Prognostic implications of thyroid disease in patients with atrial fibrillation

This study constitutes a post hoc analysis of patients enrolled in the MISOAC-AF trial [18] (Motivational Interviewing to Support Oral AntiCoagulation adherence in patients with non-valvular Atrial Fibrillation, ClinicalTrials.gov identifier: NCT02941978), a prospective, two-armed, single-center, randomized controlled trial that aimed to address the effect of a motivational-educational intervention on patients’ adherence to oral anticoagulation therapy. The study was conducted in the cardiology department of AHEPA University Hospital of Thessaloniki, Greece, and a thorough analysis of the design, selection criteria, and the results have been published previously [19]. Participants were recruited in this clinical trial between December 2015 and June 2018 after providing their written, informed consent. The follow-up was completed in April 2020. The data for analysis were collected from patient–physician interviews, electronic hospital records, and the Greek National General Health Insurance System.

In this study, patients were included if they were aged over 18 years and had available data on their thyroid profile (TSH, T3, T4, or thyroid status). Based on their thyroid profile, AF patients were divided into three groups: hyperthyroidism, hypothyroidism, and euthyroidism. Further comparisons were made based solely on the serum TSH, T3, and T4 levels separately. Participants with missing data were excluded. The study complied with the ethical principles of Good Clinical Practice and the Declaration of Helsinki [20] and was approved by the Aristotle University ethics committee.

AF was defined as previously recorded in medical history or new-onset AF occurring during hospitalization. The latter concerned irregular heart rhythm for more than 30 s, without detectable P waves, captured by a 12-lead electrocardiogram or a 24-h Holter monitor. Euthyroidism was defined as normal thyroid function with normal serum TSH (0.35–4.5 mIU/mL) and T4 (12–30 pmol/L) levels [21].

Overt hyperthyroidism was characterized by low serum concentration of TSH (< 0.35 mIU/mL) and elevated levels of T3 and/or T4 hormones. The diagnosis of overt hypothyroidism was made when serum TSH levels are abnormally high (> 4.5 mIU/mL) and levels of T4 are lower than normal [22]. T3 levels were defined as normal if they were between 0.9 and 2.8 (nmol/L).

Stroke was defined as a permanent, focal, neurological deficit confirmed by imaging modalities. Transient ischemic attack was defined as new neurologic symptoms or deficit lasting less than 24 h with no new infarction on neuroimaging (if available). Cardiovascular death (CVD) was defined as a death related to cardiac cause or stroke. All data collection was based on the standard definitions as defined in the study.

The primary outcome was all-cause mortality, and the secondary outcomes of interest were CVD and hospitalizations (any cause, AF-related and HF-related). CVD was defined as sudden cardiac death or death due to other cardiovascular causes such as acute myocardial infarction, heart failure, or stroke. AF and HF-related hospitalizations were assessed separately. Patients were followed-up until death occurred or until April 2020. The vital status of all patients was verified through the Greek Civil Registration. Other clinical events were ascertained via telephonic or in-person interviews.

Continuous variables were summarized with means and standard deviations (SDs), while categorical variables were presented using frequencies and percentages. The baseline characteristics were compared between the three groups of the participants with either one-way analysis of variance (ANOVA) for continuous variables or Pearson Chi-square test for categorical variables as a part of a three-way cross-tabulation. Univariate Cox regression analysis was performed to assess the impact of thyroid status, T3, T4, or TSH levels on the outcomes of interest and unadjusted hazard ratios (HRs) were acquired. Variables univariately associated with the outcome of interest (i.e., gender, age in years, BMI, history of major bleeding, history of chronic kidney disease (CKD), and history of stroke) were inserted as covariates in the multivariate Cox proportional hazard models to obtain adjusted HRs (aHRs) for the investigated outcomes. The maximum number of independent variables that could be included in the final model was determined by the rule of one predictor variable for ten events of interest. Therefore, multivariate analysis for hyperthyroidism was not feasible due to the small number of events in a restricted group sample size. Survival analysis was also executed with Kaplan–Meier curves to analyze time-to-event data for patients with hypothyroidism compared to euthyroid patients, and log-rank tests were performed. The level of significance was set to a = 5% (p value < 0.05) while performing the analyses. The results are displayed with the corresponding 95% confidence intervals (CI). Data management and statistical analysis were performed using SPSS version 27 (SPSS Inc., Chicago, Illinois) software, Stata v15.1 (StataCorp, College station, Texas, United States) packages and R Statistical Software (v4.3.0; R Core Team 2023).

Of the 1113 patients of the MISOAC-AF study, 496 patients (mean age 73 ± 11 years; 54% male) had available data on thyroid status and were included in this post hoc analysis. A total of 16 patients (3.2%) had a history of hyperthyroidism, 141 (28.4%) had a history of hypothyroidism, and 339 (68.4%) had no history of thyroid disease. Baseline characteristics stratified by thyroid status are displayed in Table 1. Generally, there was homogeneity among groups regarding demographic characteristics and medical history. The majority of euthyroid patients were males, while more women with AF had concomitant hypothyroidism. Persistent or permanent AF seemed to be the most prevalent types among patients with thyroid disease (16.1%), followed by paroxysmal AF or atrial flutter (13.3%).

The median time interval between patient enrollment and the last date of follow-up was 31 months (interquartile range 10 to 52 months). During follow-up, death from any cause occurred in 200 (40.3%) patients, while the majority (72%) was due to cardiovascular causes. Among them, 134 (67%) were euthyroid patients, 62 (31%) had hypothyroidism, and 4 (2%) patients had hyperthyroidism.

Univariate and multivariate Cox regression analyses are displayed in Fig. 1. Multivariate Cox regression analysis indicated that patients with hypothyroidism had significantly higher hospitalization rates (Fig. 2A) during follow-up (aHR: 1.57, 95% CI 1.12 to 2.20, p = 0.025 by log-rank test) compared to those with no history of thyroid disease. More specifically, hypothyroidism was associated with an increased risk of hospitalizations due to heart failure (aHR: 3.17, 95% CI 1.61 to 6.24, p = 0.0027 by log-rank test, Fig. 2B) and hospitalizations related with AF complications, such as stroke, major and non-major bleeding and AF recurrence (aHR: 1.77, 95% CI 1.09–2.87, p = 0.026 by log-rank test, Fig. 2C). There were no statistically significant differences between the two groups of patients (hypothyroidism and euthyroidism) regarding all-cause death, CVD, and stroke.

During follow-up, there were no statistically significant differences between patients with a history of hyperthyroidism and those with no history of thyroid disease regarding the outcomes of interest: all-cause mortality (unadjusted HR: 0.56, 95% CI 0.21 to 1.52, p = 0.261), risk of CVD (unadjusted HR: 0.59, 95% CI 0.18 to 1.85, p = 0.366), and hospital admission rates (unadjusted HR: 0.79, 95% CI 0.35 to 1.79, p = 0.58).

The HRs generated from the univariate and multivariate Cox regression analyses are displayed in Fig. 3. The multivariate analysis resulted in increased risk of CVD (aHR: 1.03, 95% CI 1.01 to 1.05, p = 0.007) in patients with higher TSH levels compared to those with normal or low TSH levels. High TSH levels were, also, significantly correlated with hospitalizations during follow-up (aHR: 1.06, 95% CI 1.01 to 1.12, p = 0.03), including those due to heart failure (aHR: 1.18, 95% CI 1.03 to 1.35, p = 0.014). No significant differences were noted regarding all-cause mortality (aHR:1.32, 95% CI 0.96 to 1.81, p = 0.081) in relation to the levels of TSH hormone.

Lower T3 levels were correlated with higher all-cause mortality rates (aHR: 0.51, 95% CI 0.31 to 0.82, p = 0.006) and higher risk of CVD (aHR: 0.42, 95% CI 0.23 to 0.77, p = 0.005). Furthermore, lower T3 levels were associated with higher hospitalization admission rates due to heart failure (aHR: 0.17, 95% CI 0.04 to 0.72, p = 0.017). Results of the univariate and multivariate Cox regression analysis are depicted in Fig. 4.

Univariate Cox regression analysis showed no statistically significant relationship between T4 levels and the outcomes of interest: all-cause mortality (unadjusted HR: 0.98, 95% CI 0.92 to 1.04, p = 0.639), risk of CVD (unadjusted HR: 0.96, 95% CI 0.89 to 1.04, p = 0.411), risk of stroke (unadjusted HR: 0.84, 95% CI 0.68 to 1.04, p = 0.122), hospitalization due to heart failure (unadjusted HR: 1.01, 95% CI 0.93 to 1.08, p = 0.828), hospitalization due to AF recurrence (unadjusted HR: 1.03, 95% CI 0.99 to 1.07, p = 0.109), and hospitalization related to complications of AF (unadjusted HR: 1.01, 95% CI 0.97 to 1.06, p = 0.439).

This is a single-center, retrospective post hoc analysis of the MISOAC-AF trial. As a result, it was not designed at the outset of the trial. Hence, a considerable number of cases had to be excluded due to missing values on thyroid status, TSH, or T3, possibly resulting in selection bias. Due to its observational nature, there may be undetected biases and unmeasured or hidden confounders in the correlations of specific factors with adverse events. However, multivariable adjustments were performed for clinically relevant parameters. Our dataset consisted of older AF patients discharged from the hospital, and this may reduce the generalizability of our results to non-hospitalized patients with AF. Moreover, the proportional hazards models were formed using data on the demographics and clinical characteristics of the population as assessed at baseline. Therefore, no new comorbidities that could occur during follow-up or close to the time of the event of interest (i.e., death) were recorded. However, this single data registration at the outset of follow-up limited the retrocausality in the interpretation of our results.