Language Model Assistance for Hospitalized Patients

Age: 18+

Sex: Any

Trial Phase: Academic

Sponsor: Mayo Clinic

Disqualifiers: Minors, Non-CURE services

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: Unknown

1015 Participants Needed

Trial Summary

What is the purpose of this trial?

This pilot study aims to assess the feasibility of carrying out a full-scale pragmatic, cluster-randomized controlled trial which will investigate whether discharge summary writing assisted by a large language model (LLM), called CURE (Checker for Unvalidated Response Errors), improves care delivery without adversely impacting patient outcomes.

Will I have to stop taking my current medications?

The trial information does not specify whether you need to stop taking your current medications.

What data supports the effectiveness of the treatment CURE for hospitalized patients?

The research shows that language models like ChatGPT can assist in medical decision-making, as seen in a study where ChatGPT's recommendations for breast cancer treatment were similar to those of a tumor board in 70% of cases. This suggests that language models can be effective tools in supporting treatment decisions, potentially benefiting hospitalized patients. ^{1 2 3 4 5}

How does the treatment CURE differ from other treatments for hospitalized patients?

CURE is unique because it involves using large language models (LLMs) like ChatGPT to assist in medical documentation and patient communication, potentially improving health literacy and treatment adherence. Unlike traditional treatments, it focuses on enhancing clinical workflow and patient interaction through AI technology. ^{1 6 7 8 9}

Research Team

Xiaoxi Yao

Principal Investigator

Mayo Clinic

Eligibility Criteria

This trial is for hospitalized patients with aphasia, a condition that affects communication. The study will explore if using a large language model to assist in writing discharge summaries can improve care without negatively affecting patient outcomes.

Inclusion Criteria

Adult patients admitted to one of three participating cardiology services at Mayo Clinic in Rochester, MN

Clinicians who provide care to randomized patients included in this pilot

Exclusion Criteria

Patients admitted to a hospital service where CURE is not implemented

Clinicians who do not provide care to randomized patients included in this pilot

I am under 18 years old.

Timeline

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

Intervention

Clinicians use the CURE model to assist with discharge summary writing

3 months

Control

Clinicians continue with standard practice for discharge summary writing

3 months

Follow-up

Participants are monitored for safety and effectiveness after intervention

4 weeks

Treatment Details

Interventions

• CURE

Trial Overview The intervention being tested is CURE, a large language model designed to help write discharge summaries for patients. The trial will determine its effectiveness and safety in clinical settings through a pragmatic, cluster-randomized controlled design.

Participant Groups

Group I: Intervention Experimental Treatment 1 Intervention

Clinicians who care for patients randomized to intervention will have access to CURE to assist with discharge summary writing.

Group II: Control Active Control 1 Intervention

Clinicians who care for patients randomized to control will continue with standard practice for discharge summary writing.

Who Is Running the Clinical Trial?

Mayo Clinic

Lead Sponsor

Trials

3,427

Recruited

3,221,000+

References

1. United States pubmed.ncbi.nlm.nih.gov

Large language model (ChatGPT) as a support tool for breast tumor board. [2023]

2. United States pubmed.ncbi.nlm.nih.gov

Harnessing language models for streamlined postcolonoscopy patient management: a novel approach. [2023]

3. Netherlands pubmed.ncbi.nlm.nih.gov

Predicting health-related quality of life change using natural language processing in thyroid cancer. [2023]

4. Germany pubmed.ncbi.nlm.nih.gov

Extracting findings from narrative reports: software transferability and sources of physician disagreement. [2017]

5. United States pubmed.ncbi.nlm.nih.gov

Automating a severity score guideline for community-acquired pneumonia employing medical language processing of discharge summaries. [2018]

6. Germany pubmed.ncbi.nlm.nih.gov

[Clinical application of large language models : Does ChatGPT replace medical report formulation? An experience report]. [2023]

7. United States pubmed.ncbi.nlm.nih.gov

Why Patient Portal Messages Indicate Risk of Readmission for Patients with Ischemic Heart Disease. [2020]

8. Korea (South) pubmed.ncbi.nlm.nih.gov

Developing prompts from large language model for extracting clinical information from pathology and ultrasound reports in breast cancer. [2023]

9. United States pubmed.ncbi.nlm.nih.gov

Evaluating the Performance of Different Large Language Models on Health Consultation and Patient Education in Urolithiasis. [2023]