Analysing electrocardiographic traits and predicting cardiac risk in UK biobank

Abstract

The electrocardiogram (ECG) is a commonly used clinical tool that reflects cardiac excitability and disease. Many parameters are can be measured and with the improvement of methodology can now be quantified in an automated fashion, with accuracy and at scale. Furthermore, these measurements can be heritable and thus genome wide association studies inform the underpinning biological mechanisms. In this review we describe how we have used the resources in UK Biobank to undertake such work. In particular, we focus on a substudy uniquely describing the response to exercise performed at scale with accompanying genetic information.

Sex-specific cardiovascular risk factors in the UK Biobank

“…XGBoost has been applied extensively for cardiovascular disease diagnosis [19,[38][39][40][41]. Prior statistical or deep learning models of cardiovascular disease focused on lifestyle factors [19,22,41,42], medical history [19,41], sociodemographics [19,41], dietary and nutritional information [19,41], genetics [41,42], and/or one of four clinical tests: pulse wave analysis [43,44], electrocardiogram [41,45,46], carotid ultrasound [42,47], or magnetic resonance imaging [48][49][50], but not all four.…”

Cardiovascular waveforms - can we extract more from routine signals?

“…26 Similar measures have been extended to preclinical research, to more sensitively screen out cardiotoxic drugs during development. 27 In this special collection, Ramirez et al 28 reviewed the utility of various ECG morphology characteristics. Building on their previous work, that demonstrated that characteristics, such as the Tpe interval are heritable, 29 they described a data sciences case study combining various ECG waveform metrics with corresponding genetic data from a subset of UK Biobank participants, and provided a summary of associations between different genetic loci and the behaviour of the cardiac electrical system under exercise stress testing and recovery.…”

“…Building on their previous work, that demonstrated that characteristics, such as the Tpe interval are heritable, 29 they described a data sciences case study combining various ECG waveform metrics with corresponding genetic data from a subset of UK Biobank participants, and provided a summary of associations between different genetic loci and the behaviour of the cardiac electrical system under exercise stress testing and recovery. 28 Other, less routine, signals also have a part to play in supporting biomedical research, and clinical decisionmaking. In this special collection, Van Daele et al, 30 described how doppler flowmetry can be used to supplement standard arterial blood pressure measurements, providing greater detail on organ specific adverse effects, when screening compounds during medicines development.…”

“…Building on their previous work, that demonstrated that characteristics, such as the Tpe interval are heritable, 29 they described a data sciences case study combining various ECG waveform metrics with corresponding genetic data from a subset of UK Biobank participants, and provided a summary of associations between different genetic loci and the behaviour of the cardiac electrical system under exercise stress testing and recovery. 28…”

Valid inference for machine learning-assisted GWAS

“…Prominent genetic datasets, such as the UK Biobank (UKB) and All of Us, often have incomplete phenotypic data 10 . For example, as of November of 2023, proteomics 11 , brain magnetic resonance imaging (MRI) 12 , heart MRI 13 , dual-energy X-ray absorptiometry (DXA) imaging 6 , electrocardiogram 14 , and metabolomics 15 data in UKB have missing rates ranging from 45% to 94%. Similarly, All of Us has a missing rate of 96% for phenotypes in the Labs & Measurements category of the electronic health record ( Supplementary Figure 1 ).…”