Antilogic, a new supervised machine learning software for the automatic interpretation of antibiotic susceptibility testing in clinical microbiology: proof-of-concept on three frequently isolated bacterial species.

OBJECTIVE: Antibiotic susceptibility testing (AST) is necessary in order to adjust empirical antibiotic treatment, but the interpretation of results requires experience and knowledge. We have developed a machine learning software that is capable of reading AST images without any human intervention and that automatically interprets the AST, based on a database of antibiograms that have been clinically validated with European Committee on Antimicrobial Susceptibility Testing rules. METHODS: We built a database of antibiograms that were labelled by senior microbiologists for three species: Escherichia coli, Klebsiella pneumoniae, and Staphylococcus aureus. We then developed Antilogic, a Python software based on an original image segmentation module and supervised learning models that we trained against the database. Finally, we blind tested Antilogic against a validation set of 5100 photos of antibiograms. RESULTS: We trained Antilogic against a database of 18072 pictures of antibiograms. Overall agreement against the validation set reached 97% (16 855/17 281) regarding phenotypes. The severity rate of errors was also evaluated: 1.66% (287/17 281) were major errors and 0.80% (136/17 281) were very major errors. After implementation of uncertainty quantifications, the rate of errors decreased to 0.80% (114/13 451) and 0.42% (51/13 451) for major and very major errors respectively. DISCUSSION: Antilogic is the first machine learning software that has been developed for AST interpretation. It is based on a novel approach that differs from the typical diameter measurement and expert system approach. Antilogic is a proof of concept that artificial intelligence can contribute to faster and easier diagnostic methods in the field of clinical microbiology.

Comparative evaluation of the UMIC Colistine kit to assess MIC of colistin of gram-negative rods.

BACKGROUND: The recent description of the first plasmid-mediated colistin-resistant gene mcr-1, conferring transferable and low-level resistance to colistin, raised concern about the need to implement a rapid and reliable screening method to detect colistin-resistant clinical isolates. The only valid method to assess the MIC of colistin is the broth microdilution according to the joint CLSI-EUCAST Polymyxin Breakpoints Working Group. UMIC Colistine is a ready-to-use broth microdilution kit developed to easily assess colistin MIC by proposing unitary polystyrene strips containing 11 concentrations of dehydrated colistin. Here, we evaluated the UMIC Colistine kit on 235 Gram-negative rods (176 Enterobacterales, including 70 harboring a mcr gene, and 59 non-fermentative), through comparison to the reference broth microdilution method prepared in accordance with EN ISO 20776-1:2006 standard. Reproducibility of the UMIC Colistine was assayed with the three recommended quality control strains E. coli ATCC 25922, E. coli NCTC 13846 (mcr-1 positive), and P. aeruginosa ATCC 27853, as for stability testing. RESULTS: Categorical agreement was 100% with 63.4% (n = 149) of colistin-resistant strains, and 36.6% (n = 86) of colistin-susceptible strains with both methods (S ≤ 2 µg/mL and R > 2 µg/mL). No major error or very major error was reported. Essential agreement was 94.0% (n = 221), and 100% for detection of colistin-resistant strains as compared to the reference method. Pearson's correlation between UMIC Colistine and the reference method was 0.98. Reproducibility of the UMIC Colistine system was 97.8% with MICs of the quality control strains within the target ranges. However, some isolates had lower MIC with UMIC Colistine, but that did not change their categorization as colistin-susceptible, and this phenomenon should be further explored. CONCLUSIONS: The UMIC Colistine kit is an easy to perform unitary device that showed excellent results when compared to the reference method. The UMIC Colistine system is a rapid and reliable broth microdilution method that is suitable to assess the colistin MIC of clinical isolates in clinical microbiology laboratories.

No global increase in resistance to antibiotics: a snapshot of resistance from 2001 to 2016 in Marseille, France.

Since effective empirical antibiotic therapy is a key factor for survival, local antibiotic resistance epidemiology is critical. We aimed to identify current trends in antibiotic resistance for key antibiotics obtained over 16 years (2001-2016) for invasive infections corresponding to empirical treatment in a large hospital centre in Marseille, France.From January 2014 to December 2016, we have collected all data on antibiotic susceptibility from public laboratory hospitals, and a retrospective analysis was performed on key antibiotics in blood cultures since 2001. A total of 99,932 antibiotic susceptibility testings (ASTs) were analysed, and proportion of pan-drug resistant (PDR = resistant to all antibiotics tested) and extensively drug-resistant (XDR = resistant to all except for two classes) strains were < 0.03 and 0.5%, respectively. Between 2001 and 2016, we found an increase of resistance to third-generation cephalosporins for E. coli invasive strains (0% vs 17.8%; p < 10-5) and K. pneumoniae (8% vs 35.4%; p = 0.001) along with a decrease of methicillin-resistant S. aureus strains (31% vs 19.8%; p = 0.006). Moreover, during the 3-year period, a significant increase of wild-type strains, susceptible to all antibiotics tested, was observed in invasive infections. Regarding bacteraemia involving Enterobacteriaceae and S. aureus, empirical therapy is effective in > 99% cases. Active epidemiological surveillance is necessary because antibiotic resistance remains unpredictable.

New therapy from old drugs: synergistic bactericidal activity of sulfadiazine with colistin against colistin-resistant bacteria, including plasmid-mediated colistin-resistant mcr-1 isolates.

The recent emergence of colistin (COL) resistance, particularly mcr-1 plasmid-mediated COL resistance in Gram-negative bacteria, has led to renewed interest in antibiotic combinations to overcome clinical therapeutic impasses. The aim of this study was to evaluate the potential of the synergistic and bactericidal activity of COL in combination with sulphonamide compounds, including sulfadiazine (SDI), sulfamethoxazole (SMX) and trimethoprim/sulfamethoxazole (SXT), as well as trimethoprim (TMP) against clinical COL-resistant bacterial strains, including strains with the plasmid-encoded mcr-1 gene. A collection of 55 COL-resistant and -susceptible strains from different origins (Laos, Thailand and France) was used in this study. Several in vitro methods were used to determine the potential of the synergistic activity of these combinations, including Etest on agar pre-treated plates, the Etest cross method and the chequerboard assay. A time-kill assay was performed to evaluate the potential bactericidal activity of combinations in addition to synergistic activity. Significant synergistic activity was observed with all combinations tested. The combination of COL + SDI presented the highest synergistic effect against the various species of COL-resistant strains (92.7%). For the other combinations, a synergistic effect was also observed but with lower frequency for COL + SMX (33.3%), COL + TMP (47.3%) and COL + SXT (31.5%). Synergy was observed independently of the COL resistance mechanism. These in vitro results suggest that the combination of COL + SDI would appear to be justifiable in patients with multidrug-resistant bacterial infections that cannot be treated with COL monotherapy.

LBJMR medium: a new polyvalent culture medium for isolating and selecting vancomycin and colistin-resistant bacteria.

BACKGROUND: Multi-drug resistant bacteria are a phenomenon which is on the increase around the world, particularly with the emergence of colistin-resistant Enterobacteriaceae and vancomycin-resistant enterococci strains. The recent discovery of a plasmid-mediated colistin resistance with the description of the transferable mcr-1 gene raised concerns about the need for an efficient detection method for these pathogens, to isolate infected patients as early as possible. The LBJMR medium was developed to screen for all polymyxin-resistant Gram-negative bacteria, including mcr-1 positive isolates, and vancomycin-resistant Gram-positive bacteria. RESULTS: The LBJMR medium was developed by adding colistin sulfate salt at a low concentration (4 µg/mL) and vancomycin (50 µg/mL), with glucose (7.5 g/L) as a fermentative substrate, to a Purple Agar Base (31 g/L). A total of 143 bacterial strains were used to evaluate this universal culture medium, and the sensitivity and specificity of detection were 100% for the growth of resistant strains. 68 stool samples were cultured on LBJMR, and both colistin-resistant Gram-negative and vancomycin-resistant Gram-positive strains were specifically detected. CONCLUSIONS: The LBJMR medium is a multipurpose selective medium which makes it possible to identify bacteria of interest from clinical samples and to isolate contaminated patients in hospital settings. This is a simple medium that could be easily used for screening in clinical microbiology laboratories.

Contemporary challenges and opportunities in the diagnosis and outbreak detection of multidrug-resistant infectious disease.

INTRODUCTION: The dissemination of multi-drug resistant bacteria (MDRB) has become a major public health concern worldwide because of the increase in infections caused by MDRB, the difficulty in treating them, and expenditures in patient care. Areas covered: We have reviewed challenges and contemporary opportunities for rapidly confronting infections caused by MDRB in the 21st century, including surveillance, detection, identification of resistance mechanisms, and action steps. Expert commentary: In this context, the first critical point for clinical microbiologists is to be able to rapidly detect an abnormal event, an outbreak and/or the spread of a MDRB with surveillance tools so that healthcare policies and therapies adapted to a new stochastic event that will certainly occur again in the future can be implemented.

Evaluation of the Scan® 1200 as a rapid tool for reading antibiotic susceptibility testing by the disc diffusion technique.

OBJECTIVES: In clinical microbiology, some instruments are able to automatically read inhibition zone diameters for antibiotic susceptibility testing (AST) performed by the disc diffusion (DD) method. The actual resolution of commercial reader systems is low and high-resolution scanners have been developed for microbiology. Here, we evaluated and compared the reading and interpretation of AST by the DD method using the Scan® 1200 instrument as compared with the Sirscan® system on 211 clinical strains and the possibility to read AST after 6 or 8 h of incubation as compared with 24 h on 121 additional Gram-negative strains and 76 non-fermenter Gram-negative and Gram-positive strains. METHODS: Validation of the technique was assessed on three reference strains as requested by EUCAST for analysis of the repeatability and reproducibility of the method. RESULTS: Correlation between the two methods, assessed using 211 clinical isolates (n = 2439 zones of growth inhibition measured), was excellent with a correlation coefficient of 0.97. For the earlier reading experiments, preliminary results demonstrate the possibility of reading AST for drug-species combinations after 6 and 8 h for Gram-negative bacteria with rapid growth (correlation coefficient at 6 h = 0.96 and at 8 h = 0.98) and at 8 or 10 h for Gram-positive bacteria. CONCLUSIONS: The Scan® 1200 has many advantages thanks to its small size, rapidity of use and high-resolution imaging allowing the possibility to improve AST results after only 6-8 h of incubation. This AST reader system represents a robust alternative tool for routine use in clinical microbiology laboratories.

Real-time video imaging as a new and rapid tool for antibiotic susceptibility testing by the disc diffusion method: a paradigm for evaluating resistance to imipenem and identifying extended-spectrum ß-lactamases.

The disc diffusion method has long been considered the standard technique for antibiotic susceptibility testing (AST) in clinical microbiology laboratories because of its simplicity, reproducibility and low cost compared with commercial automated microdilution systems that are usually more rapid but less sensitive for detecting important mechanisms of resistance. Here we measured reading zone diameters around antibiotics in a series of 25 well-characterised Gram-negative bacteria by the disc diffusion technique in real-time using an Advencis Bio-System instrument consisting of a real-time high-resolution video imager in a dedicated incubator. The susceptibility of wild-type Gram-negative bacteria to imipenem, determined by reading the diameter of inhibition, was detectable as early as 3.5h (mean time 3.7 ± 0.45 h), whereas carbapenemase-producing Gram-negative bacteria could be correctly categorised as early as 3h (mean time 4.2 ± 0.8 h) of incubation. Similarly, the characteristic champagne cork aspect of extended-spectrum ß-lactamase (ESBL) could be detected by the system as early as 3.5 h. Moreover, we present here for the first time video movies of the appearance of the diameter of inhibition by disc diffusion in real-time. This preliminary study using a new and innovative technology provides for a renewed interest for microbiologists who wish to continue to use the disc diffusion method as a reference method for AST. New video imaging technology presents a proof of concept that could improve the real-time management of patients with AST within a very rapid turnaround time and can provide a large financial saving for hospitals.