In this work, the hierarchical cluster analysis (HCA) of time series is proposed for monitoring antimicrobial resistance (AMR). The purpose was to develop a new mathematical tool or new applications that could allow the identification of pairs of bacteria/antimicrobial agents that require early intervention or monitoring.

Time series data of resistant bacteria/antimicrobial agent pairs was obtained from the public website of ECDC in 2023.

Variables: Resistant isolate percentage values in Portugal from 2012 to 2021 were obtained. Methods: HCA was performed directly on the time series data using the standard Affinity coefficient and Euclidean distance. The complete linkage was used as the aggregation criteria. Results: Enterococcus faecium|Aminopenicillins presents very high resistance (87,96% +/- 3,12 [min-83,66/max-94.44] and Escherichia coli |Carbapenems presenting the lowest resistance values (0,18% +/-0,17 [min-0,02/max-0,53]; this is quite curious, since the WHO considers this pair as a critical priority. The HCA with the (1-standard Affinity coefficient) distance showed three different clusters. The first cluster includes K. pneumoniae |Carbapenems and E. coli |Carbapenems. The third cluster includes E. faecium|Vancomycin, E. faecium| Gentamicin, Enterococcus faecalis|Vancomycin, Acinetobacter spp.|Aminoglycosides, Fluoroquinolones, and Carbapenems. The second cluster is a heterogeneous group showing some clusters not analyzed here. The first cluster includes two Enterobacteria considered Critical Priority by the WHO. These bacteria can produce several resistance mechanisms, namely carbapenemase enzymes. The third cluster includes Acinetobacter spp. with all antimicrobial agents studied, including Carbapenems, Enterococcus with Vancomycin and Gentamicin. Conclusions: This was the first time that the Affinity coefficient and Euclidean distance were applied to the AMR profile. The standard Affinity coefficient (Silhouette coefficient’s (0.8)) presented to be more suitable than Euclidean distance (Silhouette coefficient’s value (0.4)) for identifying patterns and trends in resistance values in AMR data because it was shown to be more robust in capturing similar behaviors in resistance values over time and a better cluster compactness and separability.