Dynamics of Antimicrobial Resistance and Antimicrobial Consumption in a Secondary-Care Hospital in Serbia, 2019–2022: A Retrospective Observational Study

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Title & Abstract
1- Study design (retrospective observational) is not explicitly stated in the title.
2- The abstract does not clearly specify inclusion/exclusion criteria for isolates or patients.
3- Methods in the abstract insufficiently describe how correlations between antimicrobial use and resistance were assessed.
4- Abstract conclusions imply causality (“pandemic-associated surges… drove resistance”) despite observational design.

Introduction
5- The knowledge gap specific to secondary-care hospitals is not clearly articulated.
6- Rationale for focusing exclusively on E. coli and K. pneumoniae is not sufficiently justified.
7- Study hypotheses are not explicitly stated; aims are descriptive rather than hypothesis-driven.

Methods
8- No clear justification for the chosen study period (2019–2022) beyond pandemic relevance.
9- Changes in hospital structure/workflow during COVID-19 are mentioned later but not prespecified here.
10- Inclusion and exclusion criteria for “clinically significant isolates” are not defined.
11- Criteria for distinguishing infection vs colonization (especially for non-invasive isolates) are unclear.
12- No explicit definition of primary and secondary outcomes.
13- Potential confounders (ward type, ICU admission, COVID-19 status, patient demographics) are not defined or measured.
14- Colistin testing strategy (restricted testing) introduces selection bias but is not addressed methodologically. (No formal discussion of selection bias, information bias, or confounding in the Methods section.)
15- No sample size or power considerations are provided.
16- No adjustment for multiple comparisons despite extensive year-by-year testing.
17- No multivariable analyses to control for confounding factors.
18- Trend analyses are inconsistently specified (chi-square vs correlation vs descriptive).

Results
19- Patient-level characteristics (age, sex, ward, ICU vs non-ICU) are completely absent.
20- Number of isolates is reported, but number of patients contributing isolates is unclear.
21- Flow diagram of isolates through the study (screened, included, excluded) is not shown.
22- Resistance rates are reported without confidence intervals.
23- Invasive vs non-invasive isolate analyses are not consistently stratified across outcomes.
24- Results repeatedly imply temporal “effects” without accounting for confounding.
25- Correlation results are reported without effect sizes or correlation coefficients.
26- Subgroup analyses (e.g., invasive vs non-invasive) are limited and not prespecified.
27- DU75% analysis lacks justification for its interpretive relevance to resistance outcomes.

Discussion
28- Key findings are selectively emphasized (K. pneumoniae) without balanced discussion of null findings (E. coli).
27- Overinterpretation of associations as causal relationships persists.
28- Important limitations are acknowledged but insufficiently analyzed: lack of patient-level data, absence of appropriateness-of-use assessment, and confounding by infection control practices. The impact of restricted colistin and novel agent testing is underestimated.
29- Alternative explanations (clonal spread, outbreaks, diagnostic intensity) are not explored.
30- Contextual differences between Serbia and other countries are not critically discussed.

Conclusions
31- Conclusions overstate causal inferences beyond observational data support.
32- Recommendations for stewardship are generic and not directly derived from study-specific findings.

Author Response

Response to reviewers

 

We sincerely thank the editor and reviewers for their thorough and insightful evaluation of our manuscript and for their valuable suggestions. Detailed point-by-point responses to the reviewers’ comments are provided below.

 

Reviewer #1

Title & Abstract
1- Study design (retrospective observational) is not explicitly stated in the title.

Response:
We thank the reviewer for this valuable suggestion. We have revised the title to explicitly state the study design by adding “A Retrospective Observational Study”. The updated title now reads: lines 2-4; “Dynamics of Antimicrobial Resistance and Antimicrobial Consumption in a Secondary-Care Hospital in Serbia, 2019–2022: A Retrospective Observational Study.”

2- The abstract does not clearly specify inclusion/exclusion criteria for isolates or patients.

Response: We thank the reviewer for this comment. Accordingly, the abstract has been revised: lines 21-23; “All bacterial isolates were included for species-frequency analyses. For AMR and specimen distribution, repeat isolates of E. coli and K. pneumoniae per patient were excluded”. This clarification aligns the abstract with the study design described in the Materials and Methods section.

3- Methods in the abstract insufficiently describe how correlations between antimicrobial use and resistance were assessed.

Response: We agree with the reviewer and have revised the abstract to specify that associations between antimicrobial consumption and resistance rates were assessed using Spearman’s rank correlation, consistent with the statistical methods described in the Materials and Methods section: line 21; “…and antibiotic use (Spearman’s correlation).”

4- Abstract conclusions imply causality (“pandemic-associated surges… drove resistance”) despite observational design.

Response: We appreciate this observation. The abstract conclusions have been revised to remove causal language and to suggest temporal concurrence: lines 31-33; “Temporal concurrence of increased antimicrobial use during COVID-19 and escalating resistance underscores need for strengthened surveillance and stewardship.”. Other changes in the Abstract section were made to retain the maximum word count.

Bottom of Form

Top of FormBottom of Form

Introduction

5- The knowledge gap specific to secondary-care hospitals is not clearly articulated.

Response: We thank the reviewer for this helpful comment. We have accordingly revised the Introduction to explicitly highlight the knowledge gap in secondary-care hospitals in Serbia, noting that most published hospital AMR data originate from tertiary-care centers, whereas secondary-care hospitals remain underrepresented despite their major role in inpatient care and antibiotic use: lines 81-83; “Moreover, most published hospital AMR data in Serbia originate from tertiary-care centers [12,13] whereas secondary-care hospitals remain underrepresented despite their important role in inpatient care and antibiotic use”

6- Rationale for focusing exclusively on E. coli and K. pneumoniae is not sufficiently justified.

Response: We thank the reviewer for this important comment. We have revised the Introduction to better justify the exclusive focus on Escherichia coli and Klebsiella pneumoniae, emphasizing that they are the most common clinically significant Enterobacterales isolated from hospitalized patients and represent key targets for empirical therapy and antimicrobial stewardship. This rationale has now been explicitly stated in the revised manuscript: lines 64-68; “In addition to their recognized public health importance, E. coli and K. pneumoniae were selected because they are the most common clinically significant Enterobacterales isolated from hospitalized patients [7] and key targets for empirical therapy and antimicrobial stewardship.”.

7- Study hypotheses are not explicitly stated; aims are descriptive rather than hypothesis-driven.

Response: We thank the reviewer for this comment. We have accordingly revised the Aims section to be more explicit and added a dedicated Hypotheses statement.: lines 84-93; “The aims of this study were to evaluate the following: i) the prevalence of causative agents of infections among hospitalized patients; ii) the frequency of first isolates of E. coli and K. pneumoniae and their distribution by specimen type (invasive and non-invasive) and hospital ward; iii) antimicrobial resistance rates and multidrug resistance trends in E. coli and K. pneumoniae; iv) hospital antimicrobial consumption during 2019–2022; and v) the association between antimicrobial consumption and resistance in E. coli and K. pneumoniae.

Hypotheses: We hypothesized that antimicrobial resistance rates in E. coli and K. pneumoniae increased over the study period (2019–2022) and were correlated with annual hospital antimicrobial consumption.”

8- No clear justification for the chosen study period (2019–2022) beyond pandemic relevance.

Response: We thank the reviewer for this comment. The rationale for selecting the 2019–2022 study period has been clarified in the Methods section: lines 99-104; “The study period (2019–2022) was selected to include a pre-pandemic baseline year (2019), the pandemic years (2020–2021), and the subsequent year (2022), enabling assessment of temporal changes in antimicrobial consumption and resistance trends as routine microbiology and stewardship monitoring indicated notable changes in resistance patterns and antimicrobial consumption during these years.”

 

9- Changes in hospital structure/workflow during COVID-19 are mentioned later but not prespecified here.

Response: We thank the reviewer for this comment. We have updated the Methods section to explicitly describe the temporary hospital reorganization during the COVID-19 pandemic (capacity expansion and reallocation of wards/ICU resources), and workflow. We also expanded the hospital description by including the number of ICU beds (38) alongside the total bed capacity, to better contextualize potential changes in case-mix and microbiological sampling patterns: lines 105-110; “, including 38 Intensive care unit (ICU) beds. During the COVID-19 pandemic, the hospital capacity was temporarily expanded by 152 beds in an auxiliary facility (COVID hospital) from July 2020. Hospital workflows were reorganized to accommodate COVID-19 care, including the establishment of dedicated COVID-19 wards and partial reallocation of ICU, surgical, and maternity capacities. These structural changes may have influenced patient case-mix and microbiological sampling patterns during the study period.”

10- Inclusion and exclusion criteria for “clinically significant isolates” are not defined.

Response: We thank the reviewer for this comment. “Clinically significant isolates” were defined as isolates considered causative agents of infection according to routine clinical microbiology criteria. Isolates regarded as commensal/colonizing flora or recovered in non-significant quantities were excluded. This definition has now been added to the Materials and Methods section: lines 111-114 “Only isolates deemed clinically significant and likely causative agents of infection, according to routine clinical microbiological assessment, were included. Isolates representing expected physiological microbiota at the sampling site and/or recovered in non-significant quantities were excluded.”

11- Criteria for distinguishing infection vs colonization (especially for non-invasive isolates) are unclear.

Response: We thank the reviewer for this comment. We have clarified in the Materials and Methods that infection vs. colonization (for non-invasive specimens) was determined using routine clinical microbiology interpretation and that isolates interpreted as colonizers were excluded lines 114-118 ”For non-invasive specimens, the distinction between infection and colonization was determined using routine clinical microbiology criteria, taking into account relevant factors for each specimen type, including species identity, number of isolates, leukocyte presence, and dominance of the isolate in culture. Isolates interpreted as colonizers were excluded. Formal clinical diagnostic criteria were not applied”

12- No explicit definition of primary and secondary outcomes

Response: We thank the reviewer for this comment. We have revised the Materials and Methods section to explicitly define the co-primary, reflecting the dual focus of the study and the study title, and secondary outcomes of the study as now stated in Section 2.3 (Outcomes): lines 169–175; “The co-primary outcomes were (i) antimicrobial resistance patterns and temporal trends among E. coli and K. pneumoniae isolates and (ii) hospital antimicrobial consumption patterns, expressed as DDD/100BD and DU75%. Secondary outcomes included specimen-type distribution (invasive/non-invasive classification), ward-level distribution, prevalence of multidrug resistance, and associations between antimicrobial consumption and resistance.”

13- Potential confounders (ward type, ICU admission, COVID-19 status, patient demographics) are not defined or measured.

Response: We thank the reviewer for this important comment. In response, we have expanded the Results section by adding the distribution of E. coli and K. pneumoniae isolates by hospital ward, including the Intensive Care Unit, to better contextualize potential differences in case-mix and ward-related exposure (lines 247-278; “3.3. Ward distribution of Escherichia coli and Klebsiella pneumoniae)

In addition, we further addressed ward-level heterogeneity by providing ward- and year-specific distributions of multidrug-resistant (MDR) isolates in the Supplementary Material. Specifically, a detailed ward- and year-specific distribution of MDR E. coli isolates is provided in Table S1. For MDR K. pneumoniae, we included a detailed ward- and year-specific distribution in Table S2, and we also present the ward- and year-specific distribution of the three most common MDR phenotypes in Table S3 and Figure S1.

However, due to the retrospective design and the structure of the microbiology database, we were not able to reliably extract or link isolate-level data to patient-level variables such as COVID-19 status, demographic characteristics, or ICU admission status beyond ward attribution. Therefore, these potential confounders could not be formally measured or included in multivariable analyses. This limitation has been explicitly acknowledged in the revised manuscript: lines 641-647; “…due to the retrospective design and limitations of the microbiology database, we were unable to reliably link all isolates to patient-level clinical data (e.g., demographics, comorbidities, COVID-19 status, prior antibiotic exposure, or clinical outcomes). Consequently, temporal changes in resistance and antimicrobial consumption should be interpreted as ecological trends, with possible residual confounding, …”

14- Colistin testing strategy (restricted testing) introduces selection bias but is not addressed methodologically. (No formal discussion of selection bias, information bias, or confounding in the Methods section.)

Response: We thank the reviewer for this important observation. We have now clarified the colistin susceptibility testing strategy in the Methods section, including the criteria for selective testing and the improved test availability in 2021–2022: lines 129-135; “Due to limited resources, colistin testing was not performed routinely for all isolates; it was conducted primarily for isolates resistant to all other tested agents and/or isolates recovered from critically ill patients where colistin was considered a potential therapeutic option. Test availability improved in 2021–2022, resulting in a higher number of isolates tested in these years. Therefore, colistin results were interpreted descriptively and were not used for temporal trend inference.”

In addition, we revised the interpretation of colistin resistance in the Results section to ensure it is presented descriptively only, without implying temporal trends: lines 327-328; “Colistin susceptibility was assessed in a targeted subset of isolates; among tested isolates, the highest resistance rate (39.3%) was observed in 2020 (Figure 3; Table A3). “

15- No sample size or power considerations are provided.

Response: We thank the reviewer for this comment. A formal sample size or power calculation was not performed because this was a retrospective observational study that included all available microbiology records and hospital-level antimicrobial consumption data during 2019–2022, rather than a sampled population with a predefined target size. This has been clarified in the Methods section (lines 98–99): “No a priori sample size calculation was performed, as all available isolates were included.” Nevertheless, the large number of isolates enabled robust estimation of resistance patterns and assessment of differences across the study period.

16- No adjustment for multiple comparisons despite extensive year-by-year testing.

Response: We thank the reviewer for this comment. To address the issue of multiple comparisons, we have added a description of the Bonferroni adjustment for year-to-year pairwise testing in the Methods section (lines 181–183; “ Pairwise comparisons of column proportions for year-to-year resistance proportions were adjusted for multiple testing using the Bonferroni correction.”) and reported the corrected pairwise results in the Results section (lines 298–302; “In E. coli isolates, notable year-to-year variations in resistance proportions (adjusted with Bonferroni correction, p<0.05) were observed for gentamicin between 2019-2020 and 2019-2022; for ampicillin between 2019-2020, 2019-2022, and 2020-2022; for ciprofloxacin between 2020-2022 and 2021-2022; for levofloxacin, cefuroxime, ceftriaxone, and cefotaxime only between 2020-2022; and for cefepime between 2019-2020.” for E. coli and 318-325; In K. pneumoniae isolates, notable year-to-year variations in resistance proportions (adjusted using Bonferroni correction, p<0.05) were seen for the years 2019-2021, 2019-2022, 2020-2021, and 2021-2022 concerning amikacin, ciprofloxacin, levofloxacin, cefepime, imipenem, meropenem, ertapenem, and PIT. The proportions for gentamicin significantly varied between 2019-2022 and 2020-2022, whereas for AMC, the variation occurred only between 2020-2022. For TRS and cefuroxime, the resistance proportion varied for 2019-2021 and 2019-2022, whereas for ceftriaxone and cefotaxime, the data varied for 2019-2021, 2019-2022, and 2020-2022, as well. “ for K. pneumoniae)

17- No multivariable analyses to control for confounding factors.

Response: We thank the reviewer for this valuable suggestion. We agree that multivariable analyses would strengthen the assessment of potential confounding. However, such analyses were not feasible in the present study because it was not designed to evaluate predictors of resistance and the available dataset lacked key patient-level covariates (e.g., demographics, comorbidities, COVID-19 status, ICU admission, and prior antibiotic exposure) required for robust multivariable modelling. This limitation has been acknowledged in the revised manuscript: lines 644-645; “Therefore, multivariable analyses to control for confounding could not be performed.”

18- Trend analyses are inconsistently specified (chi-square vs correlation vs descriptive).

Response: We thank the reviewer for this comment. We have clarified the analytical approach in the Methods section: lines 177-185; “Descriptive statistics were used to summarize categorical variables as absolute and relative frequencies, n (%). Differences in proportions across study years were assessed using Pearson’s chi-squared test (overall 2019–2022 and selected pairwise year-to-year comparisons where relevant). Associations between antimicrobial consumption and resistance rates were assessed using Spearman’s rank correlation coefficient. Pairwise comparisons of column proportions for year-to-year resistance proportions were adjusted for multiple testing using the Bonferroni correction. Statistical analysis was performed using IBM SPSS Statistics for Windows, version 23.0 (IBM Corp., Armonk, NY, USA). The significance level was set at α = 0.05.”

Briefly, temporal changes across 2019–2022 were assessed using Pearson’s χ² test for differences in proportions, while associations between antimicrobial consumption and resistance rates were evaluated using Spearman’s rank correlation. Where formal inferential testing was not appropriate (e.g., limited numbers or selectively tested agents), results were presented descriptively without reporting p-values.

Results
19- Patient-level characteristics (age, sex, ward, ICU vs non-ICU) are completely absent.

Response: We thank the reviewer for this comment. Ward-level distribution of E. coli and K. pneumoniae isolates (including the ICU) has been added to the Results section to provide additional clinical context and allow comparison between ICU and non-ICU wards. lines 247-278; “3.3. Ward distribution of Escherichia coli and Klebsiella pneumoniae…” These ward-level distributions were also briefly discussed to contextualize resistance dynamics and to emphasize that temporal trends may reflect changes in ward case-mix and sampling intensity rather than true epidemiological shifts: lines 436-454 in the Discusion section.

Unfortunately, patient-level demographic characteristics (age and sex) were not available in the microbiology database, as explained in our response to comment #13, and therefore could not be included in the analyses.

20- Number of isolates is reported, but number of patients contributing isolates is unclear.

Response: We thank the reviewer for this comment. We have now added in the Results section data on the annual number of hospitalized patients and hospital bed-days during the study period and COVID-19-positive hospitalized patients: lines 188-193; “The total number of hospitalized patients was 27,067 in 2019, 24,503 in 2020, 28,946 in 2021, and 27,760 in 2022. The number of COVID-19-positive hospitalized patients was 0 in 2019, 3,263 in 2020, 7,278 in 2021, and 1,988 in 2022. Total hospital bed-days were 150,756, 133,332, 177,217, and 138,652, respectively. The lowest number of patients was recorded in 2020, representing a 9.5% decrease (2020) and a 6.9% increase (2021), respectively, compared with 2019. In addition, we note that: lines 194-196; “The microbiology dataset did not allow reliable linkage of all isolates to unique patient identifiers across the full study period; therefore, patient counts are reported at the hospital level.”

For E. coli and K. pneumoniae, susceptibility and specimen distribution analyses were restricted to the first isolate per patient; therefore, the number of included first isolates corresponds to the number of patients contributing to these species-specific analyses.

21- Flow diagram of isolates through the study (screened, included, excluded) is not shown.

Response: We thank the reviewer for this comment. We have added a flow diagram (Figure A1) in the, Appendix section, summarizing the isolate selection process (screened, excluded repeated isolates, and included first isolates per patient) for E. coli and K. pneumoniae.

22- Resistance rates are reported without confidence intervals.

Response: We thank the reviewer for this comment. Resistance rates were presented to summarize observed proportions within the available datasets. As the goal of our study and presented results is mainly descriptive reporting, we deemed point estimates alone as acceptable.

23- Invasive vs non-invasive isolate analyses are not consistently stratified across outcomes.
Response: We thank the reviewer for this comment. Invasive vs non-invasive stratification was applied to specimen-type distribution analyses in the results section (lines 223; 3.2. Specimen distribution of Escherichia coli and Klebsiella pneumoniae) and discussion: lines 417-423. Resistance outcomes were analysed using the overall first-isolate-per-patient datasets of E. coli and K. pneumoniae (invasive and non-invasive isolates), because the number of invasive isolates per year was low and would limit the robustness and statistical power of stratified analyses. We have clarified this in the Materials and Methods section: lines 142-145; “Antimicrobial resistance rates were analysed using the overall E. coli and K. pneumoniae first-isolate-per-patient datasets (invasive and non-invasive isolates), because the number of invasive isolates per year was low and would limit the robustness of stratified analyses.”.

24- Results repeatedly imply temporal “effects” without accounting for confounding.

Response: We thank the reviewer for this important observation. We agree that, given the retrospective observational design and the lack of multivariable adjustment for potential confounders, causal inference is not appropriate. Accordingly, we have revised the entire Results section to systematically remove causal or effect-implying language (e.g., “impact,” “due to,” “driven by”) and to describe findings strictly in terms of year-to-year changes and observed temporal trends. In addition, we have strengthened the Limitation section to more explicitly acknowledge the potential influence of unmeasured confounding and other biases inherent to retrospective analyses: line 645-647; “Consequently, temporal changes in resistance and antimicrobial consumption should be interpreted as ecological trends, with possible residual confounding, …”. These revisions clarify that the reported associations should be interpreted as descriptive rather than causal.

25- Correlation results are reported without effect sizes or correlation coefficients.

Response: We thank the reviewer for this thoughtful comment. The correlation analyses were conducted as an exploratory screening tool to identify the presence and direction of potential associations rather than to provide precise estimates of their magnitude. In several cases, these analyses were based on ordinal or aggregated measures, for which reporting correlation coefficients could introduce interpretive bias by suggesting false precision. For this reason, we prioritized reporting the statistical significance and direction of the associations.

26- Subgroup analyses (e.g., invasive vs non-invasive) are limited and not prespecified.

Response: We thank the reviewer for this comment. The methods section was expanded and revised according to reviewer suggestion: lines 141-145; “Invasive vs non-invasive stratification was prespecified and applied to specimen-type distribution analyses. Antimicrobial resistance rates were analysed using the overall E. coli and K. pneumoniae first-isolate-per-patient datasets (invasive and non-invasive isolates), because the number of invasive isolates per year was low and would limit the robustness of stratified analyses.”.

 

27- DU75% analysis lacks justification for its interpretive relevance to resistance outcomes.

Response: We thank the reviewer for this constructive comment. The DU75% analysis was included as a standardized drug-utilization metric to identify the limited number of antimicrobials accounting for the majority (75%) of total consumption, thereby characterizing prescribing concentration and highlighting priority targets for antimicrobial stewardship interventions. Importantly, DU75% was not intended to serve as a direct explanatory measure of resistance outcomes. Rather, it provides contextual information on hospital-level prescribing patterns and helps interpret which agents may plausibly exert the greatest selective pressure over time. To address this concern, we have expanded and clarified the description of DU75% in the Methods section: lines 163-166; “The Drug Utilization 75% (DU75%) metric was applied to identify the antimicrobials accounting for 75% of total consumption and to describe prescribing concentration, thereby supporting stewardship prioritization [10]. DU75% was not intended as a direct measure of resistance drivers.” and explicitly avoided any causal interpretation linking DU75% directly to resistance outcomes in the Results (lines 376-386; “Antimicrobials accounting for 75% (DU75) of annual hospital consumption.”

Detailed annual hospital consumption of antimicrobials is shown in Table 2. Ceftriaxone was the most frequently used antimicrobial across all years. During 2020–2021, particularly in 2021, antimicrobial consumption was largely dominated by a limited number of agents—ceftriaxone, meropenem, amikacin, vancomycin, and metronidazole—which together accounted for approximately 75% of total use (Table 2, Figure A2).

Table 2. Antimicrobials accounting for 75% (DU75) of annual hospital consumption, expressed as defined daily doses per 100 bed-days (DDD/100 BD) and relative proportions, 2019–2022”) and Discussion (lines 579-582; “To complement DDD/100 bed-days, we used the Drug Utilization 75% (DU75%) metric to identify the antimicrobials accounting for 75% of total consumption, as an indicator of prescribing concentration and stewardship priorities, rather than a direct driver of resistance.”

 

Discussion
28- Key findings are selectively emphasized (K. pneumoniae) without balanced discussion of null findings (E. coli).

Response: We thank the reviewer for this comment. We revised the Discussion section to provide a more balanced interpretation by explicitly highlighting the more modest and less consistent resistance changes observed in E. coli compared with K. pneumoniae: from lines 436-454; 4.2. Prevalence and ward distribution of Escherichia coli and Klebsiella pneumoniae isolates. In addition, we expanded the description of consumption–resistance patterns for E. coli: lines 610-623; “Changes in E. coli resistance did not mirror trends in the consumption of the corresponding antimicrobials, with resistance remaining relatively stable despite marked fluctuations in consumption. Notably, levofloxacin consumption increased in 2020–2021, whereas E. coli resistance decreased in 2020 before rising again in 2022, indicating a non-parallel (“U-shaped”) pattern. For K. pneumoniae, resistance increased during 2020–2021 in parallel with increased antimicrobial use; however, resistance did not decline in 2022 despite reduced consumption showing only a slower rate of increase. Overall, no statistically significant correlations between antimicrobial consumption and resistance were observed. These findings should be interpreted cautiously, as consumption–resistance relationships may require longer observation periods to be reliably detected, as highlighted by Peñalva et al. [43]. In addition, resistance to one antimicrobial may be influenced by the use of another through co-selection mechanisms [43]; for example, broad-spectrum cephalosporins can co-select fluoroquinolone resistance when resistance determinants are carried on the same plasmid [44].”

29- Overinterpretation of associations as causal relationships persists.

Response: We thank the reviewer for highlighting this important concern.

We have further revised the Discussion to remove causal or effect-implying language and to ensure that all interpretations are consistently framed as observed temporal associations rather than causal relationships. In addition, we now explicitly state the constraints of causal inference in the Discussion (lines 625–627): “Given the retrospective observational design and the lack of significant correlations, causal relationships cannot be inferred.” These revisions were made to prevent overinterpretation of the findings.

30- Important limitations are acknowledged but insufficiently analyzed: lack of patient-level data, absence of appropriateness-of-use assessment, and confounding by infection control practices. The impact of restricted colistin and novel agent testing is underestimated.

Response: We thank the reviewer for this important comment. We have expanded the Limitations section to better address the implications of the main sources of bias: lines 639-666; “Several limitations should be noted. First, this was a single-center study, which may limit the generalizability of the findings to other hospitals in Serbia, particularly tertiary-care institutions. Second, due to the retrospective design and limitations of the microbiology database, we were unable to assess molecular epidemiology and reliably link all isolates to patient-level clinical data (e.g., demographics, comorbidities, COVID-19 status, prior antibiotic exposure, or clinical outcomes). Consequently, temporal changes in resistance and antimicrobial consumption should be interpreted as ecological trends, with possible residual confounding, as the consumption–resistance correlation was based on aggregated annual hospital-level data and therefore has limited sensitivity to ecological bias and time-lagged effects. Third, prescribing appropriateness was not assessed; therefore, changes in antimicrobial consumption cannot be attributed to guideline-concordant empirical therapy, overtreatment, or shifts in disease severity. In addition, COVID19-related changes in infection prevention and control practices and hospital reorganization may have influenced both sampling intensity and transmission dynamics, independently of antimicrobial use. Fourth, due to restricted material resources and national drug registration constraints, susceptibility testing was not available for all antimicrobials throughout the study period, precluding assessment of extensively drug-resistant (XDR) and pandrug-resistant (PDR) phenotypes. Susceptibility for newer β-lactam/β-lactamase inhibitor combinations was introduced after their clinical availability. Therefore, testing for ceftazidime/avibactam was introduced in 2022; therefore, resistance to these agents may therefore be underestimated.

Despite these limitations, the inclusion of both invasive and non-invasive isolates of E. coli and K. pneumoniae over a four-year period, combined with hospital antimicrobial consumption data, provides valuable insight into local resistance dynamics in a secondary-care hospital setting.”

31- Alternative explanations (clonal spread, outbreaks, diagnostic intensity) are not explored.

Response: We thank the reviewer for this important comment. We agree that alternative explanations for the observed trends should be considered. For example, the increase in resistance, particularly in K. pneumoniae, may reflect clonal spread or unrecognized ward-level outbreaks; however, molecular typing and formal outbreak investigation data were not available to assess this. Additionally, changes in diagnostic intensity and sampling practices during the COVID-19 period (e.g., a higher proportion of ICU admissions and respiratory specimens) could have influenced isolate distribution and resistance estimates independently of antimicrobial consumption. We have now added these considerations to the Discussion and further emphasized that the findings describe temporal associations rather than causal relationships: lines 625-627; “Given the retrospective observational design and the lack of significant correlations, causal relationships cannot be inferred.”.

32- Contextual differences between Serbia and other countries are not critically discussed.

Response: We thank the reviewer for this comment. While this aspect was briefly mentioned in the original Discussion regarding differences between secondary- and tertiary-care settings, we have now expanded this section to more critically discuss contextual differences between Serbia and other countries. We have clarified this in the revised manuscript (lines 489–508); “Cross-country comparisons should be interpreted cautiously, as Serbia differs from many EU/EEA settings in terms of antimicrobial prescribing practices, availability of hospital-sector consumption surveillance, limited antimicrobial stewardship capacity, inconsistent infection prevention and control practices, and patient referral patterns. Therefore, although our findings are broadly consistent with international reports of increased broad-spectrum antibiotic use and rising resistance during the COVID-19 period, the magnitude and dynamics of these changes may not be directly comparable across countries.

The marked increase in K. pneumoniae resistance, particularly to carbapenems, should be interpreted cautiously. Besides changes in antimicrobial consumption, alternative explanations may include clonal spread of a successful hospital-adapted lineage, introduction and dissemination of carbapenemase-producing strains (e.g., OXA-48-like, or NDM-producing isolates), or multiple parallel transmission events occurring across wards. In addition, the COVID-19 period was associated with major changes in hospital organization and case-mix, including increased ICU burden and intensive use of invasive devices, which may have contributed to a higher diagnostic intensity and altered microbiological sampling patterns. Because molecular typing and carbapenemase characterization were not available in this study, we cannot determine whether the observed rise reflects expansion of a single clone/outbreak or broader endemic transmission, which limits the epidemiological interpretation and the ability to propose targeted containment measures.”.

Conclusions

33- Conclusions overstate causal inferences beyond observational data support.

Response: We thank the reviewer for this important comment. We agree that causal inferences cannot be drawn from a retrospective observational study. We have therefore revised the Conclusions section to avoid causal language (e.g., “impact” and “drove”) and to clearly frame our findings as temporal patterns and associations observed over 2019–2022: lines 669-675; “This study provides insight into antimicrobial resistance and antimicrobial consumption patterns in a secondary-care hospital in Serbia during 2019–2022. Resistance in K. pneumoniae, particularly to carbapenems, increased substantially over the study period, while hospital antimicrobial consumption peaked in 2020–2021 and declined in 2022. The findings support the need for sustained local surveillance and targeted stewardship efforts focusing on high-use agents (e.g., ceftriaxone and meropenem) and high-risk settings such as the ICU. ”.

34- Recommendations for stewardship are generic and not directly derived from study-specific findings.

Response: We thank the reviewer for this comment. We have revised the Conclusions to ensure that the antimicrobial stewardship recommendations are directly informed by our study findings.

Specifically, our analyses showed that (i) overall antimicrobial consumption peaked in 2020–2021 and was concentrated in a limited number of broad-spectrum agents (as indicated by DU75%), and (ii) K. pneumoniae resistance, particularly to carbapenems, increased markedly over the study period. Accordingly, the revised recommendations focus on targeted stewardship priorities relevant to our setting, including monitoring and optimization of high-consumption agents (e.g., ceftriaxone and meropenem), strengthened surveillance of carbapenem-resistant K. pneumoniae, and reinforcement of infection prevention and control measures in high-risk wards (including the ICU) – lines 670-675; “Resistance in K. pneumoniae, particularly to carbapenems, increased substantially over the study period, while hospital antimicrobial consumption peaked in 2020–2021 and declined in 2022. The findings support the need for sustained local surveillance and targeted stewardship efforts focusing on high-use agents (e.g., ceftriaxone and meropenem) and high-risk settings such as the ICU.”

Reviewer 2 Report

Comments and Suggestions for Authors

The presented work is devoted to the analysis of the dynamics of antibiotic resistance and antimicrobial drug consumption in a large Serbian hospital in the period 2019-2022, including the COVID 19 pandemic period. The research topic is extremely relevant, especially for countries where systemic monitoring of AMR remains fragmented. The authors address the important issue of the impact of the pandemic on antibiotic consumption patterns and the subsequent evolution of resistance, making the work a significant contribution to the regional and international literature. The article is well structured, logically built and contains a large array of data, which allows us to trace the dynamics of both the prevalence of key pathogens and changes in their sensitivity. Particularly valuable is the inclusion of data on DU75%, which is rare in regional studies and provides a better understanding of which drugs form the main contribution to consumption. The study methodology is generally correct: the authors use standardized approaches to the identification of microorganisms, EUCAST criteria, as well as generally accepted indicators of antibiotic consumption. The application χ Spearman's ² test and correlation analysis corresponds to the type of data. However, the work has a number of significant limitations to consider. One of the key limitations of the study is the absence of molecular typing. The surge in carbapenem resistance among Klebsiella pneumoniae—reaching 52–69% in 2022—is too striking to interpret confidently without knowing what stands behind it. It remains unclear whether this reflects the spread of a single successful clone, the introduction of carbapenemase‑producing strains such as KPC or NDM, or several unrelated events occurring in parallel. Without genetic data, it becomes difficult not only to explain the origins of such a rapid rise in resistance but also to identify targeted measures that could help contain further dissemination. This limitation substantially reduces the epidemiological value of the conclusions. Another weakness is the lack of data stratification by clinical unit. The COVID 19 pandemic has led to a significant redistribution of patient flows, an increase in the burden on intensive care units and the widespread use of invasive ventilation. These factors could significantly affect the structure of the released pathogens and the level of resistance. Without analysis by department, it is impossible to assess whether the identified trends were localized in individual divisions or reflect the general hospital situation. Correlation analysis between antibiotic consumption and resistance is performed on aggregated data, which inevitably leads to ecological error. The absence of statistically significant relationships does not mean the absence of causal relationships, but only reflects the lack of sensitivity of the chosen approach. For such problems, time lags, multiple regression or time series models are more appropriate. As I said, there are some limitations, but the paper gives useful info on how antimicrobial resistance changed during the pandemic. The research shows a clear increase in Klebsiella pneumoniae among all the samples, along with a similar rise in MDR and carbapenem-resistant strains. The manuscript is pretty good overall, and it could be ready for publication after a few small tweaks. It's also important for the authors to clearly acknowledge the study's limitations.  If they've got even some of the molecular epidemiology data, including them will make the paper a lot stronger and help clarify some of the trends they describe.

Author Response

Response to reviewers

 

We sincerely thank the editor and reviewers for their thorough and insightful evaluation of our manuscript and for their valuable suggestions. Detailed point-by-point responses to the reviewers’ comments are provided below.

Reviewer #2

1. However, the work has a number of significant limitations to consider. One of the key limitations of the study is the absence of molecular typing. The surge in carbapenem resistance among Klebsiella pneumoniae—reaching 52–69% in 2022—is too striking to interpret confidently without knowing what stands behind it. It remains unclear whether this reflects the spread of a single successful clone, the introduction of carbapenemase‑producing strains such as KPC or NDM, or several unrelated events occurring in parallel. Without genetic data, it becomes difficult not only to explain the origins of such a rapid rise in resistance but also to identify targeted measures that could help contain further dissemination. This limitation substantially reduces the epidemiological value of the conclusions.

Response: We thank the reviewer for this important and constructive comment. Accordingly, discussion section was expanded: lines 498-501; “Besides changes in antimicrobial consumption, alternative explanations may include clonal spread of a successful hospital-adapted lineage, introduction and dissemination of carbapenemase-producing strains (e.g., OXA-48-like, or NDM-producing isolates), or multiple parallel transmission events occurring across wards.”

 

2. Another weakness is the lack of data stratification by clinical unit. The COVID 19 pandemic has led to a significant redistribution of patient flows, an increase in the burden on intensive care units and the widespread use of invasive ventilation. These factors could significantly affect the structure of the released pathogens and the level of resistance. Without analysis by department, it is impossible to assess whether the identified trends were localized in individual divisions or reflect the general hospital situation.

Response: In response to the reviewer’s suggestion, the Results section was expanded to include ward-level distribution of E. coli and K. pneumoniae isolates, including those from the ICU: lines 437-454; “Ward-level distribution provided additional clinical context for interpreting resistance dynamics during 2019–2022. E. coli was most frequently recovered from Internal Medicine and Neurology wards across all years, with relatively stable proportions, consistent with its predominant role as a uropathogen and a common cause of infections managed outside critical care settings. ICU contribution showed a U-shaped pattern, likely reflecting pandemic-related changes in hospital organization and sampling intensity rather than a true epidemiological shift. In contrast, K. pneumoniae showed a ward distribution more typical of healthcare-associated pathogens, with persistently high proportions in Surgery and the ICU. A marked increase in isolates from Infectious Diseases and Neurology ward in 2021–2022, may reflect changes in case-mix during the later pandemic period, including a higher burden of severely ill patients, longer hospital stays, and increased exposure to invasive procedures and broad-spectrum antimicrobials. However, ward-level transmission and outbreaks, or clonal spread of carbapenemase-producing lineages cannot be excluded due to the lack of molecular typing data. Overall, these findings highlight that temporal resistance trends should be interpreted in light of shifting ward structure and patient populations, warranting cautious interpretation of consumption–resistance associations in the absence of patient-level covariates.”

In addition, detailed ward- and year-specific distributions of MDR isolates are provided in the Supplementary Material (Table S1 for MDR E. coli; Table S2 for MDR K. pneumoniae; and Table S3 and Figure S1 for the three most common MDR phenotypes of K. pneumoniae).

 

3. Correlation analysis between antibiotic consumption and resistance is performed on aggregated data, which inevitably leads to ecological error. The absence of statistically significant relationships does not mean the absence of causal relationships, but only reflects the lack of sensitivity of the chosen approach. For such problems, time lags, multiple regression or time series models are more appropriate. As I said, there are some limitations, but the paper gives useful info on how antimicrobial resistance changed during the pandemic. The research shows a clear increase in Klebsiella pneumoniae among all the samples, along with a similar rise in MDR and carbapenem-resistant strains. The manuscript is pretty good overall, and it could be ready for publication after a few small tweaks.

Response: We acknowledge that the analysis of the relationship between antibiotic consumption and resistance was performed using aggregated annual hospital-level data, which may introduce a degree of ecological bias. However, this analysis was intended as an exploratory, descriptive assessment rather than a definitive evaluation of causality. Accordingly, the absence of statistically significant correlations should be interpreted cautiously and does not preclude the existence of causal relationships. More advanced analytical approaches (e.g., lagged analyses, multivariable regression, or time-series models) would be preferable; however, they were not feasible due to the structure of the available pharmacy data, which consisted of annual aggregated values without linkage to ward- or patient-level information.

 

4. It's also important for the authors to clearly acknowledge the study's limitations.  If they've got even some of the molecular epidemiology data, including them will make the paper a lot stronger and help clarify some of the trends they describe.

Response: According to reviewers’ suggestions, limitations were substantially expanded: lines 639-666; “Several limitations should be noted. First, this was a single-center study, which may limit the generalizability of the findings to other hospitals in Serbia, particularly tertiary-care institutions. Second, due to the retrospective design and limitations of the microbiology database, we were unable to assess molecular epidemiology and reliably link all isolates to patient-level clinical data (e.g., demographics, comorbidities, COVID-19 status, prior antibiotic exposure, or clinical outcomes). Consequently, temporal changes in resistance and antimicrobial consumption should be interpreted as ecological trends, with possible residual confounding, as the consumption–resistance correlation was based on aggregated annual hospital-level data and therefore has limited sensitivity to ecological bias and time-lagged effects. Third, prescribing appropriateness was not assessed; therefore, changes in antimicrobial consumption cannot be attributed to guideline-concordant empirical therapy, overtreatment, or shifts in disease severity. In addition, COVID19-related changes in infection prevention and control practices and hospital reorganization may have influenced both sampling intensity and transmission dynamics, independently of antimicrobial use. Fourth, due to restricted material resources and national drug registration constraints, susceptibility testing was not available for all antimicrobials throughout the study period, precluding assessment of extensively drug-resistant (XDR) and pandrug-resistant (PDR) phenotypes. Susceptibility for newer β-lactam/β-lactamase inhibitor combinations was introduced after their clinical availability. Therefore, testing for ceftazidime/avibactam was introduced in 2022; therefore, resistance to these agents may therefore be underestimated.

Despite these limitations, the inclusion of both invasive and non-invasive isolates of E. coli and K. pneumoniae over a four-year period, combined with hospital antimicrobial consumption data, provides valuable insight into local resistance dynamics in a secondary-care hospital setting.”

We fully agree that the inclusion of molecular epidemiology data would substantially strengthen the manuscript; however, such data were not available for the study period. This was discussed in the revised version of the manuscript: lines 491-494; “Besides changes in antimicrobial consumption, alternative explanations may include clonal spread of a successful hospital-adapted lineage, introduction and dissemination of carbapenemase-producing strains (e.g., OXA-48-like, or NDM-producing isolates), or multiple parallel transmission events occurring across wards.”

 

Bottom of Form