Sudden cardiac death accounts for 100,000 UK deaths per annum. This research takes a completely new approach to ECG interpretation, allowing people to monitor ECGs easily, potentially saving thousands of lives every year. It automates the early detection of long QT syndrome (LQTS) – a symptomless heart condition caused by commonly prescribed medications that can lead to sudden death, even in young apparently healthy people. LQTS can also be congenital or acquired (primary from pharmacological drugs, as well as secondary from cardiac and non-cardiac conditions). We also currently extending the work to also predict, and highlight the ST-T wave changes at risk of sudden cardiac death. To do this, we use human-like AI that is intuitively ‘explainable’.
Interpreting ECGs is extremely challenging. It requires years of training, and there are currently no computerised approaches reliable enough to use in clinical practice. Our multidisciplinary team combined knowledge from cognitive psychology, medicine, and computer science to produce an explainable algorithm that works with >90% accuracy and is easily understood by both clinicians and lay people, as it can be visualised using colour superimposed on the ECG signal. The AI algorithm provides a natural language indication of the risk of TdP leading to sudden death, and can be visualised in colour on the ECG, allowing people to intuitively detect and assess the risk visually.
The algorithm was tested with ECGs from a large clinical study examining four drugs that prolong the QT-interval. It is necessary to test with a wider variety of ECGs from different sources, including congenital and acquired LQTS with dives ST-T wave changes to ensure its efficacy and generalizability.
ECG signal can be affected differently by different genetic mutations, cardiac and non-cardiac conditions, and/or medications, so it is important to test and optimise the algorithm across a range of these (including different types of congenital and acquired LQTS, from drug-induced to cardiac conditions (e.g. myocardial infarction) and non-cardiac conditions (e.g. diabetes mellitus and metabolic causes (e.g. hypokalaemia)). Such diversity and complexity in the acquired form of LQTS, which is by far the most common cause of life-threatening arrhythmia attacks in hospital settings, need to be carefully considered, and to have value in clinical practice the algorithm must be optimised to deal with this using the UK-Biobank.