Why Did PK/PD Modeling Fail Doripenem in VAP?

When imipenem is dosed at 1 g q8h for serious infections, it is infused over 40-60 min. Its label states that it is indicated for “lower RTI”, an old-fashioned term from the days when bronchitis and pneumonias were still lumped together, CAP was not differentiated from HAP, and HAP and VAP were considered the same from a treatment perspective.

The body of clinical evidence suggests that imipenem at this dose is an effective regimen for nosocomial pneumonia; indeed, no other antibiotic has surpassed its efficacy in HAP/VAP either. It has good penetration into lung tissue [1]. PK/PD modeling as we know it today did not exist in those days, and imipenem dosing was established empirically. Actually, it was initially set too high: at a dose of 1g q6h many patients developed seizures.

When doripenem was developed, one could have simply used the same dosing and be done with it – after all, the two drugs are from the same antibiotic class, share the same mode of action resulting in cidal activity by blocking the same PBPs, have very similar pharmacokinetics, ADME, and antibacterial spectrum. Both are excellent antipseudomonal drugs. Overall, doripenem MICs for Gram-negative bacteria, esp. non-lactose fermenters, are a tad lower than those for imipenem, a main reason for its development.

However, since the mid-90ies, PK/PD and Monte Carlo simulations were used regularly, with attainment ratios predicting reliably an effective dosing regimen. Based on bacterial kinetic data from the Craig mouse thigh model, we learned that (1) beta-lactam antibiotics are T>MIC drugs, (2) a %T>MIC of 40-50%% was usually sufficient for penicillins and cephalosporins, but can be a bit lower for penem antibiotics. We also learned later that ignoring dosing predictions from PK/PD modeling can be very risky and lead to underdosing (see Ambrose et al for examples [2]).

Given the predictive accuracy of PK/PD modeling, a mini-industry developed that specialized in the art of of simulating the likelihood of clinical success based on organism MIC distributions, animal PK/PD, and human exposure data. We have come to rely on PK/PD for dose-finding thus avoiding patient exposure to subtherapeutic low or toxic high doses. PK/PD has become a major pillar in a streamlined drug development program or in situations when a standard large-size RCTs cannot be conducted. This is especially true for development of drugs for MDR pathogens which are too rare for NI-based efficacy testing.

Based on Phase 2 clinical data including earlier HAP trials the Institute for Clinical Pharmacodynamics did extensive PK/PD modeling [3]. They persuasively argued that doripenem dosed at 1g IV q8h, infused over 4 hours, would lead to target attainment rates of >90% for the pathogens usually encountered in late-onset VAP patients. This regimen also provided a >35% T>MIC during the dosing cycle.

However, doripenem efficacy turned out to be sub-par compared to imipenem and the study had to be terminated prematurely as NI (with a CI of -15% margin) was not achieved [4]. The FDA released a statement summarizing the findings [5]:

How could this happen? What went wrong? Why was imipenem superior despite its higher MICs and shorter T>MIC? Where is the editorial that explains to us why PK/PD failed doripenem?

Across various antibiotic classes, PK/PD modeling is most solid for dose selection of fluoroquinolones, but the beta-lactam class of antibiotic has also been studied extensively.  So, this was not ‘new territory’ for the PK experts.

Let’s cut right to the chase: we have not heard any good answers to these questions. Here a few explanations that are bandied about which don’t make much sense on closer inspection:

• the difference in treatment duration (7 days for doripenem, 10-14 days for imipenem) may have been a factor:
  I doubt that because the clinical response curves separate already within the first 10 days, with lower clinical response rates for doripenem at EOT.
• severely ill patients have significantly higher clearance and lower Vd than less ill populations and require higher doses:
  True, in the van Wart publication the need of dose adjustment for ‘severity of illness’ is not being addressed. Patients with very high Clcr certainly did poorly in the doripenem arm.   However, such changes in drug PK would presumably apply to both drugs in the same direction and magnitude and not unevenly affect only doripenem.

So, we remain perplexed and surprised but also a bit annoyed. One would have thought our PK/PD experts would want to reanalyze the data and redo the modeling in order to understand why doripenem ran into trouble. Why are they ignoring the doripenem fiasco, isn’t there a lesson to be learned here?

The gods of statistics tell us to expect a false positive result in approx. 1 out of 20 test runs, so doripenem may just have had a ‘bad VAP day’. I don’t find that a satisfactory explanation but hope that you – the reader – may have a better answer. Your comments and insights are appreciated.

Abbreviations
RTI         respiratory tract infection
HAP       hospital-acquired pneumonia
VAP       ventilator-associated pneumonia
NI           non-inferiority
Clcr        clearance creatinine
Vd          volume of distribution
PK/PD   pharmacokinetic/pharmacodynamic
T>MIC   time above MIC
PBP       penicillin-binding protein
ADME    absorption, distribution, metabolism, excretion
EOT       end-of-therapy
RCT       randomized controlled trial
MITT      microbiologic intent-to-treat population
ME         microbiologically evaluable population

 

References:
[1] Package Insert Primaxin http://www.accessdata.fda.gov/drugsatfda_docs/label/2014/050587s076lbl.pdf
[2] Ambrose. Clin Infect Dis 2008; 47; S225
[3] Van Wart.  Diagn Microbiol. Infect. Dis. 2009; 63: 409
[4] Kollef.  Crit Care 2012; 16: R218
[5] http://www.fda.gov/Drugs/DrugSafety/ucm285883.htm