LLMs are helping patients actually understand discharge summaries
Too many patients leave the hospital confused about what happened, what to do next, and which meds to take. Early pilots with large language models (LLMs) are changing that by turning dense summaries into clear, actionable instructions. The result: better comprehension, fewer unanswered questions, and less back-and-forth after discharge.
Why discharge summaries fall short
- High reading level and heavy jargon.
- Buried action items-patients can't see what matters first.
- Limited time for teach-back, especially at discharge.
- Language barriers and health literacy gaps.
What LLMs do well (when used correctly)
- Rewrite in plain language while preserving clinical accuracy.
- Reformat into simple sections: what happened, meds, red flags, next steps, contacts.
- Highlight immediate actions up front.
- Translate to the patient's preferred language and keep tone supportive.
- Add short definitions for unavoidable medical terms.
A practical workflow you can deploy in two weeks
- Define the target reading level (6th-8th grade is a solid default) and standard sections for your AVS.
- Pick a secure model with a BAA or run it in your protected environment.
- Feed the clinical summary and structured data (meds, follow-ups) to the model; generate a patient-facing version.
- Require clinician or nurse review before release. Edit in-line; approve with a single click.
- Push to the portal and print at discharge. Use teach-back for high-risk cases.
Safety and compliance guardrails
- PHI: Keep data inside approved systems only. See HIPAA guidance from HHS here.
- Human-in-the-loop: No autonomous releases for clinical content-final review is mandatory.
- Medication integrity: Lock dose/route/frequency to EHR source of truth; flag any inconsistencies.
- Proven prompts: Use templates that forbid inventing or guessing. If a field is missing, the model should say "Not provided."
- Audit trail: Store the model version, prompt, inputs, outputs, and approver ID.
Prompt template you can adapt
Rewrite the discharge summary below in plain language for the patient and caregiver. Target 6th-8th grade. Keep medical details accurate and define terms briefly in parentheses. Use these sections in this order:
- What happened (1-3 sentences)
- Your medications (name, plain instructions, reason; list changes clearly)
- Warning signs and what to do (urgent vs. non-urgent)
- Follow-up appointments and labs (dates, location, prep)
- Care team contacts (phone, hours)
- FAQs (answers to likely patient questions)
Do not add facts not present. If something is missing, write "Not provided." Use the patient's name where appropriate and "you" voice. Keep it to one page.
What to measure
- Comprehension: Teach-back success rate; short post-discharge quiz scores.
- Readability: Flesch-Kincaid or similar grade level for the final patient version.
- Utilization: Percentage of discharges using the LLM-assisted summary.
- Safety: Discrepancy rate on meds and follow-up details (pre- vs. post-review).
- Outcomes: 7-30 day readmissions, post-discharge call volume, and MyChart messages related to instructions.
- Efficiency: Time to produce patient-facing instructions; editing time per summary.
Implementation options
- EHR add-on: Use native tools or partner apps that plug into your record system.
- Vendor workflow: A secure API that ingests discharge notes and returns a reviewed patient version.
- Internal deployment: Host a model behind your firewall and expose a simple review UI.
Clinical tips that make a difference
- Lead with actions: "Today," "This week," "By your appointment."
- Convert ranges to patient-friendly language: "If your temperature is 101°F or higher, call us."
- Explain medication changes plainly: "We stopped lisinopril because it lowered your blood pressure too much."
- Use consistent icons or bullets for warning signs and phone numbers on printouts.
- Provide translations for top languages in your community and include interpreter line info.
Equity and language access
LLMs make it easier to offer clear summaries across multiple languages. Still, use professional translation review for high-stakes content and check cultural clarity, not just literal accuracy. Track outcomes by language to ensure benefits reach everyone.
Policy and training
Codify the review step, approved prompts, and disallowed content. Train nurses, case managers, and residents on quick edits and consistent wording. For teams building internal skills with LLMs, see practical programs here: AI courses by job.
Helpful reference
For proven plain-language practices, the AHRQ Health Literacy toolkit is a solid starting point: AHRQ toolkit.
Bottom line
LLM-assisted discharge summaries can make instructions clear, specific, and easier to follow. With a secure setup and mandatory review, you can improve patient comprehension without adding workload-and likely reduce downstream calls and errors.
Your membership also unlocks:
