Automated prediction of deterioration of infants in paediatric intensive care using SpO₂

Acute life-threatening events are mostly predictable in adults and children. Despite real-time monitoring these events still occur at a rate of 4%. This paper describes an automated prediction system based on the feature space embedding and time series forecasting methods of the SpO₂ signal; a pulsatile signal synchronised with heart beat. We develop an age-independent index of abnormality that distinguishes patient-specific normal to abnormal physiology transitions. Two different methods were used to distinguish between normal and abnormal physiological trends based on SpO₂ behaviour. The abnormality index derived by each method is compared against the current gold standard of clinical prediction of critical deterioration.