LLM Judges Verify Presence, Not Absence: Omission Blindness in AI Clinical Notes and What Recovers It
2026-08-31 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors studied how well AI systems called judges can spot missing information, called omissions, in clinical notes written by AI scribes. They found that usual methods struggle to reliably find these omissions, often performing no better than guessing. By changing the approach to first list all facts from the conversation and then checking the note against this list, they developed two better detection methods. One method pinpoints missing facts with few false alarms but is costlier, while the other finds more omissions with slightly more errors but is cheaper and faster. Their work includes new tools and data to improve omission detection in clinical AI notes.
Ambient AI scribesclinical notesomission errorslarge language model (LLM) judgesfact checkingprompt optimizationfalse alarmsbenchmark datasetnatural language processingclinical documentation
Authors
Sebastian Fox, Luke Markham, Ryan Lail, Michael Karotsieris
Abstract
Ambient AI scribes draft clinical notes, and published audits find their dominant error is omission: information the encounter established that the note fails to record. The standard check is an LLM judge: a second model reads the note against the transcript and flags problems. We ask whether judges detect omissions. Public corpora cannot supply the answer key: their clinician reference notes and transcripts are materially discrepant. Our benchmark has 500 single-error note pairs from audited fact sheets, 298 with a named fact certainly absent and 202 added-or-altered controls. Across eight judge designs, paired discrimination (the flawed note below its clean twin, 0.5 a coin flip) reads 0.79-0.94 on added or altered content and 0.50-0.63 on omissions. On single notes, no design flags omissions reliably more often than perfect notes. Wording changes, voting and GEPA prompt optimisation move the operating point without creating usable detection. Restructuring the task recovers it: list the facts the transcript establishes, then check the note for each. Two methods reach it independently and trade off: a per-fact pipeline, and a GEPA-evolved prompt doing the same in one call. The pipeline's flags name the missing fact and its severity at 2.7% false alarms. The single call detects more (36.9% against 24.6%, p=0.002) at 6.2% false alarms and a tenth of the cost per note. A physician author validated 70 items and, where the two routes disagree, sided with the pipeline on 10 of 10 (p=0.002). A second clinician, not an author, graded the severity rubric blind and agrees to within a grade. On real vendor notes from a companion census no benchmark threshold transfers, but the re-calibrated single call detects more than the best of the eight at half its false-alarm rate. Omissions whose fact is restated elsewhere defeat both routes. We release the benchmark, prompts and judgements.