Antimicrobial resistance surveillance: can we estimate resistance in bloodstream infections from other types of specimen?

Background: Antimicrobial resistance (AMR) is a major global health threat. Standard approaches to AMR surveillance through susceptibility testing of isolates from blood cultures are difficult in low- and middle-income countries (LMIC), where lack of laboratory capacity prevents routine patient-level antimicrobial susceptibility testing, and systematic testing of invasive specimens may not be feasible. Other specimen types could provide an alternative but effective approach to surveillance, but the relationship between resistance prevalence in these and bloodstream infections has not been systematically evaluated. Methods: We used data from Oxfordshire, UK, 1998-2018, to investigate associations between resistance rates in Escherichia coli and Staphylococcus aureus isolates from blood and other specimens, comparing proportions resistant in each calendar year using time series cross-correlations, for multiple antibiotics. We also compared the proportion of resistant isolates from blood versus other specimens across drug-years, overall and across four arbitrary resistance categories (<5%, 5-10%, 10-20%, >20%). We repeated analysis across four high-income and 12 middle-income countries, and in three hospitals/programmes in LMICs. Findings: 8102 E. coli bloodstream infections, 322087 E. coli urinary tract infections, 6952 S. aureus bloodstream infections and 112074 S. aureus non-sterile site cultures were included from Oxfordshire. Resistance trends over time in isolates from blood versus other specimens were strongly correlated (maximum cross-correlation 0.51-0.99 with strongest associations between proportions in the same year for 18/27 pathogen-drug combinations). Resistance prevalence was broadly congruent across drug-years for each species, particularly allowing for uncertainty in estimation. 207/312 (66%) species-drug-years had resistance prevalence in other specimen types within +/-5% of that blood isolates, and 276/312 (88%) within +/-10%. 215/312 (69%) species-drug-years were in the same resistance category for blood and other specimen types; 305 (98%) were the same or adjacent resistance categorisation. Results were similar across multiple countries in high- and middle-income settings, and the three LMIC hospitals/programmes. Interpretation: Resistance in bloodstream and other less invasive infections are strongly related, suggesting the latter could be a surveillance tool for AMR in LMICs. These infection sites are easier to sample and cheaper to obtain the necessary numbers of susceptibility tests, thus providing more cost-effective evidence for decisions including empiric antibiotic recommendations.