Large electrocardiogram (ECG) dataset analyses have emerged as potential game-changers in the field of personalized predictive medicine. ECG parameters have been described in cohorts of apparently healthy subjects and from primary care but seldom in community-based representative populations.

To describe the ECG phenotypes of a representative subset of the adult French general population.

ECGs recorded at inclusion in the CONSTANCES cohort were automatically analysed using the Glasgow diagnostic algorithm. Extreme values and abnormal statements were adjudicated to detect false positives. A subset of ECGs that were classified as normal were also adjudicated to estimate false negative statements. The data obtained were used to describe the prevalence and distribution of quantitative parameters and diagnostic statements.

We automatically analysed the ECGs of 143,763 subjects (54% female; mean ± standard deviation age 47.0 ± 13.5 years in females and 46.9 ± 13.5 years in males; P = 0.44) and adjudicated > 10,000 ECGs. We describe the distribution of automatic ECG interval measurements and the prevalence of different ECG statements provided by the automatic analysis, before and after adjudication. Heart rate and interval durations were dependent on both sex and age (ANOVA P < 0.0001). At the population level, the Fridericia formula appeared to be less biased than that of Bazett.

We describe the adjudicated ECG phenotypes of a representative subset of the adult French general population. The automated measurement and interpretation provided by the Glasgow algorithm proved highly efficient for epidemiological evaluation. This ECG phenotype characterization will open up the possibilities for cross-analyses, for artificial intelligence-based algorithm development and will serve as a reference to evaluate serial ECG changes during extended follow-up.