“Discharging the (Old) Discharge Summary: How AI is Rewriting the Last Chapter of the Hospital Stay”
Hospital discharge summaries have always been the awkward epilogue of the patient’s inpatient journey—something between a clinical record, a legal form, and a lengthy regurgitation of disconnected items from within the chart. Traditionally, they’ve been authored by overworked residents or APP’s, by hospitalists and attending physicians in a hurry to get back to their other work, or by whomever lost the rock-paper-scissors battle at the nurses’ station. Way too frequently they are “authored” by someone who may have never seen the patient or who first encounters their chart and completed their discharge. They are often written in a tone best described as bureaucratic Hemingway: short on adjectives, long on omissions.
Everyone who tries to read a typical discharge summary cringes at its length, the confounding and excruciating copy and paste, the seemingly endless and repetitive information regurgitation without any explanation of why it is being included, and the inability to find the key clinical items anywhere in these multipage documents.
Unfortunately, discharge summaries have primarily been reviewed as a perfunctory administrative hurdle—an artifact created more for billing and compliance than for the actual continuity of care. If you are a primary care physician trying to make sense of one, you often needed a magnifying glass, lots of (unavailable) time, a vivid imagination, and perhaps a séance to summon the spirit of the hospitalization for clarification.
Now riding into this maelstrom is what looks to be our new knight in shining armor, bringing the promise of a solution to most everything that has been wrong with the summaries. Enter Epic’s Artificial Intelligence—medical documentation’s overly polite, endlessly patient, and remarkably confident new co-worker.
The Problem with the Old Way
The traditional discharge summary tends to suffer from three endemic flaws:
- Incoherence Through Copy-Paste:
The modern electronic health record (EHR) blessed us with the “copy forward” function—meaning that the patient who had a cough in the ER may still have “chief complaint: cough” in their discharge summary three weeks later, even after having a knee replacement. - Narrative Anemia:
The human author, aware that the patient is technically leaving the hospital whether or not this summary gets read, may write something like:
“Patient admitted with chest pain. MI ruled out. Treated for pneumonia. Discharged home.”
This conveys roughly the same narrative depth as a caveman describing the history of the Roman Empire.
- Amnesia for who is the real Audience:
The intended audience for a discharge summary is supposed to be the outpatient clinician. But clinicians typically feel the really important work – taking care of the patient – has been completed successfully when the patient is ready to leave. Now they are ready to move on to care for everyone else, so the discharge summary is treated like a last bureaucratic item that needs to be whipped off before getting back to real doctoring work. The perspective of the next treating physician is rarely considered in the rush to create these summaries and check that 1 last box. It is rare to see the discharge summary author create a summary as if they are writing a story for the outpatient physician to read. But is not that actually what we are supposed to be doing?
What AI Brings to the Table
Epic Artificial Intelligence now is completely changing the rules of the game. AI is clear; AI is fast; AI is consistent; AI is thorough; AI does not miss the little things.
An AI-generated discharge summary can integrate structured data, lab results, imaging reports, and narrative notes into a coherent story arc. Think of it as transforming the random pages of a medical Choose-Your-Own-Adventure novel into something closer to a New Yorker short story (minus, ideally, the fiction).
Here’s what AI can do better:
- Synthesize Multiple notes and commentary into One
Rather than presenting comments from the attendings, the consultants, the APP’s, the dietitian, the wound care nurse, the discharge planner, and others as disconjugate pieces being linked together artificially, AI can harmonize them into a single narrative. - Contextualize the Hospital Course
AI can not only state that a patient was “treated for pneumonia,” but can also outline the rationale for therapy, relevant diagnostic challenges, and follow-up needs—without devolving into a pathophysiology lecture. - Error Checking and Completeness
The machine doesn’t get called away to answer a page mid-sentence. It can scan for missing discharge medications, unresolved abnormal labs, or contradictory statements before finalizing the note. Just as importantly, late changes in the hospital course from vital signs to lab results that are easily missed never escape the AI “eye”
Potential Pitfalls
Of course, no love letter to AI in medicine is complete without a list of caveats.
- Hallucination Hazard
AI occasionally produces “creative” results—attributing tests that weren’t performed or results that don’t exist. It’s the clinical equivalent of a student padding an essay with confident but fabricated citations. - Loss of Clinician Voice
1 potential criticism is that a discharge summary is not just a recitation of facts; it is supposed to encodes a clinician’s gestalt—their professional judgment, sense of urgency, and priorities for follow-up. However with augmented authorship, it is easy for the clinician to add in any qualitative descriptors that AI may miss. - Over-Dependence
A generation of clinicians might grow less skilled at writing their own summaries, the same way GPS has made us collectively worse at reading maps. But is this really bad? Look carefully at our current default standard before deciding.
The Future: Human-AI Co-Authorship
The ideal role for AI in discharge summaries is not full automation but augmented authorship. Think of it as a collaboration: the clinician as the director, the AI as the screenwriter, and the EHR as the slightly neurotic production assistant who keeps reminding you about billing codes.
In this model, the discharge summary becomes less of a bureaucratic hurdle and more of a genuinely useful document—one that the outpatient provider reads, the patient understands, and the insurance company grudgingly respects.
A Closing Note (Pun Intended)
If the discharge summary is the last chapter of a hospital stay, then AI is our new ghostwriter—polite, efficient, and suspiciously uncomplaining. It can turn fragmented clinical notes into something approaching literature (or at least coherent prose), and it can help ensure that what leaves the hospital with the patient is not just a bag of medications and a follow-up appointment, but also a clear record of what happened and why.
And perhaps the greatest irony?
The patient may not remember much about their hospital stay, but thanks to AI, their discharge summary might finally be something worth reading.
