Applying Cefepime Population Pharmacokinetics to Critically Ill Patients Receiving Continuous Renal Replacement Therapy

ABSTRACT

Patients admitted to the intensive care unit (ICU) may need continuous renal replacement therapy (CRRT) due to acute kidney injury or worsening of underlying chronic kidney disease. This will affect their antimicrobial exposure and may have a significant impact on the treatment. We aim to develop a cefepime pharmacokinetic (PK) model in CRRT ICU patients and generate the posterior predictions for a group and assess their therapy outcomes. Adult patients, who were admitted to the ICU, received cefepime, and had its concentration measured while on CRRT were included from three different data sets. In two data sets, samples were collected from the predialyzer, postdialyzer ports, and effluent fluid at different times within the same dosing interval. The third data set had only cefepime plasma concentration measured as part of clinical service. Patients’ demographics, cefepime regimens and concentration, CRRT parameters, and therapy outcomes were recorded. NPAG was used for population PK and posterior predictions. A total of 125 patients were included. Cefepime was described by a five-compartment model, and the CRRT flow rates described the rates of cefepime transfer between compartments. The posterior predictions were generated for the third data set and the median (range) fT_>MIC was 100% (27%–100%) and fT_>4×MIC was 64% (0%–100%). The mortality rate was 53%. There was no difference in target attainment in terms of clinical cure and 30-day mortality. This model can be used as a precision dosing tool in CRRT patients. Future studies may address other PK/PD targets in a larger population.

INTRODUCTION

Sepsis in critically ill patients is associated with 25% mortality and may cause 50% of renal injuries in the intensive care unit (1–4). As a result, patients may be started on renal replacement therapy while still receiving antibiotics. This will add to the variability in antimicrobial exposure in this population (5–7). Continuous renal replacement therapy (CRRT) is one of the modalities used to manage renal insufficiency in ICU and may provide tolerability (8). The variable CRRT flow rates, antimicrobial characteristics, and patient’s pharmacokinetics can all affect the antimicrobial exposure in those receiving such type of renal replacement therapy (9–11). Severe outcomes, including resistance, treatment failure, and even death, can occur when the antimicrobial exposure is not optimized to eradicate the pathogen (12, 13).

Beta-lactams are commonly prescribed in the ICU and their therapy is optimized by maximizing the time the free concentration is above the MIC (fT_>MIC) (14). Cefepime, a fourth generation cephalosporin, is one of the most commonly administered antimicrobials to patients admitted to hospitals in the United States (15). It is of great importance that septic patients receive early and appropriate antimicrobial therapy to reduce the risk of mortality (16). This includes appropriate antimicrobial exposure (17). Currently, there are few data published in the literature on the optimal PK/PD target in CRRT patients and what cefepime regimen is needed to achieve such target (18–22), which highlights the need for more investigation in this area. We aim to assess cefepime target attainment in a group of CRRT patients based on population PK modeling and evaluate their therapy outcomes.

RESULTS

Patient characteristics.

One-hundred and 25 patients contributed 390 plasma and 95 effluent samples. Table 1 shows the baseline characteristics for each data set. Among the three data sets, the mean age ranged from 53 to 57 years and weight from 96 kg to 119 kg. Sixty-three percent were males. Two patients received continuous venovenous hemodialysis (CVVHD) and seven received sustained low efficiency dialysis (SLED), while the rest received continuous venovenous hemofiltration (CVVH).

TABLE 1.

Baseline characteristics^a

	First dataset (n = 10)	Second dataset (n = 4)	Third dataset (n = 111)
Age, yrs	53 (11)	57 (15)	57 (15)
Male, n (%)	7 (70)	3 (75)	69 (62)
wt, kg	119 (27)	97 (8)	96 (39)
CRRT modality, n (%)  CVVH  CVVHD  SLED	8 (80) 2 (20) 0	4 (100) 0 0	109 (98) 0 7 (6)
Samples, n  Predialyzer serum  Postdialyzer serum  Effluent filtrate	82 80 79	16 16 16	196 0 0
SOFA score	-	-	12 (4)
Source of infection, n  Lung  Intra-abdominal  Blood  Skin & soft tissue  Bone & joint  Urinary tract  Vascular  Central nervous  Device related	-	-	68 24 21 12 7 5 3 1 1
Bacteria isolated, n  Pseudomonas aeruginosa  Enterobacter cloacae  Klebsiella pneumoniae  Escherichia coli  Proteus mirabilis  Acinetobacter baumannii  Citrobacter freundii	-	-	31 16 9 8 3 3 2
MIC, mg/L	-	-	5 (12)

^aData presented as mean (SD) unless specified.

CRRT, continuous renal replacement therapy; CVVH, continuous venovenous hemofiltration; CVVHD, continuous venovenous hemodialysis; SLED, sustained low-efficiency dialysis; SOFA, sequential organ failure assessment score.

Population pharmacokinetic model.

By minimally adjusting the ranges of central volume of distribution (V_d) and rates of transfer between the peripheral and the central compartments in the previously published model (23), cefepime was still described well by a five-compartment model; two compartments for the patient and three for the CRRT machine (Fig. 1). The CRRT flow rates adequately described the cefepime transfer through the CRRT machine, consistent with our previous observation (23). Table 2 shows the updated parameter estimates for the final model. The mean (SD) for the rate of elimination is 0.12 h⁻¹ (0.17), the rate of transfer from the central to the peripheral compartment is 1.65 h⁻¹ (1.04), the rate of transfer from the peripheral to the central compartment is 2.58 h⁻¹ (2.75), and the V_d in the central compartment is 28.09 L (18.49). Fig. 2 shows the observed versus population and individual predicted predialyzer, postdialyzer, and effluent concentrations.

FIG 1.

Cefepime five-compartment model in patients on CRRT. *When CRRT was off, all these flow rates were set to zero. CRRT, continuous renal replacement therapy; IV, intravenous; k_cp, transfer rate from the central to the peripheral compartment; k_e, rate of elimination; k_pc, transfer rate from the peripheral to the central compartment.

TABLE 2.

Population parameter estimates for cefepime final model

Parameter	Median	95% credibility interval	Mean	SD	Shrinkage (%)
ke, hr−1	0.06	0.04–0.11	0.12	0.17	66.48
kcp, hr−1	1.40	0.84–2.61	1.65	1.04	55.74
kpc, hr−1	0.74	0.52–4.32	2.58	2.75	56.05
Vcentral, L	21.89	17.76–41.00	28.09	18.49	52.61
Veffluent, L	0.25	0.06–0.77	0.62	0.77	67.74
VCRRT, L	0.18	0.04–3.72	1.77	2.50	63.56
Vpostdialyzer, L	0.17	0.04–2.12	1.17	1.52	56.51

CRRT, continuous renal replacement therapy compartment; k_cp, rate of transfer from the central to the peripheral compartment; k_e, elimination rate constant; k_pc, rate of transfer from the peripheral to the central compartment; SD, standard deviation; V, volume of distribution.

FIG 2.

Observed versus population and individual predicted cefepime concentrations in (A) predialyzer, (B) effluent, and (C) postdialyzer compartments. The solid line represents the regression line and the dashed line represents the unity line.

Posterior prediction and therapy outcomes.

Fig. 3 shows the MIC distribution for the University of Florida (UF) Health data set. The individual predictions of cefepime concentration were generated for all UF Health patients and the fT_>MIC and fT_>4×MIC were calculated. The median (range) of fT_>MIC was 100% (27%–100%) and fT_>4×MIC was 64% (0%–100%). Most of the patients (83%) achieved 100% fT_>MIC and 33% achieved 100% fT_>4×MIC. Fifty-two patients (47%) had clinical cure, and the 30-day mortality was reported in 59 patients (53%).

FIG 3.

MIC distribution for UF data set.

Fig. 4 shows the distribution of fT_>MIC, fT_>4×MIC, and free trough to MIC ratio (fC_min/MIC) against clinical cure and 30-day mortality. The target attainment looks similar between patients who had clinical cure and those who did not, and those who had 30-day mortality and those who did not. The results of the logistic regression showed that fT_>MIC, fT_>4×MIC, and fC_min/MIC were not associated with clinical cure or mortality in this group of patients. The results were the same after controlling for Sequential Organ Failure Assessment (SOFA) score.

FIG 4.

fT_>MIC, fT_>4×MIC, and fC_min/MIC versus clinical cure and 30-day mortality.

DISCUSSION

We presented cefepime population PK model in critically ill patients receiving CRRT. Using the flow rates to describe cefepime transfer rate between patient and the machine, the model described the data well regardless of the CRRT modality (CVVH, CVVHD, or SLED). Using this model as a precision dosing tool may provide the clinicians with better predictions because the CRRT parameters are known. We calculated the PK/PD target attainment for 111 patients from UF who achieved good target attainment for fT_>MIC but not for fT_>4×MIC. The mortality rate was high in this group (53%) which is similar to the values reported in the literature in this population.

Few articles were published on cefepime population PK in CRRT patients. Our model was an extension to a previously developed model in CRRT patients which emphasized the importance of CRRT parameters and downtime and their impact on cefepime concentration in this population. The simulations showed that cefepime 2 g loading followed by 2 g extended infusion every 8 h achieved 100% fT_>MIC at MIC ≤16 mg/liter with very low probability of reaching neurotoxicity threshold of trough of 70 mg/liter (23). In another model, ultrafiltration rate was used as a covariate on the volume of distribution and clearance. To achieve >90% probability of target attainment (100% and 60% T_>MIC) at MIC 8 mg/liter, cefepime 1 g every 6 h or 2 g every 8 h was needed (20). Another study evaluated the PK of different beta-lactams, including cefepime, in CRRT patients. Eight out of 53 patients received cefepime 2 g every 12 h and the median (range) total trough concentration was 11 mg/liter (3–22). The authors evaluated the number of patients who achieved the target T_>4×MIC and found no patient achieved that target at MIC ≥8 mg/liter (19). Some other studies evaluated the impact of CRRT intensity on cefepime elimination. In a retrospective study of 50 ICU patients who received beta-lactam therapy, nine received cefepime. There was a weak correlation between the clearance and CRRT intensity (22). In other studies, authors used simulations using published PK values and found that CRRT intensity may not have significant impact on cefepime exposure (18, 24). The differences in the impact of CRRT intensity on cefepime exposure between what is published in the literature and our previous results are most likely due to the way of incorporating the CRRT parameters as covariates in the model (23). In addition, there are differences between the simulation approaches performed. Simulations based on parametric models sample from the continuous normal or lognormal probability distributions of the parameters estimated in the population analysis, while our simulations were semiparametric, in that a normal distribution was assumed around each support point and the weight of each distribution is equal to the probability of the support point or number of patients (25).

Preclinical studies suggested that cephalosporins may have static effect at fT_>MIC 40% but 70% is needed for the maximum kill, and even higher target for suppression of new resistance (i.e., trough:MIC ratio of 4) (14, 26, 27). In cases of critical illness, and with the assumption that the PK/PD targets should be achieved at the site of infection, clinicians at our institution usually target a range of 100% fT_{>1× − 4×MIC} in plasma. Although most of the patients in our study achieved 100% fT_>MIC, the mortality rate was >50% which is similar to other reported percentages in the literature (28, 29). We were not able to detect a difference in target attainment in terms of clinical cure and 30-day mortality as it was similar among these groups. This might be due to several factors, including (1) the small sample size and (2) the high percentage of patients who achieved 100% fT_>MIC, which in turn might have limited the ability to compare the exposure among the groups. Other factors might include (3) using a fixed protein binding fraction (20%), which might have varied in these patients, (4) the high mortality rate normally reported in similar populations, and (5) using breakpoints instead of actual MICs for some of the patients.

One of the limitations of our study is that UF data were clinical and sparse which increased the shrinkage in the model. This means that the individual parameter estimates might have shrunk toward the population parameter estimates, which in turn may have resulted in inaccurate individual PK profiles, especially when calculating area under the concentration-time curve. However, given that most of the sparse data were trough samples, which were mainly used for assessing individual target attainment, the fT_>MIC generated using this model may still hold. Another limitation is measuring the total cefepime concentration in two out of three data sets, assuming an unbound fraction of 80%, which might have added to the variability in our model. Also, the UF sample size was relatively small to detect a difference in the clinical outcome based on PK/PD target attainment and the mortality rate was high. In addition, we used breakpoints in case there was no MIC reported or microorganism growing. Future studies should combine having a larger sample size and assessing the clinical outcome based on PK/PD target attainment in this special population.

Conclusions.

Cefepime was well described using a five-compartment model and CRRT flow rates. The majority of patients achieved 100% fT_>MIC while one third achieved 100% fT_>4×MIC. The mortality rate was >50%.

MATERIALS AND METHODS

This was a PK study combining different cefepime concentration data sets. The first data set was from a prospective, PK study at the University of Cincinnati Medical Center (NCT02458261) and included adult ICU patients who received cefepime 2 g IV every 8 h via a 4-h infusion while on CVVH or CVVHD. Exclusion criteria were pregnancy, cystic fibrosis, incarceration, admission for burns, and unmeasured or >400 mL of urine output in the last 24 h. Age, sex, weight, urine output, blood, dialysate, therapy fluid, and ultrafiltrate flow rates were reported. Two sets of predialyzer serum, postdialyzer serum, and effluent samples were collected after both the first and the fourth to sixth cefepime doses. The samples were collected at times 1 h, 2 h, 3 h, 4 h, and 8 h after initiating cefepime infusion. Cefepime unbound concentration was measured using high performance liquid chromatography with UV detection (298 nm), and Microcon-30kDa filters (Millipore, Cork, Ireland) were used to obtain the protein-free ultrafiltrate for drug quantification (23, 30).

The second data set was from a prospective, open-label PK study at University of Cincinnati Medical Center including adult patients admitted to the medical or surgical ICU, receiving CVVH or CVVHD, and received cefepime 2 g IV over 30 min every 8 h. Exclusion criteria were pregnancy and incarceration. Age, sex, weight, urine output, blood, dialysate, therapy fluid, and ultrafiltrate flow rates were reported. Predialyzer serum, postdialyzer serum, and effluent samples were collected at 0.5-h, 2-h, 4-h, and 8-h postinfusion after the patient received at least three doses of cefepime. Total cefepime drug concentration was measured using high-performance liquid chromatography (31).

The final data set was from a retrospective chart review of ICU patients admitted to UF Health between 2016 and 2021, on CRRT, received cefepime therapy, and had cefepime plasma concentration measured. Typically, patients will receive cefepime 2 g IV every 12 h infused over 30 min and have their cefepime peak and trough plasma concentration measured. Sometimes, trough only samples will be tested. Patients who had CRRT started after measuring cefepime concentration were excluded. Age, sex, weight, blood, therapy fluid, and ultrafiltrate flow rates were collected. Total cefepime plasma concentration was measured at the Infectious Disease Pharmacokinetics Lab (http://idpl.pharmacy.ufl.edu) using validated liquid chromatography with tandem mass spectrometry assays. For the second and third data sets with total cefepime concentration, cefepime was assumed to be 20% protein bound (32). The reported MIC was used for the fT_>MIC calculation in this data set and the European Committee on Antimicrobial Susceptibility Testing (EUCAST) breakpoint for the growing bacteria was used in case the MIC was not reported. If no microorganism grew, we used an MIC of 8 mg/liter corresponding to Pseudomonas aeruginosa breakpoint.

CRRT machines.

NxStage CRRT machines with PUREMA high permeability polysulfone membrane filters (NxStage Medical Inc., Lawrence, MA) were used to provide the renal replacement therapy at both sites of the study.

Population PK model building and posterior predictions.

Cefepime PK model was built using Pmetrics v1.9.7 nonparametric adaptive grid (33). Initially, the PK model was built using the first and second data sets (University of Cincinnati) having predialyzer, postdialyzer, and effluent cefepime concentration. Next, the third data set (UF Health) containing only plasma samples was added as predialyzer samples. Given our previous experience with the CRRT parameters impact on cefepime concentration (23), we kept real-time changes in the blood and total effluent (the sum of therapy fluid and ultrafiltrate) flow rates in the model to describe the transfer of cefepime through the CRRT machine:

$$k=\frac{\mathit{Flow\hspace{0.17em}}\mathit{rate}}{V_d}$$

where k is the rate of transfer, V_d is the volume of distribution of the compartment, and the flow rate will be the blood flow rate in case describing the transfer of cefepime from the patient to the CRRT machine or vice versa, or the total effluent rate in case describing the transfer from the CRRT machine to the effluent compartment and the drain (Fig. 1). Similar to the flow rates, downtimes were recorded. In case the CRRT machine was down, all flow rates were set to zero for the downtime period, then flow rates were set to new values when machine was back on.

The ranges of non-CRRT PK parameters, including the patients’ central V_d, rate of elimination, and rate of transfer between the central and peripheral compartments were optimized. The best final model was chosen based on the lowest Akaike information criteria, highest coefficient of determination (R²) of observed versus predicted concentration plots, and lowest imprecision and bias. The assay error (standard deviation [SD]) and environmental noise was accounted for using error polynomials as a function of observed concentration (standard deviation = C₀ + [C₁ × observed concentration]) using C₀ (intercept) and C₁ (slope) values of 1 and 0.1, respectively. Gamma multiplicative error model was used to estimate residual error (error = SD × gamma) and used a value of 2 (34).

The posterior predictions for the patients in UF Health data set were generated. The fT_>MIC and fC_min/MIC were calculated for each patient from time 0 to 24 h of cefepime therapy and correlated with the final therapy outcome. Clinical cure was defined as the resolution of infection-related symptoms at the end of therapy (including normalization of body temperature and WBC count and taking the patient off mechanical ventilation or vasopressors) without change or addition of antibiotic therapy and noninitiation of a new antibiotic within 48 h of stopping the original one. De-escalation to a narrower spectrum antibiotic was not considered clinical failure.

Statistical analysis.

Continuous variables were summarized as means and standard deviations or medians and ranges or interquartile ranges, while categorical data were summarized as counts and percentages. Logistic regression was performed using clinical cure and 30-day mortality as outcomes, and fT_>MIC, fT_>4×MIC, fC_min/MIC, and SOFA score as predictors. JMP Pro v15.0 was used for statistical analysis.