Article Text

Download PDFPDF

Original research
Risk of intracranial hemorrhage associated with pregnancy in women with cerebral arteriovenous malformations
  1. Songmi Lee1,
  2. Youngran Kim1,
  3. Babak B Navi2,
  4. Rania Abdelkhaleq1,
  5. Sergio Salazar-Marioni1,
  6. Spiros L Blackburn3,4,
  7. Arvind B Bambhroliya5,
  8. Victor Lopez-Rivera1,
  9. Farhaan Vahidy6,
  10. Sean I Savitz1,4,
  11. Annika Medhus1,
  12. Hooman Kamel2,
  13. James C Grotta7,
  14. Louise McCullough1,
  15. Peng Roc Chen3,4,
  16. Sunil A Sheth1,4
  1. 1 Neurology, UTHealth McGovern Medical School, Houston, Texas, USA
  2. 2 Neurology, Weill Cornell Medical College, New York City, New York, USA
  3. 3 Neurosurgery, UTHealth McGovern Medical School, Houston, TX, USA
  4. 4 Institute for Stroke and Cerebrovascular Disease, UTHealth McGovern Medical School, Houston, TX, USA
  5. 5 University of Texas School of Public Health, Houston, Texas, USA
  6. 6 Center for Outcomes Research, Houston Methodist, Houston, Texas, USA
  7. 7 Stroke Research and Mobile Stroke Unit, Memorial Hermann Hospital-Texas Medical Center, Houston, TX, USA
  1. Correspondence to Dr Sunil A Sheth, Neurology, University of Texas McGovern Medical School, Houston, TX 77030, USA; ssheth{at}post.harvard.edu

Abstract

Background Prior studies on rupture risk of brain arteriovenous malformations (AVMs) in women undergoing pregnancy and delivery have reported conflicting findings, but also have not accounted for AVM morphology and heterogeneity. Here, we assess the association between pregnancy and the risk of intracranial hemorrhage (ICH) in women with AVMs using a cohort-crossover design in which each woman serves as her own control.

Methods Women who underwent pregnancy and delivery were identified using DRG codes from the Healthcare Cost and Utilization Project State Inpatient Databases for California (2005–2011), Florida (2005–2014), and New York (2005–2014). The presence of AVM and ICH was determined using ICD 9 codes. Pregnancy was defined as the 40 weeks prior to delivery, and postpartum as 12 weeks after. We defined a non-exposure control period as a 52-week period prior to pregnancy. The relative risks of ICH during pregnancy were compared against the non-exposure period using conditional Poisson regression.

Results Among 4 022 811 women identified with an eligible delivery hospitalization (median age, 28 years; 7.3% with gestational diabetes; 4.5% with preeclampsia/eclampsia), 568 (0.014%) had an AVM. The rates of ICH during pregnancy and puerperium were 6355.4 (95% CI 4279.4 to 8431.5) and 14.4 (95% CI 13.3 to 15.6) per 100 000 person-years for women with and without AVM, respectively. In cohort-crossover analysis, in women with AVMs the risk of ICH increased 3.27-fold (RR, 95% CI 1.67 to 6.43) during pregnancy and puerperium compared with a non-pregnant period.

Conclusions Among women with AVM, pregnancy and puerperium were associated with a greater than 3-fold risk of ICH.

  • arteriovenous malformation
  • hemorrhage
  • congenital
  • stroke
  • vascular malformation

Data availability statement

Data are available in a public, open access repository. Data are publicly available at the Healthcare Cost and Utilization Project (HCUP) website.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Brain arteriovenous malformations (AVMs) are uncommon but potentially deadly congenital lesions. With the increased usage of non-invasive brain imaging, they are being detected more often prior to rupture and intracranial hemorrhage (ICH).1 One of the unanswered questions regarding their natural history is the effect of pregnancy and puerperium on the risk of rupture for women with AVMs.

Prior studies examining the effect of pregnancy on ICH risk in women with AVMs have shown inconsistent results. Some studies suggest an increased risk of ICH from AVM rupture during pregnancy, while others show no difference compared with the non-pregnant period.2–6 However, a major limitation of these prior studies has been the failure to control for the heterogeneity of AVM morphology. The risk of rupture of an AVM varies considerably based on its angioarchitecture, with a number of morphologic features associated with a greater risk of ICH.7 8 As such, the aggregation of findings for all women with AVMs in prior studies may be an inaccurate assessment of true ICH risk by not accounting for the variability based on AVM size, location, and morphology.

In this study, we examined the relative risk for ICH during pregnancy and puerperium compared with the prior non-pregnancy period among women with AVMs. To account for the potential heterogeneity of AVM morphology, we used a cohort-crossover design in which each woman serves as her own control. We hypothesize that pregnancy confers additional risk of ICH in women with AVMs when compared with an equivalent non-pregnant epoch.

Methods

Data sources

We used data from the Healthcare Cost and Utilization Project State Inpatient Databases (SID) sponsored by the Agency for Healthcare Research and Quality (https://www.hcup-us.ahrq.gov). The SID contain almost 97% of all US hospital discharges and include data regardless of the expected payer. We selected the SID for California (2005–2011), Florida (2005–2014), and New York (2005–2014) because of their large, demographically, and socioeconomically diverse population in urban and rural areas. In addition, these three states provide synthetic person-level identifiers that allow tracking sequential visits for an individual patient within a state and across hospitals over time.9 This study used de-identified administrative discharge data and did not require patient consent. The study protocol was reviewed and approved by the Committee for the Protection of Human Subjects at The University of Texas Health Science Center at Houston.

Study design

We used a cohort-crossover design to estimate the relative risk of ICH during pregnancy and puerperium compared with the non-pregnancy period. The hospitalization associated with delivery was defined as described below. From that hospitalization, we defined an exposure period consisting of 40 weeks of pregnancy prior to the delivery and 12 weeks of puerperium following the delivery, consistent with prior studies.3 4 In addition, we constructed a 52-week pre-pregancy period as a non-exposure period (figure 1). In this cohort-crossover design, each participant was used as her own control in the non-exposure period and comparisons were made between the non-exposure period and exposure period within a participant.

Figure 1

Study design. To ensure 92 wks (52+40) of follow-up prior to delivery, the delivery hospitalization needs to occur no earlier than 10/07/2006. To ensure 12 wks of postpartum follow-up after the delivery, the delivery hospitalization needs to occur no later than 10/08/2011 for California (CA) and 10/08/2014 for Florida (FL) and New York (NY). Since the actual date was not available in the datasets, we used the discharge quarter instead and included delivery hospitalizations between 1/1/2007 and 9/30/2011 for CA and between 1/1/2007 and 9/30/2014 for FL and NY.

We chose the non-exposure period to be prior to the index pregnancy, as ICH during the pregnancy/delivery could lead to mortality or definitive AVM treatment, which would then modify the risk of ICH if the non-exposure period occurred after the exposure period. The advantage of using this design was a more precise estimate of the effect of pregnancy on the risk of ICH accounting for the potential heterogeneity of AVM morphology and allows us to compare the risk of ICH across time periods instead of comparing two different cohorts of women (ie, pregnant women vs non-pregnant women).

Study population

We identified delivery hospitalizations using diagnosis-related group (DRG) codes: DRG codes 370–375 for discharges prior to October 1, 2007, and Medicare Severity (MS)-DRG codes 765–768, 774 775 for discharge afterward.10 11 To ensure that all patients were observed for the full duration of the non-exposure and exposure periods depending on the states’ available data years, delivery hospitalizations were limited to those occurring between January 1, 2007, and September 30, 2011, for California and between January 1, 2007, and September 30, 2014, for Florida and New York. We excluded delivery hospitalizations if patients were out-of-state residents or died during the delivery hospitalization, or if the linkage information was missing, to minimize loss to follow-up. We limited our studies to deliveries for women of standard childbearing age between 15 and 44 years old at the time of delivery.12 13 For women with multiple deliveries, only the first pregnancy in our dataset was used in the analyses.14 15 If there were multiple labor- and delivery-related hospitalizations within 12 weeks, the latest hospitalization was used to determine the date of delivery, to exclude visits for false labor (figure 2). The presence of AVM was defined as having any records with the International Classification of Diseases (ICD) 9 diagnosis code for AVM (747.81) in the dataset.12 16 Because AVMs are congenital lesions,17 18 we considered a woman to have a brain AVM even if she was ultimately diagnosed after the index pregnancy and delivery.

Figure 2

Derivation of study population.

Outcome

The primary outcome was the incidence of ICH defined as a hospitalization with any of the following ICD-9 diagnosis codes: 430 for subarachnoid hemorrhage, 431 for intracerebral hemorrhage, and 432.0, 432.1, and 432.9 for unspecified intracranial hemorrhage.19 We report incidence rates of ICH per 100 000 person-years during the non-pregnancy period and pregnancy and puerperium period stratified by AVM status.

Statistical analysis

Demographic characteristics were compared between patients with AVM and non-AVM using univariate analyses (Chi-square for categorical variable; t-test and Kruskal–Wallis test for continuous age variables). Population attributable risks (%) were calculated by subtracting the incidence of ICH in patients without AVM from the incidence of ICH in the total population of each period, then dividing it by the incidence of ICH in the total population and multiplying 100 to obtain a percentage. To estimate the relative risk (RR) of ICH during pregnancy and puerperium compared with the non-pregnancy period, we conducted conditional Poisson regression, separately for women with and without AVM.20 All statistical analyses were performed using StataMP 14 (StataCorp LLC, College Station, TX) and significances were tested at P<0.05.

Results

Characteristics of the study population

The study population included 4 022 811 women, and of them, 568 (0.014%) women were identified as having an AVM (figure 2). Women with AVM were slightly older, had lower percentages of teen and younger mothers, and were more likely to be white compared with non-AVM women (table 1). The percentage of women with cesarean delivery was almost twice as high among women with AVM compared with non-AVM women (61% vs 35%). Comorbidities were not significantly different between AVM and non-AVM women except higher prevalences of hypertension and smoking among AVM women. (table 1)

Table 1

Characteristics of pregnant women by AVM status

Risks of ICH associated with pregnancy by AVM status

During the non-pregnancy period, incidence rates of ICH were 1942 (95% CI 794.4 to 3089.5) and 2.1 (95% CI 1.7 to 2.6) per 100 000 person-years for women with AVM and women without AVM respectively (table 2). Women with AVM only accounted for 0.014% of the total study population, but 11.5% of ICH incidence during the non-pregnancy period and 5.8% of ICH incidence during pregnancy and puerperium were attributable to AVM.

Table 2

Incidence of intracranial hemorrhage during the non-pregnancy period and pregnancy/postpartum period by AVM status

The risk of ICH during pregnancy and puerperium increased regardless of AVM status but the relative increase was greater among women without AVM because their baseline incidence rate during the non-pregnancy period was very low. Among women with AVM, the risk of ICH during pregnancy and puerperium was more than three times the baseline rate during the non-pregnancy period in conditional Poisson regression (RR 3.27, 95% CI 1.67 to 6.43) (table 2). Among women without AVM, the risk of ICH increased almost 7-fold during pregnancy and puerperium (table 2). The risk of ICH increased mostly during pregnancy (RR 3.55, 95% CI 1.78 to 7.07) rather than puerperium (RR 2.36, 95% CI 0.87 to 6.39) among women with AVM. In contrast, the increased risk of ICH associated with pregnancy was largely attributable to ICH occurring in the puerperium period (RR 19.98, 95% CI 15.81 to 25.27) among women without AVM.

Discussion

In this study of over 4 million women, we identified 568 women with AVM, representing one of the largest cohorts examining this population in the literature. We estimated a rate of ICH resulting in hospitalization of close to 2000 per 100 000 person-years during the non-pregnancy period in women without AVM. This rate increased to 6885.1 per 100 000 person-years during pregnancy and to 4590.1 per 100 000 person-years during puerperium. In cohort-crossover analysis, we identified a relative risk for ICH of over 3-fold associated with pregnancy and delivery in women with AVM.

Our estimates are within the range of rates seen in other studies.6 12 21–26 We identified in our cohort a prevalence for AVM of 14 cases per 100 000 women, a rate that is within the range identified in prior epidemiological studies.21 22 Further, the incidence of ICH that we identified across pregnancy/delivery of 15.3 cases per 100 000 person-years is also within the range reported from prior studies.12 23 Finally, our reported incidence of 6.4 ICHs/100 person-years in women with AVM who underwent pregnancy and delivery in our study is within the range of rates seen in other studies. These prior analyses were all single-center studies that reported rates ranging from 3.1% to 8.1%.6 25 26 It is worth noting that this incidence is substantially higher than the annual hemorrhage incidence in the general AVM population (approximately 2.2% for unruptured AVM and 4.5% for ruptured AVM).27

Whether the risk of AVM-related ICH is greater during pregnancy and puerperium is an important question, as it can affect a woman’s plan for pregnancy and drive clinical decision making around preventive intervention, but it remains controversial. We found an increased risk of ICH for women with AVM during pregnancy and puerperium, consistent with some prior studies.25 26 Other studies, however, found no increased risk of ICH during pregnancy and puerperium among women with AVM.4 6 12 This discrepancy may be secondary to the fact that prior studies on this topic were generally limited to relatively small cohorts from single centers. Since the risk of ICH is already high before pregnancy among women with AVM, the excess risk due to pregnancy may become easily insignificant with the small sample size. However, more recent studies or studies including more contemporary cohorts found an increased risk of ICH during pregnancy and puerperium.28 Our study finds the risk of ICH increased more than three times during pregnancy and puerperium, and estimates were similar to recent estimates by Porras et al25 (annualized rate 1.3% vs 7.0% for non-pregnancy and pregnancy and puerperium).

One limitation of prior studies was their inability to account for angioarchitectural features that are known to influence rupture risk. A number of features have been identified, including the number of arterial feeders, AVM nidus diameter and location, venous drainage pattern, and the presence of feeding artery or nidal aneurysms.8 29 Given the relative rarity of AVMs and small sample sizes generated in these studies, explicitly adjusting for these features would not be possible. Using a cohort-crossover analysis, however, we are able to better account for these features, by using each woman and her unique AVM as her own control.

Cesarean delivery has been preferred for women with untreated ruptured AVM to avoid potential risks during labor and delivery.21 In our study, the percentage of women with cesarean delivery among women with AVM was substantially higher compared with women without AVM (61% and vs 35%). This finding may reflect that AVMs are being detected more often prior to rupture due to the increased usage of non-invasive brain imaging. We further explored by analyzing the subset of women who were diagnosed with brain AVM after the study period (ie, after pregnancy, delivery, and post-partum), assuming that they may have undergone pregnancy/delivery/post-partum care without the patient or care team knowing that they harbored a brain AVM. In this population, the percentage of women with cesarean delivery was lower (42%) and the rates of ICH between those who underwent caesarian vs vaginal delivery were 33 vs 24 per 1000 women during pregnancy and the postpartum, but the difference was not statistically significant (P=0.692).

Our study has limitations, many of which are inherent to administrative databases. The ICD-9 code 747.81 used to define AVM is not validated, however, it is the only available code to identify these cerebrovascular malformations, and has been used to indicate AVM in previous studies.12 In addition, we were unable to provide granular data on the type or location of AVM nor the severity or location of ICH. Furthermore, although we identified over 4 million women undergoing pregnancy and delivery, given the rarity of AVMs, our total sample size remains relatively modest, limiting subgroup analyses. Also, it is possible that ICH cases around the time of pregnancy were more intensively investigated for an underlying AVM, which would artificially inflate the association between pregnancy and AVM-related ICH. Finally, it is possible that we are more likely to detect symptomatic rather than asymptomatic AVMs. It is also possible that we could be undercounting AVM prevalence in certain ethnic and socioeconomic groups with less access to care. However, given that our overall AVM prevalence is very similar to those seen in prior epidemiological studies, we do not believe we are missing significant AVM populations.

Conclusion

In this observational study of over 4 million women, we identified an increased risk of ICH of over 3-fold during pregnancy in women with AVMs. These findings suggest the need for further study on the mechanisms of AVM development and rupture, and careful consideration of management strategies in these women.

Data availability statement

Data are available in a public, open access repository. Data are publicly available at the Healthcare Cost and Utilization Project (HCUP) website.

Ethics statements

Ethics approval

The study protocol was reviewed and approved by the Committee for the Protection of Human Subjects (CPHS) at The University of Texas Health Science Center at Houston.

References

Footnotes

  • SL and YK contributed equally.

  • Contributors SL, YK, ABB, FV, AM, SAS: conceived, designed, analyzed data, helped prepare drafts, and gave final approval of the manuscript. Remaining authors: interpreted data, revised manuscript, and approved the final manuscript.

  • Funding SAS reports funding from the American Academy of Neurology and the National Institutes of Health. SLB has effort supported through an NIH K23 grant

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.