Background
Trastuzumab carries a black box warning for cardiotoxicity, and HER-2 positive breast cancer patients who are treated with trastuzumab are recommended to have an echocardiogram (TTE) every three month while on therapy and even every six months for the first two years after therapy based on the package insert. However, the need and cost-effectiveness of such practice is unclear. In the past we have shown that an artificial intelligence (AI) model can detect a low ejection fraction using 12-lead electrocardiography (ECG). The applicability of this technique for the detection of trastuzumab cardiotoxicity is unknown.
Methods
We studied female patients (n=330) who received Trastuzumab at three Mayo Clinic campuses (Minnesota, Arizona, Florida) from 2001 and 2019 and has TTEs during or one year after Trastuzumab therapy as well as ECGs within 2 weeks of the echocardiograms. All ECGs that were recorded during and up to 1 year after Trastuzumab therapy were scored using the model and compared to echocardiography-derived ejection fraction values.
Results
Overall, we had 515 TTE and ECG pairs ≤ 14 days apart or less from 330 unique patients. Of these, 24 TTEs from 18 patients showed an EF ≤ 40% and 15 TTEs from 11 patients showed an EF ≤ 35%. The AUC of the AI-ECG model for the detection of an EF ≤ 35% was 0.92 and for an EF ≤ 40% was 0.83. The sensitivity, specificity, and accuracy of the AI-ECG model to detect an EF ≤ 40% were 79%, 77%, and 77.1%, respectively. Accordingly, using the AI-ECG algorithm for the screening of EFs ≤ 40% in trastuzumab-treated patients with a very sensitive threshold, 44% of screening TTEs could be avoided without missing a single patient.
Conclusion
An AI-augmented ECG algorithm to detect an EF ≤ 40% in patients treated with trastuzumab is highly accurate. The AI ECG coupled with a smartphone may allow at home, inexpensive, self-administered long-term monitoring to enhance trastuzumab safety.