Electrocardiographic Predictors of Mortality: Data from a Primary Care Tele-Electrocardiography Cohort of Brazilian Patients

Gabriela Paixão, Emilly M. Lima, Paulo R. Gomes, Derick M. Oliveira, Manoel Horta Ribeiro, Jamil S. Nascimento, Antônio H. Ribeiro, Peter W. MacfarlaneHearts 2021 · DOI

TL;DR

This paper leverages a massive dataset of over 1.6 million electrocardiograms (ECGs) from primary care patients in Brazil to identify specific ECG abnormalities that predict overall mortality. The findings demonstrate that common ECG findings like atrial fibrillation and bundle branch blocks are significant predictors of death, highlighting the prognostic value of ECGs in routine clinical practice.

What Problem Does This Paper Address?

Cardiovascular diseases are a leading cause of death globally and in Brazil. While the electrocardiogram (ECG) is a low-cost and accessible tool for assessing cardiovascular health, its prognostic value, particularly in large, diverse primary care populations, needs further exploration. Previous studies often relied on smaller cohorts or specialized settings, limiting their generalizability. This research aims to address this by creating and analyzing a large-scale electronic cohort from a primary care telehealth network, linking ECG data with mortality records to identify robust predictors of mortality.

Methods

The study developed the "Clinical Outcomes in Digital Electrocardiography" (CODE) dataset by linking ECG data from the Telehealth Network of Minas Gerais (TNMG) with national mortality information. For ECG abnormality detection, a computational linguistics program was employed to process unstructured text reports, followed by a self-supervised learning classification model based on artificial intelligence (a recurrent neural network). This AI model was trained and validated against manual interpretations by cardiologists. Concordance between the AI's interpretation and either a cardiologist's report or automatic systems (Glasgow/Minnesota codes) was used for diagnosis, with manual review for discrepancies. Statistical analysis involved Kaplan-Meier survival curves and Cox proportional regression models (or Log-normal models for AVB) to assess the prognostic value of ECG abnormalities, adjusted for age, sex, and comorbidities.

Datasets

  • CODE Dataset: This is a newly developed dataset for this study, comprising over 1.6 million valid ECG recordings from 2010-2017, linked to national mortality data. It's derived from the Telehealth Network of Minas Gerais (TNMG) and represents a large primary care population.
  • National Mortality Information System: Used to obtain mortality data for linkage with the ECG cohort.

Key Insights and Ideas

  • Prognostic Power of Routine ECGs: The study strongly reinforces that even common ECG findings, often assessed in primary care, carry significant prognostic weight for overall mortality. This underscores the importance of thorough ECG interpretation beyond just diagnosing acute conditions.
  • AI for Unstructured ECG Reports: The successful application of AI and computational linguistics to extract meaningful diagnostic information from free-text ECG reports is a significant methodological contribution. This opens doors for leveraging vast amounts of existing, unstructured clinical data.
  • Telehealth as a Data Source: The use of a telehealth network's digital ECG database demonstrates its potential as a rich source for large-scale epidemiological research, bridging gaps in data accessibility.
  • Differential Risk of AVB: The study highlights that not all atrioventricular blocks (AVBs) carry the same mortality risk. While second and third-degree AVBs showed a significantly reduced survival, first-degree AVB and Mobitz I AVB did not show a significant association with increased mortality in this cohort.

Key Results

  • The CODE cohort included 1,666,778 valid ECGs from 1,558,421 patients over 16 years old.
  • The overall mortality rate was 3.31% over a mean follow-up of 3.7 years.
  • Significant Mortality Predictors:
    • Atrial Fibrillation (AF): Hazard Ratio (HR) 2.10 (95% CI 2.03–2.17) for overall mortality.
  • Left Bundle Branch Block (LBBB): HR 1.69 (95% CI 1.62–1.76) for overall mortality.
  • Right Bundle Branch Block (RBBB): HR 1.32 (95% CI 1.27–1.36) for overall mortality.
  • Second-degree AVB (2:1): Relative Survival (RS) 0.21 (95% CI 0.09–0.52).
  • Third-degree AVB: RS 0.36 (95% CI 0.26–0.49).
  • Non-Significant Mortality Predictors:
    • Ventricular pre-excitation: HR 1.41 (95% CI 0.56–3.57).
  • Mobitz I second-degree AVB: RS 0.65 (95% CI 0.34–1.24).
  • AF and LBBB were also identified as independent predictors of higher cardiovascular mortality.

Limitations and Future Work

  • Self-Reported Comorbidities: The study acknowledges that comorbidities were self-reported, which may lead to under-reporting. Additionally, the questionnaire was not specifically tailored for the study, potentially omitting important variables like heart failure.
  • AI Classifier Errors: While the AI classifier showed good accuracy, it can still make errors. The study mitigated this by requiring concordance with manual or automatic diagnoses and performing manual revisions on a significant number of ECGs.
  • Mortality Data Heterogeneity: The quality of mortality data from the national system varies by region, potentially leading to misclassification of the cause of death. Probabilistic linkage also has inherent limitations in sensitivity and specificity.
  • Future Work: The authors suggest that large ECG databases, especially when combined with AI, can be powerful tools for developing population-based studies, providing stronger evidence for clinical practice. They are also exploring AI-driven concepts like "electrocardiographic age" as potential new cardiac biomarkers for risk prediction and health promotion.

Summary updated

Abstract

Computerized electrocardiography (ECG) has been widely used and allows linkage to electronic medical records. The present study describes the development and clinical applications of an electronic cohort derived from a digital ECG database obtained by the Telehealth Network of Minas Gerais, Brazil, for the period 2010–2017, linked to the mortality data from the national information system, the Clinical Outcomes in Digital Electrocardiography (CODE) dataset. From 2,470,424 ECGs, 1,773,689 patients were identified. A total of 1,666,778 (94%) underwent a valid ECG recording for the period 2010 to 2017, with 1,558,421 patients over 16 years old; 40.2% were men, with a mean age of 51.7 [SD 17.6] years. During a mean follow-up of 3.7 years, the mortality rate was 3.3%. ECG abnormalities assessed were: atrial fibrillation (AF), right bundle branch block (RBBB), left bundle branch block (LBBB), atrioventricular block (AVB), and ventricular pre-excitation. Most ECG abnormalities (AF: Hazard ratio [HR] 2.10; 95% CI 2.03–2.17; RBBB: HR 1.32; 95%CI 1.27–1.36; LBBB: HR 1.69; 95% CI 1.62–1.76; first degree AVB: Relative survival [RS]: 0.76; 95% CI0.71–0.81; 2:1 AVB: RS 0.21 95% CI0.09–0.52; and RS 0.36; third degree AVB: 95% CI 0.26–0.49) were predictors of overall mortality, except for ventricular pre-excitation (HR 1.41; 95% CI 0.56–3.57) and Mobitz I AVB (RS 0.65; 95% CI 0.34–1.24). In conclusion, a large ECG database established by a telehealth network can be a useful tool for facilitating new advances in the fields of digital electrocardiography, clinical cardiology and cardiovascular epidemiology.
BibTeX
@inproceedings{paixao2021electrocardiographic,
  author = {Gabriela Paixão and Emilly M. Lima and Paulo R. Gomes and Derick M. Oliveira and Manoel Horta Ribeiro and Jamil S. Nascimento and Antônio H. Ribeiro and Peter W. Macfarlane and Antônio Luiz Pinho Ribeiro and Antonio L. P. Ribeiro and Antonio L. P. Ribeiro},
  title = {{Electrocardiographic Predictors of Mortality: Data from a Primary Care Tele-Electrocardiography Cohort of Brazilian Patients}},
  booktitle = {Hearts},
  year = {2021},
  doi = {10.3390/hearts2040035},
}
Figure