🗣 AI Discharge Summary Simplification Simulator

The Complex Discharge Summary

Hospital notes are written for clinicians, not patients.

• 11–15: Avg. discharge FK grade (college reading level)
• 7–8: Avg. US adult reading level (grade equivalent)
• ~40%: Patients misunderstanding instructions (after discharge)
• high: 30-day readmission link (to low comprehension)

Why discharge notes are so dense

Clinicians write for other clinicians, not patients.

Jargon hides in plain sight

Terms like "afebrile" feel routine to staff, foreign to patients.

The cost of confusion

Misunderstood instructions drive missed doses and readmissions.

The AI Simplification Model

A language model rewrites sentences while preserving medical meaning.

• LLM: Model type (fine-tuned for clinical text)
• sentence: Rewrite unit (preserves clinical intent)
• 0–100%: Simplification strength (user-tunable)
• >95%: Meaning preservation target (semantic fidelity)

Sentence-level rewriting

The model shortens sentences and swaps register, not just words.

Strength as a dial

Higher strength pushes further toward everyday phrasing.

Guardrails on meaning

Clinical facts must survive the rewrite unchanged.

Jargon Substitution

Medical terms map one-to-one onto everyday equivalents.

• 24: Jargon terms tracked (per sample note)
• 1,000+: Substitution dictionary size (clinical-to-plain pairs)
• ~8%: Ambiguous term rate (need context-aware choice)
• 2.1: Avg. syllables saved/term (plain vs. clinical word)

One term, one meaning

"Myocardial infarction" becomes "heart attack" everywhere.

Context matters

Some terms need surrounding words to pick the right swap.

Fewer syllables, same fact

Shorter plain words cut both grade level and reading time.

Readability Scoring

The Flesch-Kincaid grade level formula rescores the simplified text.

• 2: FK formula inputs (words/sentence, syllables/word)
• ~4–7: Grade drop per pass (grade levels typical)
• live: Rescoring frequency (after every edit)
• 5th–8th: Target band (recommended for patients)

The Flesch-Kincaid formula

0.39×(words/sentence) + 11.8×(syllables/word) − 15.59.

Two levers, one score

Shorter sentences and simpler words both cut the grade level.

Scoring against a target

The gauge tracks distance from the chosen reading grade.

The Patient-Ready Summary

The final text lands at the reader's chosen grade level.

• 6–8: Final FK grade (typical) (matches target slider)
• 0: Jargon terms remaining (fully substituted)
• +30%: Comprehension gain (in patient studies)
• preserved: Clinical fidelity (facts unchanged)

A readable handoff

Patients leave with instructions they can actually follow.

Same facts, plainer words

Doses, dates, and diagnoses carry over exactly.

A tunable target

Care teams can raise or lower the target grade per patient.