Optimizing species identification after antimicrobial pretreatment in Enterobacterales: a prospective cohort study

Abstract

Background

Initiation of antimicrobial treatment before specimen collection reduces bacterial recovery rate for cultures and alters Gram-stain morphology of the microbe, but very little is known about the effects to the subsequent species identification once the culture becomes positive. This prospective cohort study assessed the impact of antimicrobial pretreatment effect on species identification of the closely related members of Enterobacterales, using conventional and automated phenotypic (Vitek 2) methods, and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) in terms of overall accuracy and cost-effectiveness.

Results

A total of 240 Enterobacterales isolates, with visible antimicrobial pretreatment effect (hereafter, the pretreatment effect), were included in the study. On primary culture, the MALDI-TOF MS with a standard cutoff had an inferior success rate (75.0% vs. 97.3%; P < 0.001). Reducing the MALDI-TOF MS cutoff to 1.8 significantly increases its success rate (93.8%) without compromising accuracy (99.6%, vs. benchmark of 100.0%, P = 1). The Vitek 2 automated system maintained high accuracy comparable to the benchmark. Isolates with pretreatment effects had lower prevalence of extended-spectrum β-lactamase-producing Escherichia coli (25.2% vs. 43.4%; P < 0.001) and Klebsiella pneumoniae (9.5% vs. 30.6%; P < 0.001), and carbapenem-resistant E. coli (0.7% vs. 4.3%; P = 0.027) and K. pneumoniae (3.2% vs. 24.5%; P < 0.001) than the general population. Based on a cost-utility analysis, the conventional phenotypic method was the least cost-effective, while the optimized MALDI-TOF MS method was the most cost-effective.

Conclusions

MALDI-TOF MS with reduced cutoffs and the automated phenotypic assay are the most cost-effective and accurate methods for isolates with antimicrobial pretreatment effect. Additionally, this effect may serve as an early indicator of lower antimicrobial resistance.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-026-04954-4.

Background

Culture remains the gold standard for diagnosing bacterial infections [1], confirming organism viability and enabling phenotypic antimicrobial susceptibility testing (AST), another gold standard for therapeutic guidance [2]. While specimen collection before antimicrobial therapy is ideal, immediate empiric treatment for serious infections often necessitates collection post-treatment [3]. Additionally, tertiary hospitals that receive patient referrals also frequently encounter specimens from patients who have already received antimicrobial treatment.

Antimicrobial pretreatment significantly reduces culture sensitivity [4], and induces aberrant microscopic morphologies [5]. Even when bacterial growth occurs, pretreatment may alter phenotypic traits, affecting conventional identification methods [6, 7], particularly among closely related species within the Order Enterobacterales [8, 9], where a single aberrant result can lead to major misidentification [10]. Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS), the current standard, may also be influenced because the technique depends on the organism’s proteomic profile [11], which can be altered as a part of the stress response to the antimicrobial exposure [7]. It is thus hypothesized that the accuracy of species identification methods may be lowered for bacteria isolated from patients who were pretreated with antimicrobials.

In routine laboratory workflow, laboratory technologists usually do not access patients’ clinical information and are unaware of any prior treatment. Occasionally, they may observe what is termed the “antimicrobial pretreatment effect” on the primary culture plate: there is little or no growth at the primary inoculum site (where the concentration of antimicrobial agents is presumed highest) and normal growth on subsequent streaks (Fig. 1). To minimize the influence of this effect, technologists typically select colonies farthest from the inoculum for further testing; however, aberrant results may still occur.

Fig. 1.

The antimicrobial effect on the primary blood agar plate from a culture of a bile specimen. There was no bacterial growth on the primary inoculum (arrow heads) where there was the highest concentration of the antimicrobial agent, while there was normal growth on the streak lines up to the last quadrant (numerous organisms). Routinely, the farthest colonies from the primary inoculum (arrow) were selected for further phenotypic tests in the attempt to minimize the antimicrobial effects on the subsequent media. Even then, this organism ended up having aberrant TSI (1) and LDC (2) results. The organism was later identified as Escherichia coli. Phenotypic tubes from left to right: indole (also served as a suspension for inoculation of other tubes), triple sugar iron (TSI), motility-indole-lysine decarboxylase/deaminase, Voges-Proskauer, malonate, mannitol, urease, citrate

Despite anecdotal recognition of this effect in practice, quantitative data on how antimicrobial pretreatment alters identification accuracy remain scarce. This study aims to quantify the impact of antimicrobial pretreatment on species identification using both conventional and automated phenotypic methods, and MALDI-TOF MS [10, 11]. The investigation focused on Enterobacterales, one of the most frequently isolated pathogens in bacteriology laboratories [12], and among the most misidentified by phenotypic testing. Using the cost-utility analysis incorporating the accuracy of each method and the established data on direct and indirect costs, this study shows that MALDI-TOF MS with a reduced cutoff is the most cost-effective approach for species identification in this cohort, despite the high direct cost.

Materials and methods

Study population

This prospective study was conducted at the Department of Microbiology, Faculty of Medicine Siriraj Hospital, Mahidol University, Thailand, between May and November 2025. All specimens were processed according to standard protocols. Briefly, all specimens were directly inoculated onto a 5% sheep blood agar (Biomedia, Thailand) and a MacConkey agar (Becton Dickinson, USA). Respiratory and miscellaneous specimens were also inoculated onto a chocolate agar (Biomedia, Thailand). All agar plates were incubated at 35 ± 2 °C in ambient atmosphere overnight before visualization. All plates were inspected for the presence of an antimicrobial pretreatment effect, defined as at least an 80% reduction in growth at the primary inoculation site (top of the T-streak for urine culture) compared to the distal quadrants (as demonstrated in Fig. 1). The presence of this inhibition is initially flagged by bench technologists and subsequently independently verified by a senior microbiologist prior to study inclusion. All oxidase-negative Gram-negative bacilli with colonies resembling Enterobacterales were provisionally categorized as “possible Enterobacterales organisms” and included in the study. Once the final identification was confirmed, all non-Enterobacterales organisms were excluded. The results are reported following the Standards for Reporting of Diagnostic Accuracy (STARD) guidelines [13]. There was no missing data in this dataset.

Sample size determination

Following Clark et al.‘s recommendation [14], the sample size of > 200 isolates was targeted; formal power calculations were not performed.

Species identification

The colonies farthest from the inoculation site (Fig. 1) on the sheep blood agar (SBA) were selected for identification using three methods: conventional phenotypic method, automated phenotypic method using the Vitek 2 platform (bioMérieux, France), and MALDI-TOF MS using the Biotyper Sirius Microflex LT/5H (Bruker Daltonics, Bremen, Germany). One colony was further inoculated onto a new SBA, incubated aerobically at 35 ± 2 °C overnight, and the subcultured colonies underwent the three identification methods. A summary of the workflow is shown in Fig. S1.

For the conventional phenotypic method, a 0.5 McFarland suspension was prepared from colonies on the sheep blood agar and inoculated into the following tubes: triple sugar iron [TSI], indole, motility, lysine decarboxylase [LDC], lysine deaminase [LDM], Voges-Proskauer [VP], malonate, citrate, urease, and mannitol. Additionally, lactose fermentation was recorded from the bacterial growth on the MacConkey agar plate. The local database was derived from Murray et al. [10]. This reference was chosen because it provides the most comprehensive compilation of phenotypic tests for the Order Enterobacterales. The database was also supplemented with additional data, as well as updates on the nomenclatures, from the latest update by Carroll et al. [15]. Data for Acinetobacter baumannii complex and Stenotrophomonas maltophilia were also included to help exclude these common oxidase-negative non-Enterobacterales organisms [15]. The heatmap showing phenotypic similarity among Enterobacterales is shown in Fig. S2.

The probability of each candidate species was predicted using a naïve Bayes model. Each phenotypic test result (0 = negative, 1 = positive) was compared with the expected reaction frequencies [15]. Likelihoods were computed as the product of test-specific probabilities, adjusted with ε = 10⁻⁶ to avoid log(0) underflow. For TSI reactions (K/A vs. A/A), only the higher-probability phenotype contributed to the likelihood. To account for the baseline prevalence of each species, the naïve Bayes model was extended to include empirical priors derived from the institutional microbiology database. Two analyses were performed: one using uniform priors (all species equally likely) and another using empirical priors proportional to the total number of routine clinical isolates of each Enterobacterales species recovered during the study period, irrespective of pretreatment effects. Posterior probabilities were normalized across all candidate species, and the species with the highest posterior probability was reported as the predicted identification. An example spreadsheet for this calculation is provided (see Data availability). The calculation was also validated using a Python script.

For the automated phenotypic method, a separate 0.5 McFarland suspension was prepared from colonies, inoculated, and incubated on the Vitek 2 platform according to the manufacturer’s protocol. For the MALDI-TOF MS method, a single colony was applied to the target slide, 1 µl of 70% formic acid (Millipore, Burlington, MA, USA; IVD grade) was added, followed by 1 µl of α-cyano-4-hydroxycinnamic acid matrix (Bruker Daltonics, Bremen, Germany; IVD grade), and analyzed by MALDI-TOF MS using the Biotyper Sirius Microflex LT/5H (Bruker Daltonics, Bremen, Germany) according to the manufacturer’s protocol.

The reference identification for each isolate was derived from the consensus or the majority results of the three methods (prior-based conventional phenotypic tests, Vitek 2, and MALDI-TOF MS) from the subcultured colonies. If no majority was achieved, the subcultured MALDI-TOF MS result would serve as a reference identification.

Cutoff values for species-level identification

All cutoff values for species-level identification were pre-specified. For the MALDI-TOF MS method, the log score of ≥ 2.0, indicating the reliable species-level identification according to the manufacturer’s protocol, was selected. For the Vitek 2 method, a percent confidence of ≥ 85% was considered sufficient for species-level identification based on a previous study [16]. This cutoff was also applied to the conventional phenotypic method.

Calibration of cutoff values for species-level identification

To account for the potential log score degradation from antimicrobial stress, a post-hoc exploratory analysis was conducted to evaluate the performance of alternative cutoffs ranging from 1.7 to 2.0 with a 0.1 increment being evaluated. At each candidate’s cutoff, the success rate of identification—the proportion of isolates correctly identified at the species level—was calculated. A similar analysis was also performed on the conventional phenotypic tests, with a reduction of the cutoff to ≥ 80%.

Benchmarking the accuracy of the identification methods

To verify baseline performance of the three identification methods, concurrent 150 Enterobacterales isolates and 50 oxidase-negative non-Enterobacterales (primarily A. baumannii and S. maltophilia) without pretreatment effects isolated between October and November 2025 were randomly selected from the daily workflow and were tested using the same workflow. Species distribution of this benchmark group is available in Table S1. The test parameters were compared, and the results served as benchmarks for subsequent comparisons.

Comparison of antimicrobial-resistant organisms

All isolates with AST results were classified based on their drug-resistant patterns, focusing on organisms listed by the CDC [17]. An isolate was presumed to produce extended-spectrum β-lactamases (ESBLs) if it was (i) resistant to at least one of the third-generation cephalosporins while susceptible to cefoxitin or (ii) resistant to cefepime regardless of cefoxitin susceptibility [18]. The isolate was confirmed to be carbapenem-resistant Enterobacterales (CRE) if it was nonsusceptible to at least one carbapenem [18]. The prevalence of resistance was compared to the overall prevalence at the study site during the study period (institutional surveillance data, unpublished) for organisms with at least 30 isolates; the minimum number recommended by the CLSI [2].

Exploring the impact of misidentification

To evaluate the real-world impact of misidentification, a cost-utility analysis was performed. First, each misidentification was classified into three categories, in keeping with the categories of AST errors defined by Clark et al. [14]. Next, the direct cost per unit for each method and the loss of quality-adjusted life year (QALY) were calculated based on published data [19–25]. Finally, a Bayesian model was run based on calculated parameters. The details of the parameters used in the model, and the associated explanation, are available in Text S1 (Tables S3-S6, Fig. S3).

Statistical analysis

All statistical analyses were performed using Python 3.13.9. Raw data were compiled, cleaned, and summarized using Microsoft Excel (Microsoft Corporation, Redmond, WA, USA) before statistical analysis. The primary endpoint of this study was the accuracy of the three identification methods, which was compared using a pairwise McNemar test with Holm-Bonferroni correction. The secondary endpoint is the cost-utility of these methods in a clinical setting, which was calculated as stated above and detailed in Text S1. This study also includes an exploratory analysis to optimize the MALDI-TOF MS cutoffs to maximize sensitivity. The distribution of the log score was visualized in a box plot, and a pairwise Mann-Whitney U test with Holm-Bonferroni correction was used to compare the differences between cohorts. A receiver operating characteristic curve analysis was also performed to determine the optimal cutoff. For all statistical analyses, statistical significance was achieved at P < 0.05.

Results

Study population

Of 40,743 non-blood specimens, 240 Enterobacterales isolates with a visible pretreatment effect on sheep blood agar (Fig. 1) were included. Species distribution was as follows: Escherichia coli (146, 60.8%), Klebsiella pneumoniae (63, 26.3%), Proteus mirabilis (11, 4.6%), Enterobacter cloacae complex (8, 3.3%), and others (12, 5.0%). The definitive identification of all 240 isolates was established by the majority consensus of the three methods. Consequently, the pre-specified tie-breaker protocol, which relied solely on the MALDI-TOF MS result from the subcultured colony, was not required for this cohort. The distribution of specimen types in the study cohort was compared to the overall positive specimen types at the study site (Table S2). Urine specimens were slightly over-represented compared to the overall population (61.7% vs. 49.7%), likely reflecting the predominance of E. coli in the study cohort. Although the difference reached statistical significance (χ² = 19.07, P < 0.001), the effect size was negligible (Cramér’s V = 0.035), indicating that the specimen type was not a strong determinant of study inclusion.

Benchmarking the accuracy of the identification methods

The benchmark analysis confirmed the baseline performance of each method (Table S7). All methods had relatively high specificity (> 96.0%, P = 0.109). However, the conventional phenotypic method without priors demonstrated significantly lower sensitivity (84.7%) and accuracy (88.5%) compared to other methods (both P < 0.001) and were excluded from downstream analyses. The other three methods (conventional phenotypic tests with priors, Vitek 2, and MALDI-TOF MS) all had high, comparable sensitivity of 93.3%-99.9% (χ² = 1.78, P = 0.410), and accuracy of 95.0%-99.9% (χ² = 1.77, P = 0.412).

Performance of identification methods on isolates with antimicrobial pretreatment effect

For the MALDI-TOF MS method, there was a significant reduction of the log score in the original culture group (median 2.11 vs. benchmark median 2.25; U=-7.00, P < 0.001, Fig. 2A), resulting in a decrease in the success identification at the standard cutoff of ≥ 2.0 (75.0% vs. 97.3%, P < 0.001). Subculturing the colonies helped reduce the gap (median 2.17; U = 4.13, P < 0.001), but still significantly lower than the benchmark (U=-3.85, P < 0.001). Further inspection of entries with the log score below the cutoff revealed that the majority (96.7%) of isolates have correct species-level identification. Reducing the cutoff to 1.9 and 1.8 improved the success rate to 87.9% and 93.8%, respectively, without compromising accuracy (100.0% and 99.6%, P = 1 vs. benchmark, Fig. 2B). The success rate and the accuracy of the other two methods were not significantly different from the benchmark (Fig. 3).

Fig. 2.

Performance of MALDI-TOF MS in isolates with antimicrobial pretreatment effect. A A box plot comparing the log score between the original and subcultured colonies of the study group and the benchmark group. Asterisks indicate statistical significance at P < 0.05. B ROC curve of the original and subcultured colonies of the study group using different cutoff log scores. The four dots represent the clinically significant cutoffs (1.7, 1.8, 1.9, and 2.0)

Fig. 3.

Success rate and accuracy of each species identification method. A Percent of successful identification (confidence values above the cutoff compared to the total number of isolates) and (B) Percent accuracy (correct identification compared to the total number of successful identifications) are shown for five calculation approaches from three identification methods: conventional phenotypic test without the use of species-specific priors, conventional phenotypic test with the use of species-specific priors, Vitek 2 automated phenotypic assay per manufacturer’s protocol, MALDI-TOF MS with a standard 2.0 cutoff value, and MALDI-TOF MS with reduced 1.9 and 1.8 cutoff values. The error bars indicate a 95% confidence interval calculated using the Wilson Score to account for the values near or at 100%. Asterisks indicate statistical significance at P < 0.05

Exploring the misidentifications by the conventional phenotypic tests

A total of 17 isolates (7.1%, Table S8) were initially misidentified by the conventional phenotypic method. Of these, two-thirds were due to confidence levels falling below the ≥ 85% cutoff; lowering the cutoff to ≥ 80% would have correctly identified two additional isolates, though this change was not statistically significant (χ² = 0.03, P = 0.855). Most misidentifications (n = 11, 64.7%) occurred among members of the KES group (Fig. S2). Subculturing led to minor changes in the phenotypic profile of 7 isolates but did not significantly affect the overall accuracy (Table S9). Similarly, slight changes in the Vitek 2 Bionumber were observed without a significant impact on accuracy (Table S10).

Impact of misidentification through cost-utility analysis

Conventional phenotypic tests had the highest error cost per unit, making it the least cost-effective alternative. Vitek 2 automated phenotypic assay had the highest accuracy, but was associated with a few very major errors, resulting in a high error cost per unit. The MALDI-TOF MS method, especially with the reduced cutoff of 1.8, was slightly less accurate than the automated assays, but was not associated with any very major errors, resulting in the lowest error and overall costs (Fig. 4A and C). Sensitivity analysis performed by varying the error costs (Table S5) did not result in the relative change in the effectiveness of the methods (Fig. 4D).

Fig. 4.

Cost-utility analysis results. A The cost-utility plane showing the expected total cost per unit of test on the X-axis and the accuracy of test on the Y-axis. MALDI-TOF MS with a reduced cutoff value of 1.8 has the lowest cost per unit with comparably high accuracy. B Breakdown of the total cost per unit of test (actual unit cost + estimated cost associated with errors). C Estimated annual budget impact, based on the approximated 15,000 tests per year (internal estimate). The number on the X-axis shows the total cost per 1,000 tests. D Sensitivity analysis was performed by varying the WTP for the QALY loss and the budget range was shown by the error bars

Comparison of antimicrobial resistance prevalence

AST results were available for 143 and 63 E. coli and K. pneumoniae isolates, respectively, larger than the recommended number by the CLSI [2]. The study population had a significantly lower prevalence of ESBL-producing E. coli (25.2% vs. 43.4%; χ² = 17.75, P < 0.001) and K. pneumoniae (9.5% vs. 30.6%; χ² = 12.06, P < 0.001), as well as the prevalence of carbapenem-resistant E. coli (0.7% vs. 4.3%, P = 0.027) and K. pneumoniae (3.2% vs. 24.5%, P < 0.001) than the general population at the study site (Fig. 5).

Fig. 5.

Comparison of antimicrobial resistance between isolates with antimicrobial pretreatment effect and the general population. Abbreviations: EC; Escherichia coli, KP; Klebsiella pneumoniae, ESBL; extended spectrum β-lactamase, CRE; carbapenem-resistant Enterobacterales. The error bars indicate a 95% confidence interval calculated using the Wilson Score to account for the values near or at 0%. Asterisks indicate statistical significance at P < 0.05

Discussion

This study quantifies the impact of the “visible antimicrobial pretreatment effect” on species identification and subsequent costs [1, 15]. The effect was most prominent for the MALDI-TOF MS, likely due to a stress-induced proteomic shift that obscures the proteomic fingerprint without altering it enough to cause misidentification [6, 7]. Consequently, while the success rate at the standard cutoff dropped to 75%, the underlying accuracy of ‘failed’ isolates remained high (96.7%). We recommend lowering the MALDI-TOF cutoff to either ≥ 1.9 or ≥ 1.8 when this effect is observed, which restores the success rate (> 93.0%) without compromising accuracy. Alternatively, Vitek 2 system remained robust, likely due to its extensive panel of > 40 biochemical tests. The conventional method (11 tests) was brittle, with a single aberrant result leading to possible misidentification, especially among closely related groups [8, 9], and should be avoided.

Variations in culture media composition can also alter bacterial phenotypic expression and growth characteristics [26]. To control for this external variable, this study included a concurrent benchmark group of 150 Enterobacterales, mirroring the species distribution of the study group, and 50 non-Enterobacterales cultured on identical media sets. The high accuracy (> 99%) and the normal MALDI-TOF MS log score suggest that the deviation observed in the study group were primarily driven by the antimicrobial stress rather than media formulation or laboratory conditions.

While lowering the cutoff value offers a practical solution for clinical laboratories, a more definite solution would be to develop a specific Mass Spectrum Profile (MSP) library derived from antimicrobial-exposed organisms. Creating such a library could theoretically allow for accurate identification without lowering confidence thresholds. However, generating and validating custom MSP libraries requires significant resources and reference-level expertise [27], which may be beyond the scope of many routine clinical laboratories. Future studies by reference centers could explore this avenue to further optimize identification workflows for pre-treated specimens.

Isolates with pretreatment effect showed significantly lower resistance (ESBL-producers and CRE) than the general hospital population. This likely reflects that such isolates are susceptible to the antimicrobial agents present in the patient, as evidenced by the observed growth inhibition. This observation generates a compelling hypothesis: the “visible pretreatment effect” could serve as a surrogate phenotypic marker for in vivo susceptibility. If validated by prospective outcome studies, reporting this effect might provide clinicians with early assurance that empiric therapy is active, potentially facilitating antimicrobial de-escalation decisions [28].

Misidentification has immediate clinical consequences. With phenotypic AST typically lagging identification by 24 h [29], early treatment decisions rely mainly on species identity [30]. Misidentifying organisms with intrinsic resistance (e.g., AmpC β-lactamase-producing species [31]) or species known to be more resistant (e.g., K. pneumoniae [32]) can lead to inadequate therapy, while false identification of resistance species drives unnecessary broad-spectrum treatment, exposing the patients to their adverse effects.

The cost-utility analysis confirms that, despite low unit costs, the conventional phenotypic method is the least cost-effective due to high error rates. The optimized MALDI-TOF MS method (cutoff ≥ 1.8) minimized these errors and was the most cost-effective strategy in this scenario.

This study has several limitations. First, molecular sequencing (e.g. 16 S rRNA), the theoretical gold standard [33], was not available to resolve discordant results. However, the risk of misclassification was mitigated by the use of a composite reference standard derived from subcultured colonies. Given the established high accuracy (> 98%) of these methods on unstressed cultures (the benchmark group), the majority consensus of these methods provides a robust reference standard in place of the missing molecular gold standard. Second, as a single-center study, the findings regarding organism prevalence and error rates may not be generalizable, and the cost-utility analysis was specific to the study site’s cost structure. Third, the sample size was pragmatic rather than based on a formal power calculation, introducing a potential risk of Type II error for secondary comparisons. Fourth, the reduction of the MALDI-TOF MS log score was proposed based on a post-hoc analysis, and external validation is required before implementing it in the clinical laboratories. Fifth, the primary inclusion criterion – the “visible pretreatment effect” – is a subjective visual assessment by technologists, which may introduce selection bias. Finally, the study did not have access to patient clinical information. Consequently, it was not possible to correlate the observed effect with the actual clinical outcomes. Thus, the cost-utility analysis was not fully validated.

In conclusion, the “visible antimicrobial pretreatment effect” – although subjective – is an important laboratory finding. We recommend that laboratories use this effect to guide technical workflows and consider reporting its presence to clinicians, as it may suggest a susceptible infection. When this effect is observed, conventional phenotypic tests are unreliable and should be avoided. The most cost-effective strategy is to use the MALDI-TOF MS with a validated, reduced cutoff of 1.8 or 1.9 with the Vitek 2 automated assay as a possible alternative.

Brief key points

• Antimicrobial pretreatment can affect the species identification, both by phenotypic and proteomic-based methods.

• The proteomic-based MALDI-TOF MS with a reduced cutoff (either 1.8 or 1.9) is the most cost-effective identification method, although external validation is needed before implementation.

• Organisms with pretreatment effect tend to be less resistant than those without. If validated with clinical studies, this finding may facilitate early antimicrobial optimization.

Authors’ contributions

UK: Conceptualized the study, designed and validated the methods, acquired the funding, contributed to data collection and interpretation, and wrote the manuscript. YS: Contributed to data analysis and visualisation, guarantor of the data, and revised the manuscript. SY, TW, PD, and SP: Contributed to data collection and interpretation, and revised the manuscript. KI: Conceptualized the study, contributed to method validation, performed data analysis and visualisation, supervised the project, guarantor of the data, and wrote the manuscript.

Funding

Open access funding provided by Mahidol University. This research project was supported by the Siriraj Research Development Fund (Managed by Routine to Research: R2R) Grant No. (IO) R016835036, Faculty of Medicine Siriraj Hospital, Mahidol University. The funder had no role in the design and conduct of the study.

Data availability

All the codes used in this study are available on GitHub (https://github.com/korakritimw/Enterobacterales_Identification_Optimization). A ready-to-use Excel spreadsheet for the calculation of the posterior probability for the conventional phenotypic method is available on Figshare (https://doi.org/10.6084/m9.figshare.30471284). Other data related to this study are available from the corresponding author upon a reasonable request.

Declarations

Ethics approval and consent to participate

Ethics approval and consent to participate were waived by the Siriraj Institutional Review Board (SIRB; internal document number 78.071/EC01883). Declaration of Helsinki was adhered throughout the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.