ECG-based quantification/staging of atrial cardiomyopathy: a novel standardized training tool for consistent annotations of amplified P-wave duration

Atrial fibrillation (AF) is associated with an increased risk for stroke and multiple comorbidities (particularly stroke and heart failure). Recent studies revealed that atrial cardiomyopathy (ACM) is an important factor contributing to new-onset AF and the progression from paroxysmal to persistent AF forms. Furthermore, ACM is hypothesized to increase the risk for cardioembolic stroke and heart failure independently of AF. Early detection of ACM might therefore help to prevent AF onset, cardio-embolic stroke and heart failure, thereby improving treatment. The electrical conduction slowing in ACM substrate can be non-invasively quantified by measurement of a prolonged P-wave duration in the electrocardiogram (ECG). Our group demonstrated that analysis of the amplified P-wave duration (APWD) by experts can be used to diagnose and stage ACM with a high sensitivity and specificity. However, a widely available training tool that can be used also by inexperienced physicians and that enables consistent annotations of APWD is currently lacking.

We propose a novel training tool for consistent annotations of APWD. This tool provides 99 anonymized ECGs and their corresponding P-wave ground truth annotations. The ECGs cover APWDs from 100 ms to 210 ms and were chosen based on patients with different extents of bipolar low voltage areas (<0.5 mV) assessed in high-density electro-anatomical voltage maps in sinus rhythm. The user can compare his/her annotations to a ground truth acquired by three experts to improve his/her annotation skills. All three experts had an interobserver agreement > 0.9 for APWD, and annotations that differed by more than 10 ms were re-examined to reach consensus among the three experts.

In this study, three subjects with different skill levels (non-medical postdoc, medical student, first year cardiology resident) annotated 50 ECGs before using the training tool and 50 different ECGs afterward. No ECG was presented twice to the subjects. 1/3 of the ECGs for this study consist of patients without AF and ACM, 1/3 of patients with persistent AF after electrical cardioversion and 1/3 of patients with a thrombus in the left atrial appendage, representing a more extensive stage of ACM. The subjects were given a flowchart with rules to guide the annotation. The following settings were used: high-pass filter with a cut-off frequency of 0.05 Hz and low-pass filter with a cut-off frequency of 40 Hz, sweep speed of 175 mm/s and amplification of 80 mm/mV. The amplified P-wave onset and amplified P-wave offset were annotated by the subjects (non-medical postdoc, medical student and cardiology resident) and compared to the expert annotations.

After using the novel training tool, all three subjects improved their P-wave onset annotation error from -6.8, 2.3, 5.1 ms to -4.5, 1.2, -0.4, respectively. For the P-wave offset, subjects improved from -24, -9.7, -29.1 ms to -7.7, -5.3, -5 ms. The total APWD measurement improved from -17.2, -11.9, -34.3 ms to -3.2, -6.5, -4.6 ms. The improvement in the amplified P-wave offset and APWD was statistically significant in 2/3 subjects. The improvement in the amplified P-wave onset was significant in one subject.

The novel standardized training tool has significantly improved amplified P-Wave analysis in three inexperienced subjects. This training tool can be used to improve annotation skills of future physicians, leading to more accurate and consistent ECG-based quantification/staging of ACM.