Using Generative Artificial Intelligence (AI) to Create Patient-Friendly Discharge Summaries

Using Generative Artificial Intelligence to Transform Discharge Summaries Into Patient-Friendly Language and Format

This will be an open-label, parallel-group, randomized trial. Patients will be randomized to review the patient-friendly discharge instructions at the time of discharge (intervention group) vs the standard of care. The intervention differs from the standard of care in that patients will be given additional medical documentation in the intervention group.

Study Overview

• Status: Completed
• Conditions: Hospital Discharge Summary Understanding
• Intervention / Treatment: Behavioral: AI-Generated Discharge Instructions; Behavioral: Standard of Care Discharge Instructions

• Study Type: Interventional
• Enrollment (Actual): 61
• Phase: Not Applicable

Contacts and Locations

Study Locations

• United States
  • New York
    • New York, New York, United States, 10016
      • NYU Langone Health

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Admitted to Inpatient Medicine unit at NYULH
• Able to read and understand English
• Willing and able to consent to this research

Exclusion Criteria:

• Inpatient Hospice
• Discharged on home hospice

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Treatment
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: Generative AI Discharge Instructions; At the time of discharge, patients will receive patient-friendly discharge instructions generated by GPT-4, an advanced natural language processing model, with the aim of enhancing patient readability and understandability in the discharge information.	Behavioral: AI-Generated Discharge Instructions; Patient-friendly hospital discharge instructions generated by GPT-4, an advanced natural language processing model.
Active Comparator: Standard of Care Discharge Instructions; At the time of discharge, patients will receive standard discharge instructions (i.e., not generated by GPT-4).	Behavioral: Standard of Care Discharge Instructions; Standard hospital discharge instructions.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Number of Summaries with Positive Outcome on Composite Measure of Readability and Understandability	This is a composite outcome measure. Readability of discharge summaries will be measured using the Flesch-Kincaid Grade Level assessment, which measures how challenging a text may be for readers. Scores on the Flesch-Kincaid assessment are rated as follows: 0-3: Basic reading level, (kindergarten or elementary school) 3-6: Basic reading level, (elementary school ) 6-9: Average reading level, (middle school) 9-12: Average reading level, (high school) 11-18: Skilled readers. Understandability will be evaluated by the presence of simplified description of primary diagnosis and hospital intervention. Summaries with both 1) presence of simplified description of primary diagnosis and hospital intervention, and 2) a Flesch-Kincaid grade level greater than or equal to 6 and less than 8, will be considered to have a "positive outcome."	Day 1 (At Discharge)

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Changes made to AI-Generated Content	Changes made to content will be assessed using edit distance.	Day 1 (At Discharge)
AI-Generated Content Perception Survey: Safety	Patients will be given a survey in which they are asked whether they feel safe with the discharge instructions they were provided. The total score is the number of participants who respond "Yes, definitely" or "Yes, somewhat" to the question.	Day 1 (At Discharge)
AI-Generated Content Perception Survey: Understandability	Patients will be given a survey in which they are asked whether the instructions were easy to understand. The total score is the number of participants who respond "Yes, definitely" or "Yes, somewhat" to the question.	Day 1 (At Discharge)
AI-Generated Content Perception Survey: Empathy	Patients will be given a survey in which they are asked whether the instructions were written in an empathetic manner. The total score is the number of participants who respond "Yes, definitely" or "Yes, somewhat" to the question.	Day 1 (At Discharge)

Collaborators and Investigators

• Sponsor: NYU Langone Health
• Investigators: Principal Investigator: Jonah Zaretsky, NYU Langone Health

Study record dates

Study Major Dates

• Study Start (Actual): January 29, 2025
• Primary Completion (Actual): May 22, 2025
• Study Completion (Actual): May 22, 2025

Study Registration Dates

• First Submitted: November 27, 2024
• First Submitted That Met QC Criteria: November 27, 2024
• First Posted (Actual): December 2, 2024

Study Record Updates

• Last Update Posted (Actual): June 6, 2025
• Last Update Submitted That Met QC Criteria: June 5, 2025
• Last Verified: June 1, 2025

More Information

Terms related to this study

• Other Study ID Numbers: 23-01544

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Data collected from usual clinical care; these data cannot be easily deidentified or shared.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No