Antibiotic Spectrum, Not Dose or Duration, Drives ICU Carbapenem Resistance Risk

Antibiotic exposure is a well-established risk factor for acquiring carbapenem-resistant Gram-negative bacteria (CR-GNB) in the intensive care unit, but the relative weight of its three core dimensions—spectrum, dose, and duration—has remained poorly resolved. A prospective cohort study now provides a direct comparison, and its central finding is striking: the breadth of the antibiotic spectrum, quantified by the Antibiotic Spectrum Index per antibiotic day (ASI), emerges as the principal independent driver of ICU-acquired CR-GNB, significantly outweighing both cumulative dose and length of therapy once all three are modeled together. The study, conducted at Huashan Hospital affiliated to Fudan University in Shanghai, China, was published in Annals of Intensive Care.

The research team prospectively enrolled consecutive adult patients admitted to four ICUs at the university hospital between March 2024 and January 2025. 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 hours of ICU admission and weekly thereafter, with clinical cultures obtained when infection was suspected. Patients with CR-GNB already detected before or within 48 hours of admission were excluded, as were those with fewer than two surveillance cultures, leaving 422 adults in the final analysis. The cohort had a median age of 59.0 years, included 40.3% female patients, and was dominated by the neurosurgical ICU, which accounted for 82.2% of admissions; 57.6% of admissions were emergencies.

During a median follow-up of 12.0 days (interquartile range, 8.0–17.0), 151 of the 422 patients—35.8%—acquired CR-GNB during their ICU stay. The most frequently acquired organisms were CRE, accounting for 49.0% of acquisitions, followed by CRAB at 43.7% and CRPA at 7.3%. Acquisition was defined as the first detection of CR-GNB, whether infection or colonization, occurring at least 48 hours after ICU admission in patients whose initial screening cultures were negative. Because surveillance sampling was interval-based, the timing of acquisition was treated as interval-censored between the last negative and the first positive culture, and patients who never acquired CR-GNB were right-censored at their last negative surveillance culture.

The methodological core of the study lies in how it quantified antibiotic exposure. Dose was measured as Defined Daily Doses (DDDs) according to World Health Organization methodology; duration was measured as Length of Therapy (LOT), the number of days a patient received at least one systemic antibiotic; and spectrum was measured as the Antibiotic Spectrum Index per antibiotic day, a standardized quantitative score of antibacterial breadth. Exposure data were extracted from electronic medical records, covering antibiotic name, dose, frequency, and administration start and end dates, from admission until ICU discharge or the first CR-GNB-positive culture, whichever came first. The investigators then fitted interval-censored Cox proportional hazards models incorporating time-dependent covariates—such as durations of invasive device use, colonization pressure, and use of proton pump inhibitors, histamine-2 receptor antagonists, or corticosteroids—using an Expectation-Maximization algorithm, and explored non-linear relationships with restricted cubic splines.

The results were unambiguous for spectrum. In the fully adjusted model, which included the other two exposure metrics, each 1-unit increase in ASI per antibiotic day was associated with a 14% increase in the risk of ICU-acquired CR-GNB (adjusted hazard ratio 1.14; 95% confidence interval 1.09–1.19; P < 0.001). Tertile analysis reinforced the gradient: compared with patients in the lowest ASI tertile (below 5), those in the highest tertile had nearly five-fold higher risk (hazard ratio 4.90; 95% CI 2.60–9.23), while the middle tertile showed a non-significant trend (hazard ratio 1.73; 95% CI 0.89–3.35). Restricted cubic spline modeling revealed a significant non-linear, J-shaped association (P for nonlinearity = 0.027), with risk plateauing or slightly decreasing in the lower ASI range of approximately 0 to 5 before climbing steadily—a pattern the authors interpret as a threshold effect around an ASI of 5.

Dose told a different story. DDDs were significantly associated with CR-GNB acquisition in unadjusted analysis (hazard ratio 1.64; 95% CI 1.42–1.89) and remained significant after adjustment for baseline and time-dependent covariates (hazard ratio 1.34; 95% CI 1.14–1.59). But once ASI and LOT were added to the model, the association disappeared entirely (adjusted hazard ratio 0.89; 95% CI 0.69–1.15; P = 0.374). The spline analysis nonetheless showed a notable inverted U-shaped relationship (P for nonlinearity < 0.001), with risk rising rapidly with increasing DDDs, peaking at approximately 2 to 3 DDDs, and then declining at higher levels. The authors connect this pattern to the mutant selection window hypothesis, under which moderate doses may suppress susceptible strains while allowing resistant mutants to amplify, whereas very high doses may suppress both populations more effectively.

Duration followed a similar trajectory of attenuation. LOT was marginally significant in unadjusted analysis (hazard ratio 1.04; 95% CI 1.00–1.08; P = 0.043) but lost statistical significance in both the partially adjusted model (hazard ratio 1.04; 95% CI 0.98–1.11; P = 0.175) and the fully adjusted model (hazard ratio 1.03; 95% CI 0.97–1.11; P = 0.214). Spline analysis still suggested a non-linear trend of increasing risk with longer therapy (P for nonlinearity < 0.001), and patients in the longest LOT tertile showed a hazard ratio of 2.31 (95% CI 0.98–5.71) that bordered on statistical significance. The authors note that the duration–resistance relationship is increasingly recognized as complex and context-dependent rather than a simple positive correlation, citing modeling work suggesting that shortening therapy can sometimes increase resistance carriage depending on whether the antibiotic clears colonizing resistant bacteria.

Relative importance analysis, quantified using the partial χ² statistic minus predictor degrees of freedom, placed ASI per antibiotic day clearly at the top of the model (importance 25.92; P < 0.001), while LOT (importance 0.46; P = 0.228) and DDDs (importance −0.28; P = 0.397) contributed far less and did not reach significance. Sensitivity analyses supported the robustness of the primary finding: the ASI association persisted when the outcome was restricted to CRE only (adjusted hazard ratio 1.13; 95% CI 1.08–1.18), when a standard Cox model replaced interval censoring (adjusted hazard ratio 1.11; 95% CI 1.07–1.15), when the CRAB/CRPA interval start time was shifted earlier in a worst-case scenario (adjusted hazard ratio 1.07; 95% CI 1.03–1.22), when patients with antibiotic use in the prior 30 days were excluded (adjusted hazard ratio 1.14; 95% CI 1.07–1.41), and when outcomes were stratified into colonization only (adjusted hazard ratio 1.15; 95% CI 1.07–1.26) and clinical infection only (adjusted hazard ratio 1.10; 95% CI 1.02–1.33). Associations for DDDs and LOT showed more variability across these analyses.

Mechanistically, the authors argue that broad-spectrum agents exert potent selection pressure by eradicating susceptible competitors and disrupting the gut microbiome, thereby diminishing colonization resistance and facilitating CR-GNB establishment. They suggest that antibiotic stewardship programs should prioritize optimizing spectrum selection—through appropriate initial antibiotic choice guided by rapid diagnostics, local resistance patterns, and clinical prediction tools—potentially yielding greater impact on CR-GNB acquisition than interventions focused solely on dose limitation or duration shortening. The analysis also identified ICU type as a powerful independent predictor, which the authors interpret as a proxy for unmeasured unit-level factors such as local microbial ecology and infection control practices.

The study has clear limitations. It was conducted at a single center, so the results may reflect that hospital’s prescribing practices and local microbial ecology, limiting generalizability despite the inclusion of three ICU types. Rectal surveillance cultures may have lower sensitivity for detecting non-fermentative organisms such as CRAB and CRPA compared with CRE, although sensitivity analyses accounting for potential false negatives yielded consistent results. Residual confounding from unmeasured factors cannot be excluded despite comprehensive adjustment, and the DDD metric reflects prescribed therapy rather than actual drug exposure, which in critically ill patients can deviate substantially due to pharmacokinetic variability. The study was registered with the Chinese Clinical Trial Registry (ChiCTR2400081352). Within these constraints, the message for stewardship is direct: when the goal is preventing carbapenem-resistant Gram-negative bacteria in the ICU, narrowing the antibiotic spectrum may matter more than trimming doses or shortening courses.

Subject of Research: Antibiotic spectrum, dose, and duration as risk factors for ICU-acquired carbapenem-resistant Gram-negative bacteria

Article Title: Spectrum, dose, and duration of antibiotic exposure and risk of intensive care unit–acquired carbapenem-resistant gram-negative bacteria: a prospective cohort study

Article References: Chen, Z., Wu, J., Ye, X., Li, Z., JingWang, Tian, Y., Zhou, L., Ni, J., Jin, J., & Zhang, W. (2025). Spectrum, dose, and duration of antibiotic exposure and risk of intensive care unit–acquired carbapenem-resistant gram-negative bacteria: a prospective cohort study. Annals of Intensive Care, 15(1), Article 190. https://doi.org/10.1186/s13613-025-01605-1

Image Credits: AI Generated

DOI: 10.1186/s13613-025-01605-1

Keywords: antibiotic resistance, carbapenem-resistant Gram-negative bacteria, intensive care unit, antibiotic stewardship, Antibiotic Spectrum Index, prospective cohort study, infection control, CRE, CRAB, CRPA, interval-censored Cox regression, defined daily doses