According to Greek mythology, the siren’s call was an irresistible lure which would draw sailors to veer off course and crash into rocks. Odysseus avoided this by covering his crew’s ears. However, he wanted to hear the siren’s song. To prevent himself from being seduced and jumping off the ship, he tied himself to the mast.
Some theories are so attractive that they are nearly irresistible. No matter how many times they are disproven, these theories still seem compelling. One example is double-coverage for pseudomonas. Recently, the IDSA recommended this for ventilator-associated PNA (VAP), despite openly admitting that RCTs found it to be ineffective.
Clinical evidence regarding double-coverage
Heyland et al. 2008: Randomized trial of combination versus monotherapy for the empiric treatment of suspected ventilator-associated pneumonia
This was a multicenter RCT involving 28 centers in Canada and the US. 739 patients with suspected ventilator-associated pneumonia were randomized to receive meropenem with or without ciprofloxacin. There was no difference in the primary endpoint of mortality (19% in combination therapy vs. 18% in monotherapy) or secondary endpoints including length of stay and treatment response.
Among a subgroup of 56 patients with Pseudomonas, Acinetobacter, or multi-drug resistant gram-negative bacilli, combination therapy was much more likely to provide effective coverage (84% vs. 18%). However, even among this subgroup there were no significant clinical differences between monotherapy versus combination therapy (1).
Barry et al. 1997: Empiric treatment of hospital-acquired lower respiratory tract infections with meropenem or ceftazidime with tobramycin: A randomized study
This was a multicenter RCT involving 22 centers. 211 patients with nosocomial pneumonia (70% of whom were intubated) were randomized to receive meropenem versus ceftazidime plus tobramycin. Patients treated with monotherapy had improved rates of clinical recovery and a trend toward reduced mortality. One limitation of this study is that better outcomes in the monotherapy group could reflect superiority of meropenem compared to ceftazidime. Nonetheless, this may illustrate an important point discussed further below: monotherapy with a superior beta-lactam may be more effective than dual therapy with a less effective beta-lactam (2).
Efficacy of double-coverage in general
The concept of double-coverage has been evaluated in dozens of trials across a variety of contexts (e.g. sepsis, bacteremia). Despite persistent enthusiasm for this concept, the evidence from multiple perspectives has been consistently negative (Cochrane review 2014, Tamma 2012, Hu 2013, Vardakas 2013). Meta-analyses do reveal an increased risk of nephrotoxicity when an aminoglycoside is used (Paul 2014).
Why we think double-coverage works
Appropriate empiric antibiotic use often correlates with reduced mortality.
Numerous retrospective studies have been performed correlating mortality from VAP with whether the initial antibiotic regimen covered the pathogen. Most studies have shown that appropriate empiric antibiotics correlated with huge improvements in mortality (20-40% absolute mortality difference), although some haven’t (Swanson 2013)(3). The positive studies support the importance of appropriate empiric antibiotic therapy.
There are two main possibilities to explain the occurrence of the positive studies:
- Causation: Patients whose pathogen isn’t covered initially do worse because of inadequate antibiotic therapy. If more antibiotics had been used, these patients would have a better outcome.
- Confounding: Patients whose pathogen isn’t initially covered have drug-resistant organisms. Infection with drug-resistant organisms is correlated with numerous poor prognostic factors (e.g. comorbidity, frequent hospitalization, prior antibiotic use, older age, hemodialysis, living in a facility, etc.). Thus, having a drug-resistant organism identifies a sicker group of patients with a worse prognosis (regardless of antibiotic selection).
But… VAP has an attributable mortality of ~9%
VAP undoubtedly correlates with high mortality. However, it is unclear to what extent VAP causes death, versus to what extent VAP is merely a marker of very sick patients who are likely to die. A systematic review of prospective RCTs studying interventions to prevent VAP found an attributable mortality of 9% (Melsen 2011)(4).
The mortality benefits found in retrospective studies cannot be due to causation.
If VAP only causes a 9% increase in mortality, then it is impossible that appropriate empiric antibiotic therapy could cause a 20-40% absolute reduction in mortality (as suggested by many retrospective studies). The only logical conclusion is that these huge mortality differences aren’t caused by antibiotic selection, but instead reflect confounding variables.
Understanding why double-coverage doesn’t work
One reason for ongoing enthusiasm for double-coverage is that it really seems like it ought to work. Regardless of how many studies fail to support it, the concept still seems valid. Before we can move beyond double-coverage, we must understand why it doesn’t work.
In order for double-coverage to actually help the patient, a chain of events must occur. To determine how likely this is, let’s break down these events and examine them individually (5).
Probability #1 = probability that patient actually has VAP
Diagnosing VAP is difficult. It is common for us to make a presumptive diagnosis of pneumonia and initiate treatment, only later to realize that the patient didn’t have pneumonia at all.
Two studies have been performed comparing clinical diagnosis of ventilator associated pneumonia versus autopsy findings, revealing a specificity of 36-75% (Tejerina 2010, Fabregas 1999). Heyland 2008 (discussed above) found that fully 50% of patients eventually had cultures revealing no growth, normal flora, or candida (implying the absence of pneumonia). Thus, among patients who we are starting empiric treatment, the probability that they truly have VAP might be about 50% (6):
Probability #1 ~ 0.5
Probability #2 = probability that VAP is due to a gram-negative organism
If the patient does have a pneumonia, then the likelihood that the pneumonia is due to a gram-negative organism is ~60% (e.g., Weber 2007).
Probability #2 ~ 0.6
Probability #3 = probability that gram-negative is resistant to beta-lactam
What is the likelihood that a gram-negative pathogen would resist the beta-lactam backbone? This depends on the local antibiogram. At most hospitals its possible to select a beta-lactam which will cover at least 70% of gram-negatives, making the likelihood of failure <30% (7).
Probability #3 ~ 0.3
Probability #4 = probability that this bacteria is sensitive to the 2nd antibiotic
Bacteria which are resistant to the beta-lactam are more likely to be multi-drug resistant organisms. Therefore, the likelihood that the second antibiotic will cover them is lower than for average bacteria. This evidence has been explored in detail here and is summarized in this table:
On average, if a gram-negative is resistant to a beta-lactam, the likelihood of its being sensitive to a second antibiotic is roughly 30% for fluoroquinolones, 60% for tobramycin, and 80% for amikacin:
Probability #4 for fluoroquinolone ~ 0.3
Probability #4 for tobramycin ~ 0.6
Probability #4 for amikacin ~ 0.8
Probability #5 = probability that adding an antibiotic to which the pathogen is sensitive affects the outcome
What is the likelihood that adding this second antibiotic up-front will improve the clinical outcome? One answer to this question may be found in the “90-60 rule.” This rule predicts that infections are 90% likely to improve when treated with an antibiotic to which the pathogen is sensitive. Alternatively, if an antibiotic is used to which the pathogen is resistant, then there is a 60% likelihood of clinical improvement (Rex 2002, Doern 2011).
Although a crude generalization, the 90-60 rule highlights the gap between the microbiology laboratory and clinical outcomes. For example, consider reasons that a patient might improve clinically despite use of an antibiotic to which the pathogen demonstrates in vitro resistance:
- Immunocompetent patients may be able to fend off infections on their own.
- Even under the most rigorous conditions, the drug concentration required to inhibit bacterial growth (the minimum inhibitory concentration) is accurate only to within plus or minus a two-fold dilution. In some cases, this random error may be the difference between a “sensitive” or “resistant” result.
- If the pathogen is slightly resistant, there still could be some partial suppression of bacterial growth or transient killing (e.g. at peak antibiotic concentrations immediately after drug boluses).
- Patients with impaired renal function or lower weight may accumulate higher drug concentrations, overcoming low-level resistance (8).
Based on this approximation:
Probability #5 ~ 0.3
In order for double-coverage to have a clinical effect, all of these probabilities must be satisfied simultaneously. Thus:
Probability of clinical benefit = (p1)(p2)(p3)(p4)(p5)
Based on the values estimated above:
- Probability of clinical benefit from fluoroquinolone ~0.8% (number needed to treat ~123)
- Probability of clinical benefit from tobramycin ~1.6% (number needed to treat ~62)
- Probability of clinical benefit from amikacin ~2.2% (number needed to treat ~46)
This shows that although double-coverage seems smart, it’s unlikely to actually help. The next step is a risk-benefit analysis. If antibiotics had zero side-effects, then double-coverage would be marginally beneficial. Unfortunately, there is no free lunch:
- Fluoroquinolones have numerous side effects as explored previously here. The risk of clostridium difficile alone is probably >1% when giving fluoroquinolones to a patient in the ICU, negating any benefit from double-coverage.
- Aminoglycosides are associated with a significant risk of acute kidney injury. In most cases, this may out-weigh any benefit obtained from double-coverage.
Optimal single-coverage beats suboptimal double-coverage.
Let’s consider a hypothetical antibiogram with the following coverage rates for pseudomonas:
- piperacillin-tazobactam: 84%
- cefepime: 93%
Based on evidence discussed above (Probability #4), the following coverage rates would be predicted for various combinations (9):
- piperacillin-tazobactam: 84%
- piperacillin-tazobactam + ciprofloxacin: 89%
- cefepime: 93%
- piperacillin-tazobactam + tobramycin: 93.6%
- cefepime + ciprofloxacin: 95%
- piperacillin-tazobactam + amikacin: 96.8%
- cefepime + tobramycin: 97%
- cefepime + amikacin: 98.6%
Cefepime alone is probably superior to piperacillin-tazobactam plus ciprofloxacin, and equivalent to piperacillin-tazobactam plus tobramycin. Thus, single-coverage with a more effective beta-lactam may be superior to double coverage with a less effective beta-lactam. Rather than piling on additional antibiotics, it may be better to optimize the beta-lactam backbone. More isn’t better.
Decisions about double-coverage should be individualized, not guideline-mandated.
There are situations where double-coverage is reasonable. For example, imagine a patient with septic shock due to VAP with gram negative bacteremia. It could make sense to give a dose or two of aminoglycoside while awaiting culture data, especially if the patient had normal renal function and if there is a high local rate of antibiotic resistance (10).
The intention of this post isn’t to argue that double-coverage should never be used. Occasionally it might help. However, the IDSA recommends using double-coverage for nearly all cases of VAP, which simply isn’t evidence-based (11). Even the IDSA admits that their guideline contradicts their own meta-analysis of the evidence, stating:
We made this suggestion despite the panel’s meta-analysis suggesting no difference in mortality, clinical response, adverse effects, or acquired resistance rates between regimens with one antipseudomonal agent versus two. The panel was concerned that the trial data we reviewed do not apply to all patients with VAP because most of the studies specifically excluded patients colonized with resistant pathogens and patients at increased risk for resistant pathogens.
- Prospective RCTs have not shown a benefit from double-coverage.
- In order for double-coverage to be beneficial, a chain of events must occur. The patient must truly have VAP, that VAP must be due to a gram-negative, the gram negative must be resistant to the beta-lactam, the gram negative must be sensitive to the second antibiotic, and broader antibiotic coverage must make a clinical difference. The likelihood of this entire sequence of events occurring is about 1-2%.
- Double-coverage with a fluoroquinolone is difficult to justify given rising resistance to fluoroquinolones and a significant toxicity profile.
- Double-coverage with an aminoglycoside may be considered in specific patients, but in most cases nephrotoxicity outweighs benefit.
- Monotherapy with an optimal beta-lactam may be more effective than double-coverage with a suboptimal beta-lactam. More isn’t necessarily better.
- The IDSA recommendation utilize double-coverage in nearly all patients with VAP is not evidence-based.
- 2016 IDSA HAP/VAP guidelines
- Double-coverage with a fluoroquinolone?
- Six reasons to avoid fluoroquinolones in the critically ill
- Which patients admitted for pneumonia need MRSA coverage?
- Evidence-based treatment of severe CAP
- One weakness of this study is that it excluded patients known to be colonized with MRSA or Pseudomonas. However, this exclusion makes sense because patients who are known to be colonized with pseudomonas will have known susceptibility patterns, making the issue of double-coverage less relevant.
- Ceftazidime has relatively poor gram-positive coverage compared to other cephalosporins. For this reason, for empiric coverage regimens when gram positives could be in play, its probably preferable to use cefepime (if pseudomonas / resistant GNR possible) or ceftriaxone (if low risk for pseudomonas or resistant GNR).
- Any study suggesting that a therapy causes 20-40% absolute reduction in mortality should probably be viewed with skepticism. Precious few treatments are this effective.
- Although this remains debatable, evaluation of prospective studies performed to reduce the VAP rate is the most rigorous way to answer the question.
- The probabilities in this section are based on the best data I could find, but aren’t necessarily exactly correct in every locale and specific situation. The intention here is mostly to illustrate a point, not to be numerically 100% accurate. Even if the numbers are off a bit, the conclusion will likely be the same.
- After writing this section, I came across Swanson 2013 who reported that only 40% of patients with suspected VAP actually have VAP, using an entirely different set of references. This may serve as additional evidence to support this number. On a personal note, the procalcitonin lab has been helpful in proving how poor clinical criteria are for diagnosing VAP. Procalcitonin can be helpful to disprove a VAP diagnosis, even if other indicators (e.g. tracheal culture and gram stain) support the diagnosis of VAP.
- This might be the most hospital-specific piece of data. Bear in mind that many gram-negatives are more sensitive than pseudomonas.
- If you consider the variation in weight and renal function among our patients, and the fact that many antibiotic regimens are fixed in dose (i.e. not weight-based), it seems probable that the tissue concentration of drug could vary substantially (possibly by a factor of 2-4).
- Ideally, institutions would have combination antibiograms so that this information could be know with certainty, allowing the rational selection of an empiric combination of antibiotics (if necessary). Unfortunately, combination antibiograms remain underutilized. Lack of combination antibiograms may cause practitioners to over-estimate the utility of combination therapy.
- This is the classic pro-double-coverage argument, which is that double coverage works in a small subset of patients (and we simply haven’t managed to do a study on this select group of patients thus far). There could be some merit to this argument. However, it must also be noted that if a claim cannot be disproven then it’s not scientific (all scientific claims must be testable).
- The IDSA 2016 guidelines recommend double-coverage for pseudomonas if any of the following criteria are met: (a) prior IV antibiotic use within 90 days, (b) septic shock, (c) ARDS, (d) VAP developing more than four days after hospitalization, (e) acute renal replacement therapy prior to VAP, (f) patients in units where >10% of gram-negative isolates are resistant to an agent being considered for monotherapy, (g) patients in an ICU where local antimicrobial susceptibility rates aren’t available, (h) if the patient has structural lung disease such as bronchiectasis or cystic fibrosis. In practice, the vast majority of patients with VAP will meet at least one of these criteria. This was described as a “weak recommendation, low-quality evidence.”
Image credit: Slot machine image from http://www.freeimages.com/photo/slot-machine-at-bellagio-las-v-1516244
Latest posts by Josh Farkas (see all)
- PulmCrit- How to convert a VBG into an ABG - January 16, 2017
- PulmCrit- Sedation update: IV olanzapine & combo vs. monotherapy - January 11, 2017
- PulmCrit- Dogmalysis of PCI for NSTEMI patients with a history of CABG - January 2, 2017