Using Large Language Models Such As GPT-4 to Assess Guideline Adherence in Patients With Chronic Obstructive Pulmonary Disease

NCT06410547 | Completed

• 1 other identifier: EA2/322/23
• Study Type: interventional
• Target: 78
• Locations: 1 country
• Sites: 1

Brief Summary

According to studies in the US and the Netherlands, 33-40% of patients with chronic conditions receive care that does not follow guideline recommendations. These findings have also been demonstrated in the management of COPD. This leads to under- or over-treatment of patients and, in the case of COPD, to exacerbations and hospitalisations. These exacerbations are a significant clinical problem, affecting patient's lung function, quality of life and mortality. They are also a burden on the healthcare system. Technological advances in artificial intelligence offer the opportunity to address these issues in COPD management. In the past year, there have been remarkable innovations in the field of natural language processing, especially through large language models such as GPT-4 from OpenAI and Bard or Gemini from Google. These models offer an opportunity to improve the implementation of evidence-based care in clinical practice. This study is a prospective, randomised trial that will compare therapy on discharge for patients with COPD. One arm will receive no intervention, while the other arm will receive a treatment recommendation from an LLM. The study will compare the percentage of patients treated according to the guideline.

Conditions

COPD

Outcome Measures

Primary Outcomes (2)

• Adherence to treatment guidelines at the time of hospital discharge

  The primary endpoint will assess whether the treatment at the time of discharge is consistent with the guidelines' recommendations. This is a binary outcome measure of yes or no.

  From date of admission (which is enrollment) to the date of discharge, assessed up to one month

• Percentage of patients treated in concordance with treatment guidelines at the time of hospital discharge

  This primary endpoint will assess the percentage of guideline-concordant treatments in each study arm.

  From date of admission (which is enrollment) to the date of discharge, assessed up to one month

Study Arms (2)

No Intervention

NO INTERVENTION

This arm will be treated as usual. No intervention will be performed.

LLM Assessement

ACTIVE COMPARATOR

During hospital stay and after written informed consent, an LLM is asked to indicate if the treatment the patient receives is guideline-concordant. The information is ascertained by two study physicians (human-in-the-loop) and later provided to the treating physician who can recommend a change in therapy to the patient (outside of the study).

Other: LLM

Interventions

LLM OTHER

A LLM-based comparison between treatment and guideline.

LLM Assessement

Eligibility Criteria

• Age: 18 Years+
• Sex: all
• Healthy Volunteers: No
• Age Groups: Adult (18-64), Older Adult (65+)

You may qualify if:

• Diagnosis of COPD
• Consent
• Discharge after hospitalization

You may not qualify if:

• Lack of Consent

Sponsors & Collaborators

• Charite University, Berlin, Germany: lead

Study Sites (1)

Charité University

Berlin, State of Berlin, 10117, Germany

Location

MeSH Terms

Conditions

Pulmonary Disease, Chronic Obstructive

Condition Hierarchy (Ancestors)

Lung Diseases, Obstructive; Lung Diseases; Respiratory Tract Diseases; Chronic Disease; Disease Attributes; Pathologic Processes; Pathological Conditions, Signs and Symptoms

Study Officials

• Matthias Gröschel, MD PhD

  Charité

  PRINCIPAL INVESTIGATOR

Study Design

• Study Type: interventional
• Phase: not applicable
• Allocation: RANDOMIZED
• Masking: TRIPLE
• Who Masked: PARTICIPANT, CARE PROVIDER, OUTCOMES ASSESSOR
• Purpose: TREATMENT
• Intervention Model: PARALLEL
• Sponsor Type: OTHER
• Responsible Party: PRINCIPAL INVESTIGATOR
• PI Title: Principal Investigator

Study Record Dates

• First Submitted: April 24, 2024
• First Posted: May 13, 2024
• Study Start: May 15, 2024
• Primary Completion: April 1, 2025
• Study Completion: April 1, 2025
• Last Updated: March 20, 2026

Record last verified: 2026-03

Data Sharing

• IPD Sharing: Will share

Locations

DE (1)