From 500 pages to one: clinical summarization that clinicians actually trust

A new patient arrives with six years of records: discharge summaries, specialist notes, lab panels and imaging reports — sometimes 500 pages. No clinician can read all of it before a 20-minute visit. AI summarization promises to help, but only if clinicians can trust what it says.

Why most summaries fail

Generic summaries tend to be either too long to be useful or too short to be safe. They blend old and current problems, miss medication changes, and — most importantly — give no way to check where a statement came from. A clinician who can't verify a summary won't rely on it.

What a trustworthy summary looks like

Murphi structures a pre-visit summary around what clinicians actually need: active problems, key events, trends in labs and vitals, and open items like pending referrals or overdue screenings. Every line links back to the source document and page, so a clinician can tap "LDL 188 → 162" and see both lab reports in seconds.

Reviewing summaries safely

Summaries should be read as a map, not a verdict. Good practice is to confirm anything that changes a decision — a medication, an allergy, a diagnosis — against its source. Because sources are linked, that confirmation takes seconds rather than minutes.

Where summarization helps most

The biggest gains come at transitions of care: new patients, referrals, hospital follow-ups and annual reviews. Behavioral health teams use it to review prior treatment history; correctional health teams use it to reconstruct care when records arrive from outside facilities.

Measuring the impact

Track pre-visit prep time, the number of records clinicians open during a visit, and how often important history is discovered late. Teams typically see prep time drop from tens of minutes to a few, with fewer surprises during the visit.

Five hundred pages will keep arriving. The goal is to make sure the one page clinicians read is accurate, complete and easy to verify.