Artificial Intelligence Based Risk Assessment of Patients Clinical Respiratory Progression Status

Study Design: Cross-sectional Analytical – Both Retrospective and Prospective

Study Setting: Departments of Emergency Medicine, Trauma, Medicine, medicine, Pulmonology, Anaesthesiology and Critical Care, AIIMS Rishikesh

Study Population: All patients (minimum 500 patients) admitted to the above departments at AIIMS Rishikesh (irrespective of COVID Status) during the study period satisfying the following criteria.

Inclusion Criteria:

1.     Aged 18 years and above.

2.     Able and willing to give informed written consent (Signature/Thumb impression)

Exclusion Criteria:

1.     Unable or unwilling to give informed consent

Requirement for informed consent will be waived off for retrospective patient data with IEC approval.

Methods:

Phase 1: In Phase 1 of the study, retrospective data of 10-100 patients will be analysed for Model evaluation on GEHC site, iteration and development of initial proof of concept.

Phase 2: In Phase 2, prospective data of 100 patients will be used for Testing of initial proof of concept in the clinical setting on prospectively collected datasets.

Phase 3: In Phase 3, Testing of the iterated model in the clinical setting on prospectively collected datasets will be done in 500 patients for Performance evaluation of model as a decision support tool in the clinical setting

 Phase 4: Clinical deployment across multiple hospitals (through separate MOUs)

Data Collection

1.     Patient demographics, history, monitor parameters, ventilator parameters, imaging results and lab tests for COVID related illness that are readily available in a consistent format at AIIMS Rishikesh inpatient files/records/systems/e-Hospital will be collected

2.     The associated metadata (ie DICOM tag extraction, if possible EPR data) will be collected and aggregated

3.     Data will be anonymized to remove any patient protected health information (PHI)

a.     Site to keep a PHI-rich copy of the data on-hand

b.     A PHI free copy of the data

4.     Transfer the data via secure media to GE for algorithm development and validation

Retrospective/Prospective Data Collection - De-identified datasets for minimum of 100 patients

a.     Demographic datasets including but not limited to Age, Gender, Ethnicity, Weight, Height, ICD-9 code diagnosis, ICD-9 code procedures, ICU Admit Time, ICU Discharge Time, Past Medical History, Hospital Admit Time, Hospital Discharge Time, Mortality (Yes or no, if yes then the time of death)

b.     Lab tests including but not limited to ABG tests, Urea, Creatinine, Electrolytes, Monitor parameters including RR, HR, SpO2, ECG, Blood Pressure Parameters (Invasive & Non-Invasive BP parameters), Temperature

c.     Ventilator parameters including FiO2, FeO2, PEEP, EtCO2, MV, TV, PIP, PP, Intubation Time, Extubation Time, Mode of Ventilation, Ventilator Settings Parameters

d.     Radiology imaging (CT, longitudinal X-ray)

Prospective testing and performance evaluation

1.     Deploy a version of the developed algorithm at AIIMS Rishikesh to run on prospectively recruited patient datasets

2.     Record performance of the algorithm against the expected outcome and determine if further development is required

3.     Determine the relevance of the algorithm

Clinical deployment across multiple hospitals (through MOUs)

1.     Determine quickest and most efficient deployment platform accessible for all AIIMS Rishikesh partners

 **Outcomes:**1.     Data Collection and initial proof of concept model build

a.     De-identified datasets to transferred to GEHC data scientists

b.     Proof of concept model results shared with broader collaboration

2.     Prospective evaluation

a.     The model developed is evaluated within 2 weeks of readiness in a controlled environment in AIIMS Rishikesh as a pilot

3.     Deploy algorithm on mass

a.     The deployment model identified and implemented at collaboration sites

b.     Review the option to deploy models nationally