Abstract No.: ABS0875: Utilisation of machine learning in predicting clinical outcome in COVID-19

BACKGROUND AND AIMS: The goal of this study was to develop a deep-learning artificial intelligence algorithm to identify the top predictors amongst the large array of clinical variables at admission to predict the likelihood of mortality in coronavirus disease 19 (COVID-19) patients. METHODS: Data such as complete blood count, RFT, serum electrolytes, liver function tests, interleukin-6 (IL-6), C-reactive protein, procalcitonin, ferritin, FiO(2), mode of ventilation, age, sex, and co-morbidities were collected from 250 patients admitted to the ICU. The variables missing for more than 20% of the people were removed. Chi-Square and Pearson’s correlation coefficient were used to remove the clinical variables having less importance with respect to the probability of being discharged or mortality. The LightGBM model was used for classification. The data were split into train and test in the ratio of 0.8:0.2. This gave an output of survival and mortality probability. [Image: see text] RESULTS: The model gave a result with an accuracy of 0.64, a precision of 0.73, a recall of 0.64, an F1 score of 0.66, and an area under the curve of 0.79. The model had a specificity of 66.6% and a sensitivity of 61.5%. Ferritin and IL-6 values were the top predictors of mortality. CONCLUSION: This approach has the potential to provide frontline physicians with a simple and objective tool to stratify patients based on risks so that they can triage COVID-19 patients more effectively in time-sensitive, stressful, and potentially resource-constrained environments.