Author: Luciano Rodriguez
The population-based distributions of values for all vital signs and lab result has been known for decades. However, in harmony with the current direction and future adoption of personalized medicine, the fact that individual deviations from these population mean values can be dramatic, necessitates the need to find and use subject specific average and extreme values when making diagnosis, deciding on treatment and trying to ascertain the state of the current condition. In this study we propose a novel method for achieving this goal of personalized extreme value analysis. In particular, we implement ARIMA time series methods on the residuals component of the STL decomposition of the raw vital signs and lab results time series values in order to detect all subject specific extreme values. Further, we find all significant predictors for the occurrences of these extreme values by using logistic regression model building combined with exhaustive algorithms for variable selection. Our method is a true step into the future of personalized medicine that will dramatically increase the precision of diagnosis, treatment and quality of care.
Co Author/Co-Investigator Names/Professional Title: Cyril Rakovski / Associate Professor