This was a retrospective cohort study using the Medical Information Mart for Intensive Care-IV (MIMIC-IV v3.1), a publicly available critical care database containing 94,458 ICU hospitalization records from 65,366 unique patients. Database access was authorized following completion of the Collaborative Institutional Training Initiative (CITI) program (certification ID: 12259631) by the investigator. As all patient identifiers have been anonymized, institutional ethics committee approval and informed consent were waived.18

Patients diagnosed with acute pancreatitis admitted to the ICU were included. SAP was defined by the following criteria: (1) primary discharge diagnosis of AP (ICD-9: 577.0; ICD-10: K85.x); (2) Sequential Organ Failure Assessment (SOFA) score ≥2, persisting for >48 h, with organ failure involving at least one system (respiratory, circulatory, or renal), meeting the revised Atlanta classification criteria.19 Exclusion criteria included: (1) ICU stay <48 h; (2) age <18 years; (3) initiation of antibiotic treatment for a confirmed infection ≥24 h prior to ICU admission; (4) recurrent pancreatitis with prior SAP-related ICU admission; (5) major active infection at ICU admission (e.g., severe pneumonia, urinary tract infection, or sepsis); (6) initiation of anti-MRSA therapy beyond 72 h; and (7) incomplete baseline data (<80%).

The primary exposure was defined as intravenous administration of anti-MRSA agents initiated within 72 h of ICU admission, prior to obtaining any microbial culture results. The patients in the Standard therapy group received guideline-directed empirical antimicrobial therapy without MRSA-active coverage. Baseline characteristics were extracted within 24 h of ICU admission using SQL queries, including demographics (age, sex, race), disease severity scores, and vital signs, etc. Additionally, laboratory markers and comorbidities were recorded. The primary outcomes were 28-day and 90-day all-cause mortality, while secondary outcomes included ICU length of stay (LOS) and total hospital LOS.

To evaluate independent effect of treatment on survival outcomes, Propensity score matching (PSM) was performed using a multivariable logistic regression model incorporating all baseline covariates listed in Table 1 (excluding outcomes).20 Covariate balance was assessed using standardized mean differences (SMDs), with values <0.1 indicating acceptable balance. The Cox proportional hazards model was adopted to estimate the adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) for mortality outcomes. Kaplan–Meier curves were generated for survival comparisons, and the proportional hazards assumption was evaluated with time-dependent covariates. PSM was also applied to assess differences in ICU LOS and hospital LOS. Robustness was tested via inverse probability of treatment weighting (IPTW).

Categorical variables were summarized as counts (%) and compared using chi-square or Fisher’s exact tests. Continuous variables were reported as mean ± standard deviation or median [interquartile range (IQR)], and compared using Student’s t-test or Mann–Whitney U test, as appropriate. Analyses were conducted in R (v4.3.0) and SPSS (v25.0), with two-sided p < 0.05 considered statistically significant.

A total of 494 ICU-admitted patients with severe acute pancreatitis were included (Fig. 1). The median age was 61.0 [46.0–74.0] years, and the majority were male (56.5%, 279/494). The median Charlson comorbidity index was 4.0 [2.0, 6.0]. The most prevalent comorbidities were diabetes (32.8%), followed by congestive heart failure (18.6%, 92/494), renal disease (18.0%, 89/494), and chronic pulmonary disease (15.8 %, 78/494) (Table 1).

Figure 1.

Study Population. a: empirical anti-MRSA therpy were defined as the first antibiotic used that has inhibitory effects on methicillin-resistant Staphylococcus aureus (298 cases started on vancomycin, 4 cases started on linezolid); b: Alternative empirical therapy is defined as antibiotics with the ability to suppress/eliminate pathogens, including β-lactams (penicillin, ampicillin, amoxicillin, ampicillin-sulbactam, piperacillin-tazobactam, piperacillin-tazobactam sodium); cephalosporins (cefotaxime, ceftriaxone), carbapenems (meropenem), macrolides (azithromycin, clarithromycin, erythromycin), fluoroquinolones (ciprofloxacin hydrochloride, intravenous ciprofloxacin, levofloxacin), aminoglycosides (amikacin, gentamicin, gentamicin sulfate, tobramycin), etc.

Pathogens were identified in 269 (54.5%) patients, with Escherichia coli (20.9%, 103/494) and coagulase-positive Staphylococcus aureus (9.7%, 48/494) being the most frequently isolated organisms (Table S1). MRSA screening was performed in 329 (66.7%) patients, with a positive rate of 2.6% (13/494) (Table 1). Regarding antibiotic therapy, vancomycin was the most commonly administered agent (44.3%, 219/494), followed by piperacillin–tazobactam (32.5%, 160/494) and cefepime (24.2%, 119/494) (Table S2).

Of the 494 ICU-admitted SAP patients, 61.1% (302/494) received empirical anti-MRSA therapy, primarily vancomycin (298 cases) and, less frequently, linezolid (4 cases). 66.7% of patients in the anti-MRSA group initiated treatment within 24 h of ICU admission. 38.9% (192/494) received standard empirical treatment without anti-MRSA coverage. There were no significant differences between the groups in sex distribution (male: 57.6% vs. 54.7%, p = 0.052) or Charlson Comorbidity Index (median 4.0 [IQR 2.0, 6.0] vs. 4.0 [2.0, 6.0], p = 0.31). However, patients receiving anti-MRSA therapy exhibited significantly higher baseline severity, as indicated by elevated APS III, SOFA, SIRS, and LODS scores (all p < 0.001) (Table 1).

On ICU day 1, the anti-MRSA group had higher levels of heart rate, BUN_max, aniongap_max, creatinine_max, glucose_max, potassium_max, INR_max, PT_max, PTT_max, and AST_max compared with the standard therapy group, and lower values of systolic blood pressure, hemoglobin_min, bicarbonate_min, and urine output (all p < 0.05), reflecting more severe physiological derangement (Table 1).

Additionally, the anti-MRSA group had higher rates of identified infection burden with increased rates of bacterial infection (57.6% vs. 26.6%), fungal infection (39.7% vs. 8.9%), and polymicrobial infection (37.7% vs. 7.3%) (all p < 0.001). Among the total cohort, 329 patients underwent MRSA screening, of whom 13 (2.6%) tested positive, and 6 received empirical anti-MRSA therapy (Table 1).

In unadjusted comparisons, patients receiving anti-MRSA group had significantly longer ICU stays (median 5.9 [IQR 2.6, 14.7] vs. 2.1 [1.4, 3.7] days, p < 0.001) and total hospital stays (median 17.7 [IQR 9.7, 27.0] vs. 9.4 [5.7, 13.9] days, p < 0.001). ICU readmission due to pancreatitis recurrence occurred in 16.6% (82/494) of the overall cohort, with no significant difference between groups (p = 0.60) (Table 1).

After propensity score matching, baseline characteristics between the empirical anti-MRSA group and the standard therapy group were well balanced, with adequate overlap in propensity score distribution (AUC = 92.5%) (Fig. 2, Table S3). Covariate balance was achieved across all matched subgroups after weighting (Fig. S1, Table S4).

Figure 2.

Relative Distribution of Propensity Scores for Treatment With Anti - Methicillin-Resistant Staphylococcus aureus (MRSA) therapy. Conditional density curves demonstrating relative distributions of propensity scores for treatment with anti-MRSA therapy. The standardized mean difference (SMD) between the groups was 0.04.

After adjustment, there was no significant difference between groups in ICU mortality (aRR 0.39; 95% CI, 0.11–1.38; p = 0.14), hospital mortality (aRR 0.53; 95% CI, 0.21–1.33; p = 0.18), 28-day mortality (aRR 1.28; 95% CI, 0.64-2.56; p = 0.48), or 90-day mortality (aRR 0.63; 95% CI, 0.37–1.07; p = 0.09) (Table 2). Kaplan–Meier analysis showed no significant difference in 28-day survival between the empirical anti-MRSA and standard therapy groups (P = 0.50), nor in 90-day survival (P = 0.09) (Fig. 3). However, the anti-MRSA group had significantly longer ICU stay (median 3.8 (IQR 3.0, 8.1] vs. 2.7 [2.0, 4.1], p < 0.001), and total hospital stay (median 13.0 [IQR 8.1, 19.8] vs. 9.7 [5.7, 14.5], p < 0.001) compared to the standard treatment group (Table S4).

Figure 3.

Impact of Empirical Anti-MRSA Therapy on Survival in Severe Acute Pancreatitis. Kaplan–Meier survival curves comparing empirical anti-MRSA therapy versus standard therapy in ICU-admitted SAP patients. Left: 28-day survival probability; Right: 90-day survival probability.

To validate the consistency of the findings, three sensitivity analyses were conducted using distinct statistical approaches. The estimated hazard ratios (HRs) for mortality were consistent across methods: 1.25 (95% CI: 0.66–2.36) in the PSM model, 1.21 (95% CI: 0.55–2.66) in the inverse probability of treatment weighting (IPTW) model (effective sample size [ESS] = 115), and 1.11 (95% CI: 0.53–2.31) in the doubly robust model (Table 3).

Across all models, empirical anti-MRSA therapy was not associated with a statistically significant reduction in mortality (all p > 0.05), confirming the consistency of the results. Although the IPTW model experienced a 55% effective sample size reduction due to extreme weighting, the direction and magnitude of the effect remained stable.

After propensity score matching, the incidence of acute kidney injury (AKI) was significantly higher in the anti-MRSA therapy group compared to the standard therapy group (33.6% vs. 17.2%, p = 0.002). This increased risk was most evident in patients without pre-existing kidney disease (24.2% vs. 8.6%, p < 0.001), whereas no significant difference was observed in those with baseline kidney disease (9.4% vs. 8.6%, p = 0.50) (Fig. 4, Table 4).

Figure 4.

Risk of Acute Kidney Injury (AKI) associated with empirical anti-MRSA therapy in the propensity-matched Cohort. This forest plot compares AKI incidence between patients receiving empirical anti-MRSA therapy (n = 128) versus standard therapy (n = 128). Comparison of AKI severity and recovery rate were restricted to patients who developed AKI. Estimates are presented as risk ratios (RR) with 95% confidence intervals, with the standard therapy group as the reference (RR = 1.0).

Regarding AKI severity, stage 1 AKI was the most common in both groups. The proportion of stage 3 AKI was higher (25.6% vs. 18.2%, p = 0.37), and the recovery rate of AKI was lower in the anti-MRSA group (62.8% vs. 72.7%, p = 0.30), although neither difference reached statistical significance (Table 4).

Although multiple causal inference methods (PSM, IPTW, and double robust models) were employed to adjust for confounding, several limitations remain. As 94% of patients in the anti-MRSA group received vancomycin, the findings predominantly reflect its impact and may not be generalizable to newer agents (e.g., tigecycline, oritavancin). The limited use of linezolid also precluded evaluation of its independent effects. Second, the applicability of these results to settings with high MRSA prevalence in ICU SAP patients may be limited, and region-specific strategies could be warranted. Third, despite rigorous adjustment, residual confounding cannot be excluded, as patients receiving anti-MRSA therapy had more comorbidities and higher baseline severity, which may have influenced outcomes through unmeasured factors. Selective decontamination or other forms of topical/enteral administration of vancomycin (aimed at suppressing MRSA colonization) are not captured as discrete variables in the MIMIC-IV dataset, and therefore we were unable to identify or adjust for such practices. In our cohort the vast majority of anti-MRSA exposure was systemic intravenous therapy (298 of 302 anti-MRSA cases received IV vancomycin), which is the exposure analyzed. Unmeasured use of topical or enteral decontamination could theoretically affect local microbial ecology and individual infection risk.

Given the inherent limitations of this retrospective and nonrandomized design, well-designed prospective, randomized controlled trials are warranted to elucidate the true clinical value of empirical anti-MRSA therapy in patients at high risk for severe acute pancreatitis. In parallel, large-scale, multicenter observational studies are needed to identify robust predictors of MRSA infection through comprehensive multivariable modeling, thereby facilitating risk stratification and optimizing patient selection for empirical therapy. Moreover, procalcitonin-guided antibiotic stewardship has been shown to reduce unnecessary antibiotic exposure in SAP without increasing the risk of infection or organ dysfunction underscoring its potential as a valuable tool in individualized treatment.39 Future prospective investigations should validate such biomarker-based approaches and determine whether tailored antibiotic strategies informed by predictive markers can translate into improved clinical outcomes.