We set out to investigate if insufficient concentrations at the infection site in the lung can explain the previous described failure of doripenem when treating VAP and whether longer infusion times can help to optimize target attainment both in plasma and at the target site. Due to the limited number of patient included and high variability of determined PK parameters, especially in the lung, these questions could only partially be answered. On the other hand, we were able to show that the observed variability intrinsically might be a factor that might have resulted in treatment failure in a relevant proportion of patients and that modification of the dosing regimen also reflects in target site PK.
In order to allow for a hypothesis explaining the insufficient efficacy of doripenem observed by Kollef et al. based on plasma PK, the following assumptions have to be made. First, as the actual MIC values of the pathogens that caused the VAP in that study were not available, only MIC50 levels could be used for PK/PD considerations. Second, individual PK profiles are not published for the study by Kollef et al., thus we could only estimate plasma exposure in that study by doubling plasma concentrations at every time point in all patients of the present study after the identical duration of infusion (4 h) to compensate for differences in dosing. Third, the fact that blood concentrations over four times the MIC might be needed in case of severe infections for beta-lactam antibiotics have to be considered . Taking all these assumptions into account—including measuring time over MIC in figures after doubling plasma concentrations—the prolonged infusion regimen would have resulted in mean T
>4xMIC50 of 81 and 45% for P. aeruginosa and A. baumanii, respectively. For beta-lactams, values between 40 and 60% T>MIC have been associated with optimal killing in in vivo and in vitro PK/PD models by Craig et al.,  therefore, for both strains time over MIC50 would be in or above the range of 40–60%,suggesting that good antimicrobial efficacy might be expected.
Still, also in patients suffering from VAP caused by P. aeruginosa, the clinical cure rate was numerically lower for patients with P. aeruginosa VAP in the doripenem arm compared to the imipenem-cilastatin arm (41.2 versus 60.0%). Therefore, one might speculate that the observed lack of efficacy might be due to insufficient target site penetration in the critically ill population rather than with subtherapeutic plasma levels. Although not enough data for thorough PK-PD calculations for the lung was generated in the present study, a difference between the ELF concentration-time profiles after the two investigated infusion schemes was observed. As shown in Fig. 1, differences in dosing regimens and thereby modified plasma PK indeed reflects in changes in target site pharmacokinetics in ELF. However, none of the two profiles seem to sufficiently cover concentrations above the thresholds of 4 or 8 mg/L (= 4× MIC50 of pathogens with MIC 1 mg/L or 2 mg/L) which might explain why potential benefits of a prolonged infusion scheme did not translate into improved clinical endpoints when treating VAP.
For reliable determination of the time above the MIC in ELF, more BAL measurements, either by including more patients or by repeated samples per patient, would be necessary to permit PK-PD calculations and to compensate for the high inter-individual variability, i.e. differences in concentrations up to the factor eight at a single-time-point. Different factors like organ dysfunction, septic shock, concomitant medication or capillary leakage have been attributed to this variability [4, 7, 13, 14]. However, more BAL time points per patient might be problematic from the ethical point of view and repeated BAL procedure in one patient might falsify data. Moreover, while including more patients in this study would allow for better target attainment analysis for the overall population it might not change the main outcome, i.e. that inter-individual variability has to be expected to be very high, and despite the fact that plasma concentrations are mirrored in ELF, target site concentrations in an individual patient currently cannot be predicted.
A significant correlation has been found between creatinine clearance and AUC0–8 in the present study as shown in Fig. 2. Thereby, differences in creatinine clearance (112.4 vs. 76.2 mL/min for 1 and 4-h infusion, respectively) might also have contributed to higher AUC0–8 values found for the 4-h infusion regimen. In contrast to plasma pharmacokinetics, which is highly impacted by creatinine clearance, factors that are more difficult to measure, e.g. local inflammation in the lung or atelectasis, might additionally impact the target site penetration [15, 16]. Intracellular concentrations have not been determined in the present study, because beta-lactam antibiotics are not active against intracellular pathogens and no accumulation in phagocytes has been described . Comparison of our data with pharmacokinetics in plasma and ELF determined by Justo et al.  in healthy adults after doripenem 500 mg administration as extended infusion allows a cautious conclusion on the difference between healthy subjects and patients. Plasma concentrations were up to 1.8-fold higher in patients than in healthy subjects (8.79 and 4.89 mg/L at 4.5-h time point). Likewise, Cmax in ELF of 6.93 mg/L (4-h time point) in patients was higher than in healthy volunteers 1.67 mg/dL (4-5 h time point), but most importantly variability was lower.
Our study thereby highlights the importance of determining infection site pharmacokinetics but most importantly the need for further exploration of factors impacting target site pharmacokinetics in the respective patient category as e.g. the individual lung penetration of antibiotics in patients with severe pneumonia.
In summary, our data suggest the potential benefit of prolonged infusion in terms of PK/PD indices in plasma in the investigated population of severely ill patients. While it was shown that differences in the concentration-time profile in plasma did transfer to the lung, the small sample size limits the information value of BAL data. We can only hypothesize that insufficient infection site concentrations might have contributed to a previously observed lack of efficacy. Nevertheless the study confirms the demand for assessment of target site concentrations of antibiotics as early and throughout antimicrobial drug development in order to avoid therapeutic failure despite plasma PK/PD targets are achieved.