Clinical and public genomic comparisons of carbapenem resistance in Pseudomonas aeruginosa

Background Cross-setting comparisons of carbapenem resistance may reflect clinical contexts and selective genomic sampling. We examined reproducibility on two independent scales. Methods We analysed retrospective adult cohorts from a Chinese hospital ( n = 1,362) and the Medical Information Mart for Intensive Care IV (MIMIC-IV) v3.1 ( n = 3,251), along with 26,417 publicly submitted clinical Pseudomonas aeruginosa BioSamples. Our clinical analyses compared susceptibility-defined carbapenem-resistant Pseudomonas aeruginosa (CRPA), evaluated outcomes, adjusting for associations, and conducted exploratory reciprocal model transport. Our genomic analyses examined annotated determinants, submission-year patterns, single-nucleotide polymorphism (SNP) clusters, and co-resistance, including cluster-weighted sensitivity analyses. Results CRPA proportions were 36.9 and 9.3% in the two institutions. Unadjusted in-hospital mortality and post-index hospital stay were higher among patients with CRPA. Adjusted association directions agreed for 13 of 18 covariates. Transported models showed modest discrimination (area under the curves [AUCs], 0.641 and 0.619) and marked miscalibration (calibration slopes, 0.45 and 0.52). Among the included public Biosamples, 60.7% carried a defined carbapenem-associated determinant and 24.4% carried an acquired carbapenemase. Chinese submissions had higher serine-carbapenemase positivity, whereas US submissions had higher metallo-β-lactamase and chromosomal-determinant positivity. The country OR for bla KPC changed from 8.04 in the full sample to 1.59 in the cluster-assigned weighted analysis; the OR for acquired carbapenemase changed from 1.32 to 0.67. Conclusion Clinical associations may recur without supporting a reliable transported prediction. Genomic contrasts are sensitive to sampling and analysis population. These exploratory, unlinked findings neither establish causal effects nor estimate national prevalence; genomic results do not confirm phenotypic CRPA.