Fig. 2

Computing the probability of treatment failure of disease episodes. (1) For each disease episode, we subset the resistance matrix with the drugs used during the disease episode. (2) Then, for each pathogen of the subsetted matrix, we select the minimal level (min) of resistance across the used drugs. (3) The resulting vector is weighted (element-wise multiplication) by the vector of aetiology probabilities computed for that disease episode by the naive Bayes model (Fig. 1); and, (4) The mean of that weighted vector is used as the probability of treatment failure in the disease episode