So, JC time again. Today, someone chose this paper from CCM regarding antibiotic monotherapy compared against an empirical combination strategy for treatment of severe sepsis. So hot off the press it has not even been formatted properly yet. Not the kind of topic that makes people climb over each other to make a point. But we welcome all choices here in Virchester – every day is a school day, even if the paper is not the greatest choice or design….
So, the abstract claims that administration of combination antibiotic therapy within 6 hours of severe sepsis onset reduces mortality. Nice one. Roll out the red carpet to the antibiotic cupboard and fetch me my therapeutic blunderbuss.
But what did they actually do? There are a couple of phrases in the methods section that immediately make me sit up and raise an eyebrow. ‘Secondary analysis’ makes me think this data was not collected with a particular hypothesis in mind. ‘Before and after’ study screams Hawthorne effect. ‘Waived informed consent’ implies this was a service evaluation, which can often result in less rigorous data collection than needed and variable screening of the population of interest.
However, before we go to town it is worth looking back at the EduSepsis study to see where this data comes from. The free JAMA article can be found here and is actually quite an interesting read. This is allegedly prospective data on a continuous sample of >2000 ICU patients admitted with an agreed predefined definition of severe sepsis. This could be useful after all….
In this substudy, the authors sought to define antibiotic prescription patterns and assess the impact of combination antimicrobials on mortality. First aim, no problem. This is observational data and can provide interesting insights into the Spanish use of abx on ICU. The second aim… hang on a minute. If you want to assess the therapeutic benefit of combination abx vs monotherapy, shouldn’t you be conducting a controlled RCT? Something like this for example?
Why does this aim need an RCT? Well, if you are looking prospectively at routinely managed patients at physician discretion receiving combo vs monotherapy without randomisation or blinding, you are failing to control for any confounders. You may also find significant bias in your results. For instance, maybe combo patients are given combo abx because particular intensivists like to manage sepsis far more aggressively. But this doesn’t just mean combo abx. Maybe it means driving the MAP a bit harder. Giving more blood. Earlier admission. Fastidious EGDT. Etc… Thus, a reduction in mortality within the combo abx group may actually not be a result of combo abx. It may just be a result of multiple other issues.
In the paper, the authors took data on the 1372 patients receiving abx within 6 hours (most of which are likely to be in the group directly following their 2 month educational intervention) and evaluated outcomes. The data on prescription patterns, nosocomial vs community infection and combination use are interesting. But the real headline of this paper is the mortality reduction, so that is what we need to look at.
The authors report a statistically significant 6% mortality reduction using combo abx overall. This is a NNT of 17 to prevent one death. Convinced?
Not quite. Lets look at the matching baseline characteristics in table 3. Firstly, the groups are very unbalanced number wise due to the nature of the service evaluation – 388 combo abx patients vs. 984 monotherapy. This is not ideal. Secondly, there are big differences in characteristics between these groups, including a significantly higher rate of Emergency surgical admissions, intraabdominal infections, meningitic infections, skin/soft tissue infections, respiratory organ failure and renal organ failure in the monotherapy group. These are all very BAD things to happen. Thus we start to think that the mortality benefit is not altogether explained by simple combo abx prescription.
The authors have conducted a multivariate analysis to look at the influence of these characteristics on the outcome. They don’t present all the data, but suggest that when they include factors significant on univariate analysis, combo therapy still reduces mortality overall. While this is an interesting analysis, it does not solve the methodological problems raised earlier. It is simply stating that when looking at this data, even when you take into account other factors, those patients prescribed combo abx had an OR less than 1.
What did you have to say about the paper? @Kirstychallen was more persuaded by the multivariate analysis –
I think this is a very valid point – maybe we should start broad and taper in with sensitivities. I kind of agree – this is EBM in tandem with experience/expertise. But, I think this is overall a different proposition to that made by the paper – that patients who get combo abx will always do better in terms of survival.
@EMmanchester shared concerns about confounders and face validity. He also posted a strange picture of @richardbody. I am not sure why.
(Ed – was just checking if Rick’s face had validity – you judge for yourself )
@baombeJP talked a bit about the regression model. Interesting discussions to be had there…
Regression model included "plausible relationship with the dependent variable". How reliable is that? #JC_StE— Janos Peter Baombe (@baombejp) December 7, 2012
All in all, I think we shared a similar view on this paper. Interesting observational cohort data. Significant limitations in design, conclusion, and generalizability to ED patients.
Roll on further RCT data on empirical broad vs montherapy in severe sepsis. Until then, I will stick to aggressive and targeted abx therapy, as dictated by local resistance and pathogenic patterns. This is often combo therapy, especially for intrabdominal source.
But I don’t think the data is there to start adding in a second abx when maybe one will do just fine…..
Next week – head injury and anticoagulation. Much more interesting……..