Retrieval worsens unsupported ECG criteria in AI reports despite verification
RAG-TRACE: Tracing Measurement Evidence and Diagnostic Completeness in AI-Generated ECG Reports
Computation and Language
Summary
AI tools that generate reports by looking up diagnostic criteria sometimes include information not actually supported by the patient's heart test measurements. The authors show that nearly half of the retrieved criteria in these AI-generated electrocardiogram (ECG) reports don’t match the patient’s actual data. They find that checking measurements independently can catch many unsupported claims, but the AI still tends to add unreliable details if retrieval is unrestricted. This highlights a key challenge in making AI-assisted medical reports truly accurate and trustworthy.
What this means in practice
- •For clinical informatics teams: Improve AI-generated ECG reports by integrating measurement-based verification to reduce unsupported diagnostic claims.
- •For medical ai developers: Design safer retrieval-augmented models by limiting criteria to those with numeric verification to prevent misleading medical reports.
Authors
Elyas Irankhah, Shahed Salehzehi, Parisa Hosseini
Abstract
Purpose: Retrieval-augmented generation (RAG) can introduce electrocardiogram (ECG) criteria that lack patient-specific measurement support. This study evaluates the reliability of retrieved diagnostic criteria in AI-generated ECG reports. Methods: We analyzed 1,083 human-validated PTB-XL patient records using a library of 47 criteria cards, including 33 without numeric verification rules. Language models generated structured reports with and without retrieval. Criterion applicability was evaluated using Marquette 12SL measurements, with independent assessment using Uni-G and comparison against expert-validated diagnostic labels. Results: Keyword retrieval introduced untestable or measurement-disconfirmed criteria in 46.8% of GPT-4.1-mini reports. A deterministic verification rule withdrew 647 criterion applications, including 639 without numeric checks, while preserving 268 of 269 applications independently supported by Uni-G. Restricting retrieval to 14 testable criteria reduced the initial unsupported-criterion rate to 13.8%. Under matched instructions and an identical output schema, retrieval increased unsupported-criterion reports from 17.4% to 31.1%, while hypertrophy recall remained unchanged at 62.6%. Conclusion: These findings identify a fundamental limitation of retrieval-based clinical report generation: retrieved criteria do not necessarily provide valid patient-specific evidence. Measurement-based verification can reduce unsupported criterion applications, but diagnostic validity requires independent assessment.