WIP: Vital Sign Analysis: Decompensation

Early and accurate prediction of decompensation (functional deterioration) in patients in domestic settings may help prevent deaths. Based on values that can be measured from portable devices such as heart rate, blood oxygen saturation, systolic blood pressure, temperature, and age, we study the prediction capability of machine learning algorithms to determine patient decompensation (death) in the next 24 hours. Two sources of variability influencing the performance of predictions are investigated during training and inference: (i) the effect of reducing the number of input signals from those available in our source data set (about 17 different vital signs; source: MIMIC-III) to a subset that can be collected by portable devices; (ii) the effect of the sampling frequency (how many measurements per unit of time are necessary for training a reliable prediction system). We conclude our analysis with a inference-only study on the effect of the number of measurements preceding the prediction (does having more measurements improve prediction?).