(P 120) USING AI PREDICTIONS FROM ECG TRACINGS TO PREDICT TOLERANCE TO ANTHRACYCLINE CHEMOTHERAPY IN SARCOMA

Objective: Anthracycline chemotherapy is a first line systemic treatment for most bone and soft tissue sarcomas. Anthracycline-based chemotherapy can have many adverse effects that may lead to poor treatment tolerance, requiring close monitoring. Physician estimate of performance status is often used to predict tolerance of treatment. We aim to assess the utility of artificial intelligence-electrocardiogram (AI-ECG) prediction of age and ejection fraction (EF) to assess tolerability of anthracycline-based chemotherapy. We hypothesized that patients with AI-ECG estimated age higher than chronological age (accelerated ECG aging) will have poor tolerance to therapy and that the AI-ECG low EF algorithm can identify those at risk of cardiotoxicity.

Methods: We identified adult (age > 18) sarcoma patients who were treated with anthracycline chemotherapy at Mayo Clinic Rochester from 2006-2020 and had an ECG within 1 year of treatment initiation. The electronic medical record was reviewed to capture patient and cancer demographics, treatment history, tolerance of treatment (defined as rates of dose reductions, treatment delays, treatment completions, and hospitalizations during/ within 30 days of therapy), and echocardiogram reports. Utilizing previously defined AI-ECG nomograms, we obtained age and EF predictions for each patient. ECG aging was defined as the ratio of predicted ECG age changes divided by the change in biological age between ECG tracings. A high (accelerated) aging rate was defined as a ratio greater than 1 and a low aging rate was defined as less than 1. Changes in AI-ECG age were correlated with rates of treatment toxicity. Kaplan Meier survival curves with a cox-proportional hazards regression model were used to compare overall survival and regression free survival between the high and low aging rate groups. For low EF detection, we measured the sensitivity and specificity of ECG prediction with a threshold of 0.256 to predict an EF of less than 50% prior to treatment and compared how changes in the ECG EF prediction related to changes in EF based on post-treatment echocardiogram.

Results: Forty patients (male = 27 (67.5%)) met inclusion criteria. Median age at diagnosis was 56.5 years (18-76 years). 33 patients (82.5%) had localized disease. Median anthracycline dose received was 180 mg/m2 (30-525 mg/m2). After adjusting for chronological age, we did not find differences in rates of dose reductions, therapy completion, treatment delays, or hospitalization rates between patients with an AI-ECG age higher or lower than the chronological age.

24 patients (60%) had ECGs within 1 year of treatment completion, allowing for calculation of ECG aging rates. We found a trend of increased cumulative anthracycline dose in patients with accelerated ECG aging (p=0.06) and a tendency towards higher rates of dose reductions in patients with high ECG aging (OR 5.13, p=0.32). There was no difference in overall survival or regression-free survival based on AI-ECG aging categories.

The AI-ECG algorithm for low EF had a sensitivity of 100% and specificity of 95% to identify patients with an EF of less than 50% prior to treatment. Nine patients had ECGs and echocardiograms both prior to and after treatment. There were only two cases where the EF decreased more than 10% after treatment, and these were the only two cases where there was a significant increase in the low EF prediction (>0.15).

Conclusion: Discrepancies between AI-ECG and chronological age do not predict tolerance to anthracycline based chemotherapy, but AI-ECG performed well to identify patients with a subnormal EF prior to treatment. Interestingly, we also found that patients who experienced significant decreases in EF after treatment had a corresponding increase in low EF AI-ECG probability output. These findings suggest that AI-ECG could be used to stratify patients for cardiotoxicity risk before and monitoring for EF decline after treatment with anthracycline-based chemotherapy.