Early trajectories of antibiotic exposure and colonization pressure and risk of ICU-acquired carbapenem-resistant Gram-negative bacteria: A prospective cohort study - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Ann Intensive Care . 2026 Apr 6;16:100063. doi: 10.1016/j.aicoj.2026.100063 Search in PMC Search in PubMed View in NLM Catalog Add to search Early trajectories of antibiotic exposure and colonization pressure and risk of ICU-acquired carbapenem-resistant Gram-negative bacteria: A prospective cohort study Zhihui Chen Zhihui Chen a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China b Shanghai Institute of Infectious Disease and Biosecurity, Fudan University, China Find articles by Zhihui Chen a, b, 1 , Xiangru Ye Xiangru Ye c Department of Neurocritical Care Unit, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China Find articles by Xiangru Ye c, 1 , Jing Wu Jing Wu a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China Find articles by Jing Wu a, 1 , Zhonghua Li Zhonghua Li a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China d Department of Infectious Diseases, Taicang First People's Hospital, Taicang, China Find articles by Zhonghua Li a, d , Sen Wang Sen Wang a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China Find articles by Sen Wang a , Jing Wang Jing Wang e Department of Infection Control, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China Find articles by Jing Wang e , Yueru Tian Yueru Tian f Department of Laboratory Medicine, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China Find articles by Yueru Tian f , Shirong Li Shirong Li f Department of Laboratory Medicine, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China Find articles by Shirong Li f , Lei Zhou Lei Zhou a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China Find articles by Lei Zhou a , Jie Ni Jie Ni g Department of Intensive Care Unit, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China Find articles by Jie Ni g , Yue Qu Yue Qu h Department of Infectious Diseases, School of Translational Medicine, Monash University, Melbourne, VIC, Australia Find articles by Yue Qu h, ⁎ , Jialin Jin Jialin Jin a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China Find articles by Jialin Jin a, ⁎ , Wenhong Zhang Wenhong Zhang a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China b Shanghai Institute of Infectious Disease and Biosecurity, Fudan University, China i Shanghai Sci-Tech Inno Center for Infection and Immunity, Shanghai, 200052, China Find articles by Wenhong Zhang a, b, i, 2, ⁎ Author information Article notes Copyright and License information a Department of Infectious Diseases, Shanghai Key Laboratory of Infectious Diseases and Biosafety Emergency Response, National Medical Center for Infectious Diseases, Huashan Hospital, Shanghai Medical College, Fudan University, China b Shanghai Institute of Infectious Disease and Biosecurity, Fudan University, China c Department of Neurocritical Care Unit, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China d Department of Infectious Diseases, Taicang First People's Hospital, Taicang, China e Department of Infection Control, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China f Department of Laboratory Medicine, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China g Department of Intensive Care Unit, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China h Department of Infectious Diseases, School of Translational Medicine, Monash University, Melbourne, VIC, Australia i Shanghai Sci-Tech Inno Center for Infection and Immunity, Shanghai, 200052, China ⁎ Corresponding authors. [email protected] [email protected] [email protected] 1 These authors contributed equally as co-first authors. 2 These authors contributed equally as co-corresponding authors. Revised 2026 Mar 30; Collection date 2026. © 2026 The Authors This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). PMC Copyright notice PMCID: PMC13090519 PMID: 42005530 Abstract Background Carbapenem-resistant Gram-negative bacteria (CR-GNB) pose a critical threat in intensive care units (ICUs), with antibiotic exposure and colonization pressure identified as key associated factors. Prior studies have analyzed these two time-varying factors as static cumulative variables, obscuring heterogeneity in temporal patterns and their joint evolution. How distinct early trajectories of these factors jointly relate to CR-GNB acquisition risk remains unclear. Methods In this prospective cohort study conducted at four ICUs in a tertiary-care center in China from March 2024 to January 2025, we enrolled consecutive patients with systematic rectal surveillance cultures. We used group-based multi-trajectory modeling to identify distinct joint trajectories of daily antibiotic exposure (dose, duration, spectrum) and colonization pressure during the first five ICU days. Continuous-time Markov multi-state models with Day-5 landmark analysis were adopted to assess associations between trajectory groups and subsequent ICU-acquired CR-GNB, adjusting for baseline and cumulative covariates. Results Among 533 patients entering the Day-5 landmark analysis, three distinct trajectory groups were identified: Low exposure/low pressure (34.1%), escalating exposure/intermediate pressure (54.4%), and high exposure/high pressure (11.4%). CR-GNB acquisition occurred in 124 patients (23.3%), with rates of 9.3%, 26.2%, and 50.8% across trajectory groups, respectively. Compared with the low-exposure trajectory, adjusted hazard ratios were 1.68 (95% confidence interval [CI] 1.24–2.28) for escalating-exposure and 2.82 (95% CI, 1.53–5.19) for high-exposure trajectories (both P < 0.001). Conclusions This study identified distinct early trajectories of antibiotic exposure and colonization pressure that were associated with differential CR-GNB acquisition risk. This trajectory-based framework for early association-based risk stratification may inform targeted prevention strategies, but external validation is required before clinical implementation. Trial registration Chinese Clinical Trial Registry Identifier: ChiCTR2400081352. Registered 28 February 2024. Keywords: Carbapenem-resistant Gram-negative bacteria, Intensive care unit, Antibiotic exposure, Colonization pressure, Group-based multi-trajectory modeling Introduction Carbapenem-resistant Gram-negative bacteria (CR-GNB) infections are an escalating public health threat, especially to critically ill patients in the intensive care unit (ICU) [ 1 , 2 ]. In this high-risk environment, CR-GNB acquisition is associated with high mortality and healthcare costs, making it an urgent clinical challenge [ [3] , [4] , [5] ]. CR-GNB acquisition has been associated with antibiotic exposure, which may contribute to selection pressure by disrupting the patient's microbiome [ 6 , 7 ], and high colonization pressure that increases patients' exposure to pathogens from environmental reservoirs [ 7 , 8 ]. Despite the established importance of these factors, how early dynamic, time-dependent patterns of the two factors relate to subsequent CR-GNB acquisition risk remains unclear. A major limitation of prior research is considering these inherently time-varying factors as static or cumulative variables [ 7 , [9] , [10] , [11] ]. This conventional approach may have obscured the vast heterogeneity of individual exposure histories; for instance, the risks associated with rapid therapeutic escalation during the early ICU stay may differ from those associated with stable, long-term antibiotic therapies. Furthermore, how the joint evolution of antibiotic exposure and colonization pressure trajectories in this critical early period relates to patient susceptibility remains an important and unaddressed question. Therefore, this study used group-based multi-trajectory modeling (GBMTM) to identify distinct longitudinal trajectories of antibiotic exposure and colonization pressure during the first five ICU days and to assess their joint association with the risk of subsequent CR-GNB acquisition [ 12 ]. We hypothesized that distinct combinations of these early exposure trajectories would be associated with subsequent CR-GNB acquisition risk, thereby providing a framework for early association-based risk stratification and identifying high-risk patterns that may be amenable to preemptive and risk-adapted intervention. Methods Study population and study design This prospective cohort study was conducted at Huashan Hospital, a tertiary care center affiliated to Fudan University, Shanghai, China. Consecutive patients admitted to four ICUs between March 2024 and January 2025 were enrolled. Data were collected and managed using Research Electronic Data Capture tools [ 13 ]. Upon ICU admission, all patients underwent rectal surveillance cultures for CR-GNB, including carbapenem-resistant Enterobacterales (CRE), carbapenem-resistant Acinetobacter baumannii (CRAB), and carbapenem-resistant Pseudomonas aeruginosa (CRPA), within 48 h and weekly thereafter. Microbiological culturing, antimicrobial susceptibility testing, and infection control practices are described in Methods S1. Exclusion criteria were age <18 years, detection of CR-GNB before or within 48 h of ICU admission, an ICU stay of less than 48 h, or no rectal surveillance cultures performed during the ICU stay. For patients with multiple ICU admissions during the study period, only the first qualifying admission was analyzed. The study protocol (KY2024-060) was approved by the Institutional Ethics Committee of Huashan Hospital, Fudan University, and the study was registered with the Chinese Clinical Trial Registry (ChiCTR2400081352). Written informed consent was obtained from all participants or their legally authorized representatives. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline [ 14 ] (Table S1). To address immortal time bias, we prespecified day 5 after ICU admission as the landmark [ 15 ]. Patients who were alive and in the ICU on day 5 and event-free before day 5 (no CR-GNB acquisition, discharge, or death) entered the landmark risk set. Follow-up started on day 5 (left truncation; time scale reset), and outcomes were evaluated from day 5 onward. Exposure trajectories We compiled daily longitudinal data on antibiotic exposure and colonization pressure for each patient during the first five ICU days. Daily antibiotic exposure was quantified using three metrics: (1) dose, measured as daily Defined Daily Doses (DDDs) in accordance with the World Health Organization methodology [ 16 ]; (2) duration, defined as the administration of at least one systemic antibiotic agent on a given day, corresponding to the Length of Therapy (LOT) [ 17 ]; and (3) spectrum, calculated as the daily Antibiotic Spectrum Index (ASI) for agents administered [ 18 , 19 ] (Table S2). Daily colonization pressure was calculated as the proportion of all patients in the same ICU who were colonized or infected with any CR-GNB on a given day [ 7 , 20 ] (Methods S2). Outcome Our outcome was the first detection of ICU-acquired CR-GNB. ICU acquisition was defined as a positive culture from any rectal surveillance or clinical sample obtained ≥48 h after admission for a patient who was CR-GNB negative on admission. Consistent with our Day-5 landmark design, outcome assessment began from Day 5 onward. The event time was considered interval-censored between the last negative culture (confirmed by rectal surveillance) and the first positive culture. Covariates Covariates included baseline variables and cumulative variables (measured from hospital admission through ICU day 5 and fixed at the Day-5 landmark for analysis). Baseline variables, measured at or before ICU admission, included sociodemographic characteristics (age, sex), medical history (emergency admission, recent hospitalization and antibiotic exposure, pre-ICU antibiotic use metrics [DDDs, LOT, ASI], ward transfers, surgical procedures), comorbidities (captured by individual conditions and the Elixhauser Comorbidity Index), illness severity (Sequential Organ Failure Assessment [SOFA] score within the first 24 h of ICU admission), ICU type, infection, and sepsis status. Cumulative variables reflected exposure accrued from hospital admission through the end of ICU day 5, including duration of invasive medical device use (central venous catheter, mechanical ventilation, urinary catheter, nasogastric tube) and non-antibiotic medication administration (proton pump inhibitors/H 2 -receptor antagonists, corticosteroids, and vasopressor use). Detailed covariate definitions are provided in Table S3. Statistical analysis Baseline patient characteristics were summarized using descriptive statistics; counts (percentages) for categorical variables and medians (interquartile range [IQR]) or means (standard deviation [SD]) for continuous variables. There were no missing data for the variables included in the final model. The sample size calculation is detailed in Methods S3. Fig. 1 A illustrates the study design and trajectory period. We used group-based multi-trajectory modeling (GBMTM) to identify distinct longitudinal trajectories of daily antibiotic exposure and daily colonization pressure through day 5 [ 12 ]. Models with 2–4 trajectory groups were fitted using up to cubic polynomial functions for time through day 5 in the ICU. The optimal model was selected based on the Bayesian Information Criterion (BIC), an average posterior probability of assignment >0.70, odds of correct classification (OCC) >5.0, each group comprising ≥5% of the cohort, and statistical significance of trajectory-shaping polynomial terms (Methods S4). Patients were assigned to their highest-probability trajectory group. Fig. 1. Open in a new tab Study design and patient enrollment flowchart. (A) Study design and timeline. The trajectory period encompassed the first 5 days of ICU admission, during which daily antibiotic exposure (measured by Defined Daily Doses, Length of Therapy, and Antibiotic Spectrum Index) and colonization pressure were quantified for group-based multi-trajectory modeling. Baseline variables were measured at or before ICU admission, while cumulative variables (cumulative invasive device use and medication exposure) were assessed through Day 5. Day 5 served as the landmark for initiating follow-up (left truncation with clock-reset time scale). Rectal surveillance cultures were performed at ICU admission (within 48 h) and weekly thereafter. CR-GNB acquisition events were interval-censored between the last negative surveillance culture and the first positive culture (from either surveillance or clinical specimens). Group-based multi-trajectory modeling identified distinct joint trajectories of antibiotic exposure and colonization pressure, classifying patients into discrete risk groups (Groups 1 through n). (B) Patient enrollment and landmark risk set derivation. Of 2492 consecutive patients admitted to four ICUs from March 2024 to January 2025, 1559 were excluded based on prespecified criteria, yielding 933 unique patients in the initial cohort. An additional 400 patients were deemed inappropriate for landmark analysis (CR-GNB acquisition ≤ Day 5, n = 11; discharged alive ≤ Day 5, n = 341; died ≤ Day 5, n = 48), resulting in 533 patients entering the Day-5 landmark risk set for outcome assessment. During follow-up from Day 5 onward, 124 patients (23.3%) acquired CR-GNB, 387 were discharged alive, and 22 died in the ICU. Abbreviations: CR-GNB, carbapenem-resistant Gram-negative bacteria; ICU, intensive care unit. We fitted continuous-time Markov multi-state models with left truncation on day 5 and a clock-reset time scale to estimate associations between day 5 trajectory groups and subsequent ICU-acquired CR-GNB, accounting for competing discharge and death [ 21 ]. For this analysis, ICU death was defined as death in the ICU or within 48 h of ICU discharge; patients meeting this definition were classified as state 4 (death) rather than state 3 (alive discharge). The model comprised four states: (1) at risk post-landmark (in ICU without CR-GNB on day 5, initial state); (2) ICU-acquired CR-GNB; (3) alive discharge; and (4) ICU death. Transitions were allowed from state 1 to states 2, 3, and 4, with the latter three as absorbing states (Figure S1). Since the exact timing of CR-GNB acquisition was unknown (occurring between the last negative culture and first positive culture), interval censoring was applied for the 1→2 transition, whereas precise observation times were applied for discharge and death events. We fitted three models: an unadjusted model (Model 1), a baseline-adjusted model including only baseline variables measured at or before ICU admission (Model 2), and a fully adjusted model further adding cumulative variables presented in Table 1 (Model 3). All covariates were measured on or before day 5: baseline variables were assessed at or before ICU admission, and cumulative variables were expressed as exposure accrued from hospital admission to ICU day 5. Hazard ratios (HRs) and 95% confidence intervals were estimated using maximum likelihood. Risk differences were calculated from cumulative incidence functions with delta method confidence intervals. The fully adjusted multi-state model converged successfully (convergence code = 0), with log-likelihood = -1339.85 and Akaike Information Criterion = 2739.70. The proportional hazards assumption for the trajectory group effect on CR-GNB acquisition was assessed by including a group × log(time) interaction term in the fully adjusted model; a likelihood ratio test showed no evidence of non-proportionality (χ² = 0.71, P = 0.40). Table 1. Baseline characteristics and outcomes of patients in the Day-5 landmark risk set, stratified by trajectory group. Characteristics Total Trajectory 1 Trajectory 2 Trajectory 3 N 533 182 290 61 Baseline variable Age, years 59.0 (48.0−69.0) 58.0 (47.0−66.0) 59.5 (49.0−70.0) 62.0 (52.0−69.0) Female, No. (%) 226 (42.4) 80 (44.0) 120 (41.4) 26 (42.6) Emergency admission, No. (%) 285 (53.5) 103 (56.6) 149 (51.4) 33 (54.1) ASI per antibiotic day before ICU admission 0 (0−3) 0 (0−3) 0 (0−3) 0 (0−4) DDDs before ICU admission 0 (0−0.1) 0 (0−0.1) 0 (0−0.1) 0 (0−0.2) LOT before ICU admission, days 0 (0−1) 0 (0−1) 0 (0−1) 0 (0−1) Hospitalization in the one month, No. (%) 21 (3.9) 9 (4.9) 8 (2.8) 4 (6.6) Prior antibiotic exposure, No. (%) 88 (16.5) 17 (9.3) 48 (16.6) 23 (37.7) Prior antibiotic count 0 (0−0) 0 (0−0) 0 (0−0) 0 (0−2) Ward transfers count 0 (0−1) 0 (0−1) 0 (0−1) 0 (0−1) Surgery, No. (%) 274 (51.4) 95 (52.2) 155 (53.4) 24 (39.3) Renal failure, No. (%) 21 (3.9) 1 (0.5) 15 (5.2) 5 (8.2) Infection, No. (%) 138 (25.9) 27 (14.8) 78 (26.9) 33 (54.1) Sepsis, No. (%) 34 (6.4) 4 (2.2) 19 (6.6) 11 (18.0) Chronic pulmonary disease, No. (%) 15 (2.8) 2 (1.1) 11 (3.8) 2 (3.3) Liver disease, No. (%) 57 (10.7) 11 (6.0) 35 (12.1) 11 (18.0) Cerebrovascular disease, No. (%) 209 (39.2) 91 (50.0) 105 (36.2) 13 (21.3) Diabetes mellitus, No. (%) 75 (14.1) 23 (12.6) 39 (13.4) 13 (21.3) Elixhauser Comorbidity Index, scores 3.0 (2.0−5.0) 1.0 (1.0−4.0) 3.0 (2.0−4.0) 4.0 (2.0−6.0) SOFA score, scores 3.5 (2.0−5.0) 3.0 (2.0−4.0) 3.5 (2.0−5.0) 4.5 (3.0−6.0) ICU Type, No. (%) Neurosurgical 444 (83.3) 159 (87.4) 240 (82.8) 45 (73.8) Multidisciplinary 53 (9.9) 10 (5.5) 33 (11.4) 10 (16.4) Infectious Disease 36 (6.8) 13 (7.1) 17 (5.9) 6 (9.8) Cumulative variable Central venous catheter, days 1 (0–4) 0 (0–4) 1 (0–4) 2 (0–5) Invasive mechanical ventilation, days 0 (0–3) 0 (0–1) 0 (0–3) 1 (0–2) Urinary catheter, days 1 (0–3) 0 (0–3) 1 (0–3) 2 (0–4) Nasogastric tube, days 0 (0–2) 0 (0–1) 0 (0–2) 1 (0–3) PPI or H 2 RA, days 3 (1–7) 2 (0–4) 3 (1–6) 4 (2–7) Corticosteroids, days 1 (0–3) 1 (0–2) 2 (0–3) 2 (0–5) Vasopressor use, N (%) 133 (25.0) 30 (16.5) 81 (27.9) 22 (36.1) Follow-up characteristics Median follow-up days 4.0 (1.0−10.0) 4.0 (1.0−10.0) 5.0 (1.0−10.0) 3.0 (1.0−11.0) Number of cultures 2.0 (2.0−3.0) 2.0 (2.0−3.0) 2.0 (2.0−3.0) 2.0 (2.0−3.0) Outcome ICU-acquired CR-GNB, No. (%) 124 (23.3) 17 (9.3) 76 (26.2) 31 (50.8) ICU-acquired CRE, No. (%) 56 (10.5) 8 (4.4) 35 (12.1) 13 (21.3) ICU-acquired CRAB, No. (%) 58 (10.9) 8 (4.4) 35 (12.1) 15 (24.6) ICU-acquired CRPA, No. (%) 10 (1.9) 1 (0.5) 6 (2.1) 3 (4.9) Open in a new tab Trajectory groups were identified using group-based multi-trajectory modeling of daily antibiotic exposure and colonization pressure during the first 5 ICU days. Baseline variables were measured at or before ICU admission. Cumulative variables represent exposure accrued from hospital admission through ICU Day 5 and were fixed at the Day-5 landmark for analysis; durations can exceed 5 days if exposure began prior to ICU admission. Follow-up for outcome assessment began at the Day-5 landmark and continued until CR-GNB acquisition, ICU discharge, death, or end of study period. Values expressed as n (%), median (IQR), or mean ± SD as appropriate. Abbreviations: ASI, Antibiotic Spectrum Index; CR-GNB, carbapenem-resistant Gram-negative bacteria; CRAB, carbapenem-resistant Acinetobacter baumannii ; CRE, carbapenem-resistant Enterobacterales; CRPA, carbapenem-resistant Pseudomonas aeruginosa ; DDDs, Defined Daily Doses; H2RA, histamine-2 receptor antagonist; ICU, intensive care unit; IQR, interquartile range; LOT, Length of Therapy; PPI, proton pump inhibitor; SD, standard deviation; SOFA, Sequential Organ Failure Assessment. To assess the robustness of our findings, we conducted sensitivity analyses: (1) restricting the outcome to CRE acquisition; (2) back-dating the at-risk interval for CRAB/CRPA in a worst-case scenario; (3) excluding patients with pre-ICU antibiotic exposure; (4) comparing baseline characteristics between landmark-eligible and ineligible patients; (5) repeating trajectory modeling using 4-day and 6-day observation windows; (6) recalculating daily colonization pressure by counting patients as colonized only from the date of their first positive culture onward; (7) performing a Fine–Gray subdistribution hazard sensitivity analysis; and (8) repeating the multi-state analysis defining death as ICU death only. We also calculated E-values to assess robustness to unmeasured confounding [ 22 ]. Further details can be found in Methods S5. GBMTM was performed using SAS software, version 9.4 (SAS Institute Inc); all other statistical analyses were conducted using the R software, version 4.5.1 (R Foundation for Statistical Computing), with the multi-state models fitted using the ‘msm' package. A 2-sided P < 0.05 was considered statistically significant. Results Baseline characteristics and exposure trajectories We identified an initial cohort of 933 patients, of whom 533 patients entered the Day-5 landmark risk set and were included for analysis ( Fig. 1 B). Of the 400 patients excluded from the Day-5 landmark analysis, 341 were discharged alive before Day 5, 48 died before Day 5, and 11 acquired CR-GNB before Day 5. Baseline characteristics at ICU admission were broadly similar between included and excluded patients (Table S4), with all standardized mean differences <0.15. A three-group multi-trajectory model provided the best balance of fit, classification precision, and clinical interpretability. Compared with the 4-group solution, which yielded a very small additional subgroup (2.6% of patients) and lower classification certainty, the selected 3-group model showed high classification quality (relative entropy, 0.977; group-specific average posterior probabilities, 99.3%, 99.1%, and 97.7% for Trajectories 1, 2, and 3, respectively; all OCC values >90; Tables S5–S6). Full model coefficients and polynomial specifications are reported in Tables S7–S8. As shown in Fig. 2 , Trajectory 1 (n = 182, 34.15%) exhibited persistently low and relatively stable patterns across all exposure metrics; Trajectory 2 (n = 290, 54.41%) demonstrated progressively escalating antibiotic exposure (DDDs, LOT, and ASI) with relatively stable intermediate colonization pressure; and Trajectory 3 (n = 61, 11.44%) was characterized by sustained high-level antibiotic exposure accompanied by progressively increasing colonization pressure. Fig. 2. Open in a new tab Multi-trajectory patterns of antibiotic exposure and colonization pressure during the first 5 ICU days identified by group-based multi-trajectory modeling. Three distinct trajectory groups were identified among 533 patients in the landmark risk set. Trajectory 1 (n = 182, 34.1%) showed stable low-level antibiotic exposure and colonization pressure. Trajectory 2 (n = 290, 54.4%) exhibited progressively increasing antibiotic exposure (DDDs, ASI, and LOT) with relatively stable intermediate colonization pressure. Trajectory 3 (n = 61, 11.4%) demonstrated sustained high-intensity antibiotic exposure coupled with progressively rising colonization pressure. Solid lines represent mean estimated trajectories derived from the multi-trajectory model; shaded areas indicate 95% confidence intervals. Each panel displays a distinct dimension of exposure: DDDs (antibiotic dose), ASI per antibiotic day (antibiotic spectrum weighted by daily use), colonization pressure (daily proportion of ICU patients colonized or infected with CR-GNB), and LOT (antibiotic duration, binary indicator of any systemic antibiotic use per day). Abbreviations: ASI, Antibiotic Spectrum Index; CR-GNB, carbapenem-resistant Gram-negative bacteria; DDDs, Defined Daily Doses; ICU, intensive care unit; LOT, Length of Therapy. Table 1 summarizes baseline characteristics by trajectory group. Compared with Trajectory 1, patients in Trajectory 3 had greater disease severity (median SOFA 4.5 [IQR 3.0–6.0] vs. 3.0 [2.0–4.0]), higher comorbidity burden (median Elixhauser Index 4.0 [2.0–6.0] vs. 1.0 [1.0–4.0]), more frequent prior antibiotic exposure (37.7% vs. 9.3%), and higher cumulative invasive device use. Additional clinical profiles, infection sources at ICU admission, and empirical antibiotic classes by trajectory group are provided in Table S9. Association between exposure trajectories and ICU-acquired CR-GNB Over a median follow-up of 4.0 days (IQR 1.0–10.0) from the Day-5 landmark, 124 of 533 patients (23.3%) acquired CR-GNB, comprising 56 (10.5%) CRE, 58 (10.9%) CRAB, and 10 (1.9%) CRPA. The median interval between consecutive surveillance cultures was 7 days (IQR 6–9 days); the median interval between the last negative and first positive culture among CR-GNB acquirers was 6 days (IQR 4–6), consistent across trajectory groups (Trajectory 1: 5 days; Trajectory 2: 6 days; Trajectory 3: 6 days). Crude acquisition rates demonstrated a steep gradient across trajectory groups: 9.3% (Trajectory 1), 26.2% (Trajectory 2), and 50.8% (Trajectory 3) ( Table 1 ). Fig. 3 displays cumulative incidence curves accounting for competing risks of discharge and death. Clear separation among the three trajectories was evident, with Trajectory 3 exhibiting the most rapid accumulation of CR-GNB events. Dynamic probabilities for all transitions are shown in Figure S3. The overall difference in CR-GNB acquisition across trajectory groups was highly significant (likelihood ratio test, P < 0.001). Fig. 3. Open in a new tab Cumulative incidence of ICU-acquired CR-GNB by trajectory group, accounting for competing risks. Cumulative incidence curves were derived from continuous-time Markov multi-state models with left truncation at Day 5 (landmark) and clock-reset time scale, accounting for competing risks of ICU discharge alive and death. Clear separation among the three trajectory groups was evident, with Trajectory 3 (high antibiotic exposure/high colonization pressure) exhibiting the most rapid accumulation of CR-GNB acquisition events, followed by Trajectory 2 (moderate exposure/intermediate pressure) and Trajectory 1 (low exposure/low pressure). The overall difference in CR-GNB acquisition risk across trajectory groups was highly significant (likelihood ratio test, P < 0.001). Follow-up began at the Day-5 landmark and extended until CR-GNB acquisition, ICU discharge, ICU death, or end of study period. Numbers at risk are shown below as n (cumulative incidence, %). Abbreviations: CR-GNB, carbapenem-resistant Gram-negative bacteria; ICU, intensive care unit; LRT, likelihood ratio test. Table 2 presents hazard ratios and risk differences from continuous-time Markov multi-state models across three models. In the fully adjusted model (Model 3), Trajectory 2 and Trajectory 3 were independently associated with significantly elevated hazards of CR-GNB acquisition compared with Trajectory 1, with adjusted hazard ratios (aHRs) of 1.68 (95% CI, 1.24–2.28) and 2.82 (95% CI, 1.53–5.19), respectively (both P < 0.001). These associations were consistent across all three models. Table 2. Association between trajectories of antibiotic exposure and colonization pressure and risk of ICU-acquired CR-GNB. Trajectory group Model 1 Model 2 Model 3 HR (95%CI) Risk difference, % (95%CI) HR (95%CI) Risk difference, % (95%CI) HR (95%CI) Risk difference, % (95%CI) Trajectory 1 Ref Ref Ref Trajectory 2 1.96 (1.55−2.47) 10.10 (4.43–15.78) 1.72 (1.27–2.32) 6.94 (0.46–13.41) 1.68 (1.24–2.28) 6.52 (0.13–12.97) Trajectory 3 3.84 (2.41−6.10) 24.34 (16.84–31.91) 2.95 (1.62–5.39) 16.64 (7.11–26.17) 2.82 (1.53−5.19) 15.62 (5.82–25.51) Open in a new tab Hazard ratios (HR) and 95% confidence intervals (CI) were estimated using continuous-time Markov multi-state models with left truncation at Day 5 (landmark) and clock-reset time scale, accounting for competing risks of ICU discharge alive and death. Risk differences represent the absolute difference in cumulative incidence between each trajectory group and Trajectory 1 at Day 30, estimated from the multi-state models with 95% CI calculated using the delta method. Trajectory groups were identified by group-based multi-trajectory modeling of daily antibiotic exposure (Defined Daily Doses, Length of Therapy, and Antibiotic Spectrum Index) and colonization pressure during the first 5 ICU days. Model 1 (unadjusted) included trajectory group only. Model 2 (baseline-adjusted) included baseline variables presented in Table 1 measured at or before ICU admission. Model 3 (fully adjusted) included all Model 2 covariates plus cumulative variables presented in Table 1 . Abbreviations: ICU, intensive care unit; CR-GNB, carbapenem-resistant Gram-negative bacteria; HR, hazard ratio; CI, confidence interval. Sensitivity analyses The primary findings remained robust across all sensitivity analyses. Restricting the outcome to CRE acquisition strengthened the associations (Trajectory 3: aHR 4.86, 95% CI 1.93–12.25). Results were consistent under worst-case timing assumptions for CRAB/CRPA events, after excluding patients with pre-ICU antibiotic exposure, and when colonization pressure was recalculated from the first positive culture date only (Table S10). A Fine–Gray subdistribution hazard analysis yielded consistent results (Table S10). Findings were also robust when the competing event of death was restricted to ICU death only (Table S10). Three distinct trajectory groups with similar shapes were consistently identified using 4-day and 6-day observation windows (Figure S2), with hazard ratios remaining stable across windows (Table S11). Discussion This prospective cohort study identified three distinct early exposure trajectories during the first five ICU days that were associated with differential CR-GNB acquisition risk. The highest-risk trajectory—characterized by sustained high antibiotic exposure coupled with elevated colonization pressure—was associated with an approximately three-fold higher hazard of CR-GNB acquisition (aHR, 2.82; 95% CI, 1.53–5.19) compared with the low-exposure baseline, while the moderate-exposure trajectory was independently associated with a 68% higher hazard (aHR, 1.68; 95% CI, 1.24–2.28). These findings suggest that CR-GNB acquisition risk is associated with the joint evolution of these factors, rather than individual factors in isolation during this early period. Our findings are consistent with a conceptual framework in which CR-GNB acquisition may be influenced by both external pathogen exposure and internal host susceptibility. Colonization pressure represents the external component, with higher levels increasing transmission probability for CRE, CRAB, and CRPA [ 7 , 8 , 11 , 20 ]; however, establishing colonization may also require host vulnerability; antibiotic exposure may disrupt commensal communities that inhibit pathogen proliferation via nutrient competition and antimicrobial production [ 23 , 24 ], potentially facilitating CR-GNB establishment [ [24] , [25] , [26] , [27] ]. The extent of ecological disruption may correlate with spectrum breadth and exposure intensity, and sub-inhibitory concentrations in critically ill patients with pharmacokinetic variability may further facilitate resistance selection [ [28] , [29] , [30] , [31] ]. Within this framework, the observed risk gradient across trajectories is biologically plausible: Trajectory 3 may represent patients requiring early therapeutic escalation whose intensive antibiotic exposure coincides with high colonization pressure, while Trajectory 2—the most prevalent pattern—may reflect progressive ecological disruption under escalating exposure with intermediate colonization pressure. Our approach aligns with emerging evidence that the antibiotic–resistance association is time-varying [ 32 ], that multidrug-resistant organism colonization reflects complex networks of microbial and antibiotic interactions [ 33 ], and that gut microbiota composition is associated with colonization resistance against carbapenem-resistant pathogens in ICU settings [ 34 , 35 ]. However, these mechanistic interpretations remain speculative within our observational design. The observed risk gradient across trajectory groups may partly reflect residual differences in unmeasured illness severity rather than independent biological effects of the exposure patterns themselves. A principal contribution of this study is its reframing of CR-GNB acquisition risk from static to dynamic, trajectory-based assessment. Although antibiotic exposure is an established risk factor, conventional analyses typically quantified cumulative exposure (e.g., total days of therapy), masking heterogeneity in temporal patterns [ 7 , 9 , 10 ]. By explicitly capturing heterogeneity in joint trajectories of antibiotic exposure and colonization pressure during the early ICU period, our approach suggests that distinct exposure patterns—not just cumulative magnitude—are associated with CR-GNB acquisition risk. The selection of day 5 as the trajectory endpoint was guided by converging clinical, methodological, and empirical considerations. Clinically, day 5 falls shortly after the guideline-recommended 48–72-h antimicrobial review window, providing a natural decision point when treatment patterns have stabilized [ 36 , 37 ]. Systematic reviews indicate that most ICU-acquired multidrug-resistant organisms have acquisition times ranging from 4 to 26 days [ 38 ]; indeed, only 11 of 135 CR-GNB acquisitions (8.1%) in our cohort occurred before day 5, confirming that most events remain observable during follow-up. Methodologically, the day-5 landmark balances trajectory identifiability against sample retention and selection bias, while providing a clinically relevant window: trajectories are identifiable early enough for risk stratification, yet late enough that patterns have stabilized and are prognostically meaningful. Whereas previous studies analyzed these factors as static or cumulative variables over the entire ICU stay [ 7 , [9] , [10] , [11] ], our approach suggests that their temporal evolution during this early period is associated with risk, enabling prospective risk stratification before most acquisitions occur. Although the trajectory–outcome associations reported here are observational and do not establish causality, the trajectory-based risk stratification framework may have clinical implications if validated externally: early identification of high-risk trajectories within the first five ICU days may offer a clinically relevant window for risk-adapted intervention. For patients following Trajectory 3, characterized by sustained high antibiotic exposure and elevated colonization pressure, strategies could include intensified infection control (enhanced contact precautions, dedicated nursing, increased surveillance frequency) [ 39 , 40 ] and antimicrobial stewardship interventions such as de-escalation to narrower-spectrum agents when microbiologic data permit and source control optimization to facilitate treatment shortening. For Trajectory 2—the most prevalent pattern, characterized by escalating exposure—routine 48–72-h antimicrobial stewardship review, already recommended by guidelines but inconsistently implemented, could identify opportunities to avoid further escalation through spectrum narrowing or early discontinuation when clinical improvement allows [ 36 , 37 ]. Given that Trajectory 2 represents the majority of ICU patients, even modest risk reduction through improved stewardship could have meaningful population-level implications. At present, however, these trajectories should be regarded primarily as early prognostic markers rather than as directly actionable treatment targets, and prospective validation is needed to determine whether trajectory-guided interventions can reduce CR-GNB acquisition. Future implementation would require integration of real-time exposure metrics and unit-level colonization pressure into electronic health record systems to support bedside decision-making [ 41 ]. The required data elements—daily antibiotic administrations, ICU census, and surveillance culture results—are routinely captured in hospital information systems, enabling automated trajectory calculation without additional data collection burden. A potential pathway would involve developing validated algorithms to compute trajectory metrics from existing data, creating clinical decision support dashboards that identify high-risk trajectory patterns by day 5, and linking alerts to prespecified intervention protocols. Before clinical adoption, however, external validation in diverse ICU settings with different patient populations and resistance epidemiology, and further assessment of predictive performance, are essential to establish generalizability, and the effectiveness and cost-effectiveness of trajectory-guided interventions require prospective evaluation in pragmatic cluster-randomized trials [ 42 ]. This study has several methodological strengths, including a prospective design with systematic active surveillance cultures that enabled sensitive CR-GNB acquisition detection—including asymptomatic colonization—while examining temporal relationships between early exposures and outcomes. Antibiotic exposure assessment incorporated multiple ecologically relevant dimensions (dose, duration, and spectrum) rather than simple metrics. The multi-state modeling framework accounted for competing risks and interval-censored observations, and findings were consistent across sensitivity analyses, including restriction to CRE outcomes, worst-case timing scenarios for non-fermenter detection, and exclusion of patients with pre-ICU antibiotic exposure. Several limitations merit consideration. First, as a single-center study in China, local practices and resistance epidemiology may limit generalizability, warranting multicenter validation. Second, end-of-life discharge shortly before death [ 43 ] may lead to under-ascertainment of ICU mortality; however, trajectory–CR-GNB associations were robust to alternative competing event definitions. Third, despite extensive adjustment, residual confounding by indication may persist—unmeasured illness severity may independently drive both escalating antibiotic use and CR-GNB susceptibility. Importantly, the observed trajectories may partly reflect early clinical deterioration rather than represent independent causal exposures, as sicker patients may simultaneously require more intensive therapy and face higher acquisition risk. Although we adjusted for baseline severity measures (SOFA, Elixhauser, infection, sepsis, vasopressor use) and cumulative day-5 device use, we did not incorporate additional time-varying severity markers, as such variables may lie on the causal pathway between clinical deterioration and trajectory classification, risking overadjustment. To clarify the dual role of cumulative Day-5 variables, we constructed directed acyclic graphs illustrating two competing assumptions (Figure S4): a confounding pathway, in which these variables are driven by disease severity and independently affect both trajectory group assignment and CR-GNB acquisition risk (favoring Model 3 as the primary analysis), and a mediation pathway, in which they mediate the trajectory–outcome effect (favoring Model 2 to avoid overadjustment). Empirical comparison showed <5% change in adjusted hazard ratios between Model 2 and Model 3, consistent with the confounding structure and arguing against substantial mediation. E-values of 3.48 (lower CI limit 2.02) for Trajectory 3 and 2.22 (lower CI limit 1.59) for Trajectory 2 suggest that moderate-to-strong unmeasured confounding would be needed to fully explain these associations; however, E-values do not rule out residual confounding, and the observed trajectory–outcome associations should be interpreted as observational rather than causal. Accordingly, our results should not be taken as evidence that modifying a patient's trajectory would necessarily reduce CR-GNB acquisition risk. Fourth, we did not perform resampling-based validation (e.g., bootstrap or subsampling) to assess the stability of trajectory classification across repeated samples, which may particularly limit confidence in the smaller trajectory subgroup (Trajectory 3, n = 61). Fifth, the Day-5 landmark design conditions the analysis on survival, ongoing ICU stay, and event-free status through Day 5, which may introduce selection bias. Although most exclusions were due to early discharge and baseline characteristics were similar between included and excluded patients, selection on unmeasured factors cannot be excluded; consequently, our findings are most directly generalizable to patients who remained alive, in the ICU, and CR-GNB-free at the Day-5 landmark. Sixth, trajectory classification covered only the first five ICU days; extending classification could amplify inter-group differences, although this early window enables actionable risk stratification before most acquisitions occur. Seventh, several measurement limitations should be noted: weekly surveillance introduced uncertainty in acquisition timing, though the median acquisition window was 6 days and similar across groups—consistent with non-differential misclassification that would attenuate estimates. Additionally, rectal surveillance sensitivity differs by organism: sensitivities of 76–97% are reported for CRE [ 44 , 45 ], but yield is substantially lower for non-fermenters (∼47% for CRAB [ 46 ]), and the optimal sampling strategy for CRPA remains uncertain [ 47 ]; consequently, some CRAB/CRPA acquisitions may have been detected later or incompletely. However, findings were robust under worst-case timing assumptions for CRAB/CRPA and when restricting the outcome to CRE only. Weekly sampling may have underestimated daily colonization prevalence, though midpoint imputation and sensitivity analysis using first positive culture date yielded consistent results. Colonization pressure was calculated at the aggregate CR-GNB level rather than species-specifically, potentially diluting organism-specific effects, although aggregate pressure serves as a pragmatic marker of overall ecological burden and interspecies carbapenemase transfer (e.g., blaNDM-1) has been documented [ 48 ]. Finally, the ASI, while quantifying spectrum breadth, is an imperfect surrogate for ecological impact, as agents with lower scores may disproportionately disrupt gut colonization resistance. Conclusions Our findings support a shift from static risk assessment to a dynamic, trajectory-based approach for early association-based risk stratification of ICU-acquired CR-GNB. By identifying high-risk patterns within the first five days of care, this framework may help inform earlier risk-adapted prevention strategies. However, external validation and further assessment of predictive performance are required before clinical implementation. Authors' contribution WHZ, JLJ and YQ conceived and designed the study, and critically revised the manuscript. ZHC, XRY, and JW (Jing Wu) contributed equally as co-first authors to study execution, data acquisition, and analysis. ZHL, SW, JW (Jing Wang), LZ, and JN contributed to data acquisition and data cleaning. YRT and SRL performed the microbiological analyses. ZHC, XRY, and JW (Jing Wu) performed the statistical analysis and were major contributors in writing the manuscript. All authors contributed to data interpretation and critical revision of the manuscript for important intellectual content. All authors read and approved the final manuscript. Consent for publication Not applicable. Ethics declarations The study protocol (KY2024-060) was approved by the Institutional Ethics Committee of Huashan Hospital, Fudan University, and the study was registered with the Chinese Clinical Trial Registry (ChiCTR2400081352). Written informed consent was obtained from all participants or their legally authorized representatives. Funding This research was supported by grants from the National Key Research and Development Program of China (2022YFC2009802) and the Taicang First People's Hospital Internal Research Project (2025-IIT-074). Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declaration of competing interest The authors declare that they have no conflicts of interest. Acknowledgements Not applicable. Footnotes Appendix A Supplementary material related to this article can be found, in the online version, at doi: https://doi.org/10.1016/j.aicoj.2026.100063 . Contributor Information Yue Qu, Email: [email protected]. Jialin Jin, Email: [email protected]. Wenhong Zhang, Email: [email protected]. Appendix A. Supplementary data The following is Supplementary data to this article: mmc1.pdf (1.1MB, pdf) References 1. Li Q., Zhou X., Yang R., Shen X., Li G., Zhang C., et al. Carbapenem-resistant Gram-negative bacteria (CR-GNB) in ICUs: resistance genes, therapeutics, and prevention-a comprehensive review. Front Public Health. 2024;12 doi: 10.3389/fpubh.2024.1376513. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Sati H., Carrara E., Savoldi A., Hansen P., Garlasco J., Campagnaro E., et al. The WHO Bacterial Priority Pathogens List 2024: a prioritisation study to guide research, development, and public health strategies against antimicrobial resistance. Lancet Infect Dis. 2025;25(9):1033–1043. doi: 10.1016/S1473-3099(25)00118-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Zha L., Li S., Guo J., Hu Y., Pan L., Wang H., et al. Global and regional burden of bloodstream infections caused by carbapenem-resistant Gram-negative bacteria in 2019: a systematic analysis from the MICROBE database. Int J Infect Dis. 2025;153 doi: 10.1016/j.ijid.2024.107769. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Karakonstantis S., Kritsotakis E.I., Tziolos R.N., Vassilopoulou L., Loukaki M., Kypraiou D., et al. Mortality due to carbapenem-resistant Acinetobacter baumannii bacteraemia: a 5-year cohort study in intensive care patients. Clin Microbiol Infect. 2025;31(6):1033–1039. doi: 10.1016/j.cmi.2025.02.018. [ DOI ] [ PubMed ] [ Google Scholar ] 5. Bartsch S.M., McKinnell J.A., Mueller L.E., Miller L.G., Gohil S.K., Huang S.S., et al. Potential economic burden of carbapenem-resistant Enterobacteriaceae (CRE) in the United States. Clin Microbiol Infect. 2017;23(1):48.e9–48.e16. doi: 10.1016/j.cmi.2016.09.003. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Willmann M., Vehreschild M., Biehl L.M., Vogel W., Dörfel D., Hamprecht A., et al. Distinct impact of antibiotics on the gut microbiome and resistome: a longitudinal multicenter cohort study. BMC Biol. 2019;17(1):76. doi: 10.1186/s12915-019-0692-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Hassoun-Kheir N., Hussein K., Karram M., Saffuri M., Badaan S., Peleg S., et al. Risk factors for acquisition of carbapenemase-producing versus non-carbapenemase-producing Enterobacterales: a case-control study. Clin Microbiol Infect. 2023;29(5):629–634. doi: 10.1016/j.cmi.2023.01.005. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Bonten M.J.M. Colonization pressure: a critical parameter in the epidemiology of antibiotic-resistant bacteria. Crit Care. 2012;16(4):142. doi: 10.1186/cc11417. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Marchenay P., Blasco G., Navellou J.C., Leroy J., Cholley P., Talon D., et al. Acquisition of carbapenem-resistant Gram-negative bacilli in intensive care unit: predictors and molecular epidemiology. Med Mal Infect. 2015;45(1-2):34–40. doi: 10.1016/j.medmal.2014.12.003. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Coppry M., Jeanne-Leroyer C., Noize P., Dumartin C., Boyer A., Bertrand X., et al. Antibiotics associated with acquisition of carbapenem-resistant Pseudomonas aeruginosa in ICUs: a multicentre nested case-case-control study. J Antimicrob Chemother. 2019;74(2):503–510. doi: 10.1093/jac/dky427. [ DOI ] [ PubMed ] [ Google Scholar ] 11. DalBen M.F., Basso M., Garcia C.P., Costa S.F., Toscano C.M., Jarvis W.R., et al. Colonization pressure as a risk factor for colonization by multiresistant Acinetobacter spp and carbapenem-resistant Pseudomonas aeruginosa in an intensive care unit. Clinics (Sao Paulo). 2013;68(8):1128–1133. doi: 10.6061/clinics/2013(08)11. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Nagin D.S., Jones B.L., Passos V.L., Tremblay R.E. Group-based multi-trajectory modeling. Stat Methods Med Res. 2018;27(7):2015–2023. doi: 10.1177/0962280216673085. [ DOI ] [ PubMed ] [ Google Scholar ] 13. Harris P.A., Taylor R., Thielke R., Payne J., Gonzalez N., Conde J.G. Research electronic data capture (REDCap)-a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42(2):377–381. doi: 10.1016/j.jbi.2008.08.010. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. STROBE Initiative. von Elm E., Altman D.G., Egger M., Pocock S.J., Gøtzsche P.C., Vandenbroucke J.P. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453–1457. doi: 10.1016/S0140-6736(07)61602-X. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Nicolaie M.A., van Houwelingen J.C., de Witte T.M., Putter H. Dynamic prediction by landmarking in competing risks. Stat Med. 2013;32(12):2031–2047. doi: 10.1002/sim.5665. [ DOI ] [ PubMed ] [ Google Scholar ] 16. WHO Collaborating Centre for Drug Statistics Methodology . 2025. ATC/DDD Index [Internet]. Oslo: Norwegian Institute of Public Health. [updated 2025 Jan 1; cited 2025 Oct 7]. Available from: https://atcddd.fhi.no/atc_ddd_index/ [ Google Scholar ] 17. Bennett N., Schulz L., Boyd S., Newland J.G. Understanding inpatient antimicrobial stewardship metrics. Am J Health Syst Pharm. 2018;75(4):230–238. doi: 10.2146/ajhp160335. [ DOI ] [ PubMed ] [ Google Scholar ] 18. Gerber J.S., Hersh A.L., Kronman M.P., Newland J.G., Ross R.K., Metjian T.A. Development and application of an Antibiotic Spectrum Index for benchmarking Antibiotic selection patterns across hospitals. Infect Control Hosp Epidemiol. 2017;38(8):993–997. doi: 10.1017/ice.2017.94. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Ray M.J., Strnad L.C., Tucker K.J., Furuno J.P., Lofgren E.T., McCracken C.M., et al. Influence of antibiotic exposure intensity on the risk of Clostridioides difficile infection. Clin Infect Dis. 2024;79(5):1129–1135. doi: 10.1093/cid/ciae259. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Arvaniti K., Lathyris D., Ruimy R., Haidich A.B., Koulourida V., Nikolaidis P., et al. The importance of colonization pressure in multiresistant Acinetobacter baumannii acquisition in a Greek intensive care unit. Crit Care. 2012;16(3) doi: 10.1186/cc11383. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Hazard D., Grodd M., Walzer D., Biever P., Rieg S., Wolkewitz M. How to avoid time-related types of bias in the analysis of clinical infectious diseases: demonstration and methods. Clin Microbiol Infect. 2025;31(9):1467–1474. doi: 10.1016/j.cmi.2025.04.005. [ DOI ] [ PubMed ] [ Google Scholar ] 22. VanderWeele T.J., Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med. 2017;167(4):268–274. doi: 10.7326/M16-2607. [ DOI ] [ PubMed ] [ Google Scholar ] 23. Cusumano G., Flores G.A., Venanzoni R., Angelini P. The impact of antibiotic therapy on intestinal microbiota: dysbiosis, antibiotic resistance, and restoration strategies. Antibiotics (Basel). 2025;14(4) doi: 10.3390/antibiotics14040371. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Yip A.Y.G., King O.G., Omelchenko O., Kurkimat S., Horrocks V., Mostyn P., et al. Antibiotics promote intestinal growth of carbapenem-resistant Enterobacteriaceae by enriching nutrients and depleting microbial metabolites. Nat Commun. 2023;14(1) doi: 10.1038/s41467-023-40872-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Fishbein S.R.S., Mahmud B., Dantas G. Antibiotic perturbations to the gut microbiome. Nat Rev Microbiol. 2023;21(12):772–788. doi: 10.1038/s41579-023-00933-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Lawley T.D., Walker A.W. Intestinal colonization resistance. Immunology. 2013;138(1):1–11. doi: 10.1111/j.1365-2567.2012.03616.x. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Lozupone C.A., Stombaugh J.I., Gordon J.I., Jansson J.K., Knight R. Diversity, stability and resilience of the human gut microbiota. Nature. 2012;489(7415):220–230. doi: 10.1038/nature11550. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Koulenti D., Roger C., Lipman J. Antibiotic dosing optimization in critically ill patients. Intensive Care Med. 2025;51:603–606. doi: 10.1007/s00134-025-07822-3. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Drlica K., Zhao X. Mutant selection window hypothesis updated. Clin Infect Dis. 2007;44(5):681–688. doi: 10.1086/511642. [ DOI ] [ PubMed ] [ Google Scholar ] 30. Jutkina J., Marathe N.P., Flach C.F., Larsson D.G.J. Antibiotics and common antibacterial biocides stimulate horizontal transfer of resistance at low concentrations. Sci Total Environ. 2018;616-617:172–178. doi: 10.1016/j.scitotenv.2017.10.312. [ DOI ] [ PubMed ] [ Google Scholar ] 31. Ding M., Ye Z., Liu L., Wang W., Chen Q., Zhang F., et al. Subinhibitory antibiotic concentrations promote the horizontal transfer of plasmid-borne resistance genes from Klebsiella pneumoniae to Escherichia coli. Front Microbiol. 2022;13 doi: 10.3389/fmicb.2022.1017092. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Baraz A., Chowers M., Nevo D., Obolski U. The time-varying association between previous antibiotic use and antibiotic resistance. Clin Microbiol Infect. 2023;29(3):390.e1–390.e4. doi: 10.1016/j.cmi.2022.10.021. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Wang J., Foxman B., Mody L., Snitkin E.S. Network of microbial and antibiotic interactions drive colonization and infection with multidrug-resistant organisms. Proc Natl Acad Sci USA. 2017;114(39):10467–10472. doi: 10.1073/pnas.1710235114. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Yang J., Zhou Y., Du A., Zhang Z., Wang B., Tian Y., et al. Microbiome-mediated colonization resistance to carbapenem-resistant Klebsiella pneumoniae in ICU patients. NPJ Biofilms Microbiomes. 2025;11(1):157. doi: 10.1038/s41522-025-00791-x. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Garcia E.R., Vergara A., Aziz F., Narváez S., Cuesta G., Hernández M., et al. Changes in the gut microbiota and risk of colonization by multidrug-resistant bacteria, infection, and death in critical care patients. Clin Microbiol Infect. 2022;28(7):975–982. doi: 10.1016/j.cmi.2022.01.004. [ DOI ] [ PubMed ] [ Google Scholar ] 36. Barlam T.F., Cosgrove S.E., Abbo L.M., MacDougall C., Schuetz A.N., Septimus E.J., et al. Implementing an antibiotic stewardship program: guidelines by the Infectious Diseases Society of America and the Society for Healthcare Epidemiology of America. Clin Infect Dis. 2016;62(10):e51–e77. doi: 10.1093/cid/ciw118. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. UK Health Security Agency . UK Health Security Agency; London: 2011. Start smart then focus: antimicrobial stewardship toolkit for inpatient care settings [Internet] https://www.gov.uk/government/publications/antimicrobial-stewardship-start-smart-then-focus/start-smart-then-focus-antimicrobial-stewardship-toolkit-for-inpatient-care-settings [updated 2023 Sep 12; cited 2025 Oct 10]. Available from: [ Google Scholar ] 38. Heath M.R., Fan W., Leu C.S., Gomez-Simmonds A., Lodise T., Freedberg D.E. Gut colonization with multidrug-resistant organisms in the intensive care unit: a systematic review and meta-analysis. Crit Care. 2024;28:211. doi: 10.1186/s13054-024-04999-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Health Care Infection Control Practices Advisory Committee. Siegel J.D., Rhinehart E., Jackson M., Chiarello L. 2007 Guideline for isolation precautions: preventing transmission of infectious agents in health care settings. Am J Infect Control. 2007;35(10 Suppl 2):S65–S164. doi: 10.1016/j.ajic.2007.10.007. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Healthcare Infection Control Practices Advisory Committee. Siegel J.D., Rhinehart E., Jackson M., Chiarello L. Management of multidrug-resistant organisms in health care settings, 2006. Am J Infect Control. 2007;35(10 Suppl 2):S165–S193. doi: 10.1016/j.ajic.2007.10.006. [ DOI ] [ PubMed ] [ Google Scholar ] 41. Kuper K.M., Nagel J.L., Kile J.W., May L.S., Lee F.M. The role of electronic health record and "add-on" clinical decision support systems to enhance antimicrobial stewardship programs. Infect Control Hosp Epidemiol. 2019;40(5):501–511. doi: 10.1017/ice.2019.51. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Davey P., Marwick C.A., Scott C.L., Charani E., McNeil K., Brown E., et al. Interventions to improve antibiotic prescribing practices for hospital inpatients. Cochrane Database Syst Rev. 2017;(2) doi: 10.1002/14651858.CD003543.pub4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Weng L., Hu Y., Sun Z., Yu C., Guo Y., Pei P., et al. Place of death and phenomenon of going home to die in Chinese adults: a prospective cohort study. Lancet Reg Health West Pac. 2021;18 doi: 10.1016/j.lanwpc.2021.100301. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Adler A., Navon-Venezia S., Moran-Gilad J., Marcos E., Schwartz D., Carmeli Y. Laboratory and clinical evaluation of screening agar plates for detection of carbapenem-resistant Enterobacteriaceae from surveillance rectal swabs. J Clin Microbiol. 2011;49(6):2239–2242. doi: 10.1128/JCM.02566-10. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Pournaras S., Zarkotou O., Poulou A., Kristo I., Vrioni G., Themeli-Digalaki K., et al. A combined disk test for direct differentiation of carbapenemase-producing Enterobacteriaceae in surveillance rectal swabs. J Clin Microbiol. 2013;51(9):2986–2990. doi: 10.1128/JCM.00901-13. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Nutman A., Temkin E., Lellouche J., Ben David D., Schwartz D., Carmeli Y. Detecting carbapenem-resistant Acinetobacter baumannii (CRAB) carriage: which body site should be cultured? Infect Control Hosp Epidemiol. 2020;41(8):965–967. doi: 10.1017/ice.2020.197. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Shahab S.N., van Veen A., Büchler A.C., Saharman Y.R., Karuniawati A., Vos M.C., et al. In search of the best method to detect carriage of carbapenem-resistant Pseudomonas aeruginosa in humans: a systematic review. Ann Clin Microbiol Antimicrob. 2024;23(1):50. doi: 10.1186/s12941-024-00707-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Weber R.E., Pietsch M., Frühauf A., Pfeifer Y., Martin M., Luft D., et al. IS26-mediated transfer of blaNDM-1 as the main route of resistance transmission during a polyclonal, multispecies outbreak in a German hospital. Front Microbiol. 2019;10 doi: 10.3389/fmicb.2019.02817. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials mmc1.pdf (1.1MB, pdf) Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Articles from Annals of Intensive Care are provided here courtesy of Elsevier ACTIONS View on publisher site PDF (365.7 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top