Article Text
PDF
Abstract
Background Subtle, prognostically-meaningful ECG features may not be apparent to physicians. In the course of supervised machine learning training, many thousands of ECG features are identified. These are not limited to conventional ECG parameters and morphology. These novel neural network (NN)-derived ECG features may have clinical, phenotypic, and genotypic associations and prognostic significance.
Methods and Results We extracted 5120 NN-derived ECG features from an AI-ECG model trained for six simple diagnoses and applied unsupervised machine learning to identify three phenogroups. The derivation set, the Clinical Outcomes in Digital Electrocardiography (CODE) cohort (n = 1,558,421), is a database of ECGs recorded in primary care in Brazil. The three phenogroups had significantly different mortality profiles (Figure 1). After adjusting for known covariates (including age, gender, and comorbidities), phenogroup B had a 1.2-fold increase in long-term mortality compared to phenogroup A (HR 1.20, 95% CI 1.17-1.23, p < 0.0001).
We then externally validated our findings in four diverse cohorts. The Whitehall II cohort (n = 5,066) consists of British civil servants. The UK Biobank is longitudinal study of volunteers (n = 42,386). The Longitudinal Study of Adult Health (ELSA-Brasil) cohort (n = 13,739) consists of Brazilian public servants. Lastly the São Paulo-Minas Gerais Tropical Medicine Research Center (SaMi-Trop) is a cohort (n = 1,631) of patients with chronic Chagas cardiomyopathy.
We found phenogroup B had a significantly greater risk of mortality in all cohorts (Figure 1). We performed a phenome-wide association study (PheWAS) in the UK Biobank. We found ECG phenogroup significantly associated with cardiac and non-cardiac phenotypes, including cardiac chamber volumes and cardiac output (Figure 2A). A single-trait genome-wide association study (GWAS) was conducted. The GWAS yielded four significant loci (Figure 2B). SCN10A, SCN5A and CAV1 have well described roles in cardiac conduction and arrhythmia. ARHGAP24 has been previously associated with ECG parameters, however, our analysis has identified for the first time ARHGAP24 as a gene associated with a prognostically significant phenogroup. Mendelian randomisation demonstrated the higher risk ECG phenogroup was causally associated with higher odds of atrioventricular block but lower odds of atrial fibrillation and ischaemic heart disease.
Conclusion NN-derived ECG features have important applications beyond the original model from which they are derived and may be transferable and applicable for risk prediction in a wide range of settings, in addition to mortality prediction. We have shown the significant potential of NN-derived ECG features, as a highly transferable and potentially universal risk marker, that may be applied to a wide range of clinical contexts.
(A) Genome-wide association study. Manhattan plots of genomic loci associated with ECG phenogroup. Nearest genes are annotated on the plot. (B) Mendelian randomisation analyses of associations between genetically predicted ECG phenogroup and cardiovascular outcomes/phenotypes (C) Grad-CAM is used to generate importance maps showing the sections of the ECG signal deemed most important for phenogroup determination. HF: heart failure, BMI: body mass index, SBP: systolic blood pressure. SCD: sudden cardiac death.
Conflict of Interest None