P-753. Applying artificial intelligence to understand antibiotic resistance patterns of Staphylococcus aureus causing skin and soft tissue infections in children

TL;DR

This study uses AI to analyze antibiotic resistance patterns in Staphylococcus aureus causing skin infections in children, identifying trends over time.

Contribution

The novel use of association rule mining to uncover multidrug resistance patterns in pediatric S. aureus infections.

Findings

40 distinct MDR S. aureus phenotypes were identified from 8,171 children with SSTI.

Clindamycin resistance was present in nearly half of the identified phenotypes.

Some resistance patterns persisted throughout the study period, while others emerged in later years.

Abstract

Antimicrobial resistance (AMR) continues to grow worldwide. For Staphylococcus aureus (S. aureus), antibiotic resistant phenotypes have expanded in community settings, especially for skin/soft tissue infections (SSTIs). Resistance to clindamycin and other non-beta lactam antibiotics leads to multidrug resistance (MDR)S. aureus.Association mining is an unsupervised machine learning algorithm that examines higher-order relationships between resistant antibiotics. It can quantify prevalence (support) and strength of association (lift). We apply artificial intelligence (AI) using AM to understand MDR S. aureus patterns in children with SSTIs over time.

Data on children with SSTI seen in a large pediatric healthcare system in Atlanta, GA, U.S.A. were obtained retrospectively (2002-2019). Using AI, we applied association rule mining to look for antibiotic resistant patterns meeting the criteria: S. aureus SSTI from patients < 19 Y living in Atlanta. Association rule mining applied to identify antibiotic resistant phenotypes most frequent/clinically relevant (using Apriori algorithm), based on arules package in R. The following antibiotic classes were included in the analyses:clindamycin, erythromycin, gentamicin, linezolid, oxacillin, rifampin, trimethoprim-sulfamethoxazole, tetracycline, and vancomycin. Interestingness measures, including support (prevalence of resistance patterns) and lift (strength of associations between resistances) were employed to extract frequent and reliable patterns and then stratified by year and methicillin susceptibility. Analyses conducted in R.

Preliminary analyses of subset (2002-2016) of data shows 40 distinct MDR S. aureus phenotypes from 8,171 children with SSTI. Clindamycin resistance was found in 19/40 (48%) phenotypes, 27/40 (68%) phenotypes had >3 resistant antibiotic classes, 11/40 (28%) occurred only in later years, and 4/40 (10%) had persistent phenotype through entire study period.

Trends of MDRS. aureus are clustered around specific phenotypes. Association mining of S. aureus phenotypes across time can benefit antimicrobial stewardship programs by demonstrating which resistant patterns are circulating in the community setting.

Lilly Immergluck, MD, MS, American Academy of Pediatrics: Board Member|Department of Energy: Grant/Research Support|moderna: Grant/Research Support|NIH: Grant/Research Support|Pfizer: Grant/Research Support|Sanofi: Grant/Research Support