EARLY PREDICTION OF RESPIRATORY FAILURE IN LUNG CANCER PATIENTS ADMITTED TO ICU USING TIME-SERIES DEEP LEARNING

Abstract

Respiratory failure is common in lung cancer patients admitted to an intensive care unit (ICU) and the disease can cause an acute decline in the patient's state of health, which may not be generalized by the usual monitoring tools. In this study, we introduce Respira, a deep learning framework for the early prediction of respiratory failure in the intensive care unit (ICU) for lung cancer patients, which is a time-series based framework. This proposed method integrates all the clinical parameters, physiologic parameters, laboratory parameters, oxygenation parameters, and treatment related parameters along the longitudinal axis to reflect the changes in patient condition over time. Unlike simply leveraging the characteristics of the baseline, Respira will model the temporal evolution of a patient's data in order to identify early high-risk patients prior to a significant deterioration in respiratory condition. Time-series deep learning models like recursive and attention-based models have been created to capture sequential dependency and detect early warning signs that may result in respiratory failure. It can be utilized for stratification of patients, for setting up continuous monitoring and for timely clinical decisions in high-risk oncology ICU. The findings suggest that deep learning techniques may be useful in the early detection of a change in ventilation in the intensive care unit, and may be useful in guiding the clinician to take preventive measures, optimise ventilatory support, and allocate critical care resources more effectively, based on the available data.