The Gambling Game: Clot vs Bleeding

John Mandrola


February 13, 2015

How does one apply population data to the person in front of you? The patient with atrial fibrillation and one additional risk factor for stroke presents this very challenge. You aim to provide benefit—stroke prevention—but you risk doing harm—bleeding.

Cases like these give me gray hairs:

  • A 54-year-old normal-weight man exercises regularly and reports no problems. He takes an ACE inhibitor for hypertension, which prompts him to have a yearly checkup, during which his primary-care doctor noted an irregular rhythm. The ECG showed AF with a controlled heart rate. An echocardiogram was unremarkable. He does not drink alcohol, and basic lab testing returned normal values. He is a neurosurgeon.

  • A 55-year-old woman with well-controlled diabetes reports an occasional fluttering in her chest, which she says is "not terribly bothersome." She attributes the symptoms to stress. She is the primary caregiver for her father, who suffered a debilitating stroke. Her 24-hour monitor showed episodes of atrial fibrillation that last up to 2 hours at a time. She reported no symptoms the day she was monitored. She has a normal exam (other than the strained facial expression), ECG, echocardiogram, and basic laboratory testing. She does not drink alcohol.

It seems a simple binary question: should this CHA2DS2-VASc-1 male patient and this CHA2DS2-VASc-2 female patient take an anticoagulant? Yes or no? And don't tell me about shared decision making, because as an expert advisor, you must, at least, render an opinion.

Stroke prevention forces a doctor and patient into a difficult place, a place no one likes to be: uncertainty. To alter (with drugs) the delicate balance of hemostasis is to do a lot. You cannot prevent thrombosis without increasing the risk of hemorrhage.

It's a gamble, so we need the odds. But a problem arises. The person in front of you is not a population. How do we apply stroke and bleeding data from a Swedish, Danish, Taiwanese, or even US population to the man or woman in your exam room?

And math is not the only issue. Humans do not easily calculate risk; humans feel risk. In these two patient vignettes: the neurosurgeon's work centers on bleeding complications, while the woman lives with the devastation wrought by a wayward clot. Parallax again. Same statistics, different perspective of the lens.

And now the bloody statistics are changing.

Provocative New Study

A recent study[1] from a group of Swedish researchers (led by Dr Leif Friberg [Karolinska Institute, Stockholm, Sweden]) published in the Journal of the American College of Cardiology suggests that the previous estimates of the risk of ischemic stroke in AF patients with one risk factor may have been too high.

As background, recall that European[2] and American[3] AF treatment guidelines differ on the tipping point of the net benefit of anticoagulation. European guidelines call for anticoagulation at CHA2DS2-VASc score ≥1, while North American guidelines put it at 2. These recommendations stem from previous registry studies that estimate the risk of stroke in untreated AF patients greater than 1% per year. A Danish registry analysis,[4] for instance, had it as high as 2% for patients with a CHA2DS2-VASc score of 1.

The question Friberg and colleagues asked in this new paper was whether the true incidence of stroke in untreated CHA2DS2-VASc-1 patients was actually that high.

Stricter Definition for Stroke Events

For their retrospective analysis of over 140 000 AF patients in the Swedish health registry, Friberg and colleagues defined "stroke events" in a much stricter way. They excluded patients with the diagnosis of transient ischemic attack (TIA), pulmonary embolism, or "unspecified" stroke and any patient with exposure to warfarin. They made efforts to avoid double counting of admissions for acute stroke and those for stroke sequelae.

The thinking behind their methodology is that TIA is too soft an end point for a registry study; pulmonary embolism is a different disease; and the diagnosis of "unspecified stroke" enrolls patients with complications of stroke rather than new events. Exposure to warfarin obviously biases the sample because they aimed to understand the untreated risk of stroke.


The results were provocative. Using this stricter definition of stroke, they found that the annual event rates in CHA2DS2-VASc-1 patients dropped to 0.3% per year. For women with no other risk factors, the event rate was 0.1% to 0.2%, and for men, the rate was 0.5% to 0.7%. Despite these low event rates, 46% of the men and 22% of women with CHA2DS2-VASc-1 were taking warfarin at baseline.

In the discussion section, the authors defend their methods and question the relatively high event rates (2.0%) for CHA2DS2-VASc-1 patients seen in the Danish registry. Their conclusion: the risk of ischemic stroke in patients with AF and a CHA2DS2-VASc score of 1 seems to be lower than previously reported.

Accompanying editorial

Dr Daniel Singer (Massachusetts General Hospital , Boston) and Michael Ezekowitz (Thomas Jefferson University, Philadelphia, Pennsylvania), who wrote the accompanying editorial,[5] note the wide cohort-to-cohort variation in reported CHA2DS2-VASc–stratified rates of stroke in untreated patients. The Friberg report, they write, "highlights how sensitive such big data is to the way it is analyzed."

Notwithstanding the bias inherent in database analyses, Singer and Ezekowitz also point out problems with the CHA2DS2-VASc method of risk assessment. Although this score is more nuanced than the CHA2DS2 score, its C statistic, a measure of predictive value, is 0.647, a value that could hardly be called predictive. (Note: a C statistic of 0.5 is random chance.)


The original question remains: "Doc, should I take an anticoagulant or not?"

When it comes to treating AF patients with one risk factor, knowing the untreated risk of stroke is key because anticoagulants come with a cost—bleeding. If only five out of 1000 patients with AF and a CHA2DS2-VASc score of 1 have a stroke, how could it be lowered with any drug? Even if it was lowered, would that be offset by an increased risk of intracranial hemorrhage (ICH)?

We assume the absolute risk reduction of anticoagulants is similar across risk groups (maybe), so the net benefit hinges on weighing that amount of risk reduction against the increase in bleeding risk.

When does that number tip to the good? How we translate these trade-offs to patients depends a lot on how we, as doctors, feel about the trade-offs.

I have thought a lot about this problem. Like any human being, I am biased by the lens through which I see the problem. My lens shows me, at a frequency of about three to four times per year, a CHA2DS2-VASc-1 (or even 0) patient who has suffered a major stroke. Seeing these patients (especially those with a score of 0) gets you thinking: "Wow, that is bad. Maybe we should think more about the tipping point of anticoagulation."

Or, maybe some strokes just happen and are out of our control. Or perhaps there are other factors not yet known that influence stroke risk in these patients, say left atrial appendage morphology.[6]

Expert opinions

To learn more, I reached out to experts in the area.

Dr Suneet Mittal (Valley Hospital Health System, Ridgewood, NJ) tweeted that, contrary to the typical cardiology view, not all strokes in AF patients come from the atria. Good point.

Dr Torben Larsen (Aalborg University Hospital, Denmark) said in an email, regarding the Friberg et al paper, that the "devil is in the detail of the methodology." Excluding patients exposed to warfarin creates a "conditioning-on-the-future" selection bias, wherein the "predefined" nontreated patient group is inherently lower risk, he noted.

Dr Gregory Lip (University of Birmingham, UK) directed me to other cohorts showing higher risks in CHA2DS2-VASc-1 patients.

At the 2014 ESC meeting, his group (lead author Dr TF Chao [Taipei Veterans General Hospital, Taiwan]) presented a Taiwanese registry study[7] showing that untreated CHA2DS2-VASc (CHA2DS2-VASc-1 for males, CHA2DS2-VASc-2 for females) patients have annual event rates in the range of 2% to 3% per year. These event rates are much closer to the Danish data. In the discussion portion of that abstract session, they showed a graphic illustration of ICH rate from the major trials of new oral anticoagulant (NOAC) agents, which range from 0.23 to 0.50 per year. Their conclusion, unlike that of the Friberg and colleagues, was that anticoagulation provides a net benefit in these patients.

Dr Lip also pointed me to a study[8] of 9727 AF patients from Hong Kong. In the supplemental material, the authors show ICH rates for patients on no therapy vs aspirin or warfarin. Overall, they found rates of ICH (in HAS-BLED-1 patients) were not significantly increased with warfarin.

On the risk of ICH in warfarin-treated patients, I found this analysis[9] of the ATRIA cohort in the Journal of the American Geriatric Society. This group of researchers in Boston noted that that ICH rates stayed quite stable until age 80. In their paper, patients younger than 80 years had untreated ICH rates of 0.1% to 0.2% annually, while those on warfarin had rates of 0.3% to 0.4% annually. In those 80 years of age or older, the annual ICH rate rose to about 0.6% in both warfarin and nonwarfarin groups. So yes, ICH rates do go up with anticoagulation, but not much in absolute terms.

ICH Rates in Warfarin- and Non–Warfarin-Treated Patients

Age (years) <60 (n=1453) 60–69 (n=3269) 70–79 (n=6767) ≥80 (n=3818)
Event rates on warfarin 0.34 0.40 0.37 0.76
Age (y) <60 (n= 2493 ) 60–69 (n=2946) 70–79 (n=5253) ≥80 (n=4934)
Event rates off warfarin 0.12 0.07 0.13 0.69
Adapted from Heart Rhythm 2014; 11:1401-1408. [8]

Dr Christian Ruff (Brigham and Women's Hospital, Boston, MA) emphasized to me the notion that not all CHA2DS2-VASc-1 patients have the same risk of stroke. "For example, a young patient with mild, well-controlled HTN does not have the same risk as a patient with systolic heart failure or a 74-year-old with AF."

His take was similar to Drs Lip and Larsen: "These patients, although at a generally lower risk of stroke, still have the same relative 60% to 80% reduction in their risk with anticoagulation. For patients who have a low risk of bleeding, it is certainly reasonable to offer anticoagulation. Perhaps particularly with the NOACs, where the risk of serious bleeding is half that of warfarin."

I followed up with Dr Ruff on the incremental risk of ICH with anticoagulation. He said, "In my opinion, any antithrombotic agent (anticoagulant or antiplatelet) increases the risk of intracranial hemorrhage. It is hard to know what the absolute ICH rate is in the same AF population not on anticoagulation. In the warfarin-vs-placebo trials, there were so few patients that we don't have a good estimate of the baseline ICH rate. [But] it is hard to say that warfarin does not increase the ICH rate vs placebo if we have definitive evidence that the NOACs decrease the ICH rate by >50% [vs warfarin] in approximately 72 000 patients from clinical trials."

Dr John Day (University of Utah, Salt Lake City) feels differently about anticoagulation in CHA2DS2-VASC 1-2 patients. He noted in an email that the way guidelines are written, most patients with AF are candidates for anticoagulation. Yet the "neurology literature is replete with data on anticoagulation, cerebral microbleeds, and long-term dementia risk."

His team recently published a paper[10] in the Heart Rhythm Journal associating the quality of anticoagulation management with future dementia risk. Patients with the worst INR control had more than a fivefold risk of dementia. Dr Day also had concerns about major noncranial bleeding, pointing to the ARISTOTLE study,[11]

Dr Ethan Weiss (University of California, San Francisco) echoed the concerns of Dr Day. Dr Weiss said that "we need to learn a lot more about risks of NOACs. One of the great opportunities of the NOACs is that we now have drugs with reasonable pharmacokinetics and without all the potential confounders of warfarin." Weiss also reminded me about the long-term risk of exposure to anticoagulation. Young patients with CHA2DS2-VASc scores of 1-2 face decades of exposure to anticoagulation. "One basic idea is to change the way we report risk of stroke or bleeding to 10-year risk (if it is even possible), which is what we do now for lipids," he added,

My Take-Home

I began this exercise with the bias that our two patients (the neurosurgeon and daughter of a stroke victim) would clearly benefit from anticoagulation. Stroke is horrible, bleeding less so. The Friberg paper gets you thinking about what the stroke risk is when these patients are not treated. Drs Lip, Larsen, and Ruff make compelling cases for the net benefit of anticoagulation in these patients. Dr Mittal, Day, and Weiss raise opposing concerns about the risks of anticoagulation, especially long-term anticoagulation.

This conundrum reminds me of the primary-prevention implantable cardioverter defibrillator (ICD) argument. The problem is risk stratification. We know most patients we treat will not have an event. Think about it: even if the annual stroke risk in CHA2DS2-VASc-1 is as high as 2%, that means there is a 98% chance of no event. It would be ideal, therefore, to know—through biochemical means, imaging, or other sensors—which CHA2DS2-VASc-1 patient benefits from taking anticoagulation.

Until then, doctors will have to accept uncertainty, discuss it frankly with our patients, and then be mindful of the perspective with which each patient interprets (feels) the benefits and risks of treatment—or no treatment.

The neurosurgeon and daughter of a stroke victim will likely choose to gamble on different choices.



Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.