What can data mining teach us about triptan safety that we don’t already know?

Triptans are a mainstay of migraine treatment. They were hailed as a breakthrough when they became available in the early 1990s, but excitement about their obvious value was tempered by concern about possible adverse vascular events. From a physiological standpoint this worry was well founded, since triptans have agonist effects at serotonin receptor subtypes that mediate blood vessel constriction (1). Moreover, the limited power of pre-approval clinical trials to detect safety problems was widely recognized.

The number of vascular complications identified in post-marketing surveillance after the drugs became widely available, however, proved small in comparison with their extensive use. Large-scale observational studies provided further evidence of a good record of vascular safety when physicians prescribed triptans in the course of routine practice (2). In some countries, although not the United States, triptans became available without physician prescription. Even so, no apparent epidemic of vascular complications emerged. What, then, remains to be learned about the vascular safety of triptans?

Although one might expect that any controversy has been settled, the limitations of traditional methods of pre- and post-marketing drug safety surveillance are well known. The vascular risks of commonly used nonsteroidal anti-inflammatory drugs (NSAIDs), for example, have only recently been recognized. Reassessment of their safety was prompted by complications observed in trials of Cox-2 inhibitors; eventually it became clear that these problems were not limited to newer drugs but instead were class effects (3). The NSAID situation teaches us that post-marketing surveillance is never “finished” and that previously unrecognized risks may emerge even with drugs that have been on the market for a long time. The lesson is that we should never stop looking.

For this reason, the study by Roberto et al. (4) published in the current issue of Cephalalgia provides a welcome new perspective on the matter of the vascular safety of triptans. The authors used data from the Food and Drug Administration Adverse Event Reporting System (FDA_AERS) from 2004 to 2012 to examine whether triptan use was disproportionately associated with reports of vascular events. Reports of drug-associated adverse events in the AERS database are submitted voluntarily by pharmaceutical manufacturers, healthcare providers, or patients. They typically include a description of the adverse event and the drug or drugs suspected of causing it, as well as patient characteristics such as age, sex and other prescribed medications.

Reported events are grouped using a four-level hierarchy. This study evaluated adverse events within the highest level categories of “Cardiac disorder” or “Vascular disorders.” In each category, reports describing patients who experienced vascular events in association with triptan use were classified as cases. Reports describing the same type of vascular events in association with non-triptan drugs were designated non-cases. The authors then calculated a reporting odds ratio (ROR) and 95% confidence interval (CI) of cases to non-cases, defining a disproportionality signal of interest as one in which there were more than three reported cases and the lower limit of the 95% CI for the ROR was above one. The researchers addressed the problem of confounding by pre-existing vascular risk by adjusting RORs for the proxy measure of co-prescribed cardiac medications.

Of 7800 adverse events in which a triptan was identified as the suspected drug or one of several suspected drugs, almost 2600 were categorized as cardiac or vascular disorders. Disproportionality signals were identified for the subcategories of coronary artery disorders (ROR 2.0, 95% CI 1.85–2.16), cardiac disorders signs and symptoms (ROR 1.11, 95% CI 1.03–1.20), vascular disorders not elsewhere codified (NEC) (ROR 2.01, 95% CI 1.34–2.99), and aneurysms and artery dissections (ROR 2.01, 95% CI 1.34–2.99). RORs were larger (but with wider CIs) for more specific levels of categorization, and most remained significant even after adjustment for concomitantly prescribed cardiac medications.

As the authors point out, most of these disproportionality signals are consistent with the known vascular risk profile of triptans, such as the association with coronary arteriospasm (adjusted ROR 21.58, 95% CI 16.06–23.86). Some, however, were unexpected, such as the signals for arterial dissection (ROR 12.24, 95% CI 4.42–32.25) and placental infarction (ROR 12.68, 95% CI 3.23–42.70). The authors identify other “unexpected signals” for cerebrovascular events, aneurysms and artery dissections and pregnancy-related vascular events of pregnancy-associated hypertension and pregnancy-induced hypertension. Because adjustment for pre-existing vascular risk did not appreciably change most of these signals, the authors suggest that triptans “may represent an additional risk factor for serious vascular events” and that their results are consistent with “the hypothesis of a rare subjective susceptibility…to a triptan vasoconstrictive effect, as also suggested by some cases of serious ischaemic events in patients without pre-existing risk factors.”

It is not easy to interpret the scientific or clinical implications of these results. One strength of the AERS database is the certainty that the suspected drug was used in close proximity to the adverse event, something that is not easily determined when drug safety is examined using routinely collected information such as insurance company prescription records. The unexpected disproportionality signals are indeed intriguing, and some might argue they should prompt research to verify or refute them, ideally using data from other sources of post-marketing safety surveillance. In the end, however, the findings of this study are hypothesis-generating and not evidence of causality. To the list of limitations acknowledged by the authors (selective reporting, under-reporting, duplicate reporting, use of cardiovascular drugs to treat migraine and notoriety bias) we must add problems associated with multiple comparisons (performing a large number of statistical comparisons increases the likelihood that some will be positive by chance alone) and confounding by indication (in which patients at high risk for adverse vascular outcomes are more likely to be exposed to triptans).

To give just one example of the latter problem, the true nature of the relationship of triptans to pregnancy-associated hypertensive problems is complicated by the fact that most women with migraine experience headache improvement during pregnancy. Even in those whose headaches remain a problem, our clinical experience is that most women try to avoid the use of medications, including triptans, during pregnancy. Women who continue to have migraine headaches during pregnancy are different. They are more likely to have aura than not, which in turn is known to be associated with an increased risk of pregnancy-related hypertensive and vascular complications (5–7). It is thus possible that the patients most likely to use triptans during pregnancy are at exceptionally high risk of conditions identified by the disproportionality signals in this study.

What does all of this mean when prescribing triptans for patients with migraine? Taken as a whole, this analysis of the AERS database represents a useful complementary source of information about triptan safety, but it is important to interpret the findings in the context of other safety information about triptans as well as information about the disease for which they are used. At present, the preponderance of evidence from post-marketing safety surveillance efforts does not suggest that clinical practice with regard to triptan prescribing should change.