Prognostic significance of diabetes mellitus in patients with atrial fibrillation

This is an ancillary study to MISOAC-AF trial (Motivational Interviewing to Support Oral AntiCoagulation Adherence in patients with non-valvular Atrial Fibrillation, ClinicalTrials.gov identifier: NCT02941978) focusing on patients with AF and DM. The design and main results of the MISOAC-AF registry have been previously published [8, 9]. In brief, MISOAC-AF demonstrated the impact of a patient-centered interview on improving patients’ adherence to oral anticoagulation (OAC) treatment.

The study protocol was approved by the Institutional Review Board of Aristotle University of Thessaloniki, and the trial was conducted in compliance with the principles of the Declaration of Helsinki [10].

The data used in our study were extracted from the MISOAC-AF database for the time range of December 2015 to April 2020. Medical history, baseline clinical characteristics, laboratory and echocardiographic data prior to the date of hospital discharge, as well as discharge diagnoses and follow-up data on clinical outcomes were all included in the database. All data were collected by trained independent investigators after obtaining patients’ written informed consent.

The study comprised adult patients, consecutively enrolled in the prospective cohort of the MISOAC-AF trial. Comorbidity with AF was the shared feature of all patients included, irrespective of the discharge diagnoses from the Cardiology Department of AHEPA Hospital of Thessaloniki, Greece. The presence of end-stage disease not permitting the patients’ follow-up, as well as the absence of data (unavailable or unknown) concerning DM status of patients were the exclusion criteria for the present study.

All-cause death, i.e. death from any cause, constituted the primary outcome, whereas CV death and specific causes of CV death [cardiac arrest, heart failure (HF)-related death, pulmonary embolism, stroke, hemorrhage], hospitalizations related to AF or HF, stroke and major bleeding episodes, as well as the composite outcome of hospitalization or CV death during the follow-up period were the secondary outcomes. Information was acquired by reviewing discharge letters and through telephonic or in-person interviews with the study participants. The vital status of all patients was additionally verified through the Greek Civil Registration System. A group of blinded to patient randomization individuals were responsible for the adjudication of all events investigating all available follow-up sources.

Patients with DM were defined as those who met at least one of the following criteria: (1) administration of oral anti-diabetic or insulin medication; (2) diagnostic code of DM (International Classification of Diseases-11) in at least one hospital admission; (3) verification of a previous physician-assigned DM diagnosis by study personnel with access to patients’ medical record. AF was defined as previously recorded in medical history or new-onset AF occurring during hospitalization. The latter was recorded as irregular heart rhythm for more than 30 s, without detectable P waves, by performing either a 12-lead electrocardiogram or a 24-h Holter monitor.

Categorical variables are presented as frequencies with percentages, while continuous ones as means with standard deviations. The Pearson Chi square or Fisher’s exact test were used for categorical variables, whereas continuous variables were compared using the Wilcoxon rank-sum or Student’s t-test, depending on the normality of data distributions. The Kaplan–Meier curves were traced for illustrating time-to-event outcomes (e.g. patients censored in the event of death) between patients with and without DM, applying the log rank test for the comparison of the results. Univariable analysis identified significant predictors of each outcome among the clinically relevant parameters. The Cox regression hazard model was developed by forcing clinically relevant variables, univariably associated with each endpoint, into the multivariate analysis. These covariates included: age, body mass index (BMI), prior stroke, coronary artery disease or prior coronary revascularization procedure, renal function, AF subtype (first-diagnosed, paroxysmal, persistent or permanent), use of OAC, angiotensin converting enzyme inhibitors or angiotensin II receptor blockers (ACEI-ARB) and rate control medication after discharge. On our analysis we considered the group of patients without DM as the reference category.

To calculate the risk of the secondary outcomes (stroke, major bleeding, AF- and HF- hospitalizations) during follow-up, a modified Cox regression analysis as described by Fine and Gray was utilized in a setting, accounting for the competing risk of all-cause death [11]. The adjusted hazard ratios (aHR) and subdistribution hazard ratios (extracted from competing-risks regression analyses) are presented with the respective 95% confidence intervals (CI). Moreover, in order to assess the relationship between continuous HbA1c levels and each outcome, we performed a spline curve analysis. These spline curves display the adjusted hazard ratios for the outcome on the y-axis versus the HbA1c blood levels on the x-axis, whereas dashed lines represent 95% confidence intervals. Following backward regression of candidate confounding variables, the hazard ratios were adjusted for age, gender, BMI, history of dyslipidemia, smoking history, alcohol consumption history, N-terminal pro b-type natriuretic peptide (NT-proBNP), estimated glomerular filtration rate (eGFR), and hemoglobin stages at discharge.

All analyses were conducted with the SPSS Statistics for Windows, Version 26.0 (Armonk, NY: IBM Corp) and Stata statistical software, release 13 (StataCorp) and a two-sided p value of less than 0.05 was considered statistically significant.

A total of 1140 patients were registered in our database, out of which 1109 were analyzed and 31 were omitted due to incomplete data on DM status. Comorbid DM was present in 373 (33.6%) patients.

Baseline clinical and demographic characteristics of the study participants by diabetes status are given in Table 1. Patients with DM had higher age and BMI than non-DM counterparts. Furthermore, DM was significantly associated with the prevalence of other medical conditions, such as hypertension, dyslipidemia, chronic obstructive pulmonary disease, HF, stroke, myocardial infarction, vascular and chronic kidney disease.

Overall, the most prominent AF type was persistent or permanent AF, afflicting more than half of the population (52.5% of diabetic patients and 50% of non-diabetic patients). The presence of DM was associated with a significant increase in risk stratification CHA₂DS₂-VASc score [Congestive heart failure, Hypertension, Age ≥ 75 years, Diabetes mellitus, Stroke, Vascular disease, Age 65–74 years, Sex category]-since it is included in the score per se-, as well as HAS-BLED score (Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile International Normalized Ratio, Elderly, Drugs/alcohol concomitantly) [12] (both p values < 0.001).

During a median follow-up of 2.6 years, 414 (36.3%) patients died (Table 2). In particular, 73.3% of deaths were attributed to CV cause, while the presence of DM was significantly associated with higher prevalence of CV death (34.9% versus 23.5%, p < 0.001). In both diabetics and non-diabetics, the most common cause of death was cardiac arrest (DM: 19.3%, non-DM: 10.9%, p < 0.001). HF-related death was the second more common cause of mortality (DM: 13.4%, non-DM: 9.6%, p = 0.058).

Patients with AF and comorbid DM had worse survival rate compared with those without DM (all-cause death: HR = 1.50, 95% CI 1.20–1.86, p < 0.001) (Fig. 1). Moreover, patients with DM had a significantly higher risk for CV death (HR = 1.56, 95% CI 1.21–2.02), cardiac arrest (HR = 1.94, 95% CI 1.41–2.67), death due to HF (HR = 1.53, 95% CI 1.06–2.19), stroke (HR = 1.92, 95% CI 1.12–3.27), and the composite outcome of CV-death or hospitalization (HR = 1.37, 95% CI 1.17–1.61).

After adjusting for potential covariates, DM remained significant for predicting all-cause death (aHR = 1.44, 95% CI 1.12–1.85), CV-death (aHR = 1.44, 95% CI 1.08–1.93), cardiac arrest (aHR = 1.73, 95% CI 1.19–2.52), stroke (aHR = 1.87, 95% CI 1.01–3.45) and the composite outcome of CV-death or hospitalization (aHR = 1.28, 95% CI 1.06–1.54) during follow-up (Fig. 2). Both univariable and multivariable competing-risk regression analyses on cumulative incidence of major bleeding events, AF- and HF-related hospitalizations did not yield significantly higher hazard for those events in patients with DM than in those without DM (Fig. 3).

The Kaplan–Meier curves (Fig. 1b) demonstrated difference in the survival rates according to the mode of DM treatment (p < 0.001). Specifically, DM patients under both insulin and oral medication had higher mortality rates during follow-up (aHR = 2.06, 95% CI 1.32–3.23) than those solely on insulin injection, whereas patients with no pharmaceutical intervention or just lifestyle measures had the best prognosis after discharge.

Our multivariable Cox regression models identified the following predictors of survival as important to AF patients, apart from the presence of DM: age, eGFR, and the use of rate control (b-blocker or/and digoxin), ACEI-ARB and NOAC medication after discharge (p value < 0.05) (Fig. 4).

Glycemic status affected mortality, with a 1% increase in HbA1c corresponding to higher all-cause mortality rates (aHR: 1.72, 95% CI 1.13–2.61) and CV mortality rates (aHR: 1.81, 95% CI 1.16–2.82). However, HbA1c levels, as a continuous variable, did not significantly predict hospitalization (aHR: 0.81, 95% CI 0.54–1.23) or stroke incidence (aHR: 0.31, 95% CI 0.06–1.66) during follow-up. Multivariable-adjusted restricted cubic spline analysis (Fig. 5a) suggested an almost linear association between HbA1c levels and the risk for all-cause death, in which an HbA1c value of 6.6% corresponded to aHR of 1. The risk for all-cause mortality was significantly lower at HbA1c levels less than 6.2%. Despite the existing trend towards increased mortality in HbA1c > 6.6%, only HbA1c levels 7.6–8.2% were independently associated with an increased risk of all-cause death. In the spline curve analysis on CV mortality (Fig. 5b) the overall shape of the curve remained linear. This curve had an even more positive slope than the corresponding for all-cause mortality, but the only statistically significant correlation depicted was the lower CV mortality risk (aHR < 1) in the HbA1c range of 5.6–6.2%.

This study is subject to several limitations. Firstly, DM status was evaluated only at enrollment and we did not ascertain possible new cases arising during the course of follow-up. It is also possible to have some cases of undiagnosed DM at patient enrollment, since oral glucose tolerance test was not performed in individuals not fulfilling the DM classification criteria. Further, a follow-up of 2.6 years potentially failed to depict some of the long-term effects of DM upon the assessed outcomes, whereas the onset, the type and total duration of DM was not coded in our database. We also note the unavailability of HbA1c values for the majority of the non-diabetic study participants and the reliance on a single HbA1c measurement at baseline for the subsequent evaluation of hazard for adverse events during the follow-up period. Moreover, we cannot exclude potential misclassification in registered causes of death and coding inaccuracies, whereas another unavoidable simplification was the classification of some variables (e.g. hypertension) in a merely binary fashion. Notwithstanding the fact that we adjusted for the majority of possible confounders, residual confounding may always exist, owing to the observational nature of our investigation. Additionally, this study lacked a control group of non-AF patients with DM, and therefore we are not able to investigate the influence of AF presence in the DM prognosis. We have also not extracted evidence about novel classes of antidiabetic drugs with potential cardiovascular benefit for AF patients with DM. Lastly, we should not discount the fact that all patients were enrolled in a single institution, which potentially limits the generalizability of our results.