Extract data from discharge summaries
A discharge summary is the record that hands a patient from the hospital to whatever comes next, and it is dense with the parts of an admission a downstream clinician has to act on quickly. A care-coordination nurse, a post-acute facility, and a health-information-exchange team read it for the same handoff: why the patient was admitted, what was done, what changed in the medication list, and what has to happen after discharge. Modeled in HL7 FHIR as a clinical document composition, a discharge summary pulls together the admission and discharge dates, the length of stay, the principal and secondary diagnoses coded in ICD-10, the procedures performed with their CPT codes, the medications the patient leaves on, the discharge instructions, and the follow-up appointments the receiving provider must honor. A patient admitted 2026-04-14 and discharged 2026-04-19 after a five-day stay leaves with a summary that a skilled-nursing intake or a primary-care follow-up depends on. Coded lists, repeating tables, and long narrative interleave through the whole document. Its diagnoses are a coded list where each entry pairs an ICD-10 code with a type of principal, secondary, or comorbidity, the discharge medications are a reconciliation table of drug, dose, frequency, and route where an added or stopped drug is the point, and the clinical course and discharge instructions are narrative that still has to be captured verbatim. Follow-up appointments pair a specialty with a recommended timeframe, a pulmonology visit by 2026-05-03 at an urgency of soon, and a missed one is a readmission risk. Vital signs and discharge labs carry values with units and reference ranges that a receiving system reads against the same standards a lab report uses. Talonic reads the discharge summary and returns the patient and encounter identifiers, the diagnosis and procedure lists, the discharge medication reconciliation, the instructions, and the follow-up appointments as structured rows, with narrative captured as text and handwriting scored for confidence. A summary for a five-day admission under MRN 4471982, finalized 2026-04-19, loads into a care-coordination system with its diagnosis codes and medication changes intact, whether it arrived as a FHIR-modeled export or a PDF scanned into the EMR. Nothing is interpreted along the way: the fields are structured as written, with no clinical judgment applied.
What gets extracted from discharge summaries
How extraction works for discharge summaries
Discharge summaries reach a records or care-coordination team as EHR exports modeled on HL7 FHIR, scanned dictated summaries, and faxed transfer packets, so no single template holds across hospitals or vendors such as Epic Systems, Oracle Health, and MEDITECH. Talonic classifies the summary and maps it to the clinical-document schema in the Field Registry, which separates the patient and encounter identifiers from the repeating diagnosis, procedure, medication, vital-sign, lab, and follow-up tables. Diagnoses return as rows pairing an ICD-10 code with a type of principal, secondary, or comorbidity, procedures as rows with a CPT or ICD code and a date, and the discharge medications as a reconciliation table of drug, dose, frequency, route, and refills. Lab results keep their value, unit, and reference range, and follow-up appointments pair a specialty with a recommended timeframe and an urgency. Narrative sections such as the clinical course and discharge instructions are captured verbatim, and handwritten or low-confidence fields are flagged rather than entered silently. Each value carries a confidence score and a pixel-region pointer under DIN SPEC 91491, so a records team can verify a diagnosis code or a discharge medication against the source summary. Clinical interpretation is never added; the fields return exactly as written.
Sample extraction
A hospital discharge summary for a five-day medical admission
{
"document_number": "DS-2026-558102",
"document_date": "2026-04-19",
"patient.name": "Helen Park",
"patient_identifier": "4471982",
"birth_date": "1968-02-14",
"gender": "female",
"admission_date": "2026-04-14",
"discharge_date": "2026-04-19",
"length_of_stay": 5,
"primary_diagnosis": "J18.9 Community-acquired pneumonia",
"managing_organization": "Riverside Regional Hospital",
"performer": "Dr. Alan Reyes, Internal Medicine",
"diagnoses": [
{
"diagnosis_code": "J18.9",
"diagnosis_description": "Community-acquired pneumonia",
"diagnosis_type": "primary",
"onset_date": "2026-04-14"
},
{
"diagnosis_code": "E11.9",
"diagnosis_description": "Type 2 diabetes mellitus",
"diagnosis_type": "comorbidity",
"onset_date": "2018-07-01"
}
],
"procedures": [
{
"procedure_code": "71046",
"procedure_description": "Chest radiograph, 2 views",
"procedure_date": "2026-04-14",
"performer": "Radiology"
}
],
"medications": [
{
"medication_name": "Amoxicillin-clavulanate",
"rx_code": "562251",
"dose": "875 mg",
"frequency": "twice daily",
"route": "oral",
"status": "active",
"start_date": "2026-04-19",
"refills": 0
},
{
"medication_name": "Metformin",
"dose": "500 mg",
"frequency": "twice daily",
"route": "oral",
"status": "active",
"start_date": "2026-04-19",
"refills": 3
}
],
"follow_up_appointments": [
{
"specialist_type": "Pulmonology",
"recommended_date": "2026-05-03",
"urgency": "soon",
"notes": "Reassess resolution of pneumonia"
}
],
"discharge_instructions": "Complete the full antibiotic course. Return to the emergency department if fever recurs or breathing worsens."
}Frequently asked
Does it capture the diagnosis and procedure lists with codes?
Yes. Diagnoses return as rows pairing an ICD-10 code with a type of principal, secondary, or comorbidity, and procedures as rows with their CPT or ICD code and date, so a receiving system loads a coded problem list rather than a paragraph.
How is the discharge medication reconciliation handled?
The discharge medications return as a table of drug, dose, frequency, route, and refills, so a care team sees which drugs were added, continued, or stopped, which is the part of a summary a medication reconciliation depends on.
How is protected health information handled?
Talonic handles protected health information (PHI) as an extraction step and nothing more. Every field is pulled from the source summary and returned with a confidence score and a pointer to the region it came from, so a coordination team confirms each value against the page before it enters the record, and no clinical reading is added on top.
Does it interpret the clinical course?
No. The clinical course and discharge instructions are captured verbatim as narrative. Talonic structures what the summary states and offers no clinical judgment about it.
Ready to extract from your own discharge summaries?
Author note
Reviewed by Talonic engineering · last reviewed 2026-07-07