How Well Do Rapid Blood Culture Panel Resistance Targets Predict Laboratory Susceptibility? An Open, Gene-by-Gene Concordance Benchmark of U.S. Clinical Isolates in NCBI Pathogen Detection

Rapid blood culture identification panels report resistance genes (mecA/mecC, vanA/vanB, KPC, NDM, OXA-48-like, VIM, IMP and CTX-M) hours to days before growth-based antimicrobial susceptibility testing (AST) is complete, and laboratories and prescribers act on that early result. We built an open, versioned, isolate-level benchmark that measures how often each panel target predicts the AST result that follows, using public data only. From 26,886 NCBI BioSample records carrying a submitted antibiogram for Staphylococcus aureus, Enterococcus faecalis, Enterococcus faecium and the Enterobacterales, we joined AMRFinderPlus gene calls from NCBI Pathogen Detection, kept U.S. human clinical isolates, mapped genes to panel targets through NCBI’s Reference Gene Hierarchy, and reinterpreted 19,741 isolate-drug MICs and zone diameters against CLSI M100 (35th edition) breakpoints. The benchmark contains 5,168 isolates (731 from blood or other sterile sites). mecA predicted oxacillin or cefoxitin resistance in 215/220 S. aureus (positive predictive value, PPV, 97.7%; 95% CI 94.8%–99.0%), and its absence predicted susceptibility in 329/330 (NPV 99.7%). Any panel carbapenemase predicted resistance to at least one carbapenem in 272/278 Enterobacterales (PPV 97.8%) but to meropenem in only 241/275 (PPV 87.6%), with OXA-48-like and IMP enzymes most often leaving meropenem susceptible. CTX-M predicted ceftriaxone resistance with PPV 97.7% across Enterobacterales, but a CTX-M-negative result was a weak predictor of susceptibility outside E. coli (NPV 94.2% in E. coli, 37.3% in K. pneumoniae complex, and below 10% in chromosomal AmpC species). In a secondary analysis, CTX-M-1-group enzymes conferred ceftazidime resistance in 64.2% of E. coli versus 14.5% for CTX-M-9-group enzymes. vanA predicted vancomycin resistance in 26/29 enterococci; no vanB or mecC was present. All 970 discordant isolate-target pairs are released with a rule-based likely cause; most gene-negative resistant isolates carried a non-panel ESBL or AmpC, a panel carbapenemase, or belonged to a chromosomal AmpC species. The data, gene-to-target crosswalk, discordance collection, code and a public lookup page are openly released for reuse and annual update.