The Medical Information Mart for Intensive Care IV ver 2.0 (MIMIC-IV, version 2.0) is a large and openly accessible critical care database. It includes more than 70,000 ICU admissions hospitalized at the Beth Israel Deaconess Medical Center (Boston, United States) from 2008 to 2019. Authorization to access and utilize this database was granted to one of the authors after successfully completing the Protecting Human Research Participants course offered by the National Institutes of Health (LX, certification number 12059504).

Permission for the establishment of this database was obtained from the Institutional Review Boards of both the Massachusetts Institute of Technology (MIT, Cambridge, MA, United States) and the Beth Israel Deaconess Medical Center. This project complied with the Helsinki Declaration, and no additional ethics approval was necessary. To ensure privacy, all patient data included were anonymized. This retrospective observational study adhered to the guidelines of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) initiative.

Eligible patients with AKI were those who were older than 18 years old at admission and who had been hospitalized for more than 48 h. Patients were excluded from our study if 1) >5% of their individual data were missing or 2) the baseline values exceeded the median ±1.5 times the interquartile range. AKI diagnosis followed the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines (Lameire et al., 2021), which specify changes in serum creatinine (SCr) levels and urine output. Stage 1 is defined as an increase ≥1.5 times the baseline SCr within the prior 7 days or 0.3 mg/dL in the SCr within 48 h or a urine output <0.5 mL/kg/h per 6 h. Stage 2 is characterized by an increase in SCr ≥2.0 times the baseline or a urine output <0.5 mL/kg/h per 12 h. Stage 3 is characterized by an increase in SCr ≥3.0 times baseline, SCr ≥4.0 mg/dL, initiation of renal replacement therapy (RRT), or urine output <0.3 mL/kg/h per 24 h. Patients who were diagnosed with AKI 6 h before to 48 h after ICU admission were included in the study. Baseline SCr was identified as the lowest value within 7 days or 48 h prior to AKI diagnosis.

Patient data were retrieved from the MIMIC-IV database using Structured Query Language (SQL) via PostgreSQL tools (version 15.1). The dataset included patient identifiers, clinical and laboratory parameters, comorbidities, and scoring systems. The clinical parameters included age, sex, ethnicity, heart rate, systolic blood pressure (SBP), percutaneous oxygen saturation (SPO2), diastolic blood pressure (DBP), vasopressin use, ventilator use, and RRT. The following laboratory parameters were measured: anion gap, blood urea nitrogen (BUN) level, SCr level, potassium level, red blood cell (RBC) count, red cell distribution width (RDW), and international normalized ratio (INR).

The scores, including the Sequential Organ Failure Assessment (SOFA) score and Glasgow Coma Scale (GCS) score, were calculated for each patient. Only the data from the first admission to the ICU were collected for patients who had multiple admissions to the ICU. The following comorbidities were identified: hypertension, diabetes, coronary artery disease, chronic obstructive pulmonary disease (COPD), malignancy, hematologic disease, atrial fibrillation, liver disease, shock, and sepsis based on the ninth or 10th revision of the International Classification of Diseases (ICD-9/10) code. The first measured values within 6 h before ICU admission and within 48 h after ICU admission were used as the baseline data. Amoxicillin exposure was defined as the administration of at least one dose of amoxicillin orally or via a nasogastric tube between 48 h before ICU admission and ICU discharge.

The primary outcome of our study focused on 30-day all-cause mortality. The secondary outcomes included 90-day mortality, incidence of AKD, duration of ICU stay, and total length of hospital stay. AKD diagnosis followed the 2017 Acute Disease Quality Initiative-16 (ADQI-16) criteria, characterizing AKD as AKI persisting for more than 7 days following the initial AKI event.

Baseline characteristics were categorized based on whether patients were treated with amoxicillin and are presented as frequencies (percentages) for categorical variables and means (standard deviations) or medians (interquartile ranges) for continuous variables. For group comparisons, the chi-square test was used for categorical data. For continuous variables, we employed analysis of variance or the Mann–Whitney U test as appropriate. We accounted for missing baseline data by using multiple imputation.

We used a stepwise selection method to choose variables for inclusion in the Cox proportional hazards models. This method involves iteratively adding or removing variables based on statistical criteria to identify the most significant predictors. Cox proportional hazards models were constructed to test the associations between amoxicillin administration and outcomes, with adjustment for all potential confounding variables, including age, sex, ethnicity, vasoactive use, RRT, GCS score, SOFA score, RDW, anion gap, potassium, BUN, SCr, INR, heart rate, SPO2, shock, sepsis, coronary artery disease, liver disease, atrial fibrillation, and COPD (listed in Table 1).

We conducted a series of sensitivity analyses to rigorously test the robustness of our outcomes. Initially, a preplanned multivariable Cox regression analysis was performed. Subsequently, to address potential confounding factors, we employed propensity score matching (PSM), aligning patients based on whether they received amoxicillin. A caliper width of 0.2 was used to ensure close matching of patient characteristics. The balance between the groups was evaluated using the standard mean difference (SMD), with a threshold SMD of 0.1 set as the criterion for significant baseline imbalance (Austin, 2011). After matching, we analyzed the outcomes using a Cox regression model to confirm the consistency of our findings.

Furthermore, to further refine our adjustment for potential confounders, we performed inverse probability of treatment weighting (IPTW) analysis. Weights for this analysis were derived from a multivariable Cox regression model incorporating the same covariates as the primary analysis. This method assigns weights to each patient based on the inverse probability of receiving the treatment they actually received, thus creating a synthetic sample in which the distribution of measured covariates is independent of treatment assignment (Raad et al., 2020; Chesnaye et al., 2022). The impact of amoxicillin on the risk of 30-day and 90-day mortality, as well as AKD, was then estimated by weighted Cox regression model with IPTW.

For outcomes including ICU length of stay and hospital length of stay, we employed linear regression models, adjusting for the covariates outlined earlier in our analysis.

Potential effect modifications by sex, ethnicity, age, GCS score, SOFA score, shock, sepsis, vasoactive use, and AKI stage were assessed by subgroup and interaction analyses for the primary outcome. Subgroup analyses were performed using multivariable Cox regression models, with a p-value for interactions less than 0.1 indicating a significant interaction.

Outcomes were measured as hazard ratios (HRs) for categorical variables and mean differences (MDs) for continuous variables, along with their 95% confidence intervals (CIs). Data analysis was conducted using R software version 4.2.2. A two-tailed p-value less than 0.05 was considered to indicate statistical significance.

A total of 27,571 critically ill patients with AKI were included in our study. After excluding patients according to the exclusion criteria, 24,650 patients with AKI were included in the analysis (Figure 1). Among these, 676 (2.7%) patients had been exposed to amoxicillin, while 23,974 (97.3%) had not received the antibiotic. The average age of the study participants was 67.9 years, with males comprising 58.0% (14,299) and females comprising 42.0% (10,351) of the sample.

Patient characteristics are detailed in Table 1. Notably, individuals in the non-amoxicillin group were more likely to present with sepsis and malignancy but less likely to suffer from coronary artery disease. Additionally, this group exhibited higher SOFA scores and lower levels of anion gap and RDW than did the amoxicillin group. Moreover, there were no significant differences in GCS score, BUN level, SCr level, or AKI stage between the two groups.

The 30-day mortality rate was significantly lower in the amoxicillin group (9.0%) compared to the non-amoxicillin group (15.8%) with a hazard ratio (HR) of 0.54 (95% confidence interval [CI] 0.42–0.70, p < 0.001) (Figure 2). Similarly, the 90-day mortality rate was also lower in the amoxicillin group (15.4%) compared to the non-amoxicillin group (21.2%), with an HR of 0.68 (95% CI 0.56–0.83, p < 0.001) (Figure 2).

Even after adjusting for clinically relevant covariates, the association between amoxicillin administration and reduced mortality remained significant. The adjusted HR for 30-day mortality was 0.54 (95% CI 0.42–0.69, p < 0.001), and for 90-day mortality, it was 0.64 (95% CI 0.52–0.77, p < 0.001) (Figure 2).

After PSM, 674 patients who received amoxicillin were matched with 674 patients who did not (Supplemental Table S1). This analysis demonstrated that amoxicillin administration was associated with significantly lower 30-day mortality (HR = 0.44, 95% CI 0.32–0.61, p < 0.001) and 90-day mortality (HR = 0.53, 95% CI 0.41–0.69, p < 0.001) (Figure 2). The IPTW analyses also supported these findings, showing that the amoxicillin group had lower 30-day mortality (HR = 0.50, 95% CI 0.38–0.67, p < 0.001) and 90-day mortality (HR = 0.61, 95% CI 0.49–0.75, p < 0.001) compared to the non-amoxicillin group (Figure 2).

Figure 3 shows the SMDs for all the individual covariates. After PSM and IPTW, the baseline profiles were all balanced between the two groups, with an SMD<10% for all variables.

According to the Kaplan‒Meier survival curve for 30-day mortality, mortality was lower for the amoxicillin group than for the non-amoxicillin group (HR = 0.54; p < 0.001). After PSM, the Kaplan-Meier survival curve for 30-day mortality remained lower for the amoxicillin group than for the non-amoxicillin group (HR = 0.44; p < 0.001), as noted in Figure 2. Similar results were observed for 90-day mortality.

The incidence of AKD was lower in patients exposed to amoxicillin compared to those who were not. The unadjusted HR was 0.46 (95% CI 0.35–0.61, p < 0.001). After adjusting for relevant covariates, the HR was 0.49 (95% CI 0.36–0.65, p < 0.001). Post-PSM analysis yielded an HR of 0.52 (95% CI 0.36–0.77, p < 0.001), and the IPTW analysis showed an HR of 0.49 (95% CI 0.46–0.52, p < 0.001) (Figure 2). Nevertheless, amoxicillin use was associated with a longer length of hospital stay (MD = 1.95, 95% CI 1.15–2.75, p < 0.001) (Figure 2).

The amoxicillin group can be divided into two groups based on the dose of amoxicillin: the high-dose amoxicillin group (dose ≥875 mg), which consisted of 1.4% (n = 351) of patients, and the low-dose amoxicillin group (dose <875 mg), which consisted of 1.3% (n = 325) of patients. In multivariate analysis, the amoxicillin dose was analyzed to determine whether it was independently associated with all-cause mortality (Figure 4).

In model I adjusted for age, ethnicity, and sex, the high-dose group was associated with a decreased risk of 30-day mortality (p < 0.001) and 90-day mortality (p < 0.001) (Figure 4). The HRs (95% CIs) for the high-dose group were 0.46 (0.44–0.67) and 0.54 (0.40–0.74), respectively (Figure 4). Moreover, the low-dose group had a decreased risk of 30-day mortality (p = 0.003) (Figure 4). The HR (95%CI) for the low-dose group was 0.61 (0.43–0.85) (Figure 4).

In model II, after adjusting for the aforementioned covariates, a high dose of amoxicillin remained a significant predictor of all-cause mortality at 30 days and 90 days (HR, 95% CI: 0.45, 0.31–0.67; 0.50, 0.37–0.68, both p < 0.001), while a low dose of amoxicillin was also associated with decreased 30-day mortality (HR, 95% CI: 0.62, 0.44–0.87) (Figure 4). The Kaplan‒Meier survival curve also indicated that the high-dose group had lower mortality than the low-dose group (Figure 4).

According to the subgroup analysis, the association between amoxicillin and the risk of 30-day mortality was similar in most strata (Figure 5). Although significant interactions were observed for subgroups stratified by age and SOFA score, the direction of the association between amoxicillin and the risk of 30-day mortality remained stable in most strata, except for ethnicity. For patients of white and other ethnicities, the use of amoxicillin was associated with decreased mortality (HR, 95% CI: 0.49, 0.34–0.72; HR, 95% CI: 0.27, 0.13–0.56), while no significant difference in mortality existed for black patients (HR, 95% CI: 0.57, 0.21–1.53) (Figure 5).