Presentation (CTI)

OBJECTIVES: Evidence on the clinical and economic value of artificial intelligence (AI) for disease detection remains limited. We present methodology for evaluating AI using real-world data and health economic modeling, applied to an imaging algorithm detecting incidental heart failure with preserved ejection fraction (HFpEF).
METHODS: The EchoGo Heart Failure (EchoGo) algorithm (Ultromics Ltd, UK) was applied retrospectively to transthoracic echocardiograms (TTEs) from patients in Dandelion Health (longitudinal, multimodal clinical data platform) without prior heart failure. Patients clinically diagnosed with HFpEF within 90 days of the index TTE were classified as true-positive (“Early Diagnosis”) (n=1,152). False-negative (“Delayed Diagnosis”) patients were flagged by EchoGo but not clinically diagnosed within 90 days (n=515, weighted n=1,113). After inverse probability treatment weighting, diagnosis, treatment, healthcare utilization, and cost patterns were compared between the two groups. A hybrid decision-tree Markov model with five states (undiagnosed, diagnosed, major adverse cardiac event hospitalization tunnel state, post-hospitalization tunnel state, and death) projected five-year outcomes from the payer perspective, comparing EchoGo versus Standard of Care (SOC). Model inputs incorporated sensitivity to represent “Early Diagnosis” and “Delayed Diagnosis” pathways.
RESULTS: Sensitivity of EchoGo was high (84.4%) versus literature-reported SOC (34%). Retrospectively, EchoGo detected HFpEF 263 days before clinical recognition in “Delayed Diagnosis” patients, who experienced higher acute care utilization and mortality than “Early Diagnosis” patients. EchoGo dominated SOC in the Markov model (NMB: $9,485); probabilistic analysis showed a 66.7% likelihood of cost-effectiveness at $150,000/QALY. One-way sensitivity analysis identified the likelihood of diagnosis at acute decompensation and hospitalization costs as primary drivers. Per 1,000 patients over five years, EchoGo yielded 51 QALYs gained, 42 fewer hospitalizations, 50 fewer readmissions, and 56 fewer emergency visits.
CONCLUSIONS: By diagnosing heart failure earlier, EchoGo proved cost-effective by improving patient trajectories and avoiding high-intensity care. Evaluating AI and identifying cost-effective deployment opportunities will be crucial for future clinical use.