Since the incidence of ventilator-associated pneumonia varies from 0-60%+ across centers and observational studies, and leading investigators argue in good journals about whether using guideline-compliant antibiotics cures people with VAP or kills them, it should come as no surprise that estimates of the mortality attributable to VAP have also varied to the point of meaninglessness, from 0 to 70%. The question is deceptively simple: do people die “with” VAP or “from” it? Asked another way, would anyone who ever died “with” a ventilator-associated pneumonia have survived had they not developed it, or were they all destined to die regardless — VAP being the ‘critically ill man’s friend,’ if you will?
Guessing / estimating the answer with historical data has relied on extrapolations and hypothetical leaps of faith that seek to disentwine from the VAP-mortality relationship the confounder of severity of illness (which predicts both VAP and mortality). Hats off to the masochistic epidemiologists (a redundancy?) who’ve tried.
Maarten Bekaert et al announce that they’ve developed the statistical special sauce that others couldn’t, and that in fact, the 30 day attributable mortality from ventilator associated pneumonia is about 1%. (At 60 days, it’s 1.5%, they say.)
Their methods are basically impenetrable to those untrained in statistics (me). However, here’s the gist.
- The overall VAP incidence was 15% (standard definition + required positive cultures).
- They essentially create a statistical parallel universe in which no one got VAP. People who “should” have gotten VAP but were lucky and didn’t are cloned in the database-universe (counted multiple times). Someone with a logistical-regression-modeled 67% chance of getting VAP who didn’t get VAP is counted 3 times, for example.
- The mortality of this VAP-free universe of database clones is the expected mortality if no VAP existed. The actual observed mortality was compared against this, and they came up with about a 5% attributable mortality. This was modeled and massaged some more (e.g., controlling for confounders on mortality after ICU discharge), and voila! There you have the attributable mortality of VAP of 1% at 30 days after ICU admission.
They throw rocks at the body of previous research purporting to ascertain the mortality of VAP, because previous investigators did not use techniques that could account fully for the confounding by severity of illness, evolution of illness, and events after ICU discharge.
For example, here they model in the daily changes in patient status permitted by the database (SAPS/SOFA, pressor use, renal failure, etc). They argue this is critical, because measurements on ICU admission obviously don’t tell the story of worsening illness in the ICU, which contributes to both VAP and mortality risk.
Using a different observational methodology (looking at 53 randomized trials of VAP prevention bundles), other investigators like Melsen et al have found a significantly higher attributable mortality to VAP; their most recent consensus has been that it’s about 10%.
Clinical Takeaway: I’m put in the mind of Pangloss who told Candide that despite his suffering, this was the “best of all possible worlds,” but unlike Pangloss, these authors had the French multicenter Outcomerea database and high-end PCs running SAS 9.2. Maybe they really did model and check all those other VAP-free worlds, and this represents reality. I have no way of knowing, but if it will trip up the VAP police, I’ll cite this study and pretend I know what I’m talking about.
Bekaert M et al. Attributable Mortality of Ventilator-Associated Pneumonia. A Reappraisal Using Causal Analysis. Am J Respir Crit Care Med 2011;184:1133-1139.