Where Review Work Hides
The UCI analysis shows why a single average editing rate cannot describe the burden created by ambient AI scribes. The subject is not merely how much text clinicians changed, but how the note’s section and the clinician’s own editing style shaped the work after an Abridge draft appeared. Understanding those distributions helps a reader judge where section-aware design and evaluation could matter, without treating one health system’s pattern as a universal rate.
A clinical note is not one uniform object. Its sections perform different communicative jobs, and those jobs determine what a clinician must preserve, remove, or reframe. The History of Present Illness may need a concise account of the patient’s experience and its timeline. The Assessment and Plan must connect evidence to clinical reasoning and action. Results and Physical Exam often follow more constrained conventions. If all four are averaged together, the sections that carry the most reasoning work can disappear inside a convenient whole-note number.
HPI and A&P ask for different kinds of editing
The paired-note study analyzed four sections: History of Present Illness, or HPI; Assessment and Plan, or A&P; Physical Exam; and Results. HPI was the largest category, with 34,569 sections, or 48.6% of the corpus. A&P accounted for 24,310 sections, or 34.2%. Results contributed 6,832 sections, or 9.6%, and Physical Exam 5,462, or 7.7%.
The volume of sections is not the same as the volume of transformation work. A&P accounted for 4,491 confirmed consumer-to-clinical transformation events, or 59.3% of all such events. HPI accounted for 2,981, or 39.3%. Physical Exam contributed 59 events and Results 45, less than 2% combined. A&P therefore carried a disproportionate share of the confirmed lexical professionalization relative to its share of the note-section corpus.
The difference makes sense once the sections’ jobs are separated. HPI often begins with the patient’s own phrasing, details about when a symptom appeared, and a sequence that may have emerged through conversation rather than a prearranged outline. The editing task is often to compress that account without stripping away the information that explains the clinical situation. A&P starts from a different demand. It must express what the clinician thinks is happening and what should happen next. Recasting conversational material into that structure can require more than changing vocabulary; it can require deciding which problem is primary, how evidence supports the assessment, and which action belongs in the plan.
The study’s per-note measures reflect this difference. HPI showed the highest average consumer-term deletion intensity, at 18.75 deletions per note. A&P showed 13.24 consumer-term deletions per note but the strongest clinical-term addition intensity, at 9.21 added clinical terms per note. The pattern is consistent with HPI being a zone of narrative compression and A&P being a zone of professional reframing. It does not mean every HPI or A&P note required that amount of work, or that the deletion and addition measures represent minutes of editing.
Results and Physical Exam showed lower absolute editing intensity: 5.26 consumer-term deletions per Results note and 3.43 per Physical Exam note. Their reported proportional deletion figures were higher than those for HPI and A&P, but a proportion and an absolute count answer different questions. A small, structured section can have a high share of terms removed while still requiring fewer edits per note than a long narrative section. The source’s table therefore should not be turned into a claim that Results or Physical Exam create the greatest total review burden. Absolute volume, section size, and the kind of judgment involved all matter.