Antibiotics in critically ill patients: a systematic review of the pharmacokinetics of β-lactams

Introduction Several reports have shown marked heterogeneity of antibiotic pharmacokinetics (PK) in patients admitted to ICUs, which might potentially affect outcomes. Therefore, the pharmacodynamic (PD) parameter of the efficacy of β-lactam antibiotics, that is, the time that its concentration is above the bacteria minimal inhibitory concentration (T > MIC), cannot be safely extrapolated from data derived from the PK of healthy volunteers. Methods We performed a full review of published studies addressing the PK of intravenous β-lactam antibiotics given to infected ICU patients. Study selection comprised a comprehensive bibliographic search of the PubMed database and bibliographic references in relevant reviews from January 1966 to December 2010. We selected only English-language articles reporting studies addressing β-lactam antibiotics that had been described in at least five previously published studies. Studies of the PK of patients undergoing renal replacement therapy were excluded. Results A total of 57 studies addressing six different β-lactam antibiotics (meropenem, imipenem, piperacillin, cefpirome, cefepime and ceftazidime) were selected. Significant PK heterogeneity was noted, with a broad, more than twofold variation both of volume of distribution and of drug clearance (Cl). The correlation of antibiotic Cl with creatinine clearance was usually reported. Consequently, in ICU patients, β-lactam antibiotic half-life and T > MIC were virtually unpredictable, especially in those patients with normal renal function. A better PD profile was usually obtained by prolonged or even continuous infusion. Tissue penetration was also found to be compromised in critically ill patients with septic shock. Conclusions The PK of β-lactam antibiotics are heterogeneous and largely unpredictable in ICU patients. Consequently, the dosing of antibiotics should be supported by PK concepts, including data derived from studies of the PK of ICU patients and therapeutic drug monitoring.


Introduction
Infection and sepsis, whether community-or hospitalacquired, are important causes of morbidity and mortality in ICU patients [1,2]. Despite all of the research, sepsis therapy continues to depend on supportive management of the different organ dysfunctions and failures and on specific therapy for infection with timely and appropriate antibiotics and/or focus control.
The β-lactam antibiotics, because of their large antimicrobial spectrum and low toxicity, are among the first-line therapies for critically ill patients, especially when a Gram-negative infection is suspected. However, the efficacy of antibiotics is not easily evaluated, since the clinical response is usually unnoticeable before 48 hours of therapy [3]. Moreover, the unavailability of routine therapeutic drug monitoring for the great majority of these drugs makes it difficult to distinguish clinical failure due to underdosing from lack of in vivo organism susceptibility.
Considerable evidence demonstrates that free drug time above bacteria minimal inhibitory concentration (f T > MIC) is the measure of drug exposure most closely linked to the ability of β-lactam antibiotics to kill the target bacteria [4]. T > MIC is dependent on the halflife (T 1/2 ) of β-lactam antibiotics and their serum concentration.
The serum concentration of an antibiotic depends on the dose delivered, its bioavailability and its volume of distribution (V d ). V d is a mathematical construct and refers to the size of a compartment necessary to account for the total amount of the drug, assuming that its concentration in the whole body is equal to that measured in plasma. Drugs that distribute essentially in the extracellular fluid (mainly hydrophilic) have low V d , whilst drugs that have rapid cellular uptake (lipophilic) have high V d [5,6].
Both V d and drug clearance (Cl) may be increased in ICU patients [7]. Therapeutic procedures, notably large-volume and blood products infusions, positive pressure ventilation, surgical procedures, capillary leak and reduction in albumin serum concentration all contribute to alter the concentration-time relationship of many drugs. A rise in the V d , although it reduces drug concentration, might proportionally increase T 1/2 , since T 1/2 = V d /(Cl × 0.693) [7]. On the contrary, a high Cl may reduce the exposure of antibiotics to bacteria ( Figure 1).
Renal Cl may be increased in septic patients because of increased renal blood flow. This has recently been shown to be a common finding in ICU patients, particularly surgical and trauma patients [8] but also septic medical patients [9]. Besides, in the study by Baptista et al. [9], the authors showed that commonly used formulas used to calculate glomerular filtration rate usually underestimate creatinine (Cr) Cl. Consequently, these authors recommended direct Cr Cl measurement.
Moreover in ICU patients, maldistribution of blood flow in the microcirculation [10], namely, in patients in septic shock, may further decrease the drug concentration in the infected tissue [11]. These pharmacokinetic (PK) changes are sometimes influenced by the clinical course of the infection itself [12]. Consequently, PK parameters measured in healthy volunteers may not correctly predict concentrations in septic ICU patients, particularly early in the course of a severe infection [13,14].
Ideally, individualized dosing strategies should account for the altered PK and pathogen susceptibility in each patient. Despite the fact that some studies addressed this issue, this information had not yet been aggregated. Furthermore, β-lactam antibiotic PK are rarely analysed outside clinical trials. Therefore, we performed a systematic review of studies that addressed the PK parameters of β-lactam antibiotics in ICU patients to assess the relationship between dose and schedule of β-lactam antibiotics and their adequacy according to pharmacodynamic (PD) end points. We also reviewed studies assessing the concentrations of β-lactam antibiotics in different tissues. Our primary intention was to aggregate PK information in this particular population and to contribute to the design of individualized dosing regimens of these drugs.
We also included studies that involved the development of PD models using PK of ICU patients and bacterial MICs. These techniques allow the calculation of the presumed T > MIC and therefore the percentage of patients in which the antibiotic will achieve its PD Figure 1 ICU patients present pharmacokinetic changes of antibiotics that may alter bacterial exposure. Concentration-time curve of antibiotics in healthy volunteers (left panel). A large volume of distribution (V d ) (middle panel) is often present in ICU patients, leading to decreased maximum concentration (C max ) but a longer half-life (T 1/2 ) and eventually higher time that the antibiotic concentration is above the bacteria minimum inhibitory concentration (T > MIC). The antibiotic area under the concentration time curve (AUC) remains virtually the same. An increase in drug clearance (Cl) (right) is associated with decreases in AUC, T 1/2 and T > MIC. Straight dotted lines-bacteria minimum inhibitory concentration.
target: that is, the antibiotic's probability of target attainment (PTA) [15,16]. The cumulative fraction of response (CFR) is calculated by multiplying the PTA obtained for each MIC by the MIC distribution according to a microbiological database [16].

Materials and methods
The data for this review were identified by a search of PubMed (January 1966 to December 2010) as well as bibliographic references from relevant articles, including reviews on this subject and all selected studies. The search terms used were 'antibiotic' or 'carbapenem' or 'penicillins' or 'cephalosporins', and 'intensive care' or 'critically ill' or 'critical care' or 'severe sepsis' or 'septic shock', and 'pharmacokinetics' or 'pharmacodynamics'. All relevant studies in the English-language literature that described antibiotic PK in critically ill patients were assessed ( Figure 2 (14): Only studies that described PK of antibiotics given intravenously to infected patients were selected. Studies referring to prophylactic antibiotics or to PK in patients under any type of renal replacement therapy were excluded. In fact, these studies are mainly directed to the measurement of Cl during renal replacement therapy to determine the ideal antibiotic dose and therefore are not easily compared with studies addressing the intrinsic PK of ICU patients. Furthermore, a full revision of those studies has recently been published [17].
For the purpose of our systematic review, we analysed only studies of antibiotics with at least five published references. This threshold of five referenced studies was arbitrarily chosen so that we could derive more representative and consistent data on the PK of each antibiotic. The weighted mean of the V d was calculated so that we could present a graphic representation of each analysed antibiotic ( Figure 3).

Results
A total of 57 studies assessing an aggregate of six different β-lactam antibiotics were selected.

Carbapenem Meropenem
Several studies have addressed meropenem PK in ICU septic patients. High V d and Cl have usually been reported, as well as a low binding fraction: < 10% [18]. Consequently, a large heterogeneity of PK parameters was found, exceeding a twofold variation (Table 1 and Figure 3). The larger reported V d , a mean of 34.4 L, was noted on the second day of therapy in eight ventilatorassociated pneumonia (VAP) patients [19] with a mean body weight of 73 kg. In a Thai VAP population (N = 9) with a lower mean body weight (only 54.2 kg), the mean V d was 6.0 L despite also being measured after 48 hours of therapy with meropenem [20]. This supports the hypothesis of a potential relationship between body weight and V d .
Meropenem Cl ranged from a mean of 4.7 L/hour to a mean of 15.4 L/hour and was generally found to be closely correlated to Cr Cl. In fact, in patients with severe sepsis, the six patients with the lower Cr Cl (< 50 mL/ minute) had the higher T > MIC and area under the concentration time curve (AUC) (230.2 mg × hour/L vs.   119.4 mg × hour/L; P = 0.001), despite a reduction in the dose administered, from 1 g every 8 h (tid) to 1 g every 12 h (bid) [21]. One study addressed the variability of individual meropenem PK between the first and fourth days of therapy in 11 surgical patients [22]. Despite an increase in Cr Cl from a mean of 63.9 to 79.1 mL/minute during the study period, meropenem V d , Cl and AUC remain unchanged. Nevertheless, in another study, by Taccone et al. [23], predefined targets were reached in only 75% of severe sepsis and septic shock patients after the first dose of 1 g of meropenem (Table 2), despite the inclusion of patients with acute renal failure (22%) who did not receive renal replacement therapy. These authors concluded that PK changes induced by sepsis were largely unpredictable and that none of the evaluated clinical parameters were predictive of PK adequacy: namely, age, severity, presence of shock, use of vasopressors and mechanical ventilation. Also, Roberts et al. [24] showed that the V d in patients with severe sepsis had great variability, both in the same patient (especially the central compartment: roughly 45%) and in different patients (nearly 27%). In their study, despite the fact that all patients had a serum Cr < 1.36 mg/dL, the meropenem Cl variability (in the same patient and between patients) still ranged between 10% and 20%.
The time of infusion of meropenem has also been shown to influence its T > MIC. In a cross-over study of nine Thai VAP patients [20], after 48 hours of therapy, 1 g of meropenem tid in 30-minute infusions provided an adequate T > MIC in 74.7% of the patients, for a MIC of 1 mg/L. However, with a MIC of 16 mg/L, only the meropenem regimen of 2 g tid given in an extended infusion (two hours) led to a T > MIC > 40% [20].
Meropenem tissue PK have been evaluated by microdialysis in several studies ( Table 3). The tissue-to-plasma meropenem mean ratio on the first day of antibiotic therapy was found to be 0.74 in the peritoneum [25] and 0.44 in subcutaneous fat [24]. The meropenem CFR was calculated for the 10 patients for whom serum levels were measured in this study according to the Mystic microbiological database [26]. The CFRs were 100% for Enterobacteriaceae and 40.6% for Pseudomonas  aeruginosa after bolus dosing, whilst with continuous infusion they were 100% for both bacteria, despite the use of a small daily dose (2 g/day) [27].

Imipenem
In ICU patients, increased V d and Cl of imipenem have also been reported (Table 1). Therefore, its T 1/2 and T > MIC may be difficult to predict, depending on the relative changes of these two parameters. This difficulty was shown by Belzberg et al. [28] in a cohort of ICU surgical and trauma patients with presumed Gram-negative sepsis. In this relatively young population (mean age 45.2 ± 17 years and mean body weight 79.7 ± 17.7 kg), 44% of patients presented trough levels lower than the intended 4 mg/L at steady state. A mean Cr Cl of 103.8 mL/minute was found, but with large variability: two patients had renal failure and nineteen patients had a Cr Cl >    120 mL/minute. Nevertheless, no correlation was found between PK parameters and body weight, severity of disease, blood pressure or renal function [28].
Another study compared meropenem and imipenem first-dose PK in patients with normal renal function (serum Cr < 1.5 mg/dL). Again, both V d and Cl were significantly elevated, although more so in the meropenem group [29]. However, their T > MIC for sensitive isolated pathogens were similar. Again, there was a relationship between Cr Cl and T 1/2 : Patients with a Cr Cl < 50 mL/minute had a significantly longer T 1/2 for both antibiotics.
The PD efficacy of imipenem is also influenced by the dose and the time of infusion [30]. Using PK data from a cross-over steady-state study of VAP patients, Jaruratanasirikul and Sudsai [30] showed by modelling of imipenem PD that, for a MIC of 4 mg/L, a 500-mg dose delivered every 6 hours (qid) for 30 minutes achieved a T > MIC of 64.7% and increased to 76.5% with a 2-hour infusion. However, this study excluded shock and renal failure patients (Cr Cl < 60 mL/minute). With PD modelling of PK data derived from another 20 VAP patients [31], continuous infusion led to improved PTA despite the use of lower dosages (Table 2). In this latter study, all patients had f imipenem T > MIC of 100%, but three patients died.
Tissue microdialysis had been used to assess imipenem PK, but with very dissimilar results (Table 3): namely, the tissue-to-plasma ratio. This has been found to be markedly depressed in a cohort of severe critically ill patients compared to healthy volunteers (subcutaneous tissue-to-plasma 0.14 vs. 0.43 and muscle tissueto-plasma 0.11 vs. 0.5, respectively) [32]. However, Dahyot et al. [33] disputed these results and found f imipenem in plasma and muscle to be virtually superimposed at any time, both in patients and in healthy volunteers. Some differences exist between these two studies. In the Tegeder et al. study [32], the patients had lower Cr Cl (medians 32.8 mL/minute vs. 156 mL/ minute) and samples were collected at steady state and not after the first dose. Moreover, Dahyot et al. [33] accounted only for the f imipenem in plasma and found higher imipenem V d and Cl. Different methods of calculating in vivo microdialysis recovery rates may also explain some of the diverse observed results. Nevertheless, low imipenem penetration ratios, as low as 0.06 [34,35], in bronchial secretions were reported in pneumonia patients (Table 3).

Penicillins Piperacillin
Similarly to other β-lactams, piperacillin V d and Cl have generally been found to be increased in ICU patients (Table 1). However, most studies have excluded renal failure patients.
Piperacillin Cl and trough concentrations were strongly related to Cr Cl [36][37][38]. Taccone et al. [23] showed that only 15% of patients with high Cr Cl (> 50 mL/minute) maintained piperacillin concentrations > 50% of T > 4 × MIC after the first antibiotic dose, as opposed to 71% of patients with lower Cr Cl (P = 0.03). In contrast, in 10 young burn patients (mean total burned area 40.8 ± 3.1%) with a mean Cr Cl of 119.8 mL/minute and Pseudomonas aeruginosa infection, the authors found a 20% increase in T 1/2 after the first dose of antibiotic compared to the third day of therapy, which was related to a larger V d (mean of 19.6 L vs. 16.4 L) [38]. Overall, the piperacillin AUC was similar in the two measurements (mean of 640 mg × hour/L vs. 622 mg × hour/L).
Piperacillin is stable for at least 24 hours at room temperature, making it a suitable choice for continuous infusion. With this strategy, higher steady-state concentrations are expected, theoretically providing a higher T > MIC even with the use of a lower daily dose [39]. A study by Rafati et al. [40] also supports this strategy. These authors showed that, for a MIC = 16 mg/L, the T > MIC was higher with continuous infusions (8 g/day) than with bolus dosing (3 g tid) (100% vs. 62%, respectively). However, the mortality rate was similar.
In VAP patients, piperacillin showed good penetration in bronchial secretions [41][42][43]. Nevertheless, its epithelial lining fluid (ELF) steady-state concentration was lower than the MIC for Pseudomonas aeruginosa after a 4.5-g tid dose [41]. With continuous infusion, an increase in pulmonary concentration was found, at least in the subset of patients with moderate renal failure (measured Cr Cl < 50 mL/minute), about three times higher than in the patients with normal renal function [43]. However, no relationship was found between ELF piperacillin concentration and clinical success. Similar concentrations were found in the eight patients who died or had persistent infections and in those who experienced therapeutic success [43].
Subcutaneous tissue-to-plasma ratio and PK have been assessed in microdialysis studies. In six septic shock patients (mean norepinephrine dose 0.8 μg/kg/minute) [44], the subcutaneous tissue-to-plasma AUC ratio was only 0.1, one-third of that measured in healthy volunteers. Peak tissue concentration was also delayed in patients (122 minutes in patients compared with 27 minutes in healthy volunteers), and T 1/2 in tissues was nearly nine times longer. In 13 younger patients with less severe sepsis [45], the AUC tissue-to-plasma ratio was roughly 0.2. In accordance with their serum PK (as well as PK of another five patients) [46], piperacillin/ tazobactam CFR was calculated to be 92.3% with continuous infusion (13.5 g/day) and 53.4% with bolus dosing (4.5 g qid, or 18 g/day). Again, no correlation was found between tissue concentration and outcomes. Despite the low tissue concentration levels, all patients in both groups survived [45].

Cephalosporins Cefpirome
Cefpirome PK studies have produced heterogeneous results. A 2-g dose was adequate in young trauma patients (Cr Cl ≥ 50 mL/minute) and in similar healthy volunteers. After the first dose, the mean T > MIC were 75% and 80%, respectively (with a MIC of 4 mg/L, P = 0.76) [47]. However, in 12 similar patients, a lower T > MIC (60%) was found, which was probably related to higher cefpirome Cl [48]. After four days of therapy, the cefpirome mean PK parameters remained similar (T > MIC 67% and AUC 242 mg × hour/L vs. 306 mg × hour/L at steady state). Further analyses [49] showed a strong correlation between Cr Cl and either cefpirome or cefepime Cl (r 2 = 0.81). Patients with the lower range of T > MIC had a higher Cr Cl, usually above 144 mL/minute [49]. According to these measured PK data, the authors performed a simulation to demonstrate improved CFR of cefpirome given as a continuous infusion to treat Pseudomonas aeruginosa infection, from 56.1% to 84.4% (Table 2) [50].
Cefpirome tissue PK were evaluated on the basis of microdialysis. Sauermann et al. [51] found a low subcutaneous tissue concentration in patients with severe sepsis, almost half of healthy volunteers, despite a longer plasma T 1/2 (183 minutes vs. 95 minutes; P < 0.05). Similar results were reported by Joukhadar et al. [52], who found muscle-to-plasma ratios of 0.63 in patients and 0.83 in healthy volunteers (Table 3).

Cefepime
Roughly a twofold variation of cefepime V d has been reported in PK studies ( Table 1) of severe sepsis and septic shock patients [23], elderly septic patients [53], young burn patients [54] and nosocomial pneumonia patients [55]. Cefepime Cl has also been found to be closely correlated with Cr Cl in this last listed cohort (r 2 = 0.77) [55], in another cohort of septic patients (r 2 = 0.74) [56] and in burn patients (r 2 = 0.58) [57]. Therefore, patients with renal dysfunction may experience toxicity.
In 21 septic patients receiving cefepime at a dose of 2 g bid, more than twofold peak variations and roughly 40-fold trough variations were observed. Again, the cefepime Cl correlated with Cr Cl (r 2 = 0.77). Two patients with low Cr Cl (19 and 12 mL/minute) had trough levels > 20 mg/L despite dosage adjustment. They both had neurologic symptoms (namely, confusion and muscle jerks) that were not identified as toxicity but resolved promptly after drug arrest [55].
A cefepime bolus of 2 g bid was found to be insufficient to reach a high PD target after the first dose (Table 2), both in 80% of young burn patients (burn area 21.8%) with high mean Cr Cl (119.2 mL/minute) [54] and in the Taccone et al. study [23], in which only 16% of patients achieved the intended target.
Two other studies have evaluated cefepime PK, one of which addressed the first day of therapy for 55 nosocomial pneumonia or bacteraemia patients (67% trauma) [58] and the other of which described the status of 32 VAP patients on the second day of cefepime treatment [59]. Both studies unveiled a relationship between V d and total body weight as well as between excretion, either elimination rate constant [59] or Cl [58], and Cr Cl. However, significant interpatient variability was again observed, with regard to both cefepime Cl (58%) and central compartment V d (67%) [58].
A PD model was developed with this VAP population PK data: despite a 2-g tid dose, PTA > 90% was achieved only with a MIC ≤ 8 mg/L [59]. In another cefepime PD model, the CFR of a 2-g bid dose, used to treat both Escherichia coli and Klebsiella pneumoniae, was 78.9%. However, for Pseudomonas aeruginosa, CFR was only 53.6% (Table 2) and increased with either 2 g tid or continuous infusion (4 g/day or 6 g/day) to 84.9%, 91.7% and 94.8% respectively. Nevertheless, the CFR for Acinetobacter baumanii [60], even with a continuous infusion of 6 g/day, was only 75%, reemphasizing the importance of appropriate dosing and the potential benefit of continuous infusion against difficult-to-treat bacteria.
Also, the cefepime tissue concentration was assessed in biopsy samples collected from the skin of burn patients three to five hours after a bolus dose on day 3 of antibiotic therapy. A mean biopsy-to-plasma cefepime ratio of 1.5 (range 0.4 to 5.1) was found [61]. Klekner et al. [62] were unable to detect cefepime in bronchial secretions from any of the five studied patients six hours after an 80 mg/kg dose. However, using continuous infusion (4 g/day) to treat VAP patients, Boselli et al. [63] found, at steady state, higher and similar plasma and ELF concentrations (mean of 13.5 mg/L and 14.1 mg/L, respectively). Although different sampling methods may have influenced these differences, continuous infusion seems to prolong T > MIC in the lungs. Nevertheless, no correlation with therapeutic outcomes was reported.

Ceftazidime
Several studies have shown ceftazidime PK heterogeneity in ICU septic patients with Pseudomonas infections (mostly nosocomial pneumonia) [64], severe sepsis [65,66] and burns [54]. Similarly to other β-lactams, the authors noted a large variation of both V d and Cl (Table  1) and consequently significant interpatient variability in T 1/2 and trough concentrations. Also, a correlation between Cl and Cr Cl was usually reported [65,66].
Continuous infusion of ceftazidime was compared with bolus dosing in five different studies [67][68][69][70][71]. In all, there was an increase in T > MIC with continuous infusion despite lower daily doses. However, only in severe melioidosis was this strategy associated with lower mortality (3 of 10 patients vs. 9 of 11 patients) [71]. Those patients had low Cr Cl (26.6 mL/minute) and received ceftazidime dosages adjusted to their body weight (4 mg/kg/hour or 40 mg/kg tid, for a mean body weight of 49.4 kg). Ceftazidime steady-state concentration was measured in another cohort of 92 patients receiving continuous infusions [72]. Therapeutic drug monitoring was performed on the second day of therapy. The mean serum concentration was 46.9 mg/L, but again with a very wide range of serum concentrations (7.4 to 162.3 mg/L). Therefore, dosage modification was common because of low serum levels (36.9%) and high serum levels (27.2%), with the latter being associated with lower Cr Cl (mean of 51 mL/minute compared with 103 mL/minute for patients with low serum levels). Similar results were shown in another large ceftazidime PK study assessing a mixed septic population with a higher mean Cr Cl (123 mL/minute) [73]. The lower T > MIC was found in patients with the higher Cr Cl, especially after bolus dosing (Table 2).
Ceftazidime concentration in bronchial secretions was measured in four studies of VAP patients. Very low concentrations, < 0.5 mg/L and < 0.3 mg/L, were found in two of them [62,76]. Bressole et al. [77] found a higher ratio between bronchial secretions and plasma concentration (0.76) in patients infected after abdominal surgery. A longer T 1/2 (6.1 hours) and a lower Cl (4.2 L/ hour) may explain some of these differences. With continuous infusion, a ratio of 0.21 between ELF and serum was observed [69].

Discussion
In our systematic review, we have aggregated information from 57 prospective studies related to the PK of β-lactam antibiotics, which are among the most often agents used to treat sepsis in ICU patients [78]. Overall, an increased V d of all the studied antibiotics was reported (Figure 3), which was related to total body weight [58,73], but with significant variability. Drug Cl was also increased and usually related to Cr Cl. Those changes were largely unpredictable, with important interpatient variability. However, the higher Cl values were noted in studies that excluded patients with renal dysfunction, a common strategy, which may limit the interpretation of the data reported.
Therapeutic drug monitoring was rarely performed. In addition, data on the daily variation of PK parameters in ICU patients, as well as the ideal frequency of this monitoring, are currently limited. Nevertheless, two of the reviewed studies [55,72] showed that inadequate dosing may be common in this population and may jeopardize β-lactam antibiotics efficacy or even lead to toxicity [79]. Roberts et al. [80] measured piperacillin/tazobactam concentrations and found that 50.4% of patients first measurement were low. The clinical efficacy of using drug levels to achieve adequate concentrations had never been properly evaluated. In a recent study, PD modelling was used to empirically treat 94 VAP in critically ill patients at high risk of infection with antibiotic-resistant Pseudomonas aeruginosa [81]. A three-hour infusion regimen of either cefepime or meropenem at a high dosage (2 g tid) was initiated, followed by both antibiotic and dose de-escalation whenever bacteria with a low MIC were identified. The infection-related mortality decreased from 21.6% to 8.5% (P = 0.029).
The PD targets of β-lactam antibiotics may be different in patients with severe bacterial infections. McKinnon et al. [82] evaluated ceftazidime and cefepime PD by using PK data from previous clinical trials [83]. Maintaining a T > MIC as high as 100% was associated with a significantly greater clinical cure and bacteriologic eradication than a shorter time (cefatzidime: 82% vs. 33%, P = 0.002; cefepime: 97% vs. 44%, P = 0.001). Also, in a febrile neutropenia population of 60 patients treated with meropenem, a calculated T > MIC of 83% was found in responders, whilst those with a poor clinical response had a T > MIC of only 60% [84]. It has also been suggested that, at least in vivo, maximum killing of bacteria is achieved at higher concentrations, four to five times MIC [85], accounting for antibiotic penetration in infected tissues. As such, concentrations of β-lactam antibiotics may need to be maintained well above the MIC for extended periods, especially in patients with life-threatening infections. Accordingly, different PD targets have been proposed in the different studies addressing ICU patients, which sometimes make their comparison difficult.
An improved PD profile of β-lactams may be obtained by promoting a longer exposure with more frequent dosing, extended infusions or continuous infusions [86,87]. Several of the studies that we reviewed reported PD benefits of continuous infusions (even using small daily doses) ( Table 2). Also, PD modelling tends to support this strategy. Nevertheless, almost none of the studies addressed reported a decrease in mortality. In addition, a recently published meta-analysis of 14 prospective studies did not show a significant benefit of using this strategy (odds ratio 1.00, 95% confidence interval 0.48 to 2.06; P = 1.00) [88].
An increasing number of studies have addressed β-lactam antibiotic tissue concentration. Despite the theoretical advantage of analysing the drug concentration at the site of infection, there are no data to support a relationship between these concentrations and outcomes. Furthermore, there are still controversial issues involved in interpreting these data, namely, microdialysis [32,33]. Therefore, we think that, at present, no recommendation can be made regarding antibiotic tissue PK.

Conclusions
The PK of β-lactam antibiotics are significantly changed in septic ICU patients. Dosage and schedule regimens based on data from healthy volunteers may be misleading. Therapeutic drug monitoring and PD modelling according to measured PK previously showed promising results. Continuous infusion, although theoretically useful, has not been shown to lead to improved outcomes. The clinical significance of tissue PK monitoring remains to be determined.

Key messages
• Among ICU patients, the PK of β-lactam antibiotics are markedly unpredictable.
• A large volume of distribution is commonly observed in ICU patients and contributes to a lower antibiotic concentration, but also to a greater exposure time.
• An increased glomerular filtration rate is usually associated with a short half-life of β-lactam antibiotics, whilst renal failure is associated with a greater exposure and increased risk of accumulation.
• Continuous infusion of β-lactam antibiotics commonly increases the time that the antibiotic concentration exceeds its MIC and may therefore increase efficacy.
• Therapeutic drug monitoring of β-lactam antibiotic concentration may help to improve its efficacy and prevent toxicity, but currently is unavailable in most clinical settings.