Pharmacokinetic/Toxicodynamic Analysis of Colistin-Associated Acute Kidney Injury in Critically Ill Patients

ABSTRACT

Acute kidney injury (AKI) occurs in a substantial proportion of critically ill patients receiving intravenous colistin. In the pharmacokinetic/toxicodynamic analysis reported here, the relationship of the occurrence of AKI to exposure to colistin and a number of potential patient factors was explored in 153 critically ill patients, none of whom were receiving renal replacement therapy. Tree-based modeling revealed that the rates of AKI were substantially higher when the average steady-state plasma colistin concentration was greater than ∼2 mg/liter.

TEXT

Colistin, administered parenterally in the form of its inactive prodrug colistimethate, has become an important component in the therapeutic armamentarium against infections caused by Gram-negative pathogens that are resistant to most or all of the more commonly used antibiotics. There are many aspects of the clinical pharmacology and toxicology of this old antibiotic that have emerged only over the last several years, with still more to learn (1). What has become increasingly clear already is that colistin is an antibiotic with a very low therapeutic index. Indeed, with the daily doses of colistimethate approved for clinical use, more than 50% of patients may experience acute kidney injury (AKI) (2, 3).

In a prospective observational cohort study in 102 patients receiving intravenous colistimethate, Sorlí et al. (4) used the RIFLE (risk of renal dysfunction; injury to the kidney; failure of kidney function; loss of kidney function; end-stage kidney disease) criteria (5) to assess the occurrence of AKI on day 7 and at end of treatment (EOT). AKI was defined as representing a ≥25% decrease in creatinine clearance, relative to baseline, estimated by the abbreviated “modification of diet in renal disease” equation (6). They reported that the trough plasma colistin concentration “breakpoints” that predicted AKI on day 7 and at EOT were 3.33 mg/liter and 2.42 mg/liter, respectively. More recently, the same group of investigators reported the results of a relatively small prospective validation study (64 patients) in which they assessed the trough plasma colistin concentration breakpoint of 2.42 mg/liter (7). While only 7 patients had a plasma colistin concentration of >2.42 mg/liter, the incidence of AKI in these patients was higher than in those with lower concentrations (P < 0.01). Here we report the results of an independent prospective study in a larger group of patients in which the relationship between plasma colistin exposure and AKI during colistimethate therapy has been examined.

The pharmacokinetic/toxicodynamic (PK/TD) data reported here are from analysis of a subgroup of patients in a population PK study that involved 214 critically ill patients in total (8). It was an open-label, nonrandomized study conducted in 4 centers across 3 continents. Approvals were obtained from the ethics committee of each institution, and informed consent was obtained from all patients. The inclusion and exclusion criteria for the study have been described previously, as have the procedures for colistimethate administration, collection of blood samples for quantification of colistimethate and colistin in plasma, and the population PK analysis to determine the average steady-state plasma concentration (C_ss,avg) of colistimethate and colistin for each patient (8, 9). For the PK/TD analysis described in this report, patients who were receiving renal replacement therapy at the time of initiation of colistimethate therapy were excluded but all other patients with at least 3 days of colistimethate therapy and corresponding measurements of serum creatinine were included. The characteristics of the 153 patients included in the PK/TD analysis are summarized in Table 1.

TABLE 1.

Characteristics of the 153 patients included in the PK/TD analysis

Patient parametera	Value(s)
Continuous variables [median (range)]
Age (yrs)	68 (19–101)
Wt (kg)	60 (30–122)
Ht (cm)	165 (140–193)
CLCR at baseline (ml/min/1.73 m2)	63 (7–171)
APACHE II score	20 (4–43)
Daily dose (mg colistin base activity)	200 (75–600)
Duration of colistin therapy (days)	9 (3–26)
Colistimethate Css,avg (mg/liter)	4.58 (1.0–18.9)
Colistin Css,avg (mg/liter)	2.20 (0.24–9.79)
Categorical variables (% of patients)
Male sex	66.0
With comorbidity
Malignancy	18.5
Neutropenia	2.5
Trauma	16.7
Hepatic disease	4.9
Diabetes	26.5
Transplant	16.7
Other potential nephrotoxin(s)	23.5

^aAbbreviations: APACHE II, acute physiology and chronic health evaluation II; CL_CR, creatinine clearance; C_ss,avg, average steady-state plasma concentration.

Creatinine clearance (CL_CR) was estimated from serum creatinine measurements and relevant patient-specific data for each day of treatment (10). The profiles of CL_CR versus day of therapy were plotted for each patient to inform the data analysis. The following descriptors were determined for each patient: the CL_CR at baseline, i.e., start of colistimethate therapy; the maximum CL_CR during colistimethate therapy; and the minimum CL_CR (after the time point at which the maximum value was obtained) during colistimethate therapy. This permitted calculation of the percentage of the decrease in CL_CR from baseline and also of the percentage of loss in CL_CR from the maximum observed (CL_CR_%Loss). General linear modeling and tree-based modeling (TBM) were considered for development of multivarible models for changes in CL_CR versus exposure to colistimethate and colistin and other patient covariates. Because there were some cases where CL_CR increased after starting colistimethate therapy, CL_CR_%Loss was chosen for regression tree analysis using recursive partitioning (Systat software version 13.0; Systat Software, San Jose, CA). Briefly, details of the parameters used in the TBM analysis are as follows: least-squares LOSS function (LSQ); maximum number of splits (NSPLIT), 10; minimum proportion reduction in error for any split (PMIN), 0.05; minimum split value (SMIN) at any node, 0.05; minimum count per node (NMIN), 5. Candidate covariates considered included the following: measures of drug exposure, such as daily dose, cumulative dose, days of therapy, and C_ss,avg of colistimethate and colistin; administration of other potential nephrotoxins; and patient characteristics such as age, sex, race, body size, baseline CL_CR, and acute physiology and chronic health evaluation II (APACHE II) score.

Overall, 49% of 153 patients had a CL_CR_%Loss of ≥25%, while 38.8% and 8.35% of patients had decreases of ≥50% and ≥75%, respectively. Decreases of ≥25%, ≥50%, and ≥75% correspond to “risk,” “injury,” and “failure” in the RIFLE categorization of AKI (5). There were no patients with AKI in the two highest RIFLE categories. The proportion of patients in the total cohort with a decrease of at least 25% is similar to the proportions reported in other studies of colistin-associated AKI (3, 11–15). Using CL_CR_%Loss, the TBM analysis partitioned the patients into five nodes dependent on the relationship of the magnitude of the decrease in CL_CR to both creatinine clearance at baseline (CL_CR_BL) and plasma colistin C_ss,avg (Fig. 1). A CL_CR_BL value of 80 ml/min/1.73 m² was a breakpoint value, and there was an effect related to the plasma colistin C_ss,avg for patients with CL_CR_BL values below and above this value. The TBM identified higher plasma colistin C_ss,avg values (≥1.88 or ≥2.25 mg/liter; see nodes 2 and 5 compared to nodes 1 and 4, P < 0.02); the interaction of higher CL_CR_BL values (≥80 ml/min/1.73 m²) with higher plasma colistin C_ss,avg values (see node 5, P < 0.05) (Fig. 1); and, as reported previously (16), duration of colistimethate therapy as risk factors for colistin-associated AKI. At lower plasma colistin C_ss,avg values, the CL_CR_BL value was not statistically significant (nodes 1 and 4 were not different, P > 0.05). No other variables were significant. Lower CL_CR_BL values were also associated with earlier onset of AKI (i.e., CL_CR_%Loss of ≥25%); the median times of onset were 2.5 and 2 days for patients in nodes 1 and 2, respectively, and 4.2 and 5 days for patients in nodes 4 and 5, respectively (P < 0.05). These onset times are in good agreement with the literature, where the time for onset of AKI has been reported to be in the first week of therapy in the majority of cases (12–22).

FIG 1.

Results of tree-based modeling. Nodes 1, 2, 3, 4, and 5 represent patient cohorts identified as showing an association of percentage of loss in creatinine clearance from the maximum observed (CL_CR_%Loss) with baseline creatinine clearance and the average steady-state plasma concentration of colistin (C_ss,avg). In nodes 1 and 3, there were no patients who experienced a CL_CR_%Loss ≥75%. Nodes 1, 2, 3, 4, and 5 comprised 20, 79, 18, 24, and 12 patients, respectively.

As shown in Fig. 1, plasma colistin C_ss,avg values of ≥1.88 mg/liter and ≥2.25 mg/liter for patients with lower and higher CL_CR were associated with substantially higher rates of AKI of all three categories (i.e., CL_CR loss of ≥25%, ≥50%, and ≥75%). Considering the experimental differences between the present study and that of Sorlí et al. (4), including in the current study the more intensive monitoring of changes in CL_CR determined by a method that accommodates either stable or changing serum creatinine levels (10), the breakpoints reported here are in good agreement with the value of 2.42 mg/liter proposed previously (4, 7). It is clear that the plasma colistin exposure associated with increased risk of AKI overlaps the exposure associated with antibacterial effect (8, 23, 24). In view of the low therapeutic index, measurement of the plasma colistin concentration to assist in guiding therapy may be beneficial, although care is needed in undertaking therapeutic drug monitoring because of the possible ongoing conversion of colistimethate to colistin after collection of samples unless samples are collected and processed appropriately (1, 25). Approaches to minimize the likelihood of colistin-associated AKI should be strongly considered, although the beneficial effects of these have not been proven in well-controlled prospective clinical trials (2, 3, 26).