Demonstration of the potential of white-box machine learning approaches to gain insights from cardiovascular disease electrocardiograms

We present the results from a white-box machine learning approach to detect cardiac arrhythmias using electrocardiographic data. A C5.0 is trained to recognize four classes using common features. The four classes are (i) atrial fibrillation and atrial flutter, (ii) tachycardias (iii), sinus bradycardia and (iv) sinus rhythm. Data from 10,646 subjects, 83% of whom have at least one arrhythmia and 17% of whom exhibit a normal sinus rhythm, are used. The C5.0 is trained using 10-fold cross-validation and is able to achieve a balanced accuracy of 95.35%. By using the white-box machine learning approach, a clear and comprehensible tree structure can be revealed, which has selected the 5 most important features from a total of 24 features. These 5 features are ventricular rate, RR-Interval variation, atrial rate, age and difference between longest and shortest RR-Interval. The combination of ventricular rate, RR-Interval variation and atrial rate is especially relevant to achieve classification accuracy, which can be disclosed through the tree. The tree assigns unique values to distinguish the classes. These findings could be applied in medicine in the future. It can be shown that a white-box machine learning approach can reveal granular structures, thus confirming known linear relationships and also revealing nonlinear relationships. To highlight the strength of the C5.0 with respect to this structural revelation, the results of further white-box machine learning and black-box machine learning algorithms are presented.

74Subclinical atrial fibrillation before and after acute medical illness or noncardiac surgery: insights from ASSERT

OnBehalf ASSERT Investigators Background Atrial fibrillation (AF) is frequently detected perioperatively or during acute medical illness. It is unclear if such AF is reversible and unlikely to recur, or is a manifestation of paroxysmal AF. Objective To compare the prevalence of pacemaker-detected, subclinical AF (atrial rate >190 bpm) before and after hospitalization for noncardiac surgery or medical illness in patients without a history of clinical AF. Methods ASSERT enrolled patients who were >65 years old and had hypertension but no known AF. Pacemakers and defibrillators recorded episodes of subclinical AF. We identified participants who were hospitalized for noncardiac surgery or medical illness, and created heart rhythm profiles, centred on the day of hospitalization. We compared the prevalence of subclinical AF before and after hospitalization. We blanked the 30 days before hospitalization, because of uncertainty in defining the precise onset of illness. Results Among 2580 patients, 436 had a documented surgical or medical hospitalization. In the 30 days following a first hospitalization, 43 patients (9.9%) had >1 episode of >6 minutes of subclinical AF; 20 (4.6%) had >6 hours and 13 (3%) had >24 hours. A higher proportion of patients had >1 episode of subclinical AF >6 minutes in the 30 days following a first surgical or medical hospitalization, as compared to the period between 30 and 60 days before hospitalization (9.9% versus 4.4%, P < 0.001). There was no significant difference when comparing 0-90 days after hospitalization to 30-120 days before (13.7% versus 10.6%, P = 0.1). Similar results were observed for the same comparisons with episodes >6 hours (4.6% versus 2.3%, P = 0.03 and 5.9% versus 5.6%, P = 0.8, respectively). The majority of patients with subclinical AF in the 30 days following hospitalization had at least one episode of subclinical AF of the same duration in the 6 months prior (50% for episodes >6 minutes; 69% for >6 hours and 60% for >24 hours). Those who did have subclinical AF in the 30 days following hospitalization were more likely to have had subclinical AF in the past 6 months than those who did not (OR 7.2 95%CI 3.2-15.8 for episodes >6 minutes; OR 32.6, 95%CI 10.3-103.4 for >6 hours and OR 36.3 95%CI 9.0-146.0 for >24 hours). Conclusions The prevalence of subclinical AF increased following hospitalization for noncardiac surgery or medical illness. However, most patients with subclinical AF following hospitalization had previously experienced similar episodes, particularly those with longer episodes of subclinical AF.

Atrial flutter

Atrial flutter is the term given to one of the four types of supraventricular tachycardia; in it, atrial activation occurs as a consequence of a continuous ‘short circuit’: a defined and fixed anatomical route, resulting in a fairly uniform atrial rate, and uniform atrial flutter waves on the ECG. The ventricles are not a part of this arrhythmia circuit, and ventricular activation is variable, dependent on atrioventricular (AV) nodal conduction. Given that the atrial rate is essentially uniform (e.g. 300 min−1), ventricular activation tends to be regular (i.e. 150 min−1, 100 min−1, 75 min−1, etc., if the atrial rate is 300 mins−1), or regularly irregular if changes are occurring in the fraction of conducted impulses to the ventricles. When AV nodal conduction permits only 4:1 conduction or less, atrial flutter is usually obvious, but when ventricular rates are higher (150 min−1 or more) the flutter waves can be obscured by the QRS complexes, making diagnosis more difficult. Atrial flutter is of two types, typical and atypical. Typical atrial flutter is a right atrial tachycardia, with electrical activation proceeding around the tricuspid valve annulus. This arrhythmia is dependent on a zone of slow electrical conduction through the cavotricuspid isthmus (the tissue lying between the origin of the inferior vena cava and the posterior tricuspid valve). The resulting circuit can be either anticlockwise (activation proceeds up the inter-atrial septum, across the atrial roof, down the free wall, and then through the cavotricuspid isthmus to the basal septum) or clockwise (down the inter-atrial septum and around the circuit in the opposite direction). Anticlockwise typical atrial flutter is more common. Atypical atrial flutter refers to all other atrial flutters, and this includes other right atrial flutters (e.g. pericristal flutter), left atrial flutters, post-ablation or post-surgical flutters, and pulmonary vein flutters. The feature common to all types of flutter and which differentiates flutter from other types of supraventricular tachycardia is the presence of a macro-re-entrant anatomical circuit around which the electrical impulse travels continuously and repeatedly, thereby generating the flutter. Even though typical atrial flutter has a fairly obvious and specific appearance on the ECG, atypical flutters do not, and often it is only possible to differentiate atypical flutter from atrial tachycardias by invasive electrophysiology studies, as the ECG alone may be insufficient.