1st International Congress for Innovation in Global Surgery
ABSTRACT FIRST PRESENTED: 20.04.2022
Machine Learning based mortality prediction model for COVID-19 patients admitted in a tertiary care hospital, New Delhi.
Introduction: SARS corona virus 2 is found to be responsible for the COVID-19 pandemic. It caused severe acute respiratory illness with a high mortality rate. This deadly disease infected around 4 crore population in India and costs around 5.2 lakh deaths. This enveloped RNA virus is transmitted from the aerosols of infected patients. The virus attaches to the host with its spike protein to ACE2 receptors of lung epithelial cells and begins its replication. The symptoms can range from mild severity like sore throat, fever, and cough to severe category like pneumonia and ARDS which leads to increased mortality. There is a huge interindividual variation in response to the virus. It is often unpredictable to decide the course of the disease. Comorbid conditions like diabetes, hypertension, and major organ failures have been known to increase the severity of the disease. Therefore, the aim of this study was to apply an interpretative Artificial intelligence algorithm for identifying the main features which decide the course of the disease.
Methods: It is an observational study of hospitalized COVID-19 patients. 1181 RT-PCR confirmed COVID-19 patients (both alive and dead) data was used to predict the factors responsible for the course of the disease using a ML model. Twelve features were used to train [75%] and test [25%] the datasets using logistic regression and 2 machine learning methods [ XG Boost, Random Forest (RF)]. AI model interpretability was analyzed using a SHAP summary plot.
Results: Among those patients analyzed using different AI models the Random Forest algorithms had higher sensitivity (83%) and accuracy (81%) on test datasets. Among the 12 features, AI interpretable SHAP summary plot showed age, kidney disease, heart disease, diabetes, male sex, and hypertension among the top rank features which have a high impact on the outcome of the disease (death).
Conclusions: We observed that AI interpretable model ranks the major factors that are contributing to the death of the patient and is able to predict the key ranking features with 83% sensitivity. Higher the age higher the risk of mortality. It also predicts that patients with heart disease, kidney disease, and diabetes are at higher risk for mortality. This model can be implemented in all healthcare facilities for risk stratification in COVID-19 patients. It helps in the proper management and taking necessary precautions beforehand.
Keywords: COVID-19, Artificial Intelligence
An innovative surgical method to operate the zygomatico-maxillary-orbital complex in the absence of intraoperative computer tomography or intraoperative navigation using novel zygoma analysis and virtual surgical planning software
Feasibility of the frugally-engineered ‘LeVe CPAP’ from pilot data in healthy volunteers at Mengo Hospital, Uganda; with potential for use in patients with acute hypoxaemic respiratory failure associated with COVID-19 and other respiratory illnesses