Using Machine Learning to Build Predictive Models on Patients with ARDS in medical Intensive Care Unit

OBJECTIVES l  To Explore and Analyse Variables Contributing to Average Length of Stay, Severity of ARDS and Possible Outcome related to ARDS.

l  To Facilitate AI Experts in Developing Prediction Algorithm for Average Length of Stay, Severity of ARDS and Possible Outcome related to ARDS.

l  To Pilot Test Effectiveness of Developed AI Algorithms in Predicting Average Length of Stay, Severity of ARDS and Possible Outcome related to ARDS.

Justification for study (whether of national significance with rationale)

Depending on the severity of the sickness and the duration of stay, an administrator must identify and predict any particular, frequent, and catastrophic condition that may increase hospital stays and result in greater costs, either from the patient directly or through a third party.

We hope that this study will be able to predict both the average length of stay for patients with acute respiratory distress syndrome (ARDS) and possible clinical outcomes. Additionally, we will identify the critical factors that will put ARDS patients into three severity categories: mild, moderate, and severe.Finding this would assist the administrator and the doctor in improving patient communication and preparing them for any potential ARDS-related clinical outcomes.

METHODOLOGY The study aims to investigate and evaluate variables in ICU 1, 2, and 3 with the help of clinicians. Data will be collected retrospectively from the Medical Records Department (MRD) of 462 patients with ARDS admitted in ICU 1, ICU 2, and ICU 3 from August 2021 to August 2024. The study will use descriptive analysis, Chi square test, Univariate analysis, and Multivariate analysis to determine factors contributing to the severity of ARDS and potential outcomes. After AI prediction model is build , it will be assessed using a pilot study, which will be conducted retrospectively on 100 patients from the MRD of IEC No. 591/2019 and CTRI/2019/11/021857. The pilot testing will evaluate the sensitivity and specificity of the model. The study aims to facilitate the machine learning development process with AI and machine learning experts. The data collected will be analyzed using a flowchart and will be used to develop a predictive model for predicting ARDS severity and potential outcomes.

Detailed description of procedure / processes :

 Objective 1 Methodology  :

With the assistance of clinicians, investigate and evaluate the Variables  noted in ICU 1, 2, and 3. Finalize the Data Collection Form by removing all unnecessary variables.

 After IEC and CTRI  approval ,  Data  will be collected Retrospectively gathering the information of 462 patients from Medical Records Department.

 All Patients hospitalized to the ICU for longer than 48 hours will have their data retrospectively collected from MRD for the previous three years(august2021-august2024), including patients with   ARDS only . This will be done to ensure that we don’t miss anything that could have an impact on the average duration of stay, the severity of ARDS, and the potential results.

Study settings :

 Departments including  Critical Care Department in Kasturba medical college , Manipal and Data Collected from the Medical Records Department.

 Data collection form:

 Data will be collected using excel sheet where new variables will be added depending upon the clinicians .

 Analysis of Data**:**

Analysis using descriptive analysis, nominal variables will be reported using mean ± standard deviation and median with IQR depending on the normality of the data.

Variables in Respiratory chart includes  ratio of inspired oxygen fraction (FiO2) to partial pressure of oxygen (PaO2) is less than 300 mmHg , Positive end-expiratory pressure (PEEP) >-5 AP, Respiratory rate, heart rate,  sp02, pH, Partial pressure of Carbon dioxide (PaC02) , Saturation of oxygen (Sa02), HC03, Lac, AG.APACHE, SOFA  ,  Neutrophil/Lymphocyte ratio and  CBC

Serology include Hb , Hct /PCV, platelet, WBC, DLC, PT, INR , Aptt, LDH , urea , creatinine, sodium, potassium,  calcium, TB, DB, proteins, Albumin, Globulin, AST, ALT, ALP, PCT. CRP, Bicarbonate.

To determine the factors that contribute to the severity of ARDS and potential outcomes, we will use the Chi square test, Univariate analysis, and Multivariate analysis.

While categorical data are given as percentages and figures, continuous variables are typically presented as mean ± standard deviation or median (interquartile range), depending on the situation.

 Objective 2 Methodology :

 Identifying the essential elements of ARDS and using clinicians help to discover significant variables for patient outcomes in order to accelerate the development of machine learning employing machine learning professionals.

 Objective 3 Methodology:

 To assess the effectiveness of the AI prediction model in predicting length of stay, severity of ARDS, and potential outcomes, a pilot study will be initiated. This study will retrospectively collect data from 100 patients from the MRD of IEC No. 591/2019 and CTRI/2019/11/021857, which were prospectively collected in 2019. The pilot testing aims to evaluate the sensitivity and specificity of the model.