Sepsis is one of the leading causes of mortality and morbidity in ICUs. One relevant step to adequately prevent the increased mortality caused by sepsis is to predict these episodes in order to allow preventive medical inter-ventions. In this study is introduced a methodology with the goal of finding onsets of worsening progressions from multiple physiological parameters which may have predictive value in sepsis events. The progression of each physiological parameter is modelled through the first derivative of the time-series, repre-senting the trend/slope of a given parameter. Aside the trend term the filtered value may also have predictive value in the early detection of a sepsis to occur, e.g., a low blood pressure status. These two terms, i.e. the trend and filtered value are calculated through a Savitzky-Golay filter for each varia-ble. However, for the extraction of these two terms, all measurements must be available, which is not the case, since the dataset presents a high degree of missing values in some variables, as expected in an ICU context. To tackle these challenges the variables with incomplete measurements, i.e., which do not present a measurement per hour, but present some measurements, are liner interpolated in order to extract the aforementioned terms. In the cases where not a single measurement is available, the trend and filtered values are not extracted, and an Adaboost Ensemble Tree with surrogate splits is em-ployed in order to handle this lack of variable presence. The referred meth-odology validated on a 5000 patient dataset achieves a mean score of 0.31±0.08 in a 10X10-fold cross validation. For future work is intended to explore new methods, namely the ones based in One Class Learning, in order to model a normality state and detect deviations and worsening progressions.