A multicenter evaluation of a novel microfluidic rapid AST assay for Gram-negative bloodstream infections

Abstract

Common phenotypic methods for antimicrobial susceptibility testing (AST) of bacteria are slow, labor intensive, and display considerable technical variability. The QuickMIC system provides rapid AST using a microfluidic linear gradient. Here, we evaluate the performance of QuickMIC at four different laboratories with regard to speed, precision, accuracy, and reproducibility in comparison to broth microdilution (BMD). Spiked (n = 411) and clinical blood cultures (n = 148) were tested with the QuickMIC Gram-negative panel and compared with BMD for the 12 on-panel antibiotics, and 10 defined strains were run at each site to measure reproducibility. Logistic and linear regression analysis was applied to explore factors affecting assay performance. The overall essential agreement and categorical agreement between QuickMIC and BMD were 95.6% and 96.0%, respectively. Very major error, major error, and minor error rates were 1.0%, 0.6%, and 2.4%, respectively. Inter-laboratory reproducibility between the sites was high at 98.9% using the acceptable standard of ±1 twofold dilution. The mean in-instrument analysis time overall was 3 h 13 min (SD: 29 min). Regression analysis indicated that QuickMIC is robust with regard to initial inoculum and delay time after blood culture positivity. We conclude that QuickMIC can be used to rapidly measure MIC directly from blood cultures in clinical settings with high reproducibility, precision, and accuracy. The microfluidics-generated linear gradient ensures high reproducibility between laboratories, thus allowing a high level of trust in MIC values from single testing, at the cost of reduced measurement range compared to BMD.

Importance: Increasing antimicrobial resistance underscores the need for new diagnostic solutions to guide therapy, but traditional antimicrobial susceptibility testing (AST) is often inadequate in time-critical diseases such as sepsis. This work presents a novel and rapid AST system with a rapid turnaround of results, which may help reduce the time to guided therapy, possibly allowing early de-escalation of broad-spectrum empirical therapy as well as rapid adjustments to treatments when coverage is lacking.

Keywords: diagnostics; multicenter study; rapid AST; sepsis.

Fig 1

Reproducibility of QuickMIC (green) compared to BMD (brown) showing percent of results within one twofold dilution (left) for each antibiotic, and percent of all results not yielding exactly the consensus MIC for each antibiotic. QuickMIC is overall more reproducible.

Fig 2

(A and B) Distribution of time to MIC for all results, split over antibiotics and species analyzed, indicating trends in the analysis time dependence on susceptibility category, where R/I results take longer time to result than S results. (C–E) Logistic and linear regression analysis of parameters affecting QuickMIC performance. (C) The ranges of inoculates achieved after sample preparation in comparison to likelihood of achieving accurate results (within EA) from the logistic regression analysis. Blue dashed lines indicate limits of quantitation, where estimated likelihood of EA > 90%, gray field is 95% confidence interval. The green dashed line indicates limit of detection, the lowest limit of quantifiable inoculum density. (D and E) display the same analysis for the time from blood culture detection to AST start and time from blood culture bottle unload to AST start, none of which are significantly affecting performance. The black dashed line indicates the limit of acceptable performance at 90% probability of achieving essential agreement, where the scale 0–1 represents 0%–100% probability.

Fig 3

Distribution of MIC results from all strains included in the study. Color coding according to reference MIC being within QuickMIC’s measurement range or not: green = in QuickMIC measurement range. Lines indicate I/R breakpoints (I = dashed, R = solid). (A) depicts the combined results from the clinical and spiked samples while (B) shows split by clinical or spiked sample collection.