Creatinine Clearance, Ceftazidime-Avibactam Exposure, and Clinical Outcomes in a Real-World TDM Cohort: A Retrospective Observational Study

Abstract

Purpose

Ceftazidime-avibactam (CAZ-AVI) is a key option for carbapenem-resistant Enterobacterales (CRE) infections, but whether creatinine clearance (CrCL) independently predicts outcomes or reflects disease severity is unclear. We investigated CrCL, drug exposure, and outcomes in patients with severe Gram-negative infections receiving CAZ-AVI with therapeutic drug monitoring (TDM).

Patients and Methods

This retrospective study included 224 patients receiving CAZ-AVI TDM at a tertiary hospital in China (May 2024-May 2025). Three cohorts were defined: trough-concentration (n=224) for factors influencing trough concentrations; concentration-outcome (n=194) for concentration-outcome associations using LASSO; and stable-dose (n=165) for baseline CrCL and outcomes. CrCL was analysed using restricted cubic splines and five prespecified analysis CrCL strata. Multivariable adjustment, E-values, LASSO, and bootstrap-validated ROC were used to separate CrCL from disease severity.

Results

CrCL was the dominant determinant of ceftazidime and avibactam trough concentrations (β = −0.224 and −0.047; q = 0.040 and 0.033, respectively). The unadjusted CrCL–28-day mortality association (OR 0.993, P = 0.030) was attenuated after adjusting for SOFA and APACHE II (OR 0.996, P = 0.285; E-value = 1.07) and after full adjustment (OR 0.996, P = 0.468; E-value = 1.06). LASSO retained SOFA (OR 2.18), CRRT (OR 1.88), age (OR 1.13), and ICU admission (OR 1.03), excluding CrCL. SOFA stratification showed minimal incremental prognostic value for CrCL. The exploratory multivariable model yielded an AUC of 0.760 vs 0.571 for CrCL alone (optimism-corrected AUC: 0.793 vs 0.577).

Conclusion

CrCL drives CAZ-AVI trough-concentration variability but appears to be a marker of disease severity rather than an independent predictor of outcomes. Given the observational design and absence of individual MIC data, these findings should be interpreted cautiously. TDM may be best targeted at patients with extreme pharmacokinetic phenotypes, pending prospective validation. Prospective studies with MIC data are needed to define pathogen-specific targets.

Introduction

The global spread of carbapenem-resistant Enterobacterales (CRE) has been declared a critical public-health threat by the World Health Organization, severely limiting treatment options for severe Gram-negative infections.1 Ceftazidime-avibactam (CAZ-AVI) has emerged as a pivotal therapeutic agent because of its ability to inactivate a broad range of β-lactamases, including class A carbapenemases (eg, KPC-2/3, IMI), OXA-48-like enzymes, AmpC, TEM, SHV, and CTX-M.2,3 Real-world studies have shown that CAZ-AVI achieves clinical success rates of approximately 63.4–79.0% and 30-day mortality rates of around 24% in diverse patient populations.4,5 This resistance is widely attributed to suboptimal drug exposure, underscoring the need to characterise the determinants of interpatient pharmacokinetic variability in routine practice.

The approved CAZ-AVI regimen (2.5 g every 8 h) was established in relatively stable populations. In routine clinical practice, many patients exhibit profound pathophysiological alterations, such as augmented renal clearance (ARC), increased volume of distribution, and renal replacement therapy, which can markedly alter drug disposition and lead to suboptimal target attainment.6,7 These concerns have fueled growing interest in therapeutic drug monitoring (TDM) as a tool to tailor CAZ-AVI dosing to individual patient needs.8,9 Despite this, real-world data on the interplay among renal function, achieved drug concentrations, and clinical outcomes remain scarce, and the mechanistic interpretation of CrCL-outcome associations is unsettled. It is important to distinguish CrCL as a pharmacokinetic determinant of drug clearance from CrCL as a prognostic factor for clinical outcomes. These two concepts are related but not equivalent: a strong association between CrCL and drug exposure does not necessarily imply that CrCL independently predicts mortality.

In hospitalized patients with severe Gram-negative infections, renal function, disease severity, organ support, drug dosing, and drug concentration are closely interrelated. Disease severity may act as a confounder because it is associated with both lower CrCL and higher mortality. In contrast, organ support, dose adjustment, and drug concentration may lie on the causal pathway between CrCL and mortality, acting as mediators. Adjusting for confounders is necessary to estimate the independent effect of CrCL, whereas adjusting for mediators could introduce over-adjustment bias and obscure the total effect of CrCL. Therefore, the analytical strategy must carefully distinguish between confounders and mediators.

A central unresolved question is whether CrCL predicts clinical outcomes independently through PK/PD mechanisms (ie, lower exposure leading to worse bacterial killing) or whether it merely reflects the underlying disease severity that drives mortality (sicker patients have impaired renal function, accumulate comorbidities, and receive lower clearance-adjusted doses). These two explanations carry opposite clinical implications: the first supports aggressive dose escalation in patients with preserved CrCL, while the second implies that the label-recommended dose is adequate and that outcome improvement requires interventions other than dosing. Disentangling them requires a study design that (1) adjusts rigorously for severity, (2) uses dose-stable patients to avoid indication bias, and (3) tests whether CrCL-outcome associations persist after comprehensive adjustment for disease severity, with particular attention to potential non-linear patterns. We therefore excluded patients who underwent dose adjustment during therapy to create a stable-dose cohort, ensuring that baseline CrCL and initial trough concentrations reflected a consistent exposure during the treatment period. Dose adjustments, whether triggered by renal function changes, TDM results, or clinical deterioration, alter drug exposure over time and could therefore bias the assessment of baseline exposure-outcome relationships.

In this context, we conducted a retrospective study of CAZ-AVI-treated patients who underwent TDM at our institution. The study aimed to: (1) identify independent factors influencing steady-state trough concentrations of ceftazidime and avibactam in the trough-concentration cohort (n=224); (2) evaluate the association between achieved trough concentrations and clinical outcomes in the concentration-outcome cohort (n=194); and (3) examine whether the relationship between baseline CrCL and clinical outcomes in the stable-dose cohort (n=165), including non-linear associations, persists after comprehensive adjustment for disease severity. We hypothesised that, under confounding-by-severity, CrCL would lose its apparent prognostic value once severity indicators were accounted for, supporting CrCL as a marker of disease severity rather than an independent PK/PD determinant.

Materials and Methods

Study Design and Patient Population

This was a retrospective, single-center, observational study conducted at Tongde Hospital of Zhejiang Province Affiliated to Zhejiang Chinese Medical University, Hangzhou, China. Adult patients (≥18 years) treated with ceftazidime-avibactam (CAZ-AVI) between May 2024 and May 2025 were eligible if they had at least one steady-state trough concentration measurement of both ceftazidime and avibactam. Patients were excluded if they received CAZ-AVI for less than 72 hours or if relevant clinical data were missing.

From the total cohort, three analysis populations were defined: (1) the trough-concentration cohort (n=224) included all eligible patients with available trough concentrations and was used to identify factors influencing CAZ and AVI trough concentrations; (2) the concentration-outcome cohort (n=194) included patients with a confirmed pathogen susceptible to CAZ-AVI (excluding those with resistant or unknown pathogens) and was used to evaluate the association between trough concentrations and clinical outcomes; and (3) the stable-dose cohort (n=165) was derived from the concentration-outcome cohort by excluding patients who underwent dose adjustment during therapy, and was used specifically to examine the relationship between baseline renal function (CrCL) and clinical outcomes, and to compute the target-attainment table stratified by renal function and organ support. Dose adjustments, whether triggered by renal function changes, TDM results, or clinical deterioration, alter drug exposure over time; excluding these patients ensured that baseline CrCL and initial trough concentrations were more likely to reflect a consistent exposure during the treatment period, thereby providing a more reliable assessment of baseline exposure-outcome relationships.

The study was approved by the Institutional Review Board of Tongde Hospital of Zhejiang Province (approval number: 2025160-JY). The requirement for informed consent was waived due to the retrospective nature of the study. The study was conducted in accordance with the Helsinki Declaration of 1964 and its later amendments. Patient data were anonymized and kept confidential in accordance with institutional privacy guidelines and the Declaration of Helsinki. No identifiable personal information was included in this study.

Variables and Definitions

Dosing Regimens

The initial dosing regimen was classified as label-recommended if it adhered to the approved prescribing information based on CrCL at treatment initiation. Because the label-recommended dose is directly determined by CrCL, and dose adjustments during therapy reflect dynamic changes in renal function, we did not attempt direct comparisons between different dosing strategies. Instead, CrCL was examined as the primary exposure variable in all outcome analyses, with dosing regimen treated as a descriptive covariate.

Renal Function and Clinical Covariates

CrCL was calculated using the Cockcroft-Gault formula based on actual body weight at treatment initiation, and was analysed as both a continuous variable using restricted cubic splines with three knots at the 10th, 50th, and 90th percentiles, and categorised into five strata: <30, 30–50, 50–90, 90–130, and ≥130 mL/min. Augmented renal clearance (ARC) was defined as CrCL >130 mL/min. Obesity was defined as BMI ≥28 kg/m². ICU admission was defined as patients being admitted to the intensive care unit during CAZ-AVI therapy. Hematological malignancy was defined as a diagnosis of hematological cancer. CRRT was defined as receipt of continuous renal replacement therapy during CAZ-AVI treatment. Two CrCL categorisations were used for distinct purposes: label-recommended renal dosing categories (6–15, 16–30, 31–50, 51–130, >130 mL/min) for target-attainment and trough-concentration analyses, and analysis CrCL strata (<30, 30–50, 50–90, 90–130, ≥130 mL/min) for all clinical-outcome analyses.

Weighted Comorbidity Score

The weighted comorbidity score (modified Charlson-type index) was calculated from 11 recorded dichotomous conditions. Chronic kidney disease, hyperlipidemia, and solid tumour were assigned a weight of 2; all other conditions (hypertension, diabetes, coronary artery disease, cerebral infarction, COPD, chronic liver disease, CNS disease, and immunosuppression) were assigned a weight of 1. This score is an abbreviated, locally adapted comorbidity measure and is not the validated Charlson Comorbidity Index; it should be interpreted as a general comorbidity-burden indicator rather than a formally validated index.

Ceftazidime Concentration Categories

Because antimicrobial susceptibility testing was performed using the Kirby-Bauer disk diffusion method (which does not provide exact minimum inhibitory concentration, MIC, values), we empirically defined three exploratory concentration categories based on pharmacokinetic/pharmacodynamic (PK/PD) considerations (target: free concentration above 4–5×MIC for ≥50% of the dosing interval) and the typical MIC range of susceptible Enterobacterales (4–8 μg/mL) reported in the literature:10 subtherapeutic (<16 μg/mL), moderate (16–40 μg/mL), and high (≥40 μg/mL). These categories are intended for descriptive exploration rather than as validated therapeutic targets, and their limitations are addressed in the discussion. Avibactam concentrations were analyzed as continuous variables due to the lack of established therapeutic ranges.

Clinical Outcomes

Clinical success was defined as complete or partial resolution of infection-related signs and symptoms, accompanied by discontinuation or de-escalation of empirical Gram-negative antibiotics; any other outcome was considered clinical failure. Microbiological response was classified as: documented eradication; presumed eradication; and failure of eradication. A favourable microbiological response (ie, eradication) was defined as documented or presumed eradication. 28-day mortality was defined as death from any cause within 28 days from the initiation of CAZ-AVI therapy.

Therapeutic Drug Monitoring

Steady-state trough concentrations of ceftazidime and avibactam were measured after at least four doses of the same regimen. Blood samples were collected immediately before the next infusion. Concentrations were determined using validated liquid chromatography-tandem mass spectrometry (LC-MS/MS).

Statistical Analysis

Continuous variables were presented as mean ± SD or median (IQR) after normality check (Shapiro–Wilk), and categorical data as n (%). Three analytic strategies were used, matched to the study objectives: (1) for the trough-concentration cohort (n=224), multivariable linear regression with Benjamini-Hochberg FDR correction was used to identify factors associated with ceftazidime and avibactam trough concentrations, with p < 0.10 in univariable screening as the entry criterion; (2) for the concentration-outcome cohort (n=194), LASSO-penalised logistic regression (L1, λ by 10-fold cross-validation) selected predictors that were then entered into a standard logistic model; (3) for the stable-dose cohort (n=165), restricted cubic splines with three knots at the 10th, 50th, and 90th percentiles of CrCL were used, with non-linearity tested by the Wald χ²-test on the spline non-linear basis functions. No a priori sample size calculation was performed because this was a retrospective study using routinely collected data; the sample size was determined by eligible patients treated during the study period, and all analyses were exploratory. To address confounding-by-severity, four sequential logistic models were fitted (Model 0: CrCL only; Model 1: + age, sex, BMI; Model 2: + SOFA, APACHE II; Model 3: a 10-variable clinical-driven model including CrCL, age, sex, SOFA, weighted comorbidity score, ICU admission, CRRT, bloodstream infection, multiple infection sites, and P. aeruginosa) with E-values quantifying the minimum strength of unmeasured confounding required to explain the residual association. Multicollinearity was assessed using variance inflation factors (VIFs). All VIFs were < 2.2 across the fitted models. Regression diagnostics included the Hosmer-Lemeshow goodness-of-fit test (Model 3: χ² = 10.08, df = 8, P = 0.259) and pseudo-R² (0.313), indicating acceptable calibration. Joint SOFA × CrCL stratification and a LASSO sensitivity analysis were also performed. A Firth penalised logistic regression was performed as a sensitivity analysis for the mortality model to reduce bias from sparse events. CrCL was analysed using a complete-case approach (19 patients, 11.5%, excluded for missing body weight); among the remaining 146 patients, APACHE II, BMI, and SOFA were missing in 48 (32.9%), 3 (2.1%), and 2 (1.4%) patients, respectively, and missing covariate values were imputed using the median. Model discrimination was assessed by 5-fold cross-validated ROC-AUC with 2000 bootstrap 95% CI; calibration by the Brier score. The DeLong test was used to formally compare AUCs between the CrCL-only and full models for each outcome. Given the structural collinearity between dose selection and CrCL, no direct dose-strategy comparisons were performed. Two-sided P < 0.05 was considered statistically significant. Because several complementary analytical approaches were applied, the possibility of chance findings cannot be entirely excluded, and secondary analyses should therefore be regarded as exploratory. Analyses used Python 3.11 (statsmodels 0.14, scikit-learn 1.9).

Results

Patient Cohort and Baseline Characteristics

A total of 232 patients who received CAZ-AVI and underwent TDM were initially screened from the hospital information system. Of these, 2 had major missing clinical data, 1 discontinued therapy within 72 hours due to an adverse reaction, and 5 had blood samples collected at incorrect timing. After excluding these 8 patients, 224 patients with at least one valid steady-state trough concentration measurement were included in the total TDM cohort. Among these, 15 patients with resistant pathogens and 15 with unknown pathogens were excluded, leaving 194 patients in the concentration-outcome cohort. A further 29 patients who underwent dose adjustment during therapy were excluded, resulting in 165 patients in the stable-dose cohort (Figure 1). Of the 165 patients in the stable-dose cohort, 19 (11.5%) had missing body weight and were excluded from CrCL-based analyses, leaving 146 patients. In this complete-case cohort, APACHE II, BMI, and SOFA were missing in 48 (32.9%), 3 (2.1%), and 2 (1.4%) patients, respectively, and were imputed using the median. The baseline characteristics of the three analysis cohorts are summarized in Table 1.

Figure 1.

Flowchart of patient selection and study cohorts. The trough concentration cohort (n=224) was used to identify factors influencing steady-state trough concentrations of ceftazidime and avibactam. The concentration-outcome cohort (n=194) was used to evaluate associations between trough concentrations and clinical outcomes. The stable-dose cohort (n=165), which excluded patients with dose adjustments during therapy, was used to examine the relationship between baseline renal function and clinical outcomes.

Table 1.

Baseline Characteristics of the Three Analysis Cohorts

Variable	The Trough Concentration Cohort (n=224)	Concentration-Outcome Cohort (n=194)	Stable-Dose Cohort (n=165)
Age, years	70.59±17.04	71.46 ± 16.98	70.88 ± 17.20
Male, n (%)	156 (69.64)	138 (71.13)	117 (70.91)
BMI (kg/m2)	22.04 ±3.92	22.13 ± 3.98	22.24 ± 3.95
APACHE II	25.04 ±9.29	25.03 ± 9.47	24.98 ± 8.79
SOFA	7.33 ±4.52	7.54 ± 4.52	7.37 ± 4.61
ICU admission, n (%)	118 (52.68)	108 (55.67)	85 (51.52)
Septic shock, n (%)	65 (29.02)	56 (28.87)	45 (27.27)
TBil, μmol/L	21.23 ±34.45	21.02 ± 35.36	21.80 ± 38.02
DBil, μmol/L	9.37 ±20.40	9.16 ± 20.52	9.62 ± 22.04
TP, g/L	57.03 ±8.59	57.77 ± 8.41	57.35 ± 8.27
ALB, g/L	29.56±4.30	29.40 ± 4.18	29.44 ± 4.04
ALT, U/L	68.14 ±279.35	73.85 ± 299.77	79.10 ± 324.47
BUN, mmol/L	13.24±11.29	13.51 ± 11.34	11.87 ± 9.65
UA, μmol/L	276.00±156.79	278.56 ± 157.00	259.25 ± 146.63
CrCL, mL/min	77.26±65.01	76.53 ± 69.42	83.21 ± 71.68
PT, s	13.79±6.76	13.90 ± 7.22	13.96 ± 7.81
D-dimer, mg/L	4.24±5.10	4.29 ± 5.03	4.32 ± 5.29
WBC, ×109/L	9.38±6.78	10.04 ± 6.69	9.30 ± 6.18
NEUT, ×109/L	7.62±6.27	8.17 ± 6.28	7.58 ± 5.88
PLT, ×109/L	175.60±133.25	188.78 ± 132.99	186.39 ± 134.23
HB, g/L	76.70±19.84	77.68 ± 19.81	77.19 ± 19.06
Comorbidities, n (%)
Hypertension	126 (56.25)	114 (58.76)	96 (58.18)
Diabetes	68 (30.36)	64 (32.99)	56 (33.94)
Coronary artery disease	67 (29.91)	60 (30.93)	49 (29.70)
Cerebral infarction	97 (43.30)	89 (45.88)	73 (44.24)
COPD	30 (13.39)	28 (14.43)	27 (16.36)
Chronic liver disease	14 (6.25)	14 (7.22)	13 (7.88)
Chronic kidney disease	37 (16.52)	34 (17.53)	28 (16.97)
Hyperlipidemia	9 (4.02)	8 (4.12)	7 (4.24)
CNS disease	58 (25.89)	54 (27.84)	50 (30.30)
Solid tumor	28 (12.50)	25 (12.89)	21 (12.73)
Hematological malignancy	33 (14.73)	18 (9.28)	18 (10.91)
Others	43 (19.20)	39 (20.10)	33 (20.00)
Infection site, n (%)
Pulmonary	149 (66.52)	137 (70.62)	119 (72.12)
Bloodstream	24 (10.71)	16 (8.25)	13 (7.88)
Urinary tract	11 (4.91)	11 (5.67)	8 (4.85)
Intra-abdominal	14 (6.25)	11 (5.67)	10 (6.06)
Other single site	5 (2.23)	3 (1.55)	3 (1.82)
Multiple sites	21 (9.38)	16 (8.25)	12 (7.27)
Microbiological status, n (%)
Susceptible to CAZ-AVI	194 (86.61)	194 (100.00)	165 (100.00)
Resistant pathogens	13 (5.80)	0 (0.00)	0 (0.00)
Unknown (negative culture)	17 (7.59)	0 (0.00)	0 (0.00)
Pathogen distribution among susceptible patients, n (%)
K. pneumoniae	121 (67.22)	120 (61.86)	101 (61.21)
P. aeruginosa	44 (24.44)	44 (22.68)	39 (23.64)
E. coli	9 (5.00)	8 (4.12)	8 (4.85)
B. cepacia	2 (1.11)	8 (4.12)	8 (4.85)
Others	4 (2.22)	14 (7.22)	9 (5.45)
Combination therapy, n (%)
Anti-Gram-negative	36 (16.07)	30 (15.46)	24 (14.55)
Anti-Gram-positive	44 (19.64)	40 (20.62)	34 (20.61)
Nephrotoxic agents	57 (25.45)	53 (27.32)	46 (27.88)
Organ support, n (%)
CRRT	26 (11.61)	25 (12.89)	20 (12.12)
ECMO	3 (1.34)	3 (1.55)	3 (1.82)

Note: Data are presented as mean ± SD or n (%).

Abbreviations: BMI, body mass index; SOFA, sequential organ failure assessment; TBil, total bilirubin; DBil, direct bilirubin; TP, total protein; ALB, albumin; ALT, alanine aminotransferase; BUN, blood urea nitrogen; UA, uric acid; CrCL, creatinine clearance; PT, prothrombin time; WBC, white blood cell; NEUT, neutrophil count; PLT, platelet count; HGB, hemoglobin; COPD, chronic obstructive pulmonary disease; CNS, central nervous system; CRRT, continuous renal replacement therapy; ECMO, extracorporeal membrane oxygenation.

Factors Associated with Ceftazidime and Avibactam Trough Concentrations

In the trough-concentration cohort (n=224), univariable screening at P < 0.10 followed by multivariable linear regression with Benjamini–Hochberg FDR correction identified CrCL (β = −0.224 per +1 mL/min for ceftazidime, q = 0.040; β = −0.047 per +1 mL/min for avibactam, q = 0.033), cerebral infarction (β = 23.296 for ceftazidime, q = 0.046; β = 4.891 for avibactam, q = 0.033), and hemoglobin (β = −0.796 for ceftazidime, q = 0.016; β = −0.128 for avibactam, q = 0.033) as significant predictors of trough concentrations (Table 2). Other variables including age (β = 0.646, q = 0.058), ICU admission (β = 23.203, q = 0.058), and hypertension (β = 20.748, q = 0.058) showed borderline associations but did not reach the q < 0.05 threshold after FDR correction. These results support the accepted role of CrCL as the primary driver of CAZ-AVI trough-concentration variability.

Table 2.

Multivariable Linear Regression of Factors Associated with Ceftazidime (CAZ) and Avibactam (AVI) Trough Concentrations, with Benjamini-Hochberg FDR Correction (n=224)

Variable	CAZ β (95% CI)	CAZ P (q_FDR)	AVI β (95% CI)	AVI P (q_FDR)
Age (years)	0.646 (0.067, 1.224)	0.029 (0.058)	0.088 (−0.021, 0.197)	0.114 (0.245)
APACHE II	−0.768 (−1.747, 0.211)	0.123 (0.150)	−0.114 (−0.298, 0.071)	0.226 (0.376)
SOFA	1.414 (−1.183, 4.011)	0.284 (0.310)	0.426 (−0.068, 0.920)	0.091 (0.231)
CrCL (mL/min)	−0.224 (−0.384, −0.063)	0.007 (0.040)	−0.047 (−0.078, −0.016)	0.003 (0.033)
ICU admission	23.203 (2.438, 43.967)	0.029 (0.058)	3.495 (−0.579, 7.570)	0.092 (0.231)
Hypertension	20.748 (2.533, 38.963)	0.026 (0.058)	1.700 (−1.836, 5.236)	0.344 (0.469)
Cerebral infarction	23.296 (5.272, 41.321)	0.012 (0.046)	4.891 (1.395, 8.386)	0.006 (0.033)
Chronic liver disease	−27.182 (−61.292, 6.928)	0.118 (0.150)		–
Chronic kidney disease		–	−1.314 (−5.606, 2.979)	0.547 (0.641)
Immunosuppression	−8.127 (−36.831, 20.576)	0.577 (0.577)	−0.724 (−6.309, 4.862)	0.799 (0.856)
WBC	–	–	−0.053 (−0.783, 0.678)	0.887 (0.887)
NEUT	–	–	0.230 (−0.538, 0.998)	0.556 (0.641)
PLT (×109/L)	−0.057 (−0.129, 0.016)	0.125 (0.150)	−0.013 (−0.027, 0.001)	0.068 (0.231)
HB (g/L)	−0.796 (−1.280, −0.313)	0.001 (0.016)	−0.128 (−0.220, −0.036)	0.007 (0.033)

The distribution of ceftazidime trough concentrations across patient subgroups is shown in Figure 2. Concentrations were significantly higher in patients with CrCL 51–130 mL/min than in those with CrCL >130 mL/min (P < 0.05). Among patients with CrCL 51–130 mL/min, ICU admission was associated with higher concentrations (P < 0.05). No significant differences were observed for CRRT, obesity, or hematological malignancy across any CrCL stratum.

Figure 2.

Violin plots of ceftazidime trough concentrations by label-recommended renal dosing category (6–15, 16–30, 31–50, 51–130, >130 mL/min) in subgroups: (A) overall, (B) ICU, (C) obesity, (D) CRRT, (E) Hematological malignancy. Horizontal lines indicate median and interquartile range. *P < 0.05, **P < 0.01 for the comparisons shown.

Ceftazidime Trough Concentrations by Renal Function and Organ Support

Among the 125 patients receiving label-recommended dosing without organ support (Table 3), the overall proportion achieving high ceftazidime trough concentrations (≥40 μg/mL) was 56.0%. Attainment varied by CrCL stratum: 50.0% in CrCL 6–15, 71.4% in CrCL 16–30, 66.7% in CrCL 31–50, 64.8% in CrCL 51–130, and 35.1% in CrCL >130 mL/min. The subtherapeutic rate was highest in the CrCL >130 mL/min stratum (21.6%). In the 15 patients receiving CRRT or ECMO, 80.0% achieved high concentrations. These data confirm that CrCL is a major determinant of trough exposure, with augmented renal clearance representing the greatest risk of subtherapeutic levels.

Table 3.

Steady-State Ceftazidime Trough Concentrations and Proportion Achieving High Concentrations (≥40 μg/mL) in Patients Receiving Label-Recommended Dosing, Stratified by Renal Function and Organ Support

Regimen/CrCL (mL/min)	Clinical Feature	n	Trough (μg/mL), Median [IQR]	Subtherapeutic (<16) n(%)	Moderate (16–40) n (%)	High (≥40) n (%)
A. Without organ support
0.94g qd (6–15)	Routine	4	48.95 [30.62–64.47]	0 (0.00)	2 (50.00)	2 (50.00)
0.94g q12h (16–30)	Routine	7	48.86 [10.03–109.75]	2 (28.57)	0 (0.00)	5 (71.43)
1.25g q8h (31–50)	Routine	6	60.20 [12.53–136.5]	1 (16.67)	1 (16.67)	4 (66.67)
2.5g q8h (51–130)	Routine	71	54.51 [6.76–330.79]	7 (9.86)	18 (25.35)	46 (64.79)
2.5g q8h (>130)	Routine	37	28.27 [5.40–101.62]	8 (21.62)	16 (43.24)	13 (35.14)
Subtotal		125		18 (14.40)	37 (29.60)	70 (56.00)
B. With organ support
0.94g q12h (16–30)	CRRT	1	86.68	0 (0.00)	0 (0.00)	1 (100.00)
1.25g q8h (31–50)	CRRT+ECMO	1	190.05	0 (0.00)	0 (0.00)	1 (100.00)
2.5g q8h (51–130)	CRRT	11	66.55 [15.20–174.13]	3 (27.27)	0 (0.00)	8 (72.73)
2.5g q8h (51–130)	ECMO	1	55.55	0 (0.00)	0 (0.00)	1 (100.00)
2.5g q8h (>130)	CRRT	1	161.08	0 (0.00)	0 (0.00)	1 (100.00)
Subtotal		15		3 (20.00)	0 (0.00)	12 (80.00)

Notes: CrCL categories follow the approved prescribing information for dose adjustment.

Drug Exposure and Clinical Outcomes

Multivariable logistic regression identified factors associated with clinical outcomes in the concentration-outcome cohort (Figure 3). For clinical success, higher neutrophil count (OR, 0.901; 95% CI, 0.839–0.967; P = 0.004), bloodstream infection (OR, 0.135; 95% CI, 0.036–0.513; P = 0.003), and higher avibactam concentration (OR, 0.966; 95% CI, 0.938–0.995; P = 0.023) were associated with lower odds, while higher platelet count (OR, 1.006; 95% CI, 1.001–1.010; P = 0.009) was associated with higher odds. For microbiological eradication, E. coli (vs K. pneumoniae) was associated with higher odds (OR, 14.306; 95% CI, 2.031–100.761; P = 0.008), whereas higher avibactam concentration predicted lower odds (OR, 0.875; 95% CI, 0.798–0.959; P = 0.004). Higher SOFA score (OR, 1.225; 95% CI, 1.105–1.357; P < 0.001), neutrophil count (OR, 1.074; 95% CI, 1.005–1.149; P = 0.035), CRRT (OR, 30.579; 95% CI, 3.209–291.362; P = 0.003), and avibactam concentration (OR, 1.067; 95% CI, 1.028–1.108; P < 0.001) were associated with increased mortality.

Figure 3.

Forest plots of multivariable logistic regression for (A) clinical success, (B) microbiological eradication, and (C) 28-day mortality.

Relationship Between Renal Function and Clinical Outcomes

In the stable-dose cohort (n=165), clinical outcomes were analysed by prespecified CrCL strata (Table 4). Clinical success rates did not show a significant trend across CrCL strata (P for trend = 0.932). Microbiological eradication showed a significant U-shaped relationship with CrCL (Spearman ρ = 0.179, P for trend = 0.021), with the lowest eradication rate in the CrCL 30–50 mL/min stratum (7.5%) and higher rates at both CrCL <30 (15.8%) and CrCL ≥90 (25.9–29.4%). The 28-day mortality rate was highest in the CrCL 50–90 mL/min stratum (51.6%) and lowest in the CrCL ≥130 mL/min stratum (14.8%), although the overall trend was not statistically significant (P for trend = 0.174). Ceftazidime trough concentrations decreased progressively with increasing CrCL, from a median of 75.50 μg/mL in CrCL <30 mL/min to 28.27 μg/mL in CrCL ≥130 mL/min.

Table 4.

Clinical Outcomes and Ceftazidime Trough Concentrations by Baseline CrCL Strata in the Stable-Dose Cohort (n = 165)

CrCL (mL/min)	n	Ceftazidime Trough (μg/mL), Median [IQR]	Clinical Success, n (%)	Micro. Eradication, n (%)	28-day Mortality, n (%)
<30	19	75.50 [48.86–184.41]	16 (84.21)	3 (15.78)	8 (42.11)
30-50	40	110.75[75.87–161.33]	33 (82.50)	3 (7.50)	11 (27.50)
50-90	62	82.63 [48.93–130.19]	47 (75.81)	9 (14.52)	32 (51.61)
90-130	17	48.43 [22.14–78.30]	15 (88.24)	5 (29.41)	4 (23.53)
≥130	27	28.27 [18.62–46.29]	21 (77.78)	7 (25.93)	4 (14.82)
P for trend	-	-	0.932	0.021	0.174

Restricted cubic spline analysis confirmed a significant non-linear relationship between CrCL and 28-day mortality (P for non-linearity = 0.010), while no significant non-linearity was detected for clinical success (P = 0.574) or microbiological eradication (P = 0.217).

CrCL as a “Renal-Severity” Proxy: Correlation and Multivariable Adjustment

To investigate whether the apparent CrCL-mortality association reflects severity rather than a direct PK/PD mechanism, we first examined the correlation between CrCL and disease severity markers in the stable-dose cohort with complete CrCL (n=146) (Table 5). CrCL correlated moderately with the weighted comorbidity score (Spearman ρ = −0.46, P < 0.001) and weakly-moderately with SOFA (ρ = −0.22, P = 0.007) and ICU admission (ρ = −0.33, P < 0.001). The CrCL-age correlation (Spearman ρ = −0.41; Pearson r = −0.52) is partly attributable to the age term embedded in the Cockcroft-Gault formula.

Table 5.

Correlation Matrix Between CrCL and Disease Severity Markers (Spearman ρ, n=146)

Variable	CrCL	SOFA	APACHE II	ICU	Shock	CRRT
CrCL	1.0000	−0.2227	−0.0699	−0.3222	−0.2427	−0.1087
SOFA	−0.2227	1.0000	0.4324	0.4701	0.3779	0.2151
APACHE II	−0.0699	0.4324	1.0000	0.2384	0.2253	0.1310
ICU	−0.3222	0.4701	0.2384	1.0000	0.4812	0.3276
Shock	−0.2427	0.3779	0.2253	0.4812	1.0000	0.3744
CRRT	−0.1087	0.2151	0.1310	0.3276	0.3744	1.0000
age	−0.4230	0.1039	0.1527	0.0632	0.0509	0.0807

We then fitted four sequential multivariable logistic models for each outcome (Table 6). For 28-day mortality, the unadjusted CrCL OR was 0.993 (P = 0.030); after adjusting for age, sex and BMI (Model 1), OR = 0.993 (P = 0.092); after adding SOFA and APACHE II (Model 2), OR = 0.996 (P = 0.285); and in the fully adjusted clinical-driven 10-variable model (Model 3; variables listed in Methods), OR = 0.996 (P = 0.468). The progressive attenuation of the CrCL-mortality association across adjustment models indicates that the apparent protective effect of CrCL is markedly attenuated and no longer statistically significant after adjustment for measured disease severity; residual confounding cannot be excluded. Similar patterns were observed for microbiological eradication, where the CrCL OR was attenuated from 1.004 (unadjusted, P = 0.097) to 1.001 (Model 3, P = 0.715).

Table 6.

Sequential Attenuation of the CrCL-Outcome Association Across Four Multivariable Models (n = 146)

Outcome	Model	CrCL OR	95% CI	P	E-value	E-value CI	No. of Variables	No. of Patients	No. of Events
Clinical success	M0	1.001	0.995–1.007	0.778	1.030	1.090	1	146	118
	M1	1.001	0.993–1.008	0.821	1.030	1.100	4	146	118
	M2	1.001	0.993–1.008	0.860	1.030	1.100	6	146	118
	M3	1.003	0.994–1.012	0.463	1.060	1.120	10	146	118
Microbiological eradication	M0	1.004	0.999–1.010	0.097	1.070	1.110	1	146	26
	M1	1.002	0.995–1.008	0.633	1.040	1.100	4	146	26
	M2	1.001	0.994–1.008	0.787	1.030	1.100	6	146	26
	M3	1.001	0.994–1.009	0.715	1.040	1.100	10	146	26
28-day mortality	M0	0.993	0.986–0.999	0.030	1.090	1.130	1	146	52
	M1	0.993	0.986–1.001	0.092	1.090	1.140	4	146	52
	M2	0.996	0.988–1.004	0.285	1.070	1.120	6	146	52
	M3	0.996	0.987–1.006	0.468	1.060	1.130	10	146	52

Notes: Model 0, unadjusted (CrCL only); Model 1, adjusted for age, sex, and BMI; Model 2, adjusted for age, sex, BMI, SOFA, and APACHE II; Model 3, a 10-variable clinical-driven model including CrCL, age, sex, SOFA, weighted comorbidity score, ICU admission, CRRT, bloodstream infection, multiple infection sites, and P. aeruginosa.

Joint stratification by SOFA and CrCL (Figure 4) demonstrated that CrCL had minimal incremental prognostic value within any given SOFA stratum. Among patients with SOFA ≥11, mortality was uniformly ≥67% across all CrCL strata.

Figure 4.

Joint stratification heatmap of 28-day mortality by SOFA score and CrCL (n=144).

Variable Selection and Model Performance

LASSO-penalised logistic regression on candidate variables for 28-day mortality (Table 7) retained SOFA (OR 2.18), CRRT (OR 1.88), age (OR 1.13), ICU admission (OR 1.03), and urinary tract infection site as predictors, while shrinking CrCL, weighted comorbidity score, APACHE II, and shock to zero. This independently corroborates the multivariable adjustment finding.

Table 7.

LASSO-Penalised Logistic Regression for 28-Day Mortality (n=146, λ_min=5.96)

Variable	Coefficient	OR
SOFA	0.779	2.179
CRRT	0.631	1.880
Age	0.126	1.134
ICU admission	0.033	1.034
Urinary tract infection	<0.001	1.000

Notes: LASSO was performed on 18 candidate predictors including CrCL, APACHE II, SOFA, ICU admission, CRRT, shock, age, sex, BMI, nephrotoxic agents, immunosuppression, infection site (bloodstream, intra-abdominal, multiple, urinary tract), and pathogen (K. pneumoniae, P. aeruginosa, E. coli, B. cepacia). Variables with coefficient = 0 were shrunk to zero by LASSO and are not shown. λ_min was determined by 10-fold cross-validation.

Model discrimination (Figure 5 and Table 8) showed that the exploratory multivariable model yielded a higher 5-fold cross-validated AUC than CrCL alone for 28-day mortality (0.760, 95% CI 0.664–0.844 vs 0.571, 95% CI 0.475–0.669; ΔAUC = 0.188; DeLong P = 0.003), whereas the difference was not significant for microbiological eradication (P = 0.585) or clinical success (P = 0.088). Brier scores were 0.222 for CrCL alone and 0.186 for the multivariable model. The optimism-corrected AUC was 0.577 for CrCL alone, 0.793 for the 20-variable exploratory model (optimism 0.085), and 0.812 for the 10-variable Model 3 (optimism 0.037).

Figure 5.

ROC curves comparing a CrCL-only model with an exploratory multivariable model for (A) clinical success, (B) microbiological eradication, and (C) 28-day mortality (n = 146).5-fold cross-validation with 2000 bootstrap iterations for 95% CI.

Table 8.

Five-Fold Cross-Validated ROC-AUC and Brier Scores (n = 146)

Outcome	Model	No. of Variables	No. of Events	AUC	95% CI	Brier Score
Clinical success	CrCL only	1	118	0.413	0.296–0.535	0.157
	Multivariable	20	118	0.561	0.427–0.685	0.192
Microbiological eradication	CrCL only	1	26	0.627	0.503–0.743	0.147
	Multivariable	20	26	0.672	0.539–0.796	0.155
28-day mortality	CrCL only	1	52	0.571	0.475–0.669	0.222
	Multivariable	20	52	0.760	0.664–0.844	0.186

Notes: Model A (CrCL only) includes creatinine clearance as the sole predictor. Model B (Multivariable) includes 20 candidate predictors selected a priori; one variable with no variation in this cohort was removed from this exploratory model. This exploratory model should be interpreted with caution given the low events-per-variable ratio (EPV = 2.7 for mortality). Five-fold cross-validation was used to generate AUC and Brier scores.

Sensitivity Analyses

E-values (Table 9) were consistent with the main findings. In a complete-case sensitivity analysis restricted to patients with available APACHE II (n=98), the CrCL-mortality association remained non-significant after adjustment (OR 0.996, 95% CI 0.986–1.006; P = 0.411), consistent with the primary analysis. A Firth penalised logistic regression yielded consistent results (OR 0.997, 95% CI 0.988–1.006). To address the potential selection bias introduced by excluding dose-adjusted patients, we first compared included (n=165) and excluded (n=29) patients. The excluded patients were older (74.8 vs 70.9 years), had lower CrCL (40.4 vs 83.2 mL/min), higher SOFA scores (8.5 vs 7.4), higher ICU admission rates (79.3% vs 51.5%), higher 28-day mortality (44.8% vs 35.8%), and lower microbiological eradication (3.4% vs 16.4%), confirming that they were sicker with worse renal function (Table S1). We then performed three sensitivity analyses. First, we repeated the sequential models in all patients with complete CrCL (n=173), treating dose adjustment as a mediator rather than a covariate; the CrCL-mortality association remained non-significant after full adjustment (OR 0.998, 95% CI 0.990–1.007; P = 0.688; Table S2). Second, we applied inverse probability of treatment weighting (IPTW) to reweight the stable-dose cohort to the severity distribution of the full cohort; the CrCL estimate was essentially unchanged (OR 0.997, 95% CI 0.990–1.004; P = 0.341; Table S3). Third, selection E-values were low (1.04–1.10), indicating that only weak unmeasured selection factors could explain away the observed association. All three approaches yielded consistent null findings, suggesting that the exclusion of dose-adjusted patients did not materially affect the main conclusion. The E-value for the fully adjusted mortality model (Model 3) was 1.06, indicating that only weak unmeasured confounding could plausibly explain the residual association.

Table 9.

E-Values for the CrCL-Outcome Associations (n = 146)

Outcome	Model	CrCL OR	E-value	E-value CI
Clinical success	M0	1.001	1.030	1.090
	M2	1.001	1.030	1.100
	M3	1.003	1.060	1.120
Microbiological eradication	M0	1.004	1.070	1.110
	M2	1.001	1.030	1.100
	M3	1.001	1.040	1.100
28-day mortality	M0	0.993	1.090	1.130
	M2	0.996	1.070	1.120
	M3	0.996	1.060	1.130

Notes: M0, unadjusted model; M2, adjusted for age, sex, BMI, SOFA, and APACHE II; M3, adjusted for age, sex, SOFA, weighted comorbidity score, ICU admission, CRRT, bloodstream infection, multiple infection sites, and P. aeruginosa. E-value represents the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain away the observed association.

Discussion

Principal Finding: CrCL as a Renal-Severity Proxy

Our study, encompassing 224 critically ill patients receiving TDM-guided CAZ-AVI therapy, provides convergent evidence that CrCL functions primarily as a “renal-severity” proxy rather than as an independent predictor of clinical outcomes. Although CrCL was the dominant determinant of trough concentrations (β = −0.224 per +1 mL/min for ceftazidime; β = −0.047 per +1 mL/min for avibactam; both FDR-significant), confirming the well-established role of CrCL in CAZ-AVI clearance,2,8 this does not translate into a direct causal relationship with patient outcomes. The apparent CrCL-mortality association was completely attenuated after comprehensive adjustment for severity (unadjusted OR 0.993, P = 0.030; fully adjusted OR 0.996, P = 0.468; E-value = 1.06). This contrasts with prior real-world studies that have variably reported CrCL as either a protective factor or a marker of severity, without systematically disentangling these two explanations.4,11,12

Four convergent analyses support our conclusion. First, in the stable-dose cohort (n=146), CrCL correlated moderately with the weighted comorbidity score (Spearman ρ = −0.46) and weakly-moderately with SOFA (ρ = −0.22), ICU admission (ρ = −0.33), and age (ρ = −0.41), with the CrCL-age correlation partly attributable to the age term embedded in the Cockcroft-Gault formula, a limitation previously noted by Fratoni et al13 Second, sequential multivariable adjustment demonstrated that the CrCL-mortality association was progressively attenuated from OR 0.993 (unadjusted) to 0.996 (fully adjusted), with similar patterns observed for microbiological eradication (attenuated from OR 1.004 to 1.001) and clinical success (attenuated from OR 1.001 to 1.003). Third, LASSO-penalised logistic regression independently retained SOFA (OR 2.18), CRRT (OR 1.88), age (OR 1.13), and ICU admission (OR 1.03) as mortality predictors while shrinking CrCL to zero, consistent with prior LASSO-based analyses of β-lactam outcomes.14,15 Fourth, joint stratification by SOFA showed that CrCL had minimal incremental prognostic value within any given SOFA stratum, with mortality exceeding 66% in the SOFA ≥11 stratum across all CrCL categories. This triangulation of methods, correlation, sequential adjustment, LASSO variable selection, and joint stratification, provides suggestive evidence against a direct PK/PD mechanism linking CrCL to mortality. These findings suggest that the observed association between CrCL and mortality is substantially attenuated after adjustment for measured indicators of disease severity. They demonstrate attenuation rather than absence of an association; they do not establish that disease severity fully explains the relationship, and residual confounding by unmeasured factors remains possible (E-value 1.06). This complements the recent work of Gatti et al on continuous-infusion CAZ-AVI.15,16 The consistency of findings across sequential adjustment, LASSO, spline analysis, and E-values supports the plausibility of the results, but these analyses remain exploratory and require external validation.

Implications for TDM-Guided Dosing Strategy

Our findings have important implications for the interpretation of TDM data and for ongoing debates about aggressive PK/PD targets in critically ill patients. Recent work by Cojutti et al has suggested that attaining an aggressive joint PK/PD target (Css/MIC 4–8) may benefit patients with carbapenem-resistant Gram-negative infections,2 and continuous-infusion TDM-guided CAZ-AVI has been proposed as a strategy to optimise outcomes against KPC- and OXA-48-producing Enterobacterales.17 The descriptive PK analyses of Xu et al in ARC patients7 and Zazo et al in ICU patients8 have further highlighted the limitations of fixed-dose regimens in populations with extreme pharmacokinetic alterations. Our data, while not designed to test dose intensity directly, suggest that any potential benefit of aggressive targeting may be limited to patients with documented sub-target exposure rather than to those with low CrCL per se. In our cohort, 80.0% of CRRT patients on label-recommended dosing already achieved high ceftazidime troughs (≥40 μg/mL), and our data did not identify a consistent benefit of dose escalation across CrCL strata. Our findings are consistent with recent reports that the benefits of TDM-guided β-lactam dosing may be most pronounced in critically ill patients with severely altered pharmacokinetics (eg, ARC, CRRT), rather than across all CrCL strata.7,12,14 Importantly, the U-shaped CrCL-microbiological-eradication relationship we observed (P for trend = 0.021) was attenuated after multivariable adjustment, suggesting that the apparent “U-shape” was driven by disproportionate severity and comorbidity accumulation in the CrCL 30–50 mL/min stratum, consistent with the findings of van Duin et al in the CRACKLE cohort11 and the descriptive analyses of Piroth et al in the French OZAVIE cohort,4 rather than by a true biological optimum. This is consistent with the finding that label-recommended renal dose adjustments of CAZ-AVI may not necessarily compromise clinical efficacy.16

The recent multi-centre observational study of Silva et al5 in Italian critically ill patients and the Phase 4 study of Yuan et al18 in Chinese patients with hospital-acquired pneumonia both reported complex relationships between CAZ-AVI exposure and outcomes that cannot be reduced to CrCL-based dose adjustments alone. The systematic review of Gatti and Pea emphasised that “TDM of CAZ-AVI requires more than renal function alone”,17 a conclusion our findings directly support. The Chinese real-world study of Shi et al12 and the ICU cohort of Yu et al19 similarly reported heterogeneous clinical outcomes that were not consistently explained by CrCL strata, supporting the generalisability of our finding that severity, not CrCL, drives mortality in this population. As this study was not designed to compare dosing strategies directly, this recommendation should be considered hypothesis-generating rather than a clinical recommendation.

Avibactam Concentration: Confounding by Indication and Methodological Implications

Notably, while CrCL lost its prognostic significance after severity adjustment, avibactam concentration’s association with mortality was substantially attenuated after exclusion of Adherence (a per-protocol indicator of whether the prescribed CAZ-AVI dose adhered to the label recommendation) from the candidate predictors set (OR 1.027 per 1 μg/mL increase, 95% CI 0.995–1.061, P = 0.102) compared with the original analysis (OR 1.067, 95% CI 1.028–1.108, P < 0.001). Likewise, the avibactam-microbiological-eradication association was attenuated from OR 0.875 (95% CI 0.798–0.959, P = 0.004) to OR 0.96 (P = 0.27) after exclusion of Adherence. This pattern indicates that part of the apparent avibactam concentration-outcome signal may have reflected confounding by indication mediated through patient adherence behaviour, rather than a direct concentration-efficacy or concentration-toxicity relationship. These findings support confounding by indication rather than a direct concentration-related adverse effect, and we caution against causal interpretation of avibactam concentration effects in the absence of MIC data. We caution against causal interpretation of avibactam concentration effects in the absence of MIC data, as previously emphasised by Gatti and Pea17 and the CLSI methodology guidelines;10 the alternative explanation, that residual confounding by indication persists despite adjustment, remains plausible. This is consistent with a recent study showing that disease severity modifies the CAZ-AVI exposure-response relationship in critically ill patients.20

Limitations and Future Directions

This study has several limitations. First, its retrospective, single-centre design may introduce selection bias and limit generalisability. Because susceptibility testing used the Kirby-Bauer disk diffusion method, individual MIC values were unavailable, precluding pathogen-specific PK/PD target calculations; the ≥40 μg/mL threshold is therefore an exploratory descriptive cut-off rather than a validated therapeutic target, a limitation shared by most real-world CAZ-AVI studies.12,18 Second, the number of outcome events was limited (microbiological eradication, 26 events, EPV = 2.6, with wide confidence intervals; mortality, EPV = 5.2; exploratory 20-variable ROC model, EPV = 2.7). APACHE II was missing for 33% of patients and median-imputed. Excluding 19 patients with missing body weight and 29 patients with dose adjustments may have reduced statistical power and introduced selection bias; the excluded patients were sicker, but sensitivity analyses including them, IPTW reweighting, and selection E-values yielded consistent null findings. The requirement for at least four doses before TDM sampling may have excluded very early deaths, potentially introducing survivor bias; however, this bias would tend to attenuate rather than create an inverse CrCL–mortality association. Overfitting and residual confounding cannot be fully excluded. Third, treatment allocation was non-randomised and dictated by renal function, creating structural collinearity between dose selection and CrCL that cannot be fully eliminated by multivariable adjustment, a limitation inherent to observational TDM studies.14,18 Additionally, excluding patients who underwent dose adjustment during therapy may have introduced selection bias, as dose changes could be related to clinical deterioration or TDM findings; however, sensitivity analyses including these patients and IPTW reweighting yielded consistent results (Tables S1-S3). Fourth, the Cockcroft-Gault formula intrinsically incorporates age, which contributed to the observed CrCL-age correlation (ρ = −0.41) and may have introduced age-related confounding in our severity-adjusted models. Finally, the timing and methods of microbiological culture collection varied in routine clinical practice, and adherence behaviour may have introduced unmeasured confounding through differential follow-up intensity.

Despite these limitations, our study advances the field by demonstrating, through method triangulation, that CrCL’s prognostic role in CAZ-AVI-treated patients is largely mediated by disease severity rather than by direct PK/PD mechanisms, a finding that complements recent calls by Gatti and Pea for individualised TDM strategies that go beyond renal function alone.17 Future studies should incorporate individual MIC data, time-updated severity measures, and adherence-adjusted analyses in multi-centre prospective cohorts to establish evidence-based, pathogen-specific concentration targets for CAZ-AVI TDM.

Conclusion

In this real-world TDM cohort, CrCL was the dominant determinant of CAZ-AVI trough concentrations but appeared to function predominantly as a marker of disease severity rather than as an independent predictor of clinical outcomes. Once SOFA, organ support and comorbidity were accounted for, CrCL lost its prognostic association and added no discriminatory value (AUC 0.571 vs 0.760 for mortality). These findings do not exclude a role for PK/PD mechanisms. Rather, they suggest that the observed CrCL-mortality association is largely attributable to disease severity in this cohort. Avibactam concentration was directionally associated with mortality after LASSO-based variable selection, but the association was not statistically significant after exclusion of Adherence from the candidate predictors set, suggesting confounding by indication. These findings support a TDM-guided approach that prioritises patients with extreme pharmacokinetic phenotypes, including augmented renal clearance and receipt of continuous renal replacement therapy, rather than routine dose escalation. This should be regarded as a hypothesis for future investigation rather than a clinical recommendation. Prospective multicentre studies incorporating individual MIC values, longitudinal renal function, and time-dependent drug exposure are needed before definitive dosing recommendations can be made.

Funding Statement

This work was supported by Hainan Hongji Medical Development Foundation (grant number HJ2024011009).

Abbreviations

CAZ-AVI, ceftazidime-avibactam; CrCL, creatinine clearance; CRE, carbapenem-resistant Enterobacterales; TDM, therapeutic drug monitoring; ARC, augmented renal clearance; SOFA, Sequential Organ Failure Assessment; APACHE II, Acute Physiology and Chronic Health Evaluation II; CRRT, continuous renal replacement therapy; PK/PD, pharmacokinetic/pharmacodynamic; MIC, minimum inhibitory concentration; Css, steady-state concentration; LASSO, least absolute shrinkage and selection operator; FDR, false discovery rate; IQR, interquartile range; CLSI, Clinical and Laboratory Standards Institute.

Disclosure

The authors report no conflicts of interest in this work.