Antimicrobial susceptibility concordance between colonising and bloodstream ESBL and carbapenemase-producing Enterobacterales in allogeneic stem cell transplant recipients: a 10-year retrospective cohort study

Abstract

Background

Surveillance cultures for multidrug-resistant Gram-negative bacteria are increasingly used to guide empirical therapy in allo-HSCT recipients, yet whether the antimicrobial susceptibility profile of the colonising isolate reliably predicts that of the infecting strain remains poorly characterised. The aim of the study was to evaluate antimicrobial susceptibility concordance between colonising and bloodstream extended-spectrum β-lactamase-producing Enterobacterales (ESBL-E) and carbapenem-resistant Enterobacterales (CRE) isolates in allo-HSCT patients, assess surveillance culture performance as BSI predictors, and examine whether concordance varies by timing.

Methods

Retrospective cohort of 872 adult allo-HSCT recipients (2015–2024). Weekly rectal and oropharyngeal surveillance cultures were performed to detect ESBL-E and CRE. Categorical agreement between surveillance and bloodstream isolates was evaluated using the earliest and most recent surveillance cultures as reference. Factors associated with concordance were analysed using logistic regression.

Results

Of 338 colonised patients, 47 developed BSI by the same organism. Rectal cultures showed high sensitivity (CRE 74.1%, ESBL-E 69.0%) with negative predictive value≥95%; oropharyngeal cultures were independently associated with CRE BSI. Full concordance was 33.3% with the earliest isolate and 53.3% with the most recent; specific β-lactam concordance improved from 36.7% to 70.0% (p<0.01). Each additional 7 days between the most recent surveillance culture and BSI onset independently reduced the odds of β-lactam concordance by approximately 15% (aOR 0.85 [0.74–0.98], p<0.05). ESBL-E phenotype was the dominant predictor of overall concordance (aOR 6.23 [1.1–34.1], p<0.05).

Conclusions

The most recent surveillance isolate substantially outperforms the earliest as an empirical reference. The predictive value of surveillance culture antibiograms is time-dependent and mechanism-specific and should be interpreted accordingly.

Introduction

Multidrug-resistant (MDR) Gram-negative bacteria, particularly extended-spectrum β-lactamase-producing Enterobacterales (ESBL-E) and carbapenem-resistant Enterobacterales (CRE), are associated with increased mortality, prolonged hospital stays, and limited therapeutic options [1–4]. Early appropriate antimicrobial therapy is crucial to improving outcomes in bloodstream infections (BSI) caused by these pathogens [1,5–8].

Patients with haematological malignancies, especially those undergoing allogeneic haematopoietic stem cell transplantation (allo-HSCT), are particularly vulnerable to MDR Gram-negative infections due to prolonged neutropenia, mucosal barrier injury, and profound immunosuppression [5,6,9–11]. Gram-negative bacteraemia remains one of the most severe infectious complications in this setting, and surveillance cultures for MDR detection are increasingly used to guide empirical therapy [2,7,12,13].

MDR Gram-negative colonisation is a strong risk factor for bacteraemia caused by the same pathogen [5,7,9,11,13,14,15,17]. However, whether the colonising isolate reliably predicts the antimicrobial susceptibility profile of the infecting strain remains poorly understood [18], a distinction that is clinically critical in patients under sustained antibiotic pressure.

Few studies have specifically evaluated susceptibility concordance between colonising and infecting isolates; those that have done so were conducted predominantly in non-haematological populations, used clinical cultures rather than BSI as the outcome, or focused on a single resistance mechanism [14,15,18,19]. Whether susceptibility concordance diminishes as the interval increases between the surveillance culture and the infectious episode, a question with direct implications for clinical decision-making, has not been examined [20].

We therefore evaluated the concordance of antimicrobial susceptibility profiles between colonising and bloodstream isolates of ESBL-E and CRE in allo-HSCT patients, assessed the performance of surveillance cultures as BSI predictors, and examined whether concordance differed when using the earliest versus the most recent surveillance isolate as the reference.

Material and methods

Study setting

This retrospective cohort study was conducted at the HSCT unit of a tertiary-care university hospital in Valencia, Spain, over a ten-year period from January 2015 to December 2024. All adult patients (≥18 years) who underwent allo-HSCT during the study period were eligible for inclusion. The unit comprises 20 dedicated beds and provides comprehensive care for transplant recipients, including the management of transplant-related infectious and non-infectious complications.

Clinical and microbiological data were retrieved retrospectively from electronic medical records and the institutional microbiology laboratory database. Patients identified as colonised with ESBL-E or CRE were managed under contact precautions in accordance with local infection prevention and control protocols throughout their admission.

Each patient was followed from the date of transplantation until the first occurrence of any of the following: bloodstream infection caused by the colonising organism, death, or completion of one year of post-transplant follow-up. For patients who developed BSI, surveillance cultures obtained from one year prior to infection up to the day before the event were considered for analysis.

No definite temporal clustering or outbreak-related pattern in colonisation incidence was identified over the study period. Routine epidemiological monitoring through weekly culture-based surveillance was maintained throughout the ten-year study period as part of standard institutional practice; no formal outbreak was declared at any point during the study.

Surveillance and microbiology

Surveillance cultures were obtained at admission and weekly thereafter from the perianal region and oropharynx using sterile swabs. Oropharyngeal samples were collected using CliniSwab-LTS^® with Amies medium (APTACA, Italy), whereas perianal samples were collected using CliniSwab-LTS^® with Cary-Blair medium (APTACA, Italy).

Oropharyngeal swabs were inoculated onto Columbia blood agar (bioMérieux, France) and Leeds Acinetobacter medium (MAIM, Spain); colonies with gram-negative bacillary morphology were subsequently subcultured onto CHROMID^® ESBL agar and CHROMID^® CARBA SMART agar (bioMérieux, France) for selective screening. Perianal samples were cultured directly onto Columbia blood agar, CHROMID^® ESBL agar, and CHROMID^® CARBA SMART agar.

Blood cultures were obtained when BSI was clinically suspected, in accordance with the institutional protocol. Blood culture bottles were incubated in the BACT/ALERT VIRTUO^® automated system using BACT/ALERT^® FA PLUS aerobic and BACT/ALERT^® FN PLUS anaerobic bottles (bioMérieux, France). Following detection of positivity and confirmation of Gram-negative bacilli on Gram stain, samples were subcultured onto Columbia blood agar, chocolate PVX agar (bioMérieux, France), and Levine EMB agar (MAIM, Spain). All culture plates were incubated at 37°C for 48 hours, with systematic readings at 24 and 48 hours. Chocolate PVX agar plates were incubated under a 5% CO₂ atmosphere. Bacterial isolates were identified by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS; VITEK^® MS, bioMérieux, France).

Antimicrobial susceptibility testing was performed in accordance with EUCAST guidelines using disk diffusion on Mueller-Hinton agar (bioMérieux, France) with Neo-Sensitabs^™ disks (Rosco Diagnostica, Denmark) [21]. When required, minimum inhibitory concentrations (MICs) were determined by epsilometric testing (Etest^®, bioMérieux, France) or broth microdilution using the MicroScan^® WalkAway system (Beckman Coulter, United States). Colistin susceptibility was determined by broth microdilution using the UMIC^® colistin BMD test (Bruker, United States).

ESBL production was defined in accordance with EUCAST criteria. Isolates showing reduced susceptibility to third-generation cephalosporins (ceftriaxone and/or ceftazidime) were screened for ESBL production, with phenotypic confirmation established by demonstrating clavulanate synergy (ghost-zone effect) between amoxicillin-clavulanate and ceftriaxone and/or ceftazidime.

Isolates showing reduced susceptibility to at least one carbapenem (ertapenem or meropenem) were screened for carbapenemase production using inhibitor-based disk diffusion assays with meropenem alone and in combination with phenylboronic acid and dipicolinic acid, enabling phenotypic differentiation of class A serine carbapenemases and metallo-β-lactamases, respectively. Temocillin disks were additionally used as a phenotypic marker suggestive of OXA-48-like carbapenemases. Suspected CRE isolates underwent further characterization by immunochromatographic testing (O.K.N.V.I. RESIST-5, Coris BioConcept, Belgium) and molecular assays when indicated.

Isolates in which both an ESBL and a carbapenemase were detected were classified once, as CRE, both for the description of colonisation categories and for the ESBL-E versus CRE comparison in the regression models. This classification applies to individual isolates. Co-colonisation refers instead to patients from whom a CRE isolate and a separate ESBL-E isolate were both recovered and is therefore independent of whether either isolate co-produced more than one enzyme.

Definitions

Colonisation was defined as the isolation of ESBL-E or CRE from surveillance cultures in the absence of clinical signs, symptoms, or suspicion of active infection. All surveillance cultures obtained prior to the onset of BSI were included in the analysis. Following the initial positive surveillance culture, antimicrobial susceptibility testing was repeated on all subsequent surveillance cultures from which the same organism was recovered, and this process continued until bacteraemia onset.

Patients who had one or more negative surveillance cultures intercalated between positive ones were maintained under contact precautions and managed as colonised throughout, consistent with current evidence demonstrating that MDR Gram-negative colonisation may persist for prolonged periods despite transient negative cultures.

BSI was defined as the isolation of ESBL-E or CRE from at least one blood culture, regardless of the presence or absence of clinical symptoms. Only the first blood culture isolate per bacterial species and resistance mechanism was included.

Microbiological concordance between colonising and infecting isolates was assessed at two levels: bacterial species and resistance phenotype. Full concordance was defined as the recovery of the same bacterial species with an identical categorical susceptibility profile across every antimicrobial agent tested in both the surveillance and the bloodstream isolate; a single categorical discrepancy in any agent classified the pair as discordant. Additionally, β-lactam concordance was evaluated by restricting the comparison to β-lactam agents only (amoxicillin-clavulanate, cephalosporins, aztreonam, piperacillin-tazobactam, and carbapenems), thereby excluding non-β-lactam antimicrobials from the profile comparison. For CRE isolates, concordance was further assessed by carbapenemase class as determined by immunochromatographic testing, when results were available for both surveillance and bloodstream isolates. Concordance was primarily evaluated between the bloodstream isolate and the most recent surveillance isolate obtained prior to the onset of infection. Additional analyses examined concordance with the earliest available surveillance isolate, allowing assessment of whether temporal distance from the infectious episode influenced the degree of agreement. Agent-level comparisons were restricted to antimicrobials tested consistently across the study period. Amoxicillin was excluded because resistance is intrinsic to both phenotypes studied, so agreement would be trivial by construction; agents tested only sporadically, in fewer than five pairs, were excluded because the resulting estimates would be uninterpretable. Excluded comparisons are not reflected in the totals of Supplementary Tables S2 and S3.

Concordance was assessed as a microbiological property of the isolate pair and not as a proxy for treatment appropriateness. Full concordance required recovery of the same species with the same resistance phenotype, so a surveillance isolate producing an ESBL followed by a bloodstream isolate producing a carbapenemase was classified as discordant irrespective of the categorical result obtained for any individual agent. β-lactam concordance is a distinct and deliberately phenotypic metric that quantifies whether the categorical result of the surveillance antibiogram predicts that of the bloodstream antibiogram for β-lactam agents; it does not imply that an agent reported as susceptible would be an appropriate therapeutic choice. This distinction is material for OXA-48-producing isolates, for which current guidance does not recommend carbapenems as preferred therapy despite in vitro susceptibility [1].

Discordance in antimicrobial susceptibility between colonising and infecting isolates was further characterised using three categories adapted from susceptibility testing methodology. A susceptibility discrepancy was defined as a false-susceptible result in the surveillance isolate when the corresponding bloodstream isolate was resistant. A resistance discrepancy was defined as a false-resistant result in the surveillance isolate when the bloodstream isolate was susceptible. A bidirectional discrepancy was defined as any discrepancy in susceptibility category that did not meet the criteria for either of the above, typically reflecting intermediate classification differences or borderline susceptibility results.

Ethics

The study was approved by the ethics committee of the Hospital Universitari i Politècnic La Fe (Valencia, Spain) (approval number: 2024-0635-1).

Statistical analysis

Continuous variables were summarised as medians with interquartile ranges (IQR) and categorical variables as absolute frequencies and percentages. The performance of surveillance cultures as predictors of BSI was assessed by calculating sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic odds ratios (DOR) with 95% confidence intervals (CI) for each culture site and organism. These analyses were restricted to patients with at least one surveillance culture obtained within the 7 days preceding BSI onset.

Raw contingency table data (true positives [TP], false positives [FP], true negatives [TN], and false negatives [FN]) are reported alongside all diagnostic accuracy estimates. DOR was calculated as (TP×TN)/(FP×FN) and corresponds to the ratio of the positive to negative likelihood ratio.

Differences in concordance rates between the earliest and most recent surveillance isolate were assessed using McNemar’s test for paired proportions. Agreement was additionally quantified using Cohen’s kappa (κ). Univariable and multivariable mixed-effects logistic regression models were fitted for full and β-lactam-specific concordance, incorporating days-to-infection, empirical antibiotic exposure between the most recent surveillance culture and BSI onset, and ESBL-E bacteraemia as covariates, with patient included as a random effect to account for within-patient correlation. The unit of analysis was the colonisation–BSI pair; each pair contributed one observation per model. Results are expressed as OR or adjusted OR (aOR); for the days-to-infection variable specifically, estimates are reported per 7-day increment, matching the weekly interval between surveillance cultures. Categorical time bands were not used, since the resulting strata would have been too small to interpret.

Paired analyses comparing the earliest with the most recent surveillance isolate were restricted to the subset of colonisation–BSI pairs with at least two surveillance isolates of the same organism available before BSI onset, since both references are required for the comparison. Diagnostic accuracy estimates and agreement at the level of individual antimicrobial agents were not subject to this restriction and were calculated across all eligible pairs. The regression models were fitted on all 47 colonisation–BSI pairs, since concordance with the most recent surveillance isolate can be evaluated whenever at least one surveillance isolate is available. Statistical significance was set at p<0.05 and all analyses were performed using R version 4.5.2.

Results

Study population and surveillance culture findings

Between January 2015 and December 2024, 872 adult patients undergoing allo-HSCT were screened for ESBL-E and/or CRE colonisation through systematic surveillance cultures. The median age was 51 years (IQR 38–59), and 501 (57.5%) were male.

A total of 20,984 surveillance cultures were performed over the study period (10,652 rectal and 10,332 oropharyngeal swabs), with a median of 20 cultures per patient (IQR 12–31). Of the 872 patients screened, 338 (38.8%) were colonised by at least one target organism: 122 (14.0%) by CRE only, 134 (15.3%) by ESBL-E only, and 82 (9.4%) by both. Counted by organism rather than by patient, these figures correspond to 204 patients colonised by CRE (122 plus the 82 co-colonised) and 216 colonised by ESBL-E (134 plus the same 82). Among the 20,984 cultures processed, 900 (4.3%) yielded CRE and 683 (3.3%) ESBL-E, corresponding to 4.4 and 3.2 positive cultures per colonised patient respectively. That CRE accounted for more positive cultures than ESBL-E despite a slightly smaller number of colonised patients should be read with two caveats. First, isolates co-producing an ESBL and a carbapenemase were classified once, as CRE, so a proportion of ESBL-producing isolates is counted within the CRE figure by design. Second, the number of positive cultures per patient depends on how long each patient remained under surveillance and how many samples were obtained, neither of which was held constant across groups. This comparison is therefore descriptive and was not interpreted as a measure of carriage duration. OXA-48-type carbapenemases predominated among CRE isolates, followed by VIM, KPC and NDM; the full distribution is shown in Figure 1.

Figure 1. Flow diagram of patient selection and structure of the paired isolate analysis, showing the derivation of the 47 colonisation–bloodstream infection pairs from 872 allo-HSCT recipients and the distribution of resistance mechanisms among colonising isolates.

Allo-HSCT: allogeneic haematopoietic stem cell transplantation; BSI: bloodstream infection; CRE: carbapenem-resistant Enterobacterales; ESBL-E: extended-spectrum β-lactamase–producing Enterobacterales; OXA-48: oxacillinase-48-type carbapenemase; VIM: Verona integron-encoded metallo-β-lactamase; KPC: Klebsiella pneumoniae carbapenemase; NDM: New Delhi metallo-β-lactamase.

*Only CRE and ESBL-E were considered for BSI analyses.

The median time from transplantation to first documented colonisation was 33 days (IQR 12–92), and the median time from first colonisation to BSI onset was 6 days (IQR 3–20).

Performance of surveillance cultures as predictors of BSI caused by ESBL-E or CRE

Rectal cultures showed high sensitivity for CRE (74.1%; DOR 11.2, [4.7–27.0]; p<0.01) and for ESBL-E (69.0%; DOR 8.0, [3.6–17.9]; p<0.01). The addition of oropharyngeal cultures did not change the sensitivity of rectal screening for either organism. Regarding oropharyngeal cultures specifically, a significant association with subsequent BSI was observed for CRE (DOR 10.0, [4.1–24.4]; p<0.01), but not for ESBL-E (DOR 1.3, [0.2–10.0]; p=0.81). NPV was consistently high across both pathogens and sampling sites (≥95%), reflecting the low baseline incidence of MDR-GNB BSI in this population; this same dynamic explains the low PPV values throughout, and both metrics should be interpreted in the context of local infection rates. Full performance metrics, including raw contingency data and 95% confidence intervals for all estimates, are provided in Table 1.

Table 1.

Predictive performance of active surveillance cultures for subsequent ESBL-E or CRE BSI

	Sensitivity (%)	Specificity(%)	PPV (%)	NPV (%)	LR+	LR−	DOR	95% CI	p-value
Complete surveillance for CRE+ESBL-E	75.0	65.4	13.2	97.4	2.2	0.4	5.7	3.0-10.6	<0.01
Complete surveillance for CRE	74.1	79.0	10.2	99.0	3.5	0.3	10.8	4.5-25.9	<0.01
Complete surveillance for ESBL-E	69.0	77.8	9.8	98.6	3.1	0.4	7.8	3.5-17.4	<0.01
Rectal for CRE+ESBL-E	66.7	66.0	15.0	95.6	2.0	0.5	3.9	2.3-6.5	<0.01
Rectal for CRE	74.1	79.7	10.5	99.0	3.7	0.3	11.2	4.7-27.0	<0.01
Rectal for ESBL-E	69.0	78.3	10.0	98.6	3.2	0.4	8.0	3.6-17.9	<0.01
Oropharyngeal for ESBL-E+CRE	16.4	93.5	14.8	94.2	2.5	0.9	2.8	1.3-6.0	<0.01
Oropharyngeal for ESBL-E	3.6	97.2	4.2	96.8	1.3	1.0	1.3	0.2-10.0	0.81
Oropharyngeal for CRE	29.6	95.9	19.0	97.7	7.3	0.7	10.0	4.1-24.4	<0.01

PPV: positive predictive value; NPV: negative predictive value; DOR: diagnostic odds ratio; CI: confidence interval; LR+: positive likelihood ratio; LR−: negative likelihood ratio; CRE: carbapenem-resistant Enterobacterales; ESBL-E: extended-spectrum β-lactamase–producing Enterobacterales.

Concordance between surveillance and bloodstream isolates

Among the 47 colonisation–infection pairs, 44 received at least one antimicrobial during the surveillance period; the agents most frequently administered between the most recent surveillance culture and BSI onset were trimethoprim-sulfamethoxazole (n=40), meropenem (n=25), amikacin (n=20) and piperacillin-tazobactam (n=15). Antimicrobial exposure was not examined as an exposure of interest in its own right but was retained as a potential confounder of the temporal effect, since longer intervals between the most recent surveillance culture and BSI onset are by construction associated with greater cumulative antibiotic exposure. Full details by microorganism and resistance mechanism are provided in Supplementary Table S1.

Of the 47 colonisation–BSI pairs, 30 (12 ESBL-E and 18 CRE) had at least two surveillance isolates available before BSI onset and were therefore eligible for the paired comparison between the earliest and the most recent isolate. The concordance proportions reported refer to that subset, whereas the agreement figures at the level of individual antimicrobial agents reported below were computed across all 47 pairs. Categorical agreement between surveillance and bloodstream isolates was assessed comparing both the earliest and the most recent surveillance isolate against the bloodstream isolate (Supplementary Tables S2 and S3). When the earliest surveillance isolate was compared against the bloodstream isolate, full concordance was 33.3% [19.2%–51.2%]; this increased to 53.3% [36.1%–69.8%] when the most recent surveillance isolate was used instead (p=0.07). Restricting the analysis to β-lactam agents, concordance between the earliest surveillance isolate and the bloodstream isolate was 36.7% [21.9%–54.5%], compared to 70.0% [52.1%–83.3%] when using the most recent isolate (p<0.01).

Stratification by resistance mechanism revealed that full concordance between the earliest surveillance isolate and the bloodstream isolate among ESBL-E pairs was 58.3% [32.0%–80.7%], increasing to 75.0% [46.8%–91.1%] with the most recent isolate (p=0.50). Among CRE pairs, full concordance was 16.7% [5.8%–39.2%] with the earliest isolate versus 38.9% [20.3%–61.4%] with the most recent (p=0.22); in terms of β-lactams, concordance was 22.2% [9.0%–45.2%] with the earliest isolate and 66.7% [43.7%–83.7%] with the most recent (p<0.05).

At the individual antimicrobial agent level, overall categorical agreement across the comparisons reported was 90.1% across 686 comparisons with the earliest isolate and 92.9% across 675 comparisons with the most recent isolate (Supplementary Tables S2 and S3). Among β-lactams, carbapenems showed the lowest concordance: meropenem agreement was 74.3% with the earliest isolate and 78.1% with the most recent; imipenem agreement was 75.0% and 85.7%, respectively. Because the earliest and the most recent isolate coincide in the 17 pairs with a single surveillance culture available, those comparisons contribute identically to both tables; the difference in agent-level agreement between them therefore understates the difference observed in the paired analysis restricted to the 30 pairs with two distinct isolates.

These two sets of figures answer different questions and are not in conflict. Full concordance is an all-or-nothing property of the isolate pair, so a single categorical discrepancy in any agent tested classifies the whole pair as discordant, and the metric is therefore governed by the number of agents compared and by the least stable among them. Agreement at the level of individual agents instead averages across all pairwise comparisons, most of which involve agents whose category is stable in both isolates. Per-agent agreement of 92.9% across 675 comparisons is consequently compatible with only 53.3% of pairs being fully concordant because the residual discrepancies are distributed across pairs rather than concentrated in a few of them.

Cohen’s kappa followed the same pattern. Agreement between the earliest and the most recent surveillance isolate, assessed in the 30 pairs in which the two were distinct, was κ=0.85 for full profiles and κ=0.83 for β-lactam agents. Agreement with the bloodstream isolate was consistently higher when the most recent surveillance isolate was used as reference (κ=0.85 vs. 0.78 for full profiles; κ=0.84 vs. 0.79 for β-lactams).

Influence of the interval between surveillance and bloodstream cultures on concordance

Discordant episodes had a longer interval between the most recent surveillance culture and BSI onset, for both overall (13 days [IQR 6–63.8] vs. 7 days [IQR 3–19.0]) and β-lactam concordance (19.5 days [IQR 6–86.5] vs. 7.0 days [IQR 3–18.0]).

Univariable and multivariable model results are summarised in Table 2. In the univariable analysis, each additional 7 days between the most recent surveillance culture and the onset of BSI was associated with a trend towards lower odds of overall concordance (OR 0.87 per 7 days, [0.76–1.01], p=0.08) and with a statistically significant reduction in the odds of β-lactam concordance (OR 0.87 per 7 days, [0.77–0.98], p<0.05).

Table 2.

Univariable and multivariable mixed-effects logistic regression models for full and β-lactam concordance between surveillance cultures and bloodstream infections

Variable	OR [95% CI]	p-value	Adjusted OR [95% CI]	p-value
Full concordance
Days to infection per 7 days	0.87 [0.76-1.01]	0.08	0.87 [0.74-1.01]	0.07
Empirical antibiotic exposure between surveillance culture and BSI	0.19 [0.03-1.12]	0.07	0.47 [0.10-2.24]	0.35
ESBL-E bacteraemia	5.91 [1.15-30.13]	<0.05	6.23 [1.14-34.12]	<0.05
β-lactam concordance
Days to infection per 7 days	0.87 [0.77-0.98]	<0.05	0.85 [0.74-0.98]	<0.05
Empirical antibiotic exposure between surveillance culture and BSI	0.35 [0.10-1.21]	0.09	0.77 [0.22-2.71]	0.68
ESBL-E bacteraemia	2.24 [0.77-6.54]	0.13	2.70 [0.81-8.98]	0.11

OR: odds ratio; 95% CI: 95% confidence interval; ESBL-E: extended-spectrum β-lactamase–producing Enterobacterales.

ESBL-E bacteraemia is compared against CRE bacteraemia as the reference category; isolates co-producing an ESBL and a carbapenemase are included in the CRE group. Days to infection are modelled as a continuous variable and reported per 7-day increment.

In the multivariable model for overall concordance, ESBL-E bacteraemia was independently associated with a higher probability of concordance (aOR 6.23, [1.14–34.12], p<0.05), while days-to-infection showed a consistent trend without reaching statistical significance (aOR 0.87 per 7 days, [0.74–1.01], p=0.07). Empirical antibiotic exposure between the surveillance culture and BSI onset was not independently associated with overall concordance (aOR 0.47, [0.10–2.24], p=0.35). Exploratory analyses examining piperacillin-tazobactam and meropenem individually showed that neither was independently associated with full or β-lactam concordance, though a non-significant trend towards lower β-lactam concordance was observed with piperacillin-tazobactam.

The multivariable model for β-lactam concordance revealed that temporal distance was an independent predictor (aOR 0.85 per 7 days, [0.74–0.98], p<0.05), whereas ESBL-E bacteraemia did not reach statistical significance (aOR 2.70, [0.81–8.98], p=0.11). Antibiotic exposure between surveillance and BSI was again not independently associated with β-lactam concordance (aOR 0.77, [0.22–2.71], p=0.68).

Discussion

This study provides, to our knowledge, the first quantification of susceptibility concordance dynamics between colonising and bloodstream ESBL-E and CRE isolates in allo-HSCT recipients, demonstrating that β-lactam concordance decreases independently with temporal distance from colonisation to infection and that the most recent surveillance isolate substantially outperforms the earliest as an empirical reference. The predictive performance of rectal surveillance cultures for CRE and ESBL-E BSI in our cohort is broadly consistent with published series in haematological patients [2,10,11,14,17].

The low PPV observed throughout our analysis is an expected consequence of the low baseline incidence of MDR-GNB BSI in this population and is a consistent finding across the surveillance culture literature [14,20]. A positive surveillance culture does not equate to a prediction of impending infection; rather, it identifies patients at increased risk who require heightened clinical vigilance. The high NPV, conversely, provides more actionable information at the individual patient level, suggesting that a recently negative surveillance culture may support a lower-intensity empirical approach in selected clinical contexts. These figures must nonetheless be read in the light of the low prevalence of MDR-GNB BSI in our cohort, since NPV rises mechanically as prevalence falls, and the absolute values reported here should not be transferred unchanged to settings with a substantially higher burden of MDR infection [2,14,20]. What is more likely to be transferable is the underlying relationship rather than the number: endogenous BSI requires prior mucosal colonisation by the causative organism, so a truly negative surveillance culture identifies a patient in whom that pathway is improbable, and what differs between epidemiological settings is principally the proportion of patients colonised rather than the biological link between colonisation and subsequent infection [6,9,10,11].

It is important to note that this study was not designed to evaluate the impact of surveillance culture results on empirical prescribing decisions or antimicrobial stewardship practice. Questions such as whether a negative surveillance culture should prompt de-escalation from agents like ceftazidime-avibactam, or whether a positive CRE surveillance culture should trigger their empirical use, involve clinical, pharmacological, and stewardship considerations beyond the scope of this analysis.

The more clinically consequential question is whether the colonising isolate reliably predicts how the infecting strain will behave at the antibiotic level [15,19,20,22]. Most published studies used composite clinical culture endpoints rather than restricting the outcome to BSI [14,15,18], limiting applicability to haematological patients in whom bacteraemia carries distinct prognostic implications. By restricting the outcome to BSI, our study addresses concordance where empirical antibiotic decisions carry the highest clinical stakes [16].

A further distinctive feature is that susceptibility testing was performed systematically on all surveillance isolates throughout the colonisation period. Combined with a ten-year study period, this allowed capture of phenotypic evolution across time. Full concordance proved difficult to achieve, particularly with the earliest surveillance culture, and improved meaningfully when the most recent isolate was used, a finding statistically significant for β-lactam agents.

The temporal dimension of concordance loss is the central contribution of our work. Prior studies have shown that predictive value of resistant cultures declines with time and susceptibility changes accumulate under clinical conditions [15,22–24]. However, none of these analyses specifically addressed the colonisation-to-BSI axis, restricted their scope to ESBL-E or CRE, or were conducted in allo-HSCT patients. Our study is, to our knowledge, the first to quantify this phenomenon in this specific context, identifying a statistically significant and independent reduction in the odds of β-lactam concordance of approximately 15% for every additional 7 days between the most recent surveillance culture and BSI onset. This finding should be regarded as hypothesis-generating: the rarity of MDR-GNB BSI, even across a decade-long cohort, precludes formal clinical recommendations.

The asymmetry between ESBL-E and CRE in multivariable models reflects distinct biological behaviours. ESBL phenotype was the dominant predictor of overall concordance, consistent with the relatively stable phenotypic signature of ESBL expression, a plasmid-encoded trait conferring a reproducible resistance pattern [22,25]. For CRE, the low carbapenem concordance and high error rates for meropenem and imipenem are consistent with the known phenotypic characteristics of OXA-48-type carbapenemases, which predominated in our cohort. OXA-48-type carbapenemases confer variable degrees of carbapenem resistance depending on context [26–28], and susceptibility profiles are not reliably uniform across successive cultures. Piperacillin-tazobactam concordance, while more stable in ESBL-E pairs, was also imperfect and should not be assumed reliable beyond the short term [1,29].

It should be noted that part of the discordance observed, particularly for carbapenems, may reflect intrinsic variability in susceptibility categorisation near EUCAST clinical breakpoints rather than true phenotypic differences between isolates [21]. OXA-48-type carbapenemases are known to confer MIC values that cluster near EUCAST breakpoint thresholds, making susceptibility categorisation sensitive to small technical variations between testing occasions. This caveat is particularly relevant when interpreting discrepancies for meropenem and imipenem [26–28].

Several limitations of this study must be acknowledged. The retrospective, single-centre design and the specific carbapenemase ecology of our setting limit the generalisability of our conclusions. The absence of molecular typing means we cannot definitively distinguish phenotypic evolution of the original colonising clone from acquisition of a genetically distinct strain. Empirical antibiotic therapy was not independently associated with concordance, though the short interval between the most recent surveillance culture and BSI onset in many cases may have limited the window for antibiotic pressure to measurably alter the colonising phenotype.

Antimicrobial exposure was characterised as a binary variable within defined time intervals, without accounting for indication, duration, dose, or cumulative exposure. A more granular characterisation would have allowed a more rigorous assessment of the relationship between antibiotic pressure and phenotypic concordance. The null association observed for this variable should therefore not be read as evidence that antibiotic pressure does not alter the colonising phenotype; it more probably reflects the limited resolution of a binary exposure measure.

Additionally, susceptibility categorisation near EUCAST breakpoints is subject to inter-occasion variability inherent to disk diffusion methodology, which may have contributed to apparent discordance for agents such as meropenem and imipenem without reflecting true phenotypic change.

Finally, whether discordant episodes translated into inadequate empirical therapy or worse clinical outcomes was beyond the scope of this analysis and remains an important question for prospective investigation. Nonetheless, the size of the cohort and the ten-year study period provide a degree of ecological diversity that shorter studies cannot offer, and we propose these findings as a conceptual framework that quantifies, for the first time in allo-HSCT, the mechanism-specific boundaries of antibiogram concordance between colonisation and BSI and the rate at which that concordance diminishes with temporal distance.

Supplementary material

The supplementary material is available on the website of the Revista Española de Quimioterapia.

Funding

No funding was received for the research, authorship and/or publication of this article.

Conflicts of interest

PF has received financial compensation for consultancy and speaking engagements from bioMérieux Argentina. PSU has received financial compensation for speaking engagements from DiaSorin. JPG has received financial compensation for consultancy, speaking engagements, and research activities from Pfizer, Gilead, and MSD. AVM has received financial compensation for speaking engagements from Werfen. All other authors declare no financial or non-financial conflicts of interest relevant to this work.