Predicting treatment failure in severe sepsis and septic shock: looking for the Holy Grail