Capabilities ofArtificial Intelligence Models in Externation Decision of Patient Who Followed in Intensive Care Unit () (ICU)

The Evaluation of the Effectiveness of General Artificial Intelligence Models in Extubation Decision-Making in the Intensive Care Unit

This clinical study aims to evaluate the effectiveness of General Artificial Intelligence (AI) models, specifically ChatGPT and Gemini, in assisting with the decision-making process for discharging patients from the Intensive Care Unit (ICU) to a general ward or home. The timing of ICU discharge is a critical decision that significantly impacts patient outcomes and the efficient use of ICU resources. This study seeks to determine whether AI models can accurately and efficiently predict the optimal time for patient discharge, supporting clinicians in making informed decisions.

The primary hypothesis is that AI models can improve the accuracy and speed of discharge decisions compared to traditional methods. The study will assess the agreement between the AI model predictions and the decisions made by ICU specialists. Additionally, the study will compare the performance of ChatGPT and Gemini AI models to identify which model offers the most reliable and timely discharge decisions.

By exploring the potential of AI in clinical decision-making, this research could contribute to the development of innovative tools for ICU management, ultimately enhancing patient care and optimizing ICU operations. The findings could lead to the integration of AI models into clinical decision support systems, facilitating more accurate and efficient patient management in the ICU.

Study Overview

• Status: Completed
• Conditions: Artificial Intelegence
• Intervention / Treatment: Other: decision

Detailed Description

This study is designed to assess the effectiveness of General Artificial Intelligence (AI) models, specifically ChatGPT and Gemini, in facilitating discharge decisions from the Intensive Care Unit (ICU). The focus is on evaluating these AI models' performance in predicting the appropriate timing for transitioning patients from the ICU to a general ward or discharge to home. The study will leverage machine learning techniques, including Random Forest and Decision Tree algorithms, to analyze patient data and generate predictions.

Study Design and Methodology:

This prospective study will include all patients admitted to the ICU during this time frame. The study will gather comprehensive clinical and demographic data, including indications for ICU admission, comorbid conditions, abnormal laboratory and imaging findings, physical examination results, vital signs, and daily treatments. The data will be anonymized to protect patient privacy, with only clinical information used for AI model training and evaluation.

The AI models will be trained on historical hospital data, applying machine learning algorithms to predict the need for continued ICU care or the suitability for discharge. These predictions will be compared daily with the clinical decisions made by ICU specialists. The study will utilize various statistical methods to assess the models' accuracy and alignment with clinical decisions, including Pearson Chi-Square tests, Kappa statistics, McNemar tests, and ROC (Receiver Operating Characteristic) analysis.

AI Model Training and Evaluation:

The AI models, ChatGPT and Gemini, will undergo training using anonymized patient data, with a focus on optimizing their predictive accuracy for ICU discharge decisions. The training process will involve analyzing a wide range of clinical variables, including demographic data (age, gender, comorbidities), vital signs, laboratory results, and imaging findings. The models will be evaluated based on their ability to predict ICU discharge needs accurately, with the results validated against actual clinical decisions made by ICU specialists.

Machine learning techniques, such as Random Forest and Decision Tree algorithms, will be employed to develop the predictive models. These techniques are chosen for their robustness in handling complex clinical data and their ability to provide insights into the factors most predictive of ICU discharge readiness.

Statistical Analysis:

The study will apply several statistical methods to evaluate the AI models' performance. Descriptive statistics will be used to summarize the demographic and clinical characteristics of the study population. Pearson Chi-Square tests will assess the association between AI model predictions and actual discharge decisions, while Kappa statistics will measure the agreement between AI predictions and ICU specialist decisions. The McNemar test will be used to evaluate changes in predictions over time, and ROC analysis will be conducted to assess the overall performance of the AI models, with a focus on sensitivity and specificity.

Expected Outcomes and Significance:

This study aims to determine whether AI models can enhance the accuracy and efficiency of ICU discharge decisions. The findings could have significant implications for clinical practice, potentially leading to the integration of AI-driven decision support systems in ICU management. By improving the timing and accuracy of discharge decisions, AI models could help optimize ICU resource utilization, reduce patient length of stay, and improve overall patient outcomes.

The study will also explore the comparative performance of ChatGPT and Gemini, providing insights into which AI model is better suited for integration into clinical workflows. The results could pave the way for the development of more advanced AI-driven tools tailored to the specific needs of ICU settings, contributing to the ongoing evolution of healthcare through innovative technology.

• Study Type: Observational
• Enrollment (Actual): 398

Contacts and Locations

Study Locations

• Turkey
  • Küçükçekmece
    • Istanbul, Küçükçekmece, Turkey, 34303
      • Health Science University İstanbul Prof Dr Cemil Taşcıoğlu City Hospital

Participation Criteria

Eligibility Criteria

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

• Sampling Method: Probability Sample

Study Population

The study population will consist of adult patients admitted to the Intensive Care Unit (ICU) at a tertiary care hospital during the 12-month study period. This hospital is a high-acuity facility that serves a diverse population, including patients from both urban and rural areas. The ICU is equipped with advanced medical technologies and staffed by experienced healthcare professionals, making it an ideal setting for evaluating the effectiveness of AI models in clinical decision-making.

Description

Inclusion Criteria:

• Patients aged 18 years or older.
• Patients currently admitted to the Intensive Care Unit (ICU) during the study -period.
• Patients with sufficient clinical data available in the hospital's information system, including demographic information, clinical indicators, and treatment history.
• Patients for whom a discharge decision (to a general ward or home) needs to be made during their ICU stay.

Exclusion Criteria:

• Patients younger than 18 years old. Patients with incomplete or insufficient clinical data in the hospital's information system, making it difficult to assess their condition accurately.

Patients who are in the ICU for palliative care or end-of-life care, where discharge to a general ward or home is not anticipated.

Patients who have opted out of participating in the study or whose legal representatives have declined participation.

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Decision from Experts; Decisions maded by Intensive Care doctors	Other: decision; decision-making process regarding patient discharge from the Intensive Care Unit (ICU)
Decision from AI; Decisions maded by Artificial intelegence models	Other: decision; decision-making process regarding patient discharge from the Intensive Care Unit (ICU)

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
accuracy	The primary outcome is the accuracy of the AI models (ChatGPT and Gemini) in predicting the optimal timing for ICU discharge. Accuracy will be measured by comparing the AI models' predictions with the actual decisions made by ICU specialists. This outcome will be quantified using statistical metrics such as sensitivity, specificity, and the area under the ROC curve (AUC).	1 year

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
models comparison	The study will compare the performance of the two AI models, ChatGPT and Gemini, in predicting ICU discharge. This outcome will be assessed by evaluating the accuracy, agreement with clinical decisions, and time efficiency of each model.	1 year
utilzation	The effect of AI-assisted decision-making on ICU resource utilization, such as the length of stay and bed occupancy rates, will be evaluated. This outcome will assess whether AI-driven discharge decisions lead to more efficient use of ICU resources.	1 year

Collaborators and Investigators

• Sponsor: Kanuni Sultan Suleyman Training and Research Hospital
• Investigators: Principal Investigator: Engin ihsan turan, Health Science University İstanbul Kanuni Sultan Süleyman Education and Training Hospital

Study record dates

Study Major Dates

• Study Start (Actual): September 15, 2024
• Primary Completion (Actual): May 26, 2025
• Study Completion (Actual): May 27, 2025

Study Registration Dates

• First Submitted: September 3, 2024
• First Submitted That Met QC Criteria: September 3, 2024
• First Posted (Actual): September 5, 2024

Study Record Updates

• Last Update Posted (Actual): May 31, 2025
• Last Update Submitted That Met QC Criteria: May 27, 2025
• Last Verified: May 1, 2025

More Information

Terms related to this study

• Keywords: intensive care; discharge; clinical decission; artificial intelegence
• Other Study ID Numbers: Externation of ICU

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED

Drug and device information, study documents

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