Classification trees.
Classification trees were applied in order to identify subgroup of patients who are at low or high risk for arrhythmia. Figure 4 demonstrates a classification tree incorporating baseline patients’ characteristics including prior arrhythmias, as well as clinical symptoms and disease severity. As seen, the first division for discriminating patients with and without new arrhythmia is based on age. Patients aged ≥ 70 compared to < 70 years had 18.1% vs. 2.6% prevalence of arrhythmias, respectively. The next divisions of the model include illness severity and background CHF. The highest prevalence of arrhythmias was 33.3% seen in patients aged ≥ 70 years and critical disease severity. The lowest prevalence of 1.6% was seen among patients aged < 70 years with mild to moderate disease severity. A second classification tree incorporated the same clinical parameters of the first tree with addition of hsTnI levels and CRP (Figure 5). Of note, hsTnI levels were measured in 219/390 patients (56%). As seen in Figure 5, hsTnI was the best discriminator for the new arrhythmia occurrence and obviated the contribution of other clinical parameters. Patients with negative or not measured hsTnI had an arrhythmia prevalence of 2.1% compared to patients with troponin ≥ 48 ng/L who had an arrhythmia prevalence of 34.1%.