ECG Sex Index (ESI) identified breast cancer in males with an AUC of 0.78 within 30 days of diagnosis, indicating its potential for risk assessment.
Does an AI-derived ECG Sex Index (ESI) identify men at risk for breast cancer?
An AI-derived ECG Sex Index may help identify men at risk for breast cancer, though validation in larger cohorts is needed.
Absolute Event Rate: 0% vs 0%
Abstract Background Breast cancer is rare among males, accounting for less than 1% of breast cancer cases. Research suggests a hormonal link, particularly with increased estrogen levels. Mammography-based screening is recommended for females yet there is no such screening guidelines for males. Due to rareness of breast cancer in men, repurposing of low cost and routinely collected data modalities, e.g. electrocardiogram (ECG) via artificial intelligence (AI), to identify men at risk can result in timely diagnosis and interventions as well as to AI-guided breast cancer screening for males. Purpose We previously developed a novel non-binary representation of sex, namely ECG Sex Index (ESI), which assigns a value between 0 to 1 for a given ECG, where 0 representing female and 1 representing male sex. We hypothesize that ESI-alone is informative about breast cancer risk in males. Methods/Approach We trained a CNN model on for sex detection using 80%:10%:10% random samples from over 3.5 million ECGs in the Wake Forest School of Medicine’s ECG Repository. This dataset included data from 754,761 patients (75% White, 17% Black; 51% female; mean age (SD) 61(17) years). Within the holdout set, we identified males with and without breast cancer and their corresponding ECGs. Cases comprised 18 ECGs from 5 patients, mean age at ECG: 62(22) years, while controls consisted of 2578 ECGs from 512 individuals, mean age at ECG: 61(11) years. We calculated ESI for both cases and controls and then calculated AUC values using ESI as only input and breast cancer as output. We calculated AUCs by age-matching at the time of ECG, we assessed its effectiveness across various time windows (30, 60, 90, 180 days, 1 year, and 5 years post-diagnosis until ECG acquisition). Results/Data In the holdout dataset (%10), the CNN model achieved an AUC of 0.95 (95% CI: 0.95-0.95) and an accuracy of 87% (95% CI: 0.87-0.87) for classifying patients into female or male category from their ECGs. The mean (std) ESI was 0.18 (0.24) for male, 0.11 (0.13) for correctly classified males (n=159,713), and 0.7 (0.14) for misclassified males (n=23,413). The CNN-derived ESI successfully detected breast cancer in men with an AUC of 0.78 (95% CI: 0.57-0.98) when the ECG was captured within 30 days of diagnosis. For case ECGs (n=5), the mean (std) non-binary sex value was 0.43 (0.35), compared to 0.17 (0.23) for matched control ECGs (n=287). Detailed metric values for other time intervals are illustrated in Figure 1. Conclusion ESI effectively identifies breast cancer in males with high accuracy, even without fine-tuning. Future studies are required on larger dataset also including other risk factors for higher accuraciesAUCs with CI for 30, 60, 90, 180 days, 1
Akbilgic et al. (Thu,) reported a other. ECG Sex Index (ESI) identified breast cancer in males with an AUC of 0.78 within 30 days of diagnosis, indicating its potential for risk assessment.