David M Kent & David E Thaler. American Scientist. Volume 103, Issue 1. Jan/Feb 2015.
Sudden weakness on one side of the body, slurred or incoherent speech, a droop in the muscles of the face-these are all possible signs of a stroke, an interruption in the circulation of blood throughout the brain. According to figures from the American Stroke Association, on average someone in this country suffers a stroke every 40 seconds-a rate that amounts to nearly 800,000 a year. Somewhere between 4 and 8 percent of these strokes are fatal; the remainder cover a spectrum from clinically trivial to catastrophic, depending on the size and anatomic location of the brain area affected. A stroke can produce temporary or long-lasting visual field defects, loss of sensation, paralysis, loss of the ability to speak or to understand language, or any of a wide variety of other debilitating and sometimes curious neurological symptoms and syndromes.
Just as the consequences of a stroke can vary, so too can the causes. These causes divide strokes into two broad categories: ischemic strokes, in which a cerebral blood vessel becomes blocked, and the less common hemorrhagic strokes, in which a vessel ruptures or leaks blood into the surrounding tissue. In either case prompt treatment is essential, because brain tissue dies off quickly without a continuous supply of nutrients and fresh oxygen to the neurons that control our muscles, our perceptions and thoughts, and even our ability to breathe.
Following a stroke, clinicians typically run tests to learn as much as possible about its cause. For example, in the case of an ischemic stroke, did the problem begin with the buildup of plaque in a large vessel, or with a fragment of plaque that broke off and migrated to a smaller vessel? Did a blood dot originate in the left atrium of the heart, owing to stasis from an abnormal rhythm, or somewhere else in the circulatory system-perhaps at a site quite distant from the brain? Such distinctions may provide crucial information that can guide therapy to help the patient avert another stroke in the future.
A Hole in the Heart
In approximately 30 percent of stroke cases, the cause cannot be identified even after an extensive workup. Such cases are known as cryptogenic strokes, which is simply to say that the cause remains unknown. Thus, cryptogenic stroke represents a “wastebasket” category of strokes with many different etiologies.
Fairly often, patients with cryptogenic stroke are found to have a patent foramen ovale (PFO), a tunnel-like structure in the septum that separates the right and left atria of the fetal heart. During gestation, the foramen ovale allows the circulation to bypass the dormant fetal lungs, since blood is oxygenated in the placenta. Usually the foramen ovale seals shortly after birth, when the infant’s first breath creates pressure that brings together two flaps of tissue to form an impenetrable septum. Sometimes, however, the two flaps fail to come together completely, leaving a PFO in about 25 percent of adults..
Most often, this incomplete closure apparently poses no threat throughout the individual’s lifetime. However, by providing a small channel from the right side to the left side of the heart, a PFO may, in theory, permit venous blood clots to pass into the arterial circulation, creating the conditions for an ischemic stroke from a paradoxical embolism. (This process is illustrated on the facing page.) What makes such an event paradoxical is that the contents of a vein could never travel directly into an artery under normal conditions.
As mentioned earlier, figuring out the probable cause of a stroke may help guide therapy to prevent the next one. After a stroke from atherosclerotic disease, most physicians recommend antiplatelet therapy and statins (as well as antihypertensive drugs, in cases of high blood pressure). For prevention of a second stroke from atrial fibrillation (caused by clotting from stagnant blood in the left atrium that embolizes to the brain), the appropriate treatment is usually some form of anticoagulant or blood thinner. For strokes believed to be caused by paradoxical embolism, the intuitively appealing approach is to close the PFO with a device such as the one that appears on page 56. The device is usually placed in the heart by a catheter in a minimally invasive procedure. However, because PFO is so common and is usually regarded as benign, its presence in a patient with cryptogenic stroke may be a completely incidental finding—and the stroke may have been caused by another occult mechanism altogether. In such cases, even the relatively small risks associated with the procedure and the permanent cardiac device seem poorly justified.
Although PFO closure is widely if haphazardly employed, clinical proof of its benefit has been hard to find. Three recent trials, comprising more than 2,000 patients, were unable to find a statistically significant benefit in preventing stroke recurrence for mechanical closure, as compared to clot-preventing medication alone. Nevertheless, the trials did not put the question to rest, since they showed some tantalizing (yet inconclusive) signals suggestive of benefit. Many experts think larger beneficial effects might be found if we were better able to identify the patients whose strokes are most likely to have been caused by paradoxical embolism.
For the past five years, we have led an international collaboration called the Risk of Paradoxical Embolism (RoPE) Study. One of us is a neurologist with a special interest in the role of PFO in stroke; the other is a medical doctor and research methodologist with a long-standing interest in how mathematical risk models might be applied to disaggregate the overall results of trials to provide more detailed evidence for the care of patients based on their specific characteristics. (For a discussion of how averages tend to hide individual differences in clinical trials, see the article by Kent and Hayward in the January-February 2007 issue of American Scientist.) The premise of the RoPE Study is that the benefits of PFO closure in a patient are depen- dent on the joint probability of two distinct dimensions of risk: the probability that the first, or index, cryptogenic stroke was attributable to the PFO (in other words, the PFO was not an incidental finding) and the probability that the stroke will recur. We call this joint probability the attributable recurrence risk.
What’s the Evidence?
In an individual case, it is typically not possible to determine with certainty whether a PFO discovered in the setting of a cryptogenic stroke was the cause of the stroke or just an incidental finding. However, evidence from epidemiologic studies shows that PFO is a much more common finding in patients with cryptogenic stroke than in the general population or in patients with strokes whose cause is known. Based on studies that compare the prevalence of PFO in cases of cryptogenic stroke against its prevalence in a control group (either from the general population or from patients with a stroke of known cause), it is possible to determine the fraction of cryptogenic strokes that can be attributed to the presence of a PFO; the mathematical logic is illustrated by the patients of “Stroke Ward XYZ,” on page 57. For this example, let us posit a PFO prevalence of 40 percent among those with cryptogenic stroke and 25 percent among a control population. If we assume the prevalence of PFO (in patients whose cryptogenic stroke is unrelated to PFO) to be equal to that in the general population-that is, 25 percent-these figures suggest that approximately 50 percent of PFOs discovered in the setting of cryptogenic stroke would be merely incidental.
In essence, the role of PFO in cryptogenic stroke is a question of probability. Using Bayes’s theorem to solve for the probability that PFOs are incidental yields the equation shown below on this page. Thus, as PFO prevalence in cryptogenic stroke patients decreases, the probability that a discovered PFO is merely incidental will increase. If the prevalence of PFO in cryptogenic stroke patients were equivalent to that in a control population (approximately 25 percent), the probability of a discovered PFO being merely incidental would be 100 percent-which is to say, this would be the expected rate if PFO was not a risk factor for cryptogenic stroke at all. Conversely, if we can find characteristics that identify a cryptogenic stroke population with an especially high prevalence of PFO, this would suggest a much higher probability that a stroke in this population was Attributable to the presence of the PFO.
lire equation has another interesting property as well: The right side is numerically equivalent to the inverse of the odds ratio in case-control studies. This feature makes it easy to convert the odds ratio of case-control studies associating the prevalence of PFO in cryptogenic stroke patients (versus the prevalence in control subjects) into what we really want to know: the probability that a discovered PFO is just an incidental finding, unrelated to the stroke.
When we applied this formula across 23 previously published casecontrol studies, we found that-on average-about one-third of the PFOs discovered in the setting of cryptogenic stroke are incidental; the others are considered pathogenic. But more important, we found tremendous variation across studies in these estimates; moreover, the variation seemed to correlate with the characteristics of the different populations in the different studies. This provided us with an important clue as to how we might estimate the attributable risk for an individual.
Paradoxical Data
Given that PFO is strongly associated with cryptogenic stroke, and that it can be eliminated with a minimally invasive procedure, why would a physician withhold this therapy and choose instead to wait around for a second, possibly disabling, event? Indeed, in some centers, PFOs discovered in the setting of a cryptogenic stroke are routinely closed. Despite the compelling logic, however, the consistent association of PFO with cryptogenic stroke has been accompanied by a surprising yet equally consistent finding: PFO appears not to be a risk factor for recurrent stroke. Patients with a PFO have similar (or lower) stroke recurrence risks compared with other cryptogenic stroke patients. To some researchers, this finding offers an argument against closure: Why close a PFO if the condition apparently does not increase the patient’s risk?
Even more surprising is the suggestion, from some studies, that small PFOs may be associated with a higher risk of recurrence than larger PFOs, leading some of us to propose (jokingly) that “high-risk” small PFOs should perhaps be dilated into “lowrisk” large ones.
Although several hypotheses have been proposed to explain these counterintuitive findings, we (with our colleague Issa Dahabreh, now at Brown University) have described a fundamental bias that affects all research on the causal mechanisms underlying the risk of recurrent events, when the risk factors for the recurrent episode might be similar to the risk factors for the first event, that can account for these paradoxical observations. We have named this phenomenon index event bias, because it arises in studies that select patients based on the occurrence of an index event. Selecting patients on this basis induces a negative correlation in the occurrence of risk factors, and the study can then tend to underestimate their importance in determining a future event.
PFO in cryptogenic stroke provides an excellent illustration of index event bias. Since it is a congenital anomaly, essentially distributed randomly at birth, people with and without PFO are equally likely to have vascular risk factors for stroke, such as diabetes, hypertension, and so forth. Among patients with a cryptogenic stroke, however, those with a PFO tend to be younger than those without a PFO and also tend to have much lower rates of the conventional risk factors for stroke, such as diabetes, hypertension, hypercholesterolemia, or smoking. This negative correlation between PFO and other risk factors for strokes arises in those with cryptogenic stroke because both PFO and these risk factors contribute to the same outcome-cryptogenic stroke. Heuristically, the association may be considered to emerge because patients with PFO do not require the same burden of risk factors to fall victim to a stroke; in certain circumstances, having a PFO may be sufficient. This skewing of risk factors explains why the presence of a PFO appears to “protect” cryptogenic stroke patients from the other risk factors (both known and unknown) for cryptogenic stroke-and why they might be at similar or lower risk for recurrence relative to those without a PFO, whose cryptogenic stroke is caused by other mechanisms.
The finding that stroke appears to have an equal or greater risk of recurrence in cases of small PFOs than of large ones may be similarly explained by index event bias: It may simply be a sign that the small PFOs are more likely to be incidental and coincide with other risk factors, and with mechanisms unrelated to PFO that may confer more risk than paradoxical embolism does.
The large and consistent differences in the characteristics of cryptogenic stroke patients with and without PFO hold several important implications for the prevention of future strokes. First, they add to the indirect evidence that PFO is pathogenically important in the first cryptogenic stroke; otherwise, cryptogenic stroke patients with or without this anatomical variant would be expected to have a similar burden of stroke risk factors-just as in the general population. Second, the finding suggests that PFO may be an important risk factor for stroke recurrence, since it “compensates” for the shortfall in other risk factors. Most significant, the differences in patient profiles (in comparisons of patients with and without PFO) gives us a means of predicting the probable presence or absence of PFO in a patient with a cryptogenic stroke, even before we perform any imaging on his or her heart. Although it is totally unpredictable in the general population who might have a PFO, we can use the presence or absence of vascular risk factors and other characteristics in cryptogenic stroke patients in mathematical models that can predict the probability of finding a PFO. We call this individualized estimate PFO propensity-that is, the probability that a cryptogenic stroke patient has a PFO, as calculated on the basis of other characteristics. Among cryptogenic stroke patients, this propensity is increased by lower age and by the absence of conventional stroke risk factors.
The equation on page 56 relates PFO prevalence to the probability that a PFO is incidental. The same equation can be used in an individual case to estimate the likelihood that a discovered PFO is incidental or pathogenic; the calculations here would be based on the prevalence of PFO in cryptogenic stroke patients with similar characteristics.
Calculating the RoPE Score
To get a robust estimate of these probabilities, we needed a much larger database of patients with cryptogenic stroke investigated for PFO than had previously been assembled. We therefore came up with a euphonious acronym (the RoPE Study) and formed an international collaboration to pool data from 12 different centers, including more than 3,000 cryptogenic stroke patients. After more than a year of work to harmonize the data across these different studies, we performed analyses that found the odds of detecting PFO to be diminished by older age, the presence of diabetes, coronary artery disease, hypertension, and hypercholesterolemia, as well as by current smoking and history of stroke or TLA (a transient ischemic attack or “ministroke”). If medical imaging revealed a superficial stroke (in the periphery of the cerebral cortex rather than deep within the brain), this also increased the probability of finding a PFO. These results were found to be consistent across all 12 databases.
On the basis of our mathematical model, we developed a point score, assigning a single point for the absence of each of three vascular risk factors (diabetes, hypertension, smoking), the absence of a prior stroke or TIA, and the presence of a cortical stroke on brain imaging. For age, we assigned a point for each decade younger than 80, ranging from one (for patients in their 60s) up to five (for patients in their 20s). This system yields a 10-point score of risk. A score of 10 represents the highest level of PFO-attributable risk; at the other end of the scale, a 0 or 1 represents the lowest level of PFO-attributable risk. Thus a young and apparently healthy patient who has suffered a cryptogenic stroke, such as the individual whose chart appears at the top of page 58 under the name “Krassen,” ends up with a RoPE score of 8, whereas an older patient whose risk factors include hypertension and diabetes (and whose chart bears the name “Marco”) receives a RoPE score of only 2.
Given the strength and consistency of the effects that have already been found, the presence or absence of these features should allow clinicians to identify sizable groups of cryptogenic stroke patients with very different prevalences of PFO, ranging from approximately 20 to 80 percent. This range in turn suggests clinically important variation among patients in the probability that a PFO is pathogenic (likely to have caused the stroke) rather than incidental, as estimated on the basis of easily observed characteristics. Patients with a low RoPE score (say, 0 to 3 points) have a nearzero probability that their stroke was caused by their PFO, whereas for patients with a high RoPE score (9 or 10), the likelihood that their stroke was caused by PFO is very high, near 90 percent. Even among the relatively well-selected patients who meet the conventional criteria for entry into the major randomized trials, calculating the RoPE score can uncover variations that may hold clinical significance.
A major premise of this calculation, of course, is that in a control group (a population without PFO-attributable stroke), the prevalence of PFO is totally uncorrelated with RoPE score. Such a dissociation—a correlation of PFO with risk factors in patients with cryptogenic stroke but not in other populations—is exactly what would be expected if the apparent correlation arises from the index event bias we have described. We checked this assumption in a sample of patients who had experienced strokes of known cause and found that, as predicted, the PFO prevalence was about 25 percent, regardless of their RoPE Score.
Yet Another Wrinkle
With such strong, consistent, clinically intuitive and meaningful results, one might anticipate that the problem of patient selection for mechanical closure is solved. But our RoPE analysis revealed yet another wrinkle: Recurrence risk appears to be considerably lower in patients most likely to have had a PFO-attributable stroke (that is, with a high RoPE score) than in cryptogenic stroke patients whose PFO is probably just incidental (with a low RoPE score). The bottom figure on page 58 shows the two-year recurrence rates of stroke or TIA in patients with PFO, stratified by their RoPE score. The red bars in the figure show that recurrence rates decrease dramatically as the RoPE score increases, suggesting that patients with index events most likely to be PFO-attributable are the ones least likely to experience recurrent ischemic events.
These results provide useful clinical insights into the widespread and growing disease of stroke. At the same time, they underscore the challenges of selecting the appropriate patients for PFO closure and the methodological hurdle inherent in PFO closure trials: If the RoPE score is used to select the patients with a high fraction of their strokes attributable to PFO, the number of recurrent events may be too low to provide adequate power for a trial to find any clinical benefit. Put another way, it is hard to show that PFO closure prevents stroke when the vast majority of patients remain stroke-free even without closure. Other characteristics, such as anatomical features of the PFO itself, may need to be added to select highrisk patients from among those with a high RoPE score to identify the patients most likely to benefit.
Nevertheless, the RoPE score can be useful for doctors when we counsel patients, even while the debate over mechanical closure continues. We can tell “Marco” and other patients with a low RoPE score there is a high probability that their PFO is an innocent bystander. For such patients, aspirin and statins would be indicated; these have been shown to lower recurrence rates of stroke generally, and they should be especially helpful for those with atherosclerotic risk factors. As for “Krassen” and other patients with a high RoPE score, we can reassure them that while the PFO is likely to have been involved in their stroke, they are at relatively low risk for a PFO-related recurrence, even with medical therapy alone. Whether PFO closure might further reduce their risk remains controversial.
More generally, the RoPE project provides a new way to think about patients with cryptogenic stroke: Even when the stroke cause cannot be determined with certainty, there might be indirect ways to estimate it probabilistically. The study also provides one more example of how overall results from clinical trials may be difficult to apply at the individual patient level- and how bringing together clinical, statistical, and epidemiological reasoning can ultimately help us deliver the right treatment to the right patient.