, systolic blood pressure, LDL cholesterol, HbA1c (standardised, NGSP-aligned), eGFR by the CKD-EPI equation, urinary albumin-to-creatinine ratio (UACR, log-transformed), and glucose-lowering and cardioprotective therapy at baseline: metformin, sulfonylurea, insulin, sodium–glucose cotransporter-2 (SGLT2) inhibitors, glucagon-like peptide-1 (GLP-1) receptor agonists and statins. SGLT2 inhibitors and GLP-1 receptor agonists were modelled explicitly because the glucose-lowering regimen is itself a strong determinant of variability, and because the effect of treatment intensification on outcomes is modified by glycaemic variability [20].

Sample size

Assuming a 12-month event rate of 8% in the lowest %CV quartile, a two-sided α of 0.05 and 80% power to detect a hazard ratio of 1.35 per 1-SD increment in %CV, approximately 85 primary events were required. Allowing for 10% attrition and inadequate CGM wear, a target of 1000 participants with analysable data was set.

Statistical analysis

Continuous variables are summarised as mean ± SD or median (IQR) and compared across %CV quartiles by linear trend test; categorical variables as n (%) with the Cochran–Armitage trend test. Incidence rates are expressed per 100 person-years with exact Poisson confidence intervals. Cumulative incidence was estimated by the Kaplan–Meier method and compared by log-rank test.

Cox proportional hazards models were built in four pre-specified, nested steps. Model 1 adjusted for age and sex. Model 2 added duration of diabetes, body-mass index, systolic blood pressure, LDL cholesterol, eGFR, log-UACR, smoking, established cardiovascular disease, statin, SGLT2 inhibitor, GLP-1 receptor agonist and insulin use. Model 3 added HbA1c and mean sensor glucose, isolating variability from average glycaemia. Model 4 added time below range, to test how much of any residual association is carried by hypoglycaemia rather than by variability as such. The proportional hazards assumption was examined using scaled Schoenfeld residuals and was not violated. Dose–response was modelled with restricted cubic splines using four knots at the 5th, 35th, 65th and 95th centiles, with non-linearity assessed by likelihood ratio test.

Pre-specified subgroup analyses examined effect modification by age, sex, baseline HbA1c, duration of diabetes, established cardiovascular disease, insulin use and renal function, with interaction terms tested formally. Discrimination was quantified by Harrell’s C-index and by the area under the receiver operating characteristic curve, with differences assessed by bootstrap resampling (2000 replicates). Sensitivity analyses excluded events within the first 30 days, participants with substantial hypoglycaemia exposure, insulin-treated participants, those with eGFR below 60 mL/min/1.73 m², and anyone censored before 365 days. Analyses used Python 3.13 with lifelines 0.30 and statsmodels 0.15; two-sided P < 0.05 was considered significant, and subgroup and secondary analyses are reported as exploratory without multiplicity adjustment.

Cohort and baseline characteristics

Of 1347 adults screened, 1100 were enrolled and 1000 contributed analysable CGM and follow-up data (Figure 1). Mean age was 58.0 ± 9.7 years, 59.2% were men, mean duration of diabetes was 10.1 ± 5.6 years and 17.1% had established cardiovascular disease. Mean HbA1c was 8.14 ± 1.48% and mean sensor glucose 187.0 ± 44.5 mg/dL. Mean %CV was 32.3 ± 7.9%, and 352 participants (35.2%) exceeded the 36% consensus threshold.

Figure 1. Study flow and distribution of events across baseline %CV quartiles.

Baseline characteristics by quartile of %CV are shown in Table 1. Higher variability tracked with longer diabetes duration, higher HbA1c and mean glucose, lower time in range, greater time below range, higher albuminuria, and markedly higher use of sulfonylureas and insulin. Age, sex, blood pressure, LDL cholesterol and the prevalence of established cardiovascular disease did not differ across quartiles — the latter is worth noting, because it means the gradient in events described below is not simply a gradient in pre-existing disease.

Table 1. Baseline characteristics overall and by quartile of CGM coefficient of variation.

Quartile ranges of %CV: Q1 13.0–26.9%, Q2 26.9–32.3%, Q3 32.3–37.8%, Q4 37.8–57.4%.

Incidence of cardiovascular events

Over 934 person-years, 89 participants (8.9%) reached the primary composite outcome, an incidence of 9.53 per 100 person-years (95% CI 7.65–11.73). The commonest component was non-fatal myocardial infarction (29 events), followed by hospitalisation for heart failure (16), cardiovascular death (15), non-fatal stroke (19) and urgent revascularisation (10). Incidence rose monotonically across quartiles of %CV, from 5.50 to 14.79 per 100 person-years (Table 2).

Table 2. Incidence of the primary composite outcome by quartile of %CV.

Kaplan–Meier curves separated early and continued to diverge through the follow-up year (Figure 2). Twelve-month cumulative incidence was 5.3%, 8.2%, 8.9% and 13.8% from the lowest to the highest quartile (log-rank χ² = 10.98, P = 0.012). The separation was driven predominantly by the top quartile; quartiles 2 and 3 were nearly superimposed.

Figure 2. Cumulative incidence of MACE by quartile of baseline %CV

Glycaemic variability and the primary outcome

Table 3 presents the sequentially adjusted Cox models. Unadjusted, each 1-SD (7.9%) increment in %CV was associated with a hazard ratio of 1.49 (95% CI 1.21–1.84). The estimate was essentially unchanged by adjustment for age and sex, and attenuated modestly on full clinical adjustment to 1.40 (1.11–1.76). Critically, adding HbA1c and mean sensor glucose to the model barely moved it further: 1.38 (1.10–1.74; P = 0.006). Comparing the highest with the lowest quartile gave an adjusted hazard ratio of 1.95 (1.00–3.81), with a borderline trend across quartiles (P = 0.053); dichotomised at the consensus threshold, %CV ≥36% carried an adjusted hazard ratio of 1.63 (1.06–2.50; P = 0.027).

Adding time below range (Model 4) attenuated the continuous association to 1.31 (1.00–1.71; P = 0.048) and the quartile and threshold comparisons lost significance. This pattern is informative rather than merely negative, and we return to it in the Discussion.

Table 3. Hazard ratios for the primary composite outcome by baseline %CV, under sequential adjustment.

Model 1: age, sex. Model 2: Model 1 + diabetes duration, BMI, systolic BP, LDL cholesterol, eGFR, log-UACR, smoking, established CVD, statin, SGLT2 inhibitor, GLP-1 receptor agonist, insulin. Model 3: Model 2 + HbA1c and mean sensor glucose. Model 4: Model 3 + time below range.

In the fully adjusted Model 3, the other independent predictors were age (HR 1.05 per year, 1.03–1.08), diabetes duration (1.07 per year, 1.04–1.11), eGFR (0.99 per mL/min, 0.978–0.996), log-UACR (1.23, 1.03–1.47), established cardiovascular disease (1.86, 1.16–3.00) and insulin use (1.58, 1.00–2.48). Neither HbA1c (1.11 per 1%, 0.69–1.76) nor mean sensor glucose was independently predictive once variability was in the model.

The restricted cubic spline showed a graded, broadly monotonic increase in risk across the range of %CV with no evidence of a threshold effect (P for non-linearity = 0.22; Figure 3). Risk crossed unity at approximately 25% and continued to rise through the 36% consensus threshold without inflecting at it.

Figure 3. Dose–response relationship between %CV and adjusted hazard of MACE.

Alternative glycaemic metrics and sensitivity analyses

Other measures of variability behaved concordantly (Table 4). Glucose SD and MAGE were each associated with the primary outcome per 1-SD increment, with point estimates somewhat larger than that for %CV. Time below range was independently associated with events (HR 1.22 per 1% higher, 1.03–1.46), while time in range showed a non-significant inverse association. The association of %CV with the outcome was stable across sensitivity analyses, though it lost precision in the smaller subsets: excluding insulin-treated participants and those with eGFR below 60 both widened the confidence interval across unity while leaving the point estimate near 1.25.

Table 4. Alternative glycaemic metrics (upper panel, Model 3 adjustment) and sensitivity analyses of the primary exposure (lower panel).

Subgroups and discrimination

The association was directionally consistent across every pre-specified subgroup, with no significant interaction (all P for interaction ≥ 0.45; Figure 4). The largest point estimate was in participants with eGFR below 60 mL/min/1.73 m² (HR 1.90, 1.25–2.89) against 1.18 (0.95–1.46) in those with preserved renal function, but the interaction test was not significant (P = 0.47) and the subgroup was small. Notably, the association persisted in participants with HbA1c below 7.5% (HR 1.30, 0.93–1.82), among whom mean glycaemia was at or near target.

Figure 4. Subgroup analysis of the association between %CV and the primary outcome.

Adding %CV to a Cox model containing conventional risk factors, HbA1c and mean glucose increased Harrell’s C-index from 0.744 to 0.760. As a single discriminator for 12-month events, %CV achieved an area under the curve of 0.613, numerically higher than HbA1c (0.574) but not significantly so (difference 0.039, 95% CI −0.047 to 0.117; P = 0.36); MAGE performed similarly (0.615) (Figure 5). The Youden-optimal cut-point for %CV was 30.4%, giving 79% sensitivity and 42% specificity — adequate for flagging but far too imprecise for individual risk classification.

Figure 5. Discrimination of 12-month MACE by glycaemic metrics.

Principal findings

In 1000 adults with T2DM followed for twelve months, short-term glycaemic variability measured by a single 14-day blinded CGM recording was associated with major adverse cardiovascular events independently of HbA1c, mean sensor glucose and conventional risk factors. Each 1-SD increment in %CV carried a 38% higher adjusted hazard, and participants above the 36% consensus threshold had roughly 60% higher risk than those below it. Three features of the result deserve emphasis. First, the association was graded rather than thresholded: the spline rose steadily from around 25% without inflecting at 36%, which argues that 36% is a useful operational flag rather than a biological boundary. Second, HbA1c was not independently predictive once variability was in the model, while variability remained predictive after adjustment for HbA1c — an asymmetry pointing to where the prognostic information actually sits. Third, the signal persisted among participants whose HbA1c was below 7.5%, precisely the group a clinician would otherwise consider adequately controlled.

Comparison with previous studies

Our findings extend to a short horizon what has been shown repeatedly over longer ones. Ceriello and colleagues, studying 101,533 people with T2DM and no baseline cardiovascular disease, found HbA1c variability predicted myocardial infarction, stroke and death over five years, with the risk associated with variability actually higher among those at the recommended HbA1c target [11]. Sato and colleagues reported the same pattern in the Japanese EMPATHY cohort over a median of 38 months, where the association between visit-to-visit HbA1c variability and cardiovascular events was stronger in patients with mean HbA1c below 7% than above it [14]. A meta-analysis across variability metrics found hazard ratios for all-cause mortality of 1.89 for HbA1c standard deviation, 1.47 for coefficient of variation and 1.67 for the variability score [12]. The convergence of long-term HbA1c variability studies and our short-term CGM data on the same conclusion — that instability matters most where the average looks acceptable — is the most clinically consequential point in this literature.

Our effect sizes are smaller than those reported in acute settings, which is expected. Gerbaud and colleagues found that glucose variability during hospitalisation was the strongest independent predictor of mid-term major adverse cardiac events in patients with diabetes and acute coronary syndrome, outperforming the GRACE risk score [15]; Su and colleagues similarly found admission MAGE, but not HbA1c, predicted in-hospital outcomes after percutaneous coronary intervention in non-ST-elevation acute coronary syndrome [16]. Variability measured during an acute illness partly reflects the severity of that illness. Our cohort was ambulatory and stable, so the exposure is cleaner and, unsurprisingly, the association weaker.

The subclinical evidence aligns. In 600 people with T2DM and no symptomatic cardiovascular disease followed with carotid ultrasound for 104 weeks, baseline %CV and time in range were both independently associated with change in the grey-scale median of the carotid wall, a marker of plaque composition, after adjustment for HbA1c [21]. Instability appears to be altering vascular biology before events occur.

Mechanistic considerations

The mechanism most consistently implicated is oxidative stress. Monnier’s case-control work showed that MAGE correlated with urinary 8-iso prostaglandin F2α at r = 0.86, far more strongly than HbA1c or mean glucose, and that the relationship persisted in multivariable models [8]. Intermittent high glucose appears to generate reactive oxygen species more effectively than sustained high glucose, driving endothelial dysfunction, protein kinase C activation, nuclear factor-κB signalling and inflammation [9,10]. Advanced glycation end-products, independently predictive of post-infarction heart failure, provide a plausible downstream link to the myocardium [22]. The phenomenon of metabolic memory, in which transient glycaemic insults produce durable epigenetic changes in vascular cells, offers an explanation for why a 14-day exposure window might carry information about the following year [10].

The hypoglycaemia question

The attenuation we observed on adjusting for time below range is important and should not be glossed over. Hypoglycaemia and high variability are mechanically inseparable: a wide glucose distribution with any given mean will spend more time in the low tail. Hypoglycaemia is independently associated with adverse cardiovascular outcomes through sympathoadrenal activation, QT prolongation, thrombogenesis and inflammation [7,17]. The DEVOTE 2 analysis is directly relevant: day-to-day fasting glycaemic variability was strongly associated with severe hypoglycaemia (HR 4.11) and all-cause mortality (HR 1.58), but its association with major cardiovascular events did not survive adjustment for baseline characteristics and HbA1c [24]. As Rutter noted in the accompanying commentary, these observational analyses raise awareness of the risks attached to variability and severe hypoglycaemia without establishing causation [25].

Our data sit between these positions. The association with the primary outcome persisted after adjustment for mean glycaemia but was attenuated to the margin of significance by adjustment for time below range, and the quartile contrasts lost significance. The honest reading is that part — not all — of the cardiovascular signal attributed to variability is carried by hypoglycaemia exposure, and that variability and hypoglycaemia should be understood as two views of the same unstable glucose distribution rather than as competing explanations. This is consistent with recent reviews arguing that the independent vascular role of variability is difficult to disentangle from hypoglycaemia, and that large outcome studies are needed before targets can be set [7,26].

Clinical implications

Three implications follow, stated at the strength the evidence supports. The addition of %CV to a model containing conventional risk factors and HbA1c improved the C-index from 0.744 to 0.760 — real but modest, and as a standalone discriminator %CV achieved an AUC of only 0.613. These data do not support using %CV as a risk-stratification instrument in individuals. What they do support is treating a %CV above the consensus threshold as a signal to review the regimen, particularly where HbA1c looks satisfactory. Agents that lower glucose without provoking variability or hypoglycaemia — SGLT2 inhibitors, GLP-1 receptor agonists, metformin — are preferable to sulfonylureas and aggressive insulin titration on this reasoning, and the distribution of therapy across quartiles in Table 1 shows how strongly regimen and variability travel together [23]. Structured dietary and exercise intervention lowers both average glycaemia and its amplitude, and remains the lowest-risk way to do so [28,29,30]. Finally, these data support reporting %CV and time below range alongside HbA1c and time in range when CGM is used, as standardised reporting frameworks already recommend [18,19,26]. Unblinded CGM itself reduces variability and improves cardiometabolic risk markers in T2DM, including in patients not on insulin [27].

Strengths and limitations

Strengths include the prospective design with pre-specified exposure and outcome definitions, objective CGM-derived exposure rather than self-monitored glucose, blinded sensors that avoided behaviour change, blinded event adjudication, sequential modelling that isolates variability from mean glycaemia and then from hypoglycaemia, and consistency across alternative variability metrics and multiple sensitivity analyses.

Several limitations temper the conclusions. First and most fundamental, this is an observational study; residual confounding by indication is likely, since the patients with the most unstable glucose are also those with the most advanced disease and the most complex regimens, and no adjustment fully removes that. Second, exposure was assessed once at baseline; variability changes with therapy, and a single 14-day window cannot capture that trajectory. Third, with 89 events the study is adequately powered for the continuous exposure but underpowered for quartile contrasts and subgroup analyses, which should be read as hypothesis-generating — the wide confidence intervals in Figure 4 and in the restricted sensitivity analyses make this plain. Fourth, the cohort was recruited from specialist clinics and may not represent community-managed T2DM. Fifth, twelve months is short for cardiovascular outcomes, which both limits event accrual and leaves open whether these associations strengthen or attenuate over longer follow-up. Finally, %CV is one of many variability metrics without a universally accepted standard [6,7], and the field still lacks the randomised evidence needed to show that targeting variability changes outcomes [7,26].

In adults with type 2 diabetes mellitus, short-term glycaemic variability captured by a single 14-day CGM recording was associated with major adverse cardiovascular events over the following twelve months, independently of HbA1c and mean sensor glucose. The relationship was graded across the range of %CV rather than thresholded, persisted in participants whose HbA1c was at target, and was partly — but not wholly — accounted for by hypoglycaemia exposure. The incremental gain in discrimination was modest, so %CV should not be used as a standalone risk-stratification tool. It is better understood as a modifiable marker of regimen instability: a %CV above 36%, particularly alongside a satisfactory HbA1c, identifies a patient whose glucose-lowering therapy deserves review rather than reassurance. Whether deliberately reducing variability improves cardiovascular outcomes remains unproven and requires randomised trials with variability as a pre-specified target.

Declarations

Ethics approval and consent to participate. Approved by the institutional ethics committee ([reference number]); all participants provided written informed consent.

Availability of data. De-identified data are available from the corresponding author on reasonable request.

Competing interests. [To be declared.]

Funding. [To be declared.]

Author contributions. [To be completed per ICMJE criteria.]

1. Sun H, Saeedi P, Karuranga S, et al. IDF Diabetes Atlas: global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract. 2022;183:109119. doi:10.1016/j.diabres.2021.109119
2. Rosenstock J, Perkovic V, Johansen OE, et al. Effect of linagliptin vs placebo on major cardiovascular events in adults with type 2 diabetes and high cardiovascular and renal risk: the CARMELINA randomized clinical trial. JAMA. 2019;321(1):69-79. doi:10.1001/jama.2018.18269
3. Bellido V, Pinés-Corrales PJ, Villar-Taibo R, Ampudia-Blasco FJ. Time-in-range for monitoring glucose control: is it time for a change? Diabetes Res Clin Pract. 2021;177:108917. doi:10.1016/j.diabres.2021.108917
4. Hodgson S, Cheema S, Rani Z, et al. Population stratification in type 2 diabetes mellitus: a systematic review. Diabet Med. 2022;39(1):e14688. doi:10.1111/dme.14688
5. Action to Control Cardiovascular Risk in Diabetes Study Group; Gerstein HC, Miller ME, Byington RP, et al. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med. 2008;358(24):2545-2559. doi:10.1056/NEJMoa0802743
6. Cameron FJ, Baghurst PA, Rodbard D. Assessing glycemic variation: why, when and how? Pediatr Endocrinol Rev. 2010;7(Suppl 3):432-444. PMID:20877258
7. Ajjan RA. The clinical importance of measuring glycaemic variability: utilising new metrics to optimise glycaemic control. Diabetes Obes Metab. 2024;26(Suppl 7):3-16. doi:10.1111/dom.16098
8. Monnier L, Mas E, Ginet C, et al. Activation of oxidative stress by acute glucose fluctuations compared with sustained chronic hyperglycemia in patients with type 2 diabetes. JAMA. 2006;295(14):1681-1687. doi:10.1001/jama.295.14.1681
9. Škrha J, Šoupal J, Škrha J Jr, Prázný M. Glucose variability, HbA1c and microvascular complications. Rev Endocr Metab Disord. 2016;17(1):103-110. doi:10.1007/s11154-016-9347-2
10. Darenskaya MA, Kolesnikova LI, Kolesnikov SI. Oxidative stress: pathogenetic role in diabetes mellitus and its complications and therapeutic approaches to correction. Bull Exp Biol Med. 2021;171(2):179-189. doi:10.1007/s10517-021-05191-7
11. Ceriello A, Lucisano G, Prattichizzo F, et al. HbA1c variability predicts cardiovascular complications in type 2 diabetes regardless of being at glycemic target. Cardiovasc Diabetol. 2022;21(1):13. doi:10.1186/s12933-022-01445-4
12. Qu F, Shi Q, Wang Y, et al. Visit-to-visit glycated hemoglobin A1c variability in adults with type 2 diabetes: a systematic review and meta-analysis. Chin Med J (Engl). 2022;135(19):2294-2300. doi:10.1097/CM9.0000000000002073
13. Pei J, Wang X, Pei Z, Hu X. Glycemic control, HbA1c variability, and major cardiovascular adverse outcomes in type 2 diabetes patients with elevated cardiovascular risk: insights from the ACCORD study. Cardiovasc Diabetol. 2023;22(1):287. doi:10.1186/s12933-023-02026-9
14. Sato M, Inaishi J, Saisho Y, Sato Y, Komuro I, Itoh H. Association of visit-to-visit glycemic variability with risk of cardiovascular diseases in high-risk Japanese patients with type 2 diabetes: a subanalysis of the EMPATHY trial. J Diabetes Investig. 2021;12(12):2190-2196. doi:10.1111/jdi.13597
15. Gerbaud E, Darier R, Montaudon M, et al. Glycemic variability is a powerful independent predictive factor of midterm major adverse cardiac events in patients with diabetes with acute coronary syndrome. Diabetes Care. 2019;42(4):674-681. doi:10.2337/dc18-2047
16. Su G, Zhang T, Yang H, et al. Admission glycemic variability correlates with in-hospital outcomes in diabetic patients with non-ST segment elevation acute coronary syndrome undergoing percutaneous coronary intervention. Anatol J Cardiol. 2018;19(6):368-373. doi:10.14744/AnatolJCardiol.2018.47487
17. Sun B, He F, Gao Y, et al. Prognostic impact of visit-to-visit glycemic variability on the risks of major adverse cardiovascular outcomes and hypoglycemia in patients with different glycemic control and type 2 diabetes. Endocrine. 2019;64(3):536-543. doi:10.1007/s12020-019-01893-1
18. Battelino T, Danne T, Bergenstal RM, et al. Clinical targets for continuous glucose monitoring data interpretation: recommendations from the International Consensus on Time in Range. Diabetes Care. 2019;42(8):1593-1603. doi:10.2337/dci19-0028
19. Battelino T, Alexander CM, Amiel SA, et al. Continuous glucose monitoring and metrics for clinical trials: an international consensus statement. Lancet Diabetes Endocrinol. 2023;11(1):42-57. doi:10.1016/S2213-8587(22)00319-9
20. Wang JM, Miao MY, Jia YP, et al. Effects of intensive glycemic control on microvascular outcomes in type 2 diabetes mellitus are modified by long-term HbA1c variability: a post hoc analysis of the ACCORD trial. Diabetes Res Clin Pract. 2024;208:111100. doi:10.1016/j.diabres.2024.111100
21. Mita T, Katakami N, Okada Y, et al. Continuous glucose monitoring-derived time in range and CV are associated with altered tissue characteristics of the carotid artery wall in people with type 2 diabetes. Diabetologia. 2023;66(12):2356-2367. doi:10.1007/s00125-023-06013-3
22. Raposeiras-Roubín S, Rodiño-Janeiro BK, Paradela-Dobarro B, et al. Predictive value of advanced glycation end products for the development of post-infarction heart failure: a preliminary report. Cardiovasc Diabetol. 2012;11:102. doi:10.1186/1475-2840-11-102
23. Tay J, Luscombe-Marsh ND, Thompson CH, et al. Comparison of low- and high-carbohydrate diets for type 2 diabetes management: a randomized trial. Am J Clin Nutr. 2015;102(4):780-790. doi:10.3945/ajcn.115.112581
24. Zinman B, Marso SP, Poulter NR, et al. Day-to-day fasting glycaemic variability in DEVOTE: associations with severe hypoglycaemia and cardiovascular outcomes (DEVOTE 2). Diabetologia. 2018;61(1):48-57. doi:10.1007/s00125-017-4423-z
25. Rutter MK. Devoting attention to glucose variability and hypoglycaemia in type 2 diabetes. Diabetologia. 2018;61(1):43-47. doi:10.1007/s00125-017-4421-1
26. Bombaci B, Passanisi S, Lombardo F, Salzano G. Clinical relevance of short-term glycemic variability in children and adolescents with type 1 diabetes: a narrative review. Transl Pediatr. 2024;13(7):1231-1241. doi:10.21037/tp-24-114
27. Reed J, Dong T, Eaton E, et al. Continuous glucose monitoring for glycaemic control and cardiovascular risk reduction in patients with type 2 diabetes not on insulin therapy: a clinical trial. Diabetes Obes Metab. 2024;26(7):2881-2889. doi:10.1111/dom.15608
28. Minari TP, Manzano CF, T√°cito LHB, et al. The impact of a nutritional intervention on glycemic control and cardiovascular risk markers in type 2 diabetes. Nutrients. 2024;16(9):1378. doi:10.3390/nu16091378
29. Reynolds AN, Akerman AP, Mann J. Dietary fibre and whole grains in diabetes management: systematic review and meta-analyses. PLoS Med. 2020;17(3):e1003053. doi:10.1371/journal.pmed.1003053
30. Hansen D, Niebauer J, Cornelissen V, et al. Exercise prescription in patients with different combinations of cardiovascular disease risk factors: a consensus statement from the EXPERT Working Group. Sports Med. 2018;48(8):1781-1797. doi:10.1007/s40279-018-0930-4