Summary of Findings
Although broad in scope, this SR found a sparse body of evidence addressing interventions to prevent or treat primary headaches in women who are pregnant, attempting to become pregnant, postpartum, or breastfeeding. The sparsity precluded meta-analysis (either pairwise or network) and allowed us to make few definitive conclusions about treatment effectiveness or harms of the various interventions. For prevention, while we found insufficient direct evidence regarding the harms of topiramate, indirect evidence suggested that it and other antiepileptics (carbamazepine, gabapentin, and valproate) may be associated with fetal/child adverse effects. However, lamotrigine may be associated with fewer adverse effects. Indirect evidence also suggested that use of venlafaxine, tricyclic antidepressants, benzodiazepines, β-blockers, prednisolone, or oral magnesium during pregnancy may be associated with fetal/child adverse effects, but use of calcium channel blockers or antihistamines may not be.
For treatment of acute attacks of primary headache, we found direct evidence that, when used during pregnancy, the combination of metoclopramide and diphenhydramine may be more effective than codeine in reducing migraine or tension headache severity, and may not be associated with more maternal adverse effects. We found direct and existing SR evidence that triptan use for migraine may have a low risk of adverse effects, except for increased child emotionality and hyperactivity at 3 years of age. Indirect evidence suggested that use of indomethacin, ondansetron, antipsychotics, prednisolone, acetaminophen, and intravenous magnesium during pregnancy may be associated with fetal/child adverse effects, but use of low-dose aspirin or antihistamines may not be. We found insufficient evidence to make conclusions about the benefits or harms of non-pharmacologic interventions for treatment of primary headache.
Strengths and Limitations
We followed contemporary methodological standards for SRs, including engagement with multiple stakeholders and use of recommended methods for searching, screening, data extraction, RoB assessment, data (narrative) synthesis, and SoE assessment. In anticipation of a sparse evidence base, we were very inclusive in our eligibility criteria, especially in terms of primary study designs (direct evidence).
For all interventions examined in this SR, the evidence for their use for primary headaches in pregnancy was either sparse or absent. The limited information about doses, durations, and frequencies of the interventions reported in the nonrandomized studies (especially triptans) constrains our ability to make definitive conclusions about individual triptans. The populations in the studies included in this SR were varied in terms of trimester and gestational age, which limits our ability to apply our findings specifically to particular trimesters of pregnancy. In terms of type of primary headache, all studies addressed either migraine or tension headache (or both). The lack of eligible studies on cluster headache and other TACs in pregnant women is also an important limitation.
Most primary studies were conducted in the U.S., Canada, or Europe. Various contextual factors may impact the effectiveness of treatments. It is unclear to what extent the findings of this SR might apply outside of these high-income settings.
Caveats to Indirect Evidence
We suggest caution in interpretation of the findings from the indirect evidence (SRs of harms) for various reasons. First, while the SRs included studies of pregnant women, they were not restricted to patients with primary headache. It is possible that the underlying condition may impact intervention harms in pregnancy. For example, harms of antiepileptics in pregnant patients with epilepsy may differ from those in pregnant patients with primary headache. In another example, oral magnesium may be associated with greater harms when used to prevent pre-eclampsia because that population may be at greater risk of adverse outcomes. Furthermore, drug doses and/or frequencies of use differ by indication, which may contribute to differences in harms. Second, the SRs we identified often reported harms of classes of drugs as wholes, rather than individual drugs (or doses) within a class, which may have greater (or lesser) risks of adverse effects than other drugs (or doses) in the same class. For example, different β-blockers are known to be metabolized differently[68,69] and to cross the placenta to different degrees.[70,71] Third, the SRs included variable numbers of studies, and frequently only a subset of the included studies (often one or two studies) in a given SR contributed data to estimates of specific harms for specific interventions. Finally, it is possible, even likely, that some relevant studies of harms in pregnancy were not included in the SRs that we identified.
Implications for Clinical Practice
Given the sparse information regarding effectiveness of various interventions for primary headaches in pregnancy, decision makers will need to rely on the evidence about benefits and harms of the interventions from the general population, particularly regarding potential harms to the fetus or infant. This is especially true for interventions for prevention or treatment of primary headaches other than migraine, for which we found limited studies among pregnant, postpartum, or breastfeeding women. When considering potential harms to the fetus or infant, decision makers are left to extrapolate from studies of pregnant women with mostly other conditions. However, this approach may be fraught because harms of a given drug may vary by indication and based on such factors as concomitant drug use and psychosocial behaviors (e.g., smoking, substance use, caffeine intake). An alternative is to base decisions on the levels of the drugs in maternal serum/blood or that are transmitted to the fetus through amniotic fluid and/or cord blood or to the infant through breast milk. We did not consider drug levels in maternal serum/blood or in other fluids transmitted to the fetus/infant as relevant outcomes in this SR. However, we have found that the most complete resource with data about levels of the various drugs is the National Library of Medicine's Drugs and Lactation Database (LactMed®, available at https://toxnet.nlm.nih.gov/newtoxnet/lactmed.htm).
We encourage clinicians managing patients with primary headaches during pregnancy, postpartum, or breastfeeding phases to inform patients about the limitations of existing research. Given the limitations of the evidence, the patient's values and preferences and the clinician's expertise and experience take on an even more important role.
Implications for Research
The sparseness of the direct evidence identified in this comprehensive SR, while striking, should not be interpreted as indicating that interventions are not beneficial or are harmful. It does, however, point to an important and urgent research need for direct evidence for both pharmacologic and non-pharmacologic interventions in pregnancy, postpartum, or breastfeeding phases. For example, triptans, the most studied drug class, were discussed only in the context of their harms (when used for treating migraine). Additionally, we did not find any studies addressing prevention or treatment of cluster headache and other TACs in pregnancy, which is another important research need. When enrolled as part of a larger study, subgroup-specific data for patients with these types of primary headache should be reported.
In terms of outcomes, none of the included studies reported data for some maternal outcomes, such as headache-related symptoms (e.g., photosensitivity), quality of life, functional outcomes (e.g., impact on employment/school attendance), and patient satisfaction with intervention; adverse effects on breastfeeding (e.g., decreased milk supply); and some important fetal/child adverse outcomes. These should be measured and reported in future research.
Few studies randomized patients or reported between-arm estimates of treatment effect that adequately accounted for important confounders. These aspects increased the RoB and contributed to sparsity of the evidence. In addition to randomization, studies should also, where feasible, conduct blinding of participants, care providers, and outcome assessors to minimize the likelihood of performance and detection biases. When RCTs are not feasible, observational studies should be adequately designed and analyzed to compare treatments. Such analyses should appropriately account for differences between comparison groups of patients that are inherently different. Ideally, propensity score analyses (or similar rigorous causal inference methodological techniques) should be used to adequately account for these differences.
Headache. 2021;61(1):11-43. © 2021 Blackwell Publishing