Prediabetes Is Associated With Increased Risk of Heart Failure Among Patients With Atrial Fibrillation

The global epidemic of atrial fibrillation (AF) is driven by aging populations and the rising prevalence of cardiovascular risk factors, such as hypertension, diabetes, coronary artery disease (CAD), and obesity (1). Improved stroke prevention strategies have led to a decline in AF-related stroke (2), unveiling heart failure (HF) as the leading complication of AF (3,4). Indeed, HF has emerged as the most frequent nonfatal cardiovascular event (5) and cause of death (6) in patients with AF. Identifying the major contributing factors for the development of HF in patients with AF is therefore increasingly important.

Type 2 diabetes is a shared risk factor for the development of AF and HF and represents an important target to alleviate the disease burden of AF. An antecedent of AF, prediabetes, is a prime target for diabetes prevention, given its high prevalence (7) and considerable potential for reversion (8). Moreover, prediabetes itself may constitute a high-risk state associated with an increased risk of adverse cardiovascular events, HF (9), and all-cause mortality (10). In the context of AF, one study has further demonstrated its strong association with stroke, independent of other recognized risk factors (11). Nonetheless, the added risk of HF conferred by prediabetes in patients with AF remains unexplored. The current study aimed to examine the association between prediabetes and HF in patients with AF and whether changes in glycemic status would modify the subsequent risk of HF.

This retrospective observational study was based on data from the Clinical Data Analysis Reporting System (CDARS), a territory-wide database developed by the Hong Kong Hospital Authority (12,13). As the only statutory public health care provider in Hong Kong, the Hospital Authority provides >80% of inpatient services to the local population of 7.5 million (14). CDARS prospectively collects patient information on demographics, diagnoses, procedures, laboratory tests, and medications. All patients are anonymized with a unique patient identifier in CDARS. Diagnostic data are specifically coded using the ICD-9, which has a demonstrated high degree of accuracy (15,16). This study was approved by the institutional review board of The University of Hong Kong and the West Cluster of the Hong Kong Hospital Authority.

We identified all patients (aged ≥18 years) with newly diagnosed AF (ICD-9 code 427.31) between 1 January 2015 and 31 December 2018. The date of diagnosis of AF was defined as the index date. Patients with a history of AF before 1 January 2015 were excluded to ensure that all patients had newly diagnosed AF.

Patients with valvular heart disease, rheumatic heart disease, or previous valve surgery at the time of AF diagnosis were excluded, as were those with diagnoses/procedures closely related to transient AF (pulmonary embolism, pericarditis, myocarditis, and cardiac surgery) within 90 days before the first AF diagnosis. We also excluded patients with a history of hyperthyroidism, HF before the index date, or death within 30 days of their first occurrence of AF. Patients without fasting blood glucose (FBG) or hemoglobin A_1c (HbA_1c) test results 2 years before the onset of their first AF (defined as the unscreened group) and those diagnosed with type 1 diabetes were further excluded (Supplementary Fig. 1). Patients with an ICD-9–coded diagnosis of diabetes or who had antidiabetic prescriptions within 1 year before the index date were assigned into the diabetes group. Patients without an ICD-9–coded diagnosis of diabetes and had never been prescribed antidiabetic medications at any time were subsequently assigned into three groups depending on their blood glucose status, with reference to the diagnostic criteria proposed by the American Diabetes Association (17), as follows:

• Diabetes: with ICD-9 codes 250.x; or consecutively prescribed antidiabetic drugs for ≥90 days within 1 year before the index date; or two FBG tests within a period of 6 months, both ≥126 mg/dL; or two HbA_1c tests within a period of 1 year, both ≥6.5%; or one FBG test ≥126 mg/dL and one HbA_1c test ≥6.5% within a period of 1 year.

• Prediabetes: without ICD-9 codes 250.x; and never prescribed antidiabetic medications; and no record of an FBG test ≥126 mg/dL or HbA_1c test ≥6.5% within a period of 2 years before the index AF diagnosis; and two FBG tests within a period of 6 months, both 100–125 mg/dL; or two HbA_1c tests within a period of 1 year, both 5.7–6.4%; or one FBG test 100–125 mg/dL and one HbA_1c test 5.7–6.4% within a period of 1 year.

• Normoglycemia: without ICD-9 codes 250.x; and never prescribed antidiabetic medications; and no record of an FBG test ≥100 mg/dL or HbA_1c test ≥5.7% within a period of 2 years before the index AF diagnosis.

In the prediabetes group, patients were further classified into the following groups at 2 years after the index date:

• Progression to diabetes: newly diagnosed diabetes on the basis of ICD-9 codes 250.x; or consecutively prescribed antidiabetic drugs for ≥90 days; or two FBG tests within a period of 6 months, both ≥126 mg/dL; or two HbA_1c tests within a period of 1 year, both ≥6.5%; or one FBG test ≥126 mg/dL and one HbA_1c test ≥6.5% within a period of 1 year.

• Persistent prediabetes: without ICD-9 codes 250.x; and never been prescribed antidiabetic medications; and no record of FBG test ≥126 mg/dL or HbA_1c test ≥6.5% within 2 years after the index date; and two FBG tests within a period of 6 months, both 100–125 mg/dL; or two HbA_1c tests within a period of 1 year, both 5.7–6.4%; or one FBG test 100–125 mg/dL and one HbA_1c test 5.7–6.4% within a period of 1 year.

• Reversion to normoglycemia: without ICD-9 codes 250.x; and never been prescribed antidiabetic medications and no record of an FBG test ≥100 mg/dL or HbA_1c test ≥5.7% within 2 years after the index date.

For each patient, we collected information on age, sex, smoking, comorbidities (HF, hypertension, ischemic stroke, transient ischemic attack [TIA], CAD, chronic kidney disease, peripheral vascular disease, venous thromboembolism, rheumatoid arthritis, systemic sclerosis, systemic lupus erythematosus, ankylosing spondylitis, liver cirrhosis, anemia, cancer, gastrointestinal bleeding, dyslipidemia, and obesity), and medication history (baseline use of non–vitamin K antagonist oral anticoagulant [NOAC], warfarin, aspirin, ACE inhibitor/angiotensin II receptor blocker [ACEI/ARB], β-blocker, and statins). Baseline medication use was defined as >30 days of consecutive use within 1 year after the index date. Blood glucose data, including FBG and HbA_1c, and serum creatinine were also collected. Estimated glomerular filtration rate was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (18). Procedures such as AF ablation and cardioversion were recorded within 1 year after the index date. Details of the ICD-9 codes used are summarized in Supplementary Table 1.

The primary outcome was the first hospital admission for HF (ICD-9 codes 402, 404, 425, or 428) after the index date. Patients were followed until the occurrence of HF, death, or the study end date (31 October 2021), whichever came first.

Prediabetes and normoglycemia were defined according to laboratory results, including FBG and HbA_1c levels, ensuring the accuracy of these classifications. Among patients with diabetes, the majority were diagnosed by the presence of either antidiabetic drug prescriptions (n = 7,102 [70.1%]) or laboratory results (n = 8,386 [82.8%]). Only 472 (7.2%) were classified as having diabetes solely by ICD-9 coding. Because patient information is anonymous in CDARS, we validated 110 patients classified as having diabetes using the electronic medical record system at our center (Queen Mary Hospital), of whom 109 (99.1%) had a confirmed diagnosis. This high degree of accuracy of using ICD-9 diagnosis codes in CDARS has been extensively validated (13,15,19,20).

Continuous variables are presented as means with SDs or medians with interquartile range (IQR), and categorical variables are presented as counts with proportions. Differences between groups were compared using the t test for continuous variables and the χ² test for categorical variables. Incidence rate of HF was calculated as the number of events per 100 person-years of follow-up. Cumulative incidence curves were estimated while considering all-cause mortality as a competing risk, and event rates between groups were compared using the log-rank test.

We used a propensity score approach to minimize bias due to differences in baseline covariates. Covariates that were considered prognostically significant were logistically regressed to the probability of having the glycemic status using covariate balancing propensity score (CBPS). The covariates included in the CBPS model were age, sex, comorbidities, baseline medications, and procedures. Inverse propensity of treatment weighting (IPTW) was used to generate a pseudopopulation in which the prevalence of baseline covariates was well balanced; differences in the prevalence of covariates were considered not significant if the standardized mean difference was ≤0.10 (12,19). Using a multivariable Cox proportional hazards regression model, we evaluated the association between baseline glycemic status and outcomes of interest. To minimized confounding, we used the doubly robust estimation technique, where covariates included in the CBPS model were also included in the multivariable Cox model. To account for competing risks, the Fine and Gray model was applied, with death defined as the competing event. Restricted cubic spline was performed to estimate the hazard ratio (HR) for incident HF in relation to increasing HbA_1c levels.

In the prediabetes group, the subsequent risk of HF was further evaluated according to changes in glycemic status 2 years after the index date of AF. Using a similar multivariable Cox proportional hazards regression model with the Fine and Gray competing risk model, we evaluated the risk of HF associated with progression (to diabetes) or reversion (to normoglycemia) of prediabetes, with persistent prediabetes defined as the referent.

Sensitivity analyses were conducted without considering IPTW and further adjustment for BMI and blood pressure in a subpopulation with available data. We also performed sensitivity analysis by censoring at the last clinic visit date. A two-sided P < 0.05 was considered significant. Data analyses were performed using R version 4.0.2 statistical software.

We identified 25,142 patients with AF (mean age 75.2 years, 47.1% female), among whom 3,711 (14.8%) were classified as normoglycemic, 4,105 (16.3%) as prediabetic, and 10,127 (40.3%) as diabetic at baseline. A total of 7,199 (28.6%) patients with AF had no blood glucose laboratory test results 2 years before the index date and were considered as unscreened subjects who were excluded from further analyses.

Baseline characteristics of the patient population are summarized in Supplementary Table 2. While age and sex were similarly distributed across groups, patients with prediabetes had more TIA and CAD and had the highest proportion of use of NOAC and aspirin than those with normoglycemia and diabetes. Patients with prediabetes were less often female and smokers and had a lower prevalence of venous thromboembolism, liver cirrhosis, anemia, and cancer compared with the other two groups.

The baseline characteristics among patients with normoglycemia, prediabetes, and diabetes were well balanced after IPTW (Supplementary Tables 2 and 3). During a median follow-up of 4.7 years (IQR 4.6–4.7), incident HF developed in 518 (14.0%) patients with normoglycemia, 646 (15.7%) with prediabetes, and 1,795 (17.7%) with diabetes, corresponding to incidence rates of 3.02, 3.14, and 3.38 per 100 person-years, respectively. Cumulative incidence curves demonstrated a graded increase in HF from the normoglycemia to diabetes group (Fig. 1A). After adjustment for age and sex, the risk for HF was 10% higher in patients with prediabetes (subdistribution HR [SHR] 1.10, 95% CI 0.95–1.27, P = 0.19) and 26% higher for those with diabetes compared with those with normoglycemia (SHR 1.26, 95% CI 1.13–1.40, P < 0.01). After adjustment for confounding factors, prediabetes was associated with a 12% increased risk of HF (SHR 1.12, 95% CI 1.03–1.22, P = 0.02) and diabetes with a 30% higher risk of HF (SHR 1.30, 95% CI 1.19–1.39, P < 0.01) compared with normoglycemia (Table 1). Cubic spline values (Fig. 2) showed a continuous increased risk of HF with consecutively increasing HbA_1c. When HbA_1c was presented as a continuous variable, each percentage increase of HbA_1c was associated with an 8% increased risk of HF (SHR 1.08, 95% CI 1.02–1.14, P < 0.01).

In patients with prediabetes at baseline, 3,620 received a reassessment of glycemic status, of whom 403 (11.1%) progressed to diabetes, 311 (8.6%) reversed to normoglycemia, and 2,906 (80.3%) remained prediabetic at 2 years after the index AF diagnosis (Supplementary Fig. 2). The baseline FBG of patients who progressed to diabetes was significantly higher than those who remained prediabetic and reversed to normoglycemia (112.7 [IQR 106.3–124.1] vs. 106.3 [IQR 102.7–111.5] and 104.5 [IQR 101.9–108.2] mg/dL, respectively, both P < 0.01). The cumulative incidence curves for HF based on changes in glycemic status demonstrated that patients progressing to diabetes had the highest risk of HF, followed by those who remained prediabetic and who reversed to normoglycemia, respectively (Fig. 1B). After multivariable adjustment, compared with those who remained prediabetic, patients who progressed to diabetes had a 50% higher risk of HF (SHR 1.50, 95% CI 1.13–1.97, P = 0.01), whereas those who reversed to normoglycemia incurred a 39% lower risk of HF (SHR 0.61, 95% CI 0.42–0.94, P = 0.02) (Table 2).

The association between glycemic status and HF remained consistent in the entire study population without considering IPTW for baseline covariates. After multivariable adjustment, baseline prediabetes (SHR 1.06, 95% CI 1.02–1.12, P = 0.03) and diabetes (SHR 1.23, 95% CI 1.12–1.37, P < 0.01) were associated with a higher risk of HF than normoglycemia (Supplementary Table 4). Similar results were found in change of prediabetes group in 2 years after the AF index date. Compared with the persistent prediabetes state, progression to diabetes was associated with an increased risk of HF (SHR 1.49, 95% CI 1.11–1.99, P = 0.01), while reversion to normoglycemia was associated with a decreased risk of HF (SHR 0.63, 95% CI 0.41–0.97, P = 0.03) (Supplementary Table 5). After further adjustment for BMI and blood pressure, sensitivity analysis revealed that prediabetes and diabetes were associated with a higher risk of HF compared with normoglycemia (Supplementary Tables 6 and 7). Among 17,943 patients in the entire study cohort, 322 (1.8%) did not have any clinic visits after the index date. By censoring at the last clinic visit date, sensitivity analysis showed that the association between glycemic status and the risk of HF remained consistent (Supplementary Table 8).

In this territory-wide cohort study of patients with new-onset AF, we confirmed that HF is a common complication, with an event rate of 3.2 per 100 person-years. After adjustment for known risk factors, prediabetes was associated with an increased risk of developing HF compared with normoglycemia. Furthermore, temporal changes of glycemic status in patients with prediabetes could modify the subsequent risk of HF; those who reversed to normoglycemia at 2 years had a 39% reduced risk of HF, whereas those who developed diabetes had a 50% increased risk of HF.

Perhaps because of a well-established anticoagulation strategy, HF rather than stroke/embolism has now emerged as the leading complication in patients with AF (6). In a study involving Framingham Heart Study participants who were free of HF at baseline, the incidence of HF following AF was 3.3 per 100 person-years (21), similar to findings of the current study. Similarly, the Randomized Evaluation of Long-Term Anticoagulant Therapy (RE-LY) trial (22) and Rate Control Efficacy in Permanent Atrial Fibrillation: a Comparison Between Lenient Versus Strict Rate Control II (RACE II) trial (23) have revealed HF as the primary driver of cardiovascular mortality and morbidity in patients with AF. Thus, HF prevention represents an important management priority for patients with AF (5). In this regard, preventive strategies may draw on the most relevant and potentially modifiable risk factors for HF, and diabetes has emerged as a prominent candidate for such a role in AF.

The high global prevalence of prediabetes (34.4% in the U.S. [24] and 15.5% in China [25]), alongside its assorted complications, lends itself to increasing clinical relevance as both a risk marker and risk factor in AF. In accordance with previous literature (26,27), our study confirms the adverse effect of diabetes among patients with AF and demonstrates that prediabetes is associated with the subsequent risk of HF, advocating for increased awareness of its impact on future morbidity and mortality.

Prediabetes is known to be associated with an increased risk of atherosclerotic cardiovascular disease (10), HF (9), and all-cause mortality (10,28). Mechanistically, the development of HF in prediabetes may be attributed to a higher risk of CAD (28), impaired microvascular function (29), and adverse cardiac remodeling among these patients (30).

In the context of AF, several additional mechanisms may further predispose patients with prediabetes to HF. First, one of the shared mechanisms underlying HF and AF is diastolic calcium leakage from the sarcoplasmic reticulum through dysfunctional ryanodine receptor 2 channels (31), which may be further impaired in hyperglycemic states (32). Second, patients with hyperglycemia are more likely to have reduced manifestations of AF symptoms. This could be explained by hyperglycemia-related autonomic neuropathy that may attenuate the perception of tachyarrhythmia symptoms, leading to a reduced tendency to receive appropriate medical treatment to prevent consequential adverse ventricular remodeling (33). Third, hyperglycemia may exaggerate atrial fibrosis, a uniform characteristic of AF, which may further augment diastolic dysfunction, leading to HF symptoms (34). Fourth, both AF and HF are considered inflammatory diseases that may be intensified by impaired fasting glucose, contributing to a heightened risk of clinical HF events (35). Accordingly, aggressive glycemic control may conceivably reduce HF in patients with AF.

Modifiable cardiovascular risk factors, such as prediabetes, can be influenced by multiple factors and, therefore, tend to be dynamic. Patients with prediabetes will progress to overt diabetes at a rate of ∼5–10% annually, with the same proportion reverting to normoglycemia (7). Our study reveals that changes in glycemic status could modify the subsequent risk of HF in patients with AF. Whereas progression to diabetes carried a 50% excess risk of HF compared with persistent diabetes, reversion to normoglycemia was associated with a 39% decrease in the subsequent risk of HF. Our finding is consistent with an observational study showing that reversion from prediabetes to normoglycemia is associated with a reduced risk of cardiovascular disease (36). As a result, strategies that mitigate the progression to overt diabetes and, if possible, promote reversion to normoglycemia in prediabetes is particularly desirable to reduce the burden of HF in patients with AF. Lifestyle interventions, consistently reported to reduce the risk of developing diabetes in individuals with prediabetes (37–39), may be efficacious, yet their impact on cardiovascular and microvascular complications remains contentious (40–42). In evaluating obese patients with AF, with 25% having diabetes, a randomized study demonstrated that intensive risk factor control for 15 months reduced AF symptom burden and severity (43). Considering its high prevalence and strong association with adverse outcomes such as stroke (11) and HF (as shown in the current study), prediabetes may represent a pertinent window to prevent progression to overt diabetes and its related complications in patients with AF. Future studies are warranted to investigate the impact of lifestyle modification and pharmacological intervention in patients with prediabetes and AF.

Strengths of the current study include the use of a territory-wide, well-validated electronic health care database with records of all diagnoses, laboratory results, and details of dispensed drugs, allowing the collection of the relevant information required to preclude selection and recall biases common to conventional observational studies. The validity of the current results is further enhanced by extensive adjustments for medications/procedures, such as ACEI/ARB and AF ablation, that may modify the risk of HF.

Several limitations of our study merit consideration. AF types (paroxysmal, persistent, and permanent) and treatment strategies (rhythm and rate control) were not distinguished in the current study. Echocardiographic data were not available in CDARS; thus, the differential impact of left ventricular filling pressure/function could not be evaluated. Nevertheless, previous studies have demonstrated similar rates of adverse events across various left ventricular systolic grades in patients with AF (44). Similar to other administrative databases (45,46), socioeconomic, smoking status at index date, anthropometric, and lifestyle data are not systematically available in CDARS. Further studies characterizing sociodemographic and phenotypic features may reveal additional pathophysiological insights into our findings. Longitudinal anthropometric data are required to evaluate the independent association between glycemic changes and the risk of incident HF; however, previous studies have demonstrated that prediabetes is associated with an increased risk of cardiovascular diseases independent of BMI (47,48), supporting the notion that glycemic changes per se may contribute to the risk of incident HF. In Hong Kong, >90% of the local population is under the care of public hospitals that capture relevant data in the CDARS. Although data from patients who have visited private hospitals or have moved aboard cannot be ascertained, this is rare in our locality because most patients who visit public hospitals continue to be followed in this setting. Finally, it is possible that residual confounders remain despite using propensity score matching.

In conclusion, in this population-based cohort of patients with AF, we demonstrated that prediabetes is common and associated with an increased risk of HF. The risk of HF in prediabetes was amplified in patients who progressed to overt diabetes but reduced in those who reversed to normoglycemia.

Funding. This study was supported by Sanming Project of Medicine in Shenzhen, China (grant SZSM201911020), and the HKU-SZH Fund for Shenzhen Key Medical Discipline (grant SZXK2020081).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.-Y.H. completed the initial data preparation and statistical analyses. J.-Y.H. and Y.-K.T. drafted the manuscript. J.-Y.H., G.Y.H.L., and K.-H.Y. were involved with the conception of the study. Y.-K.T., H.-L.L., C.C., C.-T.Z., M.-Y.L., M.-Z.W., Q.-W.R., S.-Y.Y., and D.H. contributed to the statistical analyses. X.-L.L., H.-F.T., G.Y.H.L., and K.-H.Y. provided the clinical expertise. All authors critically reviewed and contributed to the intellectual content of the manuscript and approved the final version of the manuscript. J.-Y.H. and K.-H.Y. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.