Long-term glycemic variability and the risk of heart failure: a meta-analysis

Introduction

Heart failure (HF) is a major global public health concern, affecting an estimated 64 million people globally (Martin et al., 2025; Savarese et al., 2023). Its burden is increasing with aging in populations and the rising prevalence of cardiovascular risk factors (Shahim et al., 2023). HF is associated with high morbidity, frequent hospitalization, and substantial mortality, with 5-year survival rates comparable to those of many cancers (Mamas et al., 2017). In addition to its effects on individuals, HF imposes a significant economic burden on healthcare systems associated with chronic disease management and recurrent admissions (Urbich et al., 2020). Although established risk factors such as hypertension, coronary artery disease, diabetes mellitus, obesity, and aging are well recognized, the prognosis of HF remains poor, and identifying novel, modifiable risk factors is crucial for early prevention and improved outcomes (Bui, Horwich & Fonarow, 2011; Lüscher, 2015; Meijers & de Boer, 2019).

Glycemic control has long been a cornerstone of diabetes management, with hemoglobin A1c (HbA1c) serving as a widely accepted marker of average glucose levels (Saudek & Brick, 2009). However, emerging evidence suggests that fluctuations in glycemic levels over time—known as glycemic variability (GV)—might have independent prognostic value beyond mean glycemia (Suh & Kim, 2015). GV can be broadly categorized as short-term (within-day or between-day fluctuations, typically assessed via continuous glucose monitoring (CGM)) or long-term (visit-to-visit fluctuations over months or years, often assessed using HbA1c or fasting plasma glucose (FPG) records) (Suh & Kim, 2015; Zhou et al., 2020). Long-term GV is commonly measured using indices such as standard deviation (SD), coefficient of variation (CV), variability independent of the mean (VIM), and average successive variability (ASV) (Zhou et al., 2020). Unlike transient glycemic excursions, long-term GV might reflect a sustained pattern of metabolic instability, potentially leading to cumulative damage to target organs (Sun, Luo & Zhou, 2021).

There is growing interest in the potential link between long-term GV and cardiovascular disease (CVD), including myocardial infarction, stroke, and cardiovascular mortality (Chen et al., 2022; Gorst et al., 2015). Several pathophysiological mechanisms could explain this relationship, such as oxidative stress, inflammation, endothelial dysfunction, and autonomic imbalance induced by repeated glucose fluctuations (Helleputte et al., 2020; Papachristoforou et al., 2020). These effects can promote myocardial remodeling, fibrosis, and neurohormonal activation, thereby contributing to the development of HF (Elendu et al., 2023). However, despite increasing research, the association between long-term GV and the risk of HF remains uncertain. Whereas some cohort studies demonstrated a significant association (Ceriello et al., 2022; Gu et al., 2018; Kwon et al., 2019; Li et al., 2020; Manosroi et al., 2023; Segar et al., 2020; Wan et al., 2020; Wang et al., 2023), others reported null findings (Kaze et al., 2020; Lin et al., 2021). In view of this uncertainty, we conducted a systematic review and meta-analysis to comprehensively evaluate the association between long-term GV and the risk of incident HF in adult populations.

Materials and Methods

This meta-analysis followed the PRISMA 2020 guidelines (Higgins et al., 2021; Page et al., 2021) and the Cochrane Handbook for Systematic Reviews and Meta-Analyses (Higgins et al., 2021) for protocol design, data extraction, statistical analysis, and results reporting. The study protocol was also registered in PROSPERO under ID CRD420251000563.

Results

Study identification

Figure 1 outlines the study selection process. Initially, 900 records were identified across three databases, and 157 duplicates were removed. After title and abstract screening, 716 articles were excluded for failing to meet the meta-analysis criteria. The full texts of the remaining 27 studies were independently reviewed by two authors, leading to the exclusion of 17 for reasons detailed in Fig. 1. Ultimately, 10 studies were included in the quantitative analysis (Ceriello et al., 2022; Gu et al., 2018; Kaze et al., 2020; Kwon et al., 2019; Li et al., 2020; Lin et al., 2021; Manosroi et al., 2023; Segar et al., 2020; Wan et al., 2020; Wang et al., 2023).

Long-term GV and HF risk

Pooled results from 11 cohorts in 10 studies (Ceriello et al., 2022; Gu et al., 2018; Kaze et al., 2020; Kwon et al., 2019; Li et al., 2020; Lin et al., 2021; Manosroi et al., 2023; Segar et al., 2020; Wan et al., 2020; Wang et al., 2023) revealed that overall, high long-term GV was associated with an increased risk of HF during follow-up (HR = 1.69, 95% CI [1.38–2.06], p < 0.001; I² = 92%; Fig. 2A). Sensitivity analysis, excluding one study at a time, found no significant impact of any individual study on the results (HR = 1.51–1.80, all p < 0.05). Specifically, the sensitivity analysis limited to studies including patients with T2DM (Ceriello et al., 2022; Gu et al., 2018; Kaze et al., 2020; Li et al., 2020; Lin et al., 2021; Segar et al., 2020; Wan et al., 2020) revealed similar results (HR = 1.96, 95% CI [1.42–2.71], p < 0.001; I² = 89%). Further sensitivity analyses also demonstrated consistent results in studies with good quality (Gu et al., 2018; Kaze et al., 2020; Kwon et al., 2019; Li et al., 2020; Lin et al., 2021; Manosroi et al., 2023; Segar et al., 2020; Wan et al., 2020; Wang et al., 2023) (NOS ≥ 7; HR = 1.78, 95% CI [1.37–2.32], p < 0.001; I² = 93%), and in studies adjusting for HbA1c levels (Ceriello et al., 2022; Gu et al., 2018; Kaze et al., 2020; Manosroi et al., 2023; Segar et al., 2020; Wan et al., 2020; Wang et al., 2023) (HR = 1.95, 95% CI [1.46–2.61], p < 0.001; I² = 88%). In subgroup analyses stratified by glycemic variability parameters, no significant difference was observed in the association with HF risk (p for subgroup difference = 0.34; Fig. 2B). In addition, further subgroup analyses recorded similar results in studies from Asian and Western countries (p for subgroup difference = 0.25; Fig. 3A), in prospective and retrospective or post-hoc studies (p for subgroup difference = 0.35; Fig. 3B), in participants with mean ages of <64 or ≥64 years (p for subgroup difference = 0.35; Fig. 4A), in populations with proportions of men of <54% or ≥54% (p for subgroup difference = 0.76; Fig. 4B), in studies with mean follow-up durations of <7 or ≥7 years (p for subgroup difference = 0.84; Fig. 5A), and in studies with the NOS scores of 6–7 and 8–9 (p for subgroup difference = 0.60; Fig. 5B).

Publication bias

Figure 6 displays the funnel plots evaluating the publication bias underlying the meta-analysis of the association between long-term GV and the risk of HF. The funnel plots were symmetrical on visual inspection, suggesting a low risk of publication bias. The findings were further supported by Egger’s regression analysis, which also did not suggest significant publication bias (p = 0.29).

Discussion

In this meta-analysis of 11 cohorts from 10 studies involving more than 4.2 million adults, we found that high long-term GV, as measured by fluctuations in HbA1c or FPG, was significantly associated with an increased risk of HF. The pooled HR indicated a 69% higher risk of HF in individuals with high long-term GV than in those with low GV, independent of traditional risk factors. This association remained robust across various sensitivity analyses, including studies restricted to patients with T2DM, studies with high methodological quality, and studies adjusting for mean HbA1c levels. Subgroup analyses further demonstrated consistent results across study designs, populations, GV measurement parameters, geographic regions, and follow-up durations, supporting the robustness and generalizability of the findings.

Several biological mechanisms might underlie the relationship between long-term GV and the development of HF. Chronic glucose fluctuations are known to promote oxidative stress (Chang et al., 2012), endothelial dysfunction (Zhang et al., 2014), and inflammatory responses (Quincozes-Santos et al., 2017), all of which contribute to cardiovascular damage. Repetitive glycemic excursions can induce myocardial fibrosis, impair myocardial energy metabolism, and accelerate left ventricular remodeling, thus predisposing individuals to HF (Hanajima et al., 2023; Zhang et al., 2022). Additionally, long-term GV has been associated with arterial stiffness (Zhang et al., 2021) and autonomic dysfunction (Xu et al., 2016), which might further exacerbate cardiac structural and functional impairment. These pathophysiological effects suggest that GV represents a form of metabolic stress capable of contributing to HF pathogenesis beyond the effects of sustained hyperglycemia alone (Tabesh et al., 2024). The molecular pathways underlying the association between higher long-term GV and an increased risk of HF warrant further investigation.

The sensitivity analyses provided valuable insight into the consistency of the observed association. Notably, when the analysis was limited to cohorts of individuals with T2DM, the risk of HF associated with high GV remained significant and even stronger. This suggests that GV is particularly detrimental in populations already at high risk for cardiovascular complications. Furthermore, the association persisted among studies that adjusted for mean HbA1c levels, indicating that the effect of GV is not merely a reflection of overall poor glycemic control. Similarly, limiting the analysis to studies with high NOS scores did not materially alter the findings, supporting the robustness of the results. Subgroup analyses revealed no significant effect modification by the type of GV parameter (HbA1c-based vs. FPG-based), geographic region, study design, age, sex, or follow-up duration, suggesting that the association between long-term GV and HF risk is consistent across diverse settings and populations.

This study had several strengths. First, it represents the most comprehensive and up-to-date synthesis of evidence on this topic, incorporating studies published up to January 2025. Second, all included studies employed longitudinal designs and multivariate models, permitting temporal assessment and control of key confounding factors. Third, our use of multiple sensitivity and subgroup analyses provided a robust evaluation of the consistency of findings across various methodological and population-level factors. The large sample size and inclusion of both diabetic and non-diabetic populations enhance the generalizability of our conclusions. Nevertheless, several limitations should be acknowledged. First, most of the included studies were retrospective cohort designs, which might have introduced selection and recall bias (Song & Chung, 2010). Second, our literature search was restricted to four major English-language databases. Although these databases, supplemented by manual reference screening, are generally adequate for systematic reviews and meta-analyses, we cannot exclude the possibility that relevant studies published in other languages or indexed in other databases (e.g., CENTRAL, Scopus) were missed. This restriction might have introduced language bias. The parameters and cutoffs used to define high GV varied across studies, limiting the comparability of the results and precluding a standardized threshold for clinical application. Although our subgroup analyses revealed no significant difference in the association according to GV metrics, the absence of a universally accepted definition underscores the need for future research to establish standardized metrics and clinically meaningful cutoffs for GV (Ajjan, 2024). Although all included studies used multivariate models, residual confounding remains possible. Important covariates, such as the use of sodium–glucose cotransporter 2 (SGLT2) inhibitors, which have been demonstrated to reduce HF risk (Pasternak et al., 2019), were not uniformly reported or included in the adjustment. Moreover, several key cardioprotective medications, particularly SGLT2 inhibitors and GLP-1 receptor agonists, were not consistently reported or adjusted for across the included studies. Important socioeconomic factors, such as income and access to healthcare, were also largely unmeasured. These unaccounted variables might have contributed to residual confounding, and they should be addressed in future research. The observational nature of the included studies prevented causal inference. Additionally, data were analyzed at the study level rather than the individual patient level, which restricted our ability to assess the association according to specific subtypes of HF, such as HF with reduced ejection fraction versus HF with preserved ejection fraction. Most included studies did not distinguish between HF subtypes such as HF with preserved versus reduced ejection fraction, as only one study (Gu et al., 2018) specifically examined HF with preserved with ejection fraction. Consequently, we were unable to evaluate whether the association between long-term GV and HF risk differed by HF subtype.

We were also unable to evaluate the influence of important clinical variables such as duration of diabetes, presence of comorbidities (e.g., chronic kidney disease, coronary artery disease), medication adherence, and lifestyle factors, which might influence both GV and HF risk. Substantial between-study heterogeneity was observed. We could not perform robust meta-regression to formally explore potential sources of heterogeneity, as only 10 studies were available and individual-participant data were not accessible. Variables such as diabetes prevalence were reported only at the study level, preventing their examination as continuous covariates.

From a clinical perspective, our findings highlight the potential importance of long-term GV as an independent indicator of cardiovascular vulnerability. Whereas GV is not yet a standard component of cardiovascular risk assessment, these results suggest that monitoring long-term GV could provide additional prognostic information, particularly in individuals with T2DM (Howsawi & Alem, 2024). Incorporating GV into routine monitoring using readily available longitudinal HbA1c or fasting glucose records could help identify individuals at higher risk of developing HF who could benefit from closer follow-up or earlier intervention. In addition to optimizing mean glycemic control, targeting reductions in GV through tailored therapeutic approaches (e.g., use of medications with more stable glucose-lowering profiles, lifestyle interventions) might further improve cardiovascular outcomes. Integrating GV metrics into electronic health records and risk prediction models might also enhance personalized risk assessment, although prospective validation is needed. Moreover, clinicians should be aware that minimizing glycemic fluctuations—not only lowering mean HbA1c—might be important in reducing the risk of HF and other diabetes-related complications (Dandona, 2017). Future research is needed to better define clinically relevant thresholds of GV, develop standardized metrics for its measurement, and investigate whether interventions targeting GV reduction can lead to improved cardiovascular outcomes. Prospective studies with detailed clinical phenotyping and individual patient-level data will be crucial to elucidate the potential differential impact of GV on HF subtypes. Moreover, randomized controlled trials or post-hoc analyses of trials evaluating glucose-lowering therapies could help determine whether reducing GV translates into HF risk reduction independent of changes in mean glycemia.

Conclusions

In conclusion, this meta-analysis demonstrated that high long-term glycemic variability is independently associated with an increased risk of incident HF in adults, regardless of the diabetic status or mean HbA1c level. These findings provide compelling evidence that long-term GV is a meaningful cardiovascular risk indicator and underscore the need for greater attention to glycemic stability in clinical practice. Further research is warranted to elucidate the mechanisms underlying this relationship and determine the clinical utility of GV in risk prediction and patient management.

Supplemental Information