Artificial intelligence and machine learning are currently having the biggest impact on imaging-related areas of cardiology, but a trio of new studies highlight AI’s potential to support significantly more cardiology treatment and therapy decisions.
Optimizing HF Treatments – University of Michigan researchers developed an algorithm that analyzes clinical data to identify HFrEF patients who would benefit from guideline-directed medical therapy (GDMT) optimization and then recommends treatment changes.
- When applied to data from the GUIDE-IT and HF-ACTION trials, the algorithm commonly recommended medication changes (34.9% to 68.1% of patients depending on drug) or dosage increases (48.8% of visits w/ ACR & ARBs; 39.4% of visits w/ beta-blockers).
- The algorithm might have been right, as patients who the AI assigned better baseline GDMT scores had much lower risks of cardiovascular death or HF hospitalization (hazard ratio: 0.41) and all-cause death and hospitalization (hazard ratio: 0.61).
Predicting Post-TAVR PPM – Mayo Clinic researchers used machine learning to predict which patients would require a permanent pacemaker (PPM) after TAVR procedure more accurately than the current standard PPM prediction model.
- Using data from 964 patients without prior PPM who underwent TAVR, the machine learning model identified 167 clinical variables to predict each patient’s PPM risk.
- The model predicted which patients would require PPM at 30 days and one year far more accurately than the standard PPM risk score model (AUROCs: 0.66 vs. 0.55 & 0.72 vs. 0.54; both (P < 0.001).
- Certain variables had greater association with PPMs, including brachiocephalic artery to aortic valve annulus distance to height ratio (the biggest predictor), pre-existing conduction abnormalities, trans-femoral access, and self-expanding valves.
High-Benefit BP Control – Hypertension treatment guidelines focus on high-risk patients, but targeting “high-benefit” patients for intensive BP therapy could have a much bigger population health impact.
- UCLA researchers used machine learning to analyze data from two BP reduction trials (n=10, 672) to identify patients who would experience the greatest benefits from intensive treatments and assess outcomes.
- They found that providing intensive treatments to “high-benefit” patients prevented one cardiovascular event per 11 patients treated over a three-year follow-up period.
- That’s more than a five-times better ratio compared to exclusively targeting high-risk patients for intensive BP treatments (~62 per prevented event).
The Takeaway
This batch of studies serves as yet another reminder that AI and high-powered computing could drive a new level of personalization and precision in cardiology care, well beyond the initial diagnostic steps where we’re currently seeing most cardiology AI activities.
