3.3. Clinical Predictors for New-Onset Arrhythmias
We analyzed predisposing factors associated with arrhythmias during hospitalization for COVID-19 comparing baseline characteristics of patients with arrhythmia and patients without arrhythmia during hospitalization for COVID-19 (Table 1). Patients with arrhythmias were older and more often had been diagnosed with hypertension and cardiovascular disease (Table 1). QTc duration at baseline was longer in patients with arrhythmia, however, median values were within the physiological range in both groups (Table 1). Furthermore, we analyzed the prevalence of potentially proarrhythmic medication administered in the context of COVID-19. In both subgroups, hydroxychloroquine was used in 44.1% and 44.6% of patients, respectively (Table 1). A smaller fraction of patients in both groups additionally received azithromycin. There was no statistically significant difference in the use of QT-prolonging drugs between both groups (Table 1). Median QTc-duration at baseline was within normal range in patients who later received hydroxychloroquine (409.0 ms; P25: 390.5 ms, P75: 421.5 ms).
Regularized logistic regression led to the selection of the variables age, cardiovascular disease and hypertension with respect to the prediction of arrhythmia incidence. The subsequently fitted multiple logistic regression model revealed significant association of age (OR 1.036; 95% CI 1.004–1.074; p = 0.036) and cardiovascular disease (OR 3.307; 95% CI 1.329–8.232; p = 0.01) with incident arrhythmia in COVID-19, whereas the effect of hypertension was not significant (OR 2.08; 95% CI 0.794–5.796; p = 0.144). As measure of the model’s performance an area under the curve (AUC) value of 0.74 (95% CI: 0.65; 0.84) was estimated by applying 5-fold cross-validation (Supplementary Figure S1). Left ventricular ejection fraction was documented in 47.1% of patients with incident arrhythmia and only 21.2% of cases without arrhythmia during hospitalization. Attempts at imputing LVEF and including the imputed dataset in the final logistic regression model hinted at a potential role of LVEF as an additional predictor for arrhythmia incidence (Table S1). However, due to the high number of missing values, LVEF was omitted from the final logistic regression model to ensure reliability of statistical analyses.
With regard to peak levels of cardiac and inflammatory biomarkers assessed during hospitalization, patients with arrhythmia displayed higher levels of hsTnT and NTproBNP (Figure 3A, B). Additionally, a more pronounced increase in IL-6 and LDH could be detected in the arrhythmia subgroup, whereas there was no statistically significant difference in peak levels of CRP between groups (Table 2, Figure 3C, D).