Exploring multidrug resistance patterns in community-acquired <i>Escherichia coli</i> urinary tract infections with machine learning

While associations of antibiotic resistance traits are not random in multidrug-resistant (MDR) bacteria, clinically relevant resistance patterns remain underexplored. This study used association-set mining to explore resistance associations within <i>E. coli</i> isolates from community-acquired urinary tract infection isolates collected from 2018 to 2022 by France's national surveillance system. Association-set mining was applied separately to extended-spectrum beta-lactamase-producing <i>E. coli</i> (ESBL-EC) and non-ESBL-EC. MDR patterns with expected support (reflecting pattern frequency) and conditional lift (reflecting association strength) higher than expected by chance (<i>P</i>-value ≤ 0.05) were used to construct resistance association networks and analyzed according to time, age, and gender. The number of isolates increased from 360,287 in 2018 to 629,017 in 2022. More MDR patterns were selected in ESBL-EC than non-ESBL-EC (2022: 1770 vs 93 patterns), with higher respective network densities (2022: 0.301 vs 0.100). Fluoroquinolone, third-generation cephalosporin, and penicillin resistance were strongly associated in ESBL-EC. Median network densities increased from 2018 to 2022 in both ESBL-EC (0.238-0.301, <i>P</i>-value = 0.06, Pearson test) and non-ESBL-EC (0.074-0.100, <i>P</i>-value = 0.04). Across all years, median densities were higher in men than in women (ESBL-EC 2022: 0.305 vs 0.271; non-ESBL-EC: 0.128 vs 0.094) and higher in individuals over 65 than under 65 (ESBL-EC: 0.289 vs 0.275; non-ESBL-EC: 0.103 vs 0.088). These findings highlight temporal, age-specific, and gender-specific variations in resistance patterns, underscoring the potential of machine learning to understand them and inform antibiotic strategies.