The largest observational study of AF in sepsis looked at admissions to Californian hospitals over a 1 year period. In this study, AF occurred in 5.7% of patients with severe sepsis and 2.6% of those patients suffered a stroke. The issue of anticoagulation (i.e. reduction in stroke risk) was not addressed.
This study (by the same lead author) is therefore both really interesting and useful. It's observational, looking at a huge patient data-set (113 511 patients with sepsis and AF) to describe outcomes from sepsis associated AF but also current practice. The study is American, but I still think there’s context validity.
Some of the more interesting findings and points (to me anyway) detailed below, in no particular order:
- 53% of patients were excluded from the study because they didn’t receive IV rate/rhythm control. The justification given was to include only those with ‘clinically significant’ AF. I can see the logic, but I’m not sure the association between IV therapy and clinical concern is a strong or logical one.
- 11% of the remaining patients had another indication for therapeutic anticoagulation (so were also excluded). This seems pretty high, and shows either that the same patients who get AF also have other associated co-morbidity, or maybe just goes to show just how common anticoagulation is these days.
- The study looked at stroke (and bleeding) in hospital. We have no idea how many of these patients went on to have a later stroke, or even in how many the AF resolved with recovery from sepsis.
- 520 hospitals were included, and in about 1/3 anticoagulation practice differed significantly from the mean hospital. The 1/3 are not ‘outliers’ as there is no ‘normal’ rate, but this is useful to demonstrate the influence of culture and practice as well as just individual clinician and patient factors. Maybe my practice is simply the practice of our unit?
- Approximately 1/3 of patients received an anticoagulant and 2/3 didn’t - that’s pretty close to equipoise.
- The patients who received anticoagulation were different from those who didn’t. The important thing here is that whether they received it was based on whether it was prescribed. What I mean is that we should look at these differences with respect to decision making, not with respect to which patients should be anticoagulated. We can only guess how those factors influence decision making. Some are relatively easy to guess; patients with prior bleeding or haematological failure were less likely to be anticoagulated. Patients with heart failure received more anticoagulation – is this because they were being looked after by cardiologists? (which is possible as if the attending physician was a cardiologist the rate of anticoagulation was higher). Patients with renal failure were less likely to be anticoagulated – could this be as simple as the need to adjust the dose making a prescription less likely? Some differences are more difficult to explain (for example women were less likely to be anticoagulated than men, patients receiving vasopressors less likely but patients in intensive care more likely).
- CHA2DS2VASc could not predict ischaemic stroke in this sample; the C statistic was 0.526 (a value of 0.5 is chance, 1.0 is perfect prediction). It’s interesting they didn’t look at HAS-BLED to predict bleeding.
- Approximately 1.5% of patients in both groups (anticoagulated and not) had an ischaemic stroke.
- Bleeding was more frequent in the anticoagulated group (8.6% vs 7.1%) giving a NNH of 67.
- The data relating to pre-existing vs. new AF is difficult to get your head around. Patients only receiving oral anticoagulation were excluded (excluding anyone with AF in whom the Warfarin is just continued). The rates of IV anticoagulation didn’t differ between those with new and existing AF, and the pre-existing AF group includes those where IV therapy was added to Warfarin but also those in whom the Warfarin was replaced. There are conflicting results in comparing new and existing AF and essentially I don’t think this paper really gives any answers.
- We don’t have any information about the population rate of bleeding. We could assume it is 7.1% (same as in the group not given anticoagulation) but that would be a big assumption. What if the background rate is 9%? We also don’t know the population stroke rate.
So what should we make of this data? My practice of not routinely prescribing anticoagulation in sepsis associated AF goes with the majority. Even if I knew which patients were at a higher risk of stroke (which I don’t from this data) anticoagulation may not help, and could harm. I’m none the wiser as to what happens to the patient post discharge, or whether anticoagulation makes any difference long-term. For patients with pre-existing AF all bets are off.
Whilst there aren’t many answers, this paper was about something more interesting than that; it describes practice, hints at how we make decisions and informs as to what is happening to our patients now rather than what might happen if the results of an RCT can ever be replicated in the ‘real world’.
As always, please leave a comment to get some discussion going......