Association Between Pregnancy and Symptomatic Hemorrhage in Women With Brainstem Cavernous Malformations and Bleeding in the Preceding Year: A Prospective Multicenter Study

Abstract

Background

The association between pregnancy and the risk of symptomatic hemorrhage (SH) in brainstem cavernous malformations remains uncertain. We aimed to quantify the risk of pregnancy‐associated SH in women with brainstem cavernous malformations.

Methods

We analyzed women with brainstem cavernous malformations in a prospective registry (2016–2022) experiencing pregnancy after enrollment. Follow‐up was divided into pregnancy‐associated segments (pregnancy plus ≤6 weeks postpartum or postabortion) and nonpregnant segments. Primary analysis used case‐crossover models; propensity score‐matched analyses assessed robustness.

Results

Among 63 women (median age, 30.0 years), 53 (84.1%) had bled before enrollment, including 45 (71.4%) with SH in the preceding year. Prospectively, they contributed 72 pregnancies and 371.7 patient‐years, observing 34 SHs. The pregnancy annual SH rate was 20.4% versus 7.2% nonpregnant (adjusted incidence rate ratio: 2.56, 95% CI, 1.11–5.93; P=0.028). Trimester‐specific and postpartum annual SH rates were 5.9%, 16.8%, 34.6%, and 38.3%; risk significantly increased only during the third trimester (P=0.004) and postpartum (P=0.007). This pregnancy effect was confined to segments with SH in the preceding year (annual SH rates 46.7% versus 9.4%; P=0.007), which was an independent risk factor. Propensity score‐matched analyses yielded consistent results.

Conclusions

Pregnancy is associated with an elevated SH risk in women with brainstem cavernous malformations, specifically among those with SH in the preceding year. Because of selection bias, generalizability to asymptomatic or incidental lesions is limited, warranting cautious extrapolation.

Registration

URL: www.chictr.org.cn; Unique Identifier: ChiCTR‐POC‐17011575.

Nonstandard Abbreviations and Acronyms

ASHR

annual symptomatic hemorrhage rate

CM

cavernous malformation

FND

focal neurological deficit

IRR

incidence rate ratio

SH

symptomatic hemorrhage

Clinical Perspective.

What Is New?

• This prospective multicenter study suggests that pregnancy is associated with an elevated risk of symptomatic hemorrhage (SH) in women with brainstem cavernous malformations, a hypothesis generating finding that appears specifically applicable to patients with SH in the preceding year, with the risk peaking during the third trimester and the early postpartum period (≤6 weeks).

What Are the Clinical Implications?

• Clinicians should provide tailored preconception counseling by advising patients with SH in the preceding year that achieving a period of clinical stability exceeding 1 year before conception may potentially reduce the likelihood of pregnancy associated rebleeding.

• For pregnant patients with brainstem cavernous malformations, particularly those with SH in the preceding year, multidisciplinary care featuring intensified clinical surveillance during the third trimester and the early postpartum period (≤6 weeks) is recommended to optimize maternal and neurological outcomes.

Brainstem cavernous malformations (CMs) account for approximately 7% to 30% of all intracranial CMs ¹ , ² , ³ , ⁴ , ⁵ and are associated with higher hemorrhage rates and more frequent focal neurological deficits (FNDs) than lesions in other locations. ⁵ , ⁶ , ⁷ Brainstem location ⁵ , ⁸ , ⁹ , ¹⁰ , ¹¹ and female sex ¹ , ¹² , ¹³ have been identified as potential risk factors for CM hemorrhage. Estrogen receptors have been demonstrated within CMs, ¹⁴ and exposure to exogenous hormones has been linked to an increased risk of hemorrhage. ¹⁵ Pregnancy‐related hemorrhage in brainstem CMs has received relatively little attention and has not been systematically studied. ¹⁶ , ¹⁷ A comprehensive review of the literature summarizing previously reported cases, ¹¹ , ¹⁴ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ identified approximately 40 pregnancies complicated by central nervous system CM hemorrhage, including 23 brainstem CMs (57.5%) (Table S1). Among these, 12 women (30%) experienced rebleeding, 10 of whom had brainstem lesions and 6 of whom rebled during pregnancy. Although these data are subject to substantial publication bias, they raise concern that pregnancy may precipitate hemorrhage in this population. Several larger cohort studies have also reported pregnancy‐associated hemorrhages, ¹¹ , ²⁹ , ³² yet collectively concluded that pregnancy does not increase the risk of cerebral cavernous malformation hemorrhage. Subsequent studies ¹¹ , ³³ , ³⁴ have discussed that these studies were constrained by factors such as incomplete imaging confirmation of some hemorrhages, ³² non‐self‐controlled designs, ²⁹ , ³² assumptions about the purely congenital nature of CMs, ²⁹ , ³² partially retrospective ascertainment, ²⁹ and relatively small sample sizes. ¹¹ These methodological considerations indicate that existing evidence, while informative, may not be sufficient to definitively exclude an increased hemorrhage risk during pregnancy, particularly for brainstem CMs.

To address this gap, we analyzed a prospective multicenter cohort using a self‐controlled design and multiple complementary statistical approaches to evaluate changes in hemorrhage risk during pregnancy. Our goal was to provide new evidence to inform clinical decision‐making and counseling and ultimately to improve outcomes for women of reproductive age with brainstem CMs.

METHODS

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Study Design and Patient Population

This was a prospective, multicenter cohort study. Patients with brainstem CMs were consecutively enrolled at 13 institutions between March 2016 and December 2022 under a registered longitudinal protocol (ChiCTR‐POC‐17011575). A case‐crossover framework was used to evaluate the effect of pregnancy on hemorrhage risk. Inclusion criteria were: radiologically confirmed brainstem CM, provision of informed consent, and at least one clinical and radiologic follow‐up. Exclusion criteria were: (1) prior surgery or radiotherapy targeting the index lesion; (2) pregnancy or breastfeeding at enrollment; (3) planned lesion‐directed treatment at baseline; (4) severe or life‐threatening symptoms or comorbidities precluding reliable neurologic assessment; (5) concurrent vascular malformations (eg, arteriovenous malformation) or neoplasms (eg, meningioma); (6) initiation after enrollment of contraceptive, statin, beta‐blocker, or anticoagulant therapy that could confound hemorrhage risk (use of these medications before enrollment was not an exclusion criterion to preserve a representative cohort, but patients starting such agents during follow‐up were excluded from the present analysis to reduce confounding by indication); (7) contraindications to magnetic resonance imaging; (8) incomplete data or lack of consent for follow‐up; and, for selection of nonpregnant controls in post‐propensity score‐matched sensitivity analyses; (9) male sex or non‐reproductive age (<15 or >45 years) (Figure 1).

Figure 1. Flowchart of cohort selection and construction of analytic samples from the multicenter prospective brainstem CM registry, including pregnancy and nonpregnancy groups and the 1:3 propensity score–matched cohort.

CCM indicates cerebral cavernous malformation; DVA, developmental venous anomaly; FND, focal neurological deficit; ICH, intracerebral hemorrhage; mRS, modified Rankin Scale; PSM, propensity score‐matched; and SH, symptomatic hemorrhage.

The study followed STROBE guidelines and was approved by institutional ethics committees at all participating centers (KYSB2016‐168 and KY2019‐034‐02) in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants (and from parents or caregivers for minors). All data sets used anonymized patient identifiers, study personnel received standardized training before initiation, and data underwent systematic cross‐checking for internal consistency, with discrepancies resolved by site investigators.

This was an exploratory, hypothesis‐generating analysis based on data from an ongoing prospective registry, with periodic evaluations during follow‐up. Because (1) no prior studies had specifically reported pregnancy cohorts with brainstem CMs, (2) reliable estimates of pregnancy‐related hemorrhage proportions or annual hemorrhage rates were unavailable, and (3) pregnancy planning and occurrence after diagnosis were not under investigator control, no formal a priori sample size calculation was performed. Instead, the sample size was determined by feasibility, and all eligible women meeting the inclusion and exclusion criteria during the study period were included.

Clinical and Radiological Baselines

Baseline clinical data included demographics, medical and family history, comorbidities, number of prior hemorrhages, presenting symptoms (classified as FNDs or nonspecific symptoms), and neurologic status assessed with the modified Rankin Scale. Pre‐enrollment obstetric history captured the number of pregnancies and, for each, the last menstrual period and termination date, the number and dates of hemorrhages during pregnancy, and pregnancy outcomes (abortion history and mode of delivery). Prior hemorrhages were confirmed by review of medical records and radiologic evidence, and hemorrhage at enrollment was validated by overt hemorrhage on magnetic resonance imaging obtained at study entry.

Radiologic variables included lesion location, maximal diameter on T1‐weighted imaging, crossing of the axial midpoint, lesion depth, presence of a developmental venous anomaly, perilesional edema, and Zabramski type. Lesion depth was defined by the thickness of parenchyma between the nidus and the pial/ependymal surface: lesions were classified as superficial if this thickness was <1 mm (exophytic or abutting the surface) and deep if ≥1 mm. Detailed procedures for baseline data collection have been described previously.

Follow‐Up

At enrollment, patients were informed that: (1) they should notify the study team of any planned or unplanned pregnancy after enrollment; (2) pregnancy‐related clinical information would be abstracted from their medical records; (3) post‐enrollment medication use and any new or worsening symptoms would be documented; and (4) all imaging‐confirmed hemorrhages and modified Rankin Scale scores at each follow‐up would be systematically recorded. The primary outcome was specifically defined as symptomatic hemorrhage (SH), characterized by an acute or subacute onset of new or worsening FNDs referable to the brainstem CM, impaired consciousness, or severe headache, accompanied by radiologic evidence of a new overt hemorrhage. Accordingly, this definition does not include patients with incidental or asymptomatic new hemorrhages.

Follow‐up assessments were scheduled at 3, 6, and 12 months after enrollment and annually thereafter, via clinic visits, telephone or video calls, mailed questionnaires, or contact with family doctors. To ensure continuity, patients and relatives provided at least three telephone numbers plus home address, email, and social media accounts, which were reviewed and updated at each visit. Magnetic resonance imaging was recommended at each follow‐up whenever feasible and was mandatory at the onset of new neurologic symptoms. Follow‐up was censored at initiation of lesion‐directed treatment (e.g., radiotherapy or surgical resection), death, attainment of age >45 years, or the last available evaluation, whichever occurred first.

Definition of Pregnancy‐Related Exposure Time

Pregnancy and the puerperium were jointly analyzed as the exposure period. Follow‐up was measured from study enrollment, rather than from age 15 years or first diagnosis (Figure 2). For each woman, pregnancy‐related exposure time after enrollment was defined as the documented duration of each pregnancy (from medical records) plus the subsequent 6 weeks postpartum or postabortion. In women who underwent surgery during pregnancy or within 6 weeks thereafter, exposure time was truncated at the date of surgery, rather than assuming a fixed gestational duration as in some previous reports. ¹¹ , ²⁹ , ³² Non‐exposed time was defined as total post‐enrollment follow‐up minus the pregnancy‐related exposure time.

Figure 2. Schematic overview of follow‐up structure and ASHR analyses.

The top panel illustrates alternative definitions of follow‐up onset in previous studies vs our cohort (starting at enrollment), censoring events, and the construction of pregnancy‐related (pink; pregnancy plus ≤6‐wk postpartum/postabortion) and nonpregnant (green) segments used to calculate ASHRs (%, per patient‐years). Gray areas indicate retrospective period before enrollment. The blue areas represent the entire prospective follow‐up period after enrollment in our cohort. The lower panels show primary and trimester‐specific comparisons of ASHRs between pregnancy and nonpregnant periods, and post‐propensity score‐matched comparisons between pregnancy and nonpregnancy cohorts. Bars represent ASHRs, and P values are derived from Poisson or generalized estimating equation Poisson models. ASHR_pregnancy and ASHR_nonpregnant were calculated as the number of SHs divided by the corresponding person‐time, expressed as % per patient‐year: ASHR_pregnancy = [number of SHs occurring during pregnancy or within ≤6 wk postpartum] ÷ [pregnancy period (patient‐years)] × 100%; ASHR_nonpregnant = [number of SHs while not pregnant] ÷ [nonpregnant period (patient‐years)] × 100%. ASHR indicates annual symptomatic hemorrhage rate; GEE, generalized estimating equation; and SH, symptomatic hemorrhage.

Statistical Analysis

Descriptive statistics were used to summarize baseline characteristics. Incidence rates of SH were calculated as the number of SHs divided by person‐years at risk and expressed per patient‐year with 95% CIs derived from the Poisson distribution.

For the main analyses, data were restructured into a person‐period format. For each woman, post‐enrollment follow‐up was divided into consecutive segments according to pregnancy status (pregnant versus nonpregnant), so that each patient contributed both pregnant and nonpregnant segments and thus served as her own control (case‐crossover framework). For each segment, we calculated person‐time and 3 time‐varying covariates: age at segment start, cumulative number of prior hemorrhages, and SH within the previous year (yes/no).

The primary analysis estimated incidence rate ratios (IRRs) for SH associated with pregnancy using generalized estimating equation Poisson regression with a log link, robust (sandwich) standard errors, and an exchangeable working correlation structure to account for within‐patient clustering. The logarithm of segment‐specific person‐years was included as an offset, and nonpregnant segments served as the reference category. A restricted cubic spline (RCS) model was constructed to visualize the continuous risk trajectory. All models were adjusted for the 3 time‐varying covariates, and their associations with pregnancy‐related SH were also examined. Sensitivity analyses, including alternative model specifications and propensity score‐matched cohorts, were conducted to assess the robustness of the primary findings (details in Data S1). Additionally, pre‐specified subgroup analyses were performed. Given the limited sample size, these subgroup analyses were considered exploratory and were not adjusted for multiple testing; thus, their results should be interpreted as hypothesis‐generating. Results are reported as IRRs with 95% CIs; 2‐sided P values <0.05 were considered statistically significant. All analyses were performed using Stata version 15.1 (StataCorp, College Station, TX, USA) and IBM SPSS Statistics version 26.0 (IBM Corp., Armonk, NY, USA).

RESULTS

Clinical Baseline Characteristics

A total of 63 women who experienced pregnancy after enrollment were included (Figure 1; Figure S1); 39 (61.9%) had no prior pregnancies and 24 (38.1%) had a pregnancy history (Table S2). The median age at enrollment was 30.0 years (interquartile range [IQR] 28.0–33.0; range 21.0–43.0). Four patients (6.3%) had hypertension and 3 (4.8%) had a family history of cerebral cavernous malformation. Fifty‐three (84.1%) had bled before enrollment, including 45 with an SH within 1 year. Over 1910 pre‐enrollment patient‐years, 70 hemorrhages were recorded, corresponding to a prior annual hemorrhage rate of 3.7% per patient‐year (95% CI, 2.9–4.6). At enrollment, 40 patients (63.5%) presented with hemorrhage, 36 (57.1%) had FND, and the median modified Rankin Scale score was 1.0 (IQR 0–1; range 0–4). The distribution of prior hemorrhage and modified Rankin Scale scores is shown in Table 1.

Table 1.

Baseline Characteristics of Women With Brainstem CMs Experiencing Post‐Enrollment Pregnancy and the Propensity Score‐Matched Control Cohort

Variable (% or median)	Pregnant cohort (n=63)	Matched non‐pregnant cohort (n=125)†
Median age (IQR), y	30.0 (28.0–33.0)	32.0 (26.0–37.0)
Hypertension	4 (6.3)	8 (6.4)
SH within 1 y before enrollment	45 (71.4)	107 (85.6)
Prior hemorrhages
0	10 (15.9)	7 (5.6)
1 hemorrhage	40 (63.5)	84 (67.2)
2 hemorrhages	10 (15.9)	26 (20.8)
3 hemorrhages	2 (3.2)	6 (4.8)
4 hemorrhages	1 (1.6)	2 (1.6)
Median no. of prior hemorrhages (IQR)	1.0 (1.0–1.0)	1.0 (1.0–2.0)
mRS on admission
0	19 (30.2)	16 (12.8)
1	33 (52.4)	81 (64.8)
2	4 (6.3)	13 (10.4)
3	4 (6.3)	8 (6.4)
4	3 (4.8)	7 (5.6)
Hemorrhage on admission	40 (63.5)	86 (68.8)
FND on admission*	36 (57.1)	91 (72.8)
Familial CCMs	3 (4.8)	3 (2.4)
Developmental venous anomaly	16 (25.4)	38 (30.4)
Location
Midbrain	16 (25.4)	39 (31.2)
Pontine	36 (57.1)	67 (53.6)
Medullary	11 (17.5)	19 (15.2)
Crossing axial midpoint	10 (15.9)	23 (18.4)
Median lesion size (IQR), cm	1.0 (0.8–1.5)	1.2 (0.9–1.6)
Depth
Superficial	25 (39.7)	58 (46.4)
Deep	38 (60.3)	67 (53.6)
Zabramski classification
Type I	18 (28.6)	32 (25.6)
Type II	21 (33.3)	57 (45.6)
Type III	24 (38.1)	36 (28.8)
Total no. of prior hemorrhage	70	162
Patient‐years prior to enrollment	1910	3930
Prior AHR, % (95% CI)	3.7 (2.9–4.6)	4.1 (3.5–4.8)

AHR indicates annual hemorrhage rate; CCM, cerebral cavernous malformation; FND, focal neurological deficit; IQR, interquartile range; mRS, modified Rankin scale; and SH, symptomatic hemorrhage.

^*FNDs were defined as cranial nerve palsies, long tract deficits, or extrapyramidal symptoms directly associated with the anatomical location of the CM. Other symptoms, such as headache, dizziness, nausea, or vomiting, were categorized as nonspecific.

^† See Tables S5 and S6 for further details on the propensity score matching.

On imaging, lesions were located in the pons in 36 patients (57.1%), the midbrain in 16 (25.4%), and the medulla in 11 (17.5%). Developmental venous anomaly was present in 16 (25.4%), and 10 (15.9%) had lesions crossing the axial midpoint. The median maximal lesion diameter was 1.0 cm (IQR 0.8–1.5; range 0.2–2.9). Zabramski types I, II, and III were observed in 18 (28.6%), 21 (33.3%), and 24 (38.1%) patients, respectively (Table 1).

Prospective SH After Enrollment in Pregnant Versus Nonpregnant Periods

After enrollment, the median duration of prospective follow‐up was 6.4 years (IQR 4.4–8.2; range 0.5–9.1). During this period, 43 women (68.3%) had no SH, 7 (11.1%) had SH only during pregnancy, 10 (15.9%) only during nonpregnant periods, and 3 (4.8%) in both periods, for a total of 34 prospective SHs (Table 2).

Table 2.

Prospective SH of 63 Brainstem CMs for Case‐Crossover Analysis

	Pregnancy period	Non‐pregnancy period
No. of patients without prospective SH	53 (84.1)	50 (79.4)
No. of patients with prospective SH§	10 (15.9)†	13 (20.6)‡
1 SH	9 (14.3)	7 (11.1)
2 SHs	1 (1.6)	4 (6.3)
3 SHs	0	0
4 SHs	0	2 (3.2)
5 SHs	0	0
No. of observation segments	72	131
No. of segments without prospective SH	62 (86.1)	118 (90.1)
No. of segments with prospective SH	10 (13.9)	13 (9.9)
No. of segments with 1 SH	9 (12.5)	7 (5.3)
No. of segments with 2 SHs	1 (1.4)	4 (3.1)
No. of segments with 3 SHs	0	0
No. of segments with 4 SHs	0	2 (1.5)
No. of segments with 5 SHs	0	0
Total no. of prospective SH	11	23
Cumulative observation duration, patient‐years	54.1	317.6
Prospective ASHR, % (95% CI)	20.4 (10.2–36.4)	7.2 (4.6–10.9)
Primary analysis of pregnancy‐associated SH risk: IRR (95% CI), P
Unadjusted GEE‐Poisson (exchangeable), robust SE	2.80 (1.32–5.93), P=0.007*	Reference
Adjusted GEE‐Poisson (exchangeable), robust SE||	2.56 (1.11–5.93), P=0.028* , ||	Reference
Sensitivity analyses of pregnancy‐associated SH risk: IRR (95% CI), P
Unadjusted GEE‐Negative binomial (exchangeable), robust SE	2.67 (1.17–6.10), P=0.020*	Reference
Adjusted GEE‐Negative binomial (exchangeable), robust SE||	2.64 (1.03–6.81), P=0.044* , ||	Reference

All models use a log link and include log (person‐years) as an offset. Robust (sandwich) standard errors clustered at the patient level unless otherwise specified. GEE estimates population‐averaged effects under an exchangeable working correlation. ASHR indicates annual symptomatic hemorrhage rate; GEE, generalized estimating equation; IRR, incidence rate ratio; and SH, symptomatic hemorrhage.

^* P<0.05.

^† 7 patients had SHs only during pregnancy.

^‡ 10 patients had SHs only during non‐pregnancy.

^§ 3 patients had SHs during both pregnancy and non‐pregnancy.

^|| Adjusted for age, number of prior hemorrhages, and SH in the preceding year, anchored at the start of each observation segment (for the baseline segment, this corresponds to the year prior to enrollment).

During follow‐up, 72 pregnancies occurred: 55 women had 1 pregnancy, 7 had 2, and 1 had 3. Of these pregnancies, 8 (11.4%) ended in vaginal delivery, 53 (75.7%) in cesarean section, and 9 (12.9%) in abortion; 61 resulted in live births, and 2 were ongoing at last follow‐up. Overall, 62 pregnancies (86.1%) were not complicated by SH, whereas 10 (13.9%) were associated with 11 SHs (2 SHs in 1 pregnancy and 1 SH in each of the remaining 9). Of the 11 SHs, 1 (9.1%) occurred in the first trimester, 3 (27.3%) in the second, 4 (36.4%) in the third, and 3 (27.3%) in the postpartum period. The median gestational/postpartum age at SH was 32.4 weeks (IQR 25.0–37.6; range, 5.1–41.3) (Table S3). Pregnancy‐related follow‐up totaled 54.1 patient‐years, with an annual SH rate (ASHR) of 20.4% per patient‐year (95% CI, 10.2–36.4). During 317.6 nonpregnant patient‐years, 13 women had 23 SHs, for an ASHR of 7.2% per patient‐year (95% CI, 4.6–10.9).

Follow‐up was censored at the last clinical visit (n=56), attainment of age >45 years (n=2), or brainstem CM resection (n=5); no deaths occurred during follow‐up. Two women underwent surgery for pregnancy‐related SH (Supplementary pregnancy timeline and material), whereas the remaining 8 pregnancy‐related SHs were managed conservatively. The other 3 resections were for nonpregnant SH (n=2) or progressive FND (n=1). Neurologic outcomes are shown in Table S3.

Case‐Crossover Analysis

In the primary analysis, an unadjusted generalized estimating equation Poisson model with an exchangeable correlation structure and robust standard errors showed a significantly higher SH risk during pregnancy than during nonpregnant periods (ASHR 20.4% versus 7.2% per patient‐year; IRR 2.80, 95% CI, 1.32–5.93; P=0.007). This association remained significant in the multivariable generalized estimating equation Poisson model (IRR 2.56, 95% CI, 1.11–5.93; P=0.028) (Table 2). Five additional sensitivity models yielded consistent results, confirming an increased risk of SH during pregnancy both before and after covariate adjustment (Table 2).

Trimester‐Specific and Subgroup Analysis

The trimester‐specific and postpartum ASHRs were 5.9%, 16.8%, 34.6%, and 38.3% per patient‐year, respectively. Although these point estimates varied widely, no pairwise differences among the four periods were statistically significant (Table S4). When each stage was compared with nonpregnant periods, SH risk during the first and second trimesters did not differ significantly from that during nonpregnancy, whereas risks in the third trimester (P=0.004) and postpartum period (P=0.007) were significantly higher (Table 3).

Table 3.

ASHR and IRRs in Nonpregnant, Trimester‐Specific, and Postpartum Segments in 63 Eligible Cases

Period	No. of SHs	Period, patient‐years	Prospective ASHR, % (95% CI)	Poisson regression
				IRR (95% CI)	P
Non‐pregnant	23	317.6	7.2 (4.6–10.9)	Reference
First trimester	1	16.9	5.9 (0.2–33.0)	0.82 (0.11–6.06)	0.844
Second trimester	3	17.8	16.8 (3.5–49.2)	2.32 (0.70–7.74)	0.170
Third trimester	4	11.6	34.6 (9.4–88.5)	4.77 (1.65–13.80)	0.004*
Postpartum	3	7.8	38.3 (7.9–112.0)	5.29 (1.59–17.62)	0.007*

ASHR indicates annual symptomatic hemorrhage rate; IRR, incidence rate ratio; and SH, symptomatic hemorrhage.

^* P<0.05.

In subgroup analyses of 171 segments with prior hemorrhage, pregnancy was associated with a higher SH risk (generalized estimating equation Poisson: IRR 2.27, 95% CI, 1.03–4.96, P=0.041). When segments were further stratified by SH within the preceding year, the pregnancy effect was confined to the 73 segments with SH within the preceding year: the ASHR during pregnancy was 46.7% versus 9.4% per patient‐year during non‐pregnant periods (generalized estimating equation Poisson, P=0.007). Among the 130 segments without SH in the prior year, SH risk did not differ significantly between pregnancy and nonpregnancy (12.2% versus 6.2% per patient‐year; P=0.254) (Table 4; Figure S2). Consistent with these findings, in a multivariable generalized estimating equation Poisson model restricted to the 72 pregnancy segments, SH within the previous year remained independently associated with pregnancy‐related SH (IRR 4.57, 95% CI, 1.19–17.57; P=0.027; Table 5). To visualize the continuous dynamic risk trajectory, a RCS model was constructed (Figure S3).

Table 4.

Subgroup Analysis of Pregnancy‐Associated SH Risk in 63 Patients With 203 Observation Segments

Subgroup	No. of pts	No. of segs	Pregnancy period		Non‐pregnancy period		GEE‐Poisson (exchangeable), robust SE, IRR (95% CI)	P
			No. of bleeding/period, pt‐y	Prospective ASHR, % (95% CI)	No. of bleeding/period, pt‐y	Prospective ASHR, % (95% CI)
Segments with prior bleeding
Yes	55	171	9/44.8	20.1 (9.2–38.1)	23/261.0	8.8 (5.6–13.2)	2.27 (1.03–4.96)	0.041*
No	10	32	2/9.2	21.7 (2.6–78.5)	0/56.7	0.0 (0.0–6.5)	NA†	NA†
Segments with SH in prior year
Yes	45	73	6/12.8	46.7 (17.1–101.6)	10/106.4	9.4 (4.5–17.3)	4.92 (1.55–15.63)	0.007*
No	55	130	5/41.1	12.2 (4.0–28.4)	13/211.3	6.2 (3.3–10.5)	1.99 (0.61–6.46)	0.254

Non‐pregnancy period after enrollment was cited as the reference. All models use a log link and include log (person‐years) as an offset. Robust (sandwich) standard errors clustered at the patient level unless otherwise specified. GEE estimates population‐averaged effects under an exchangeable working correlation. ASHR indicates annual symptomatic hemorrhage rate; IRR, incidence rate ratio; NA, not available; pts, patients; segs, segments; and SH, symptomatic hemorrhage.

^* P<0.05.

^† No SH occurred during non‐pregnancy.

Table 5.

Risk Factors for SH Within 72 Pregnancy‐Related Segments in 63 Patients: Univariable and Multivariable (Adjusted) GEE‐Poisson Models

Time‐varying covariates	No. of SH/pregnancy period, patient‐years	Prospective ASHR, % (95% CI)	Univariable (unadjusted) GEE‐Poisson		Multivariable (adjusted) GEE‐Poisson†
			IRR (95% CI)	P	IRR (95% CI)	P
Increased age	…	…	1.11 (1.02–1.21)	0.011*	1.10 (1.03–1.17)	0.006*
Prior bleeding
Yes	9/44.8	20.1 (9.2–38.1)	0.91 (0.21–3.98)	0.905	0.48 (0.09–2.71)	0.409
No	2/9.2	21.7 (2.6–78.5)	Reference		Reference
SH in prior year
Yes	6/12.8	46.7 (17.1–101.6)	3.81 (1.18–12.33)	0.026*	4.57 (1.19–17.57)	0.027*
No	5/41.1	12.2 (4.0–28.4)	Reference		Reference

ASHR indicates annual symptomatic hemorrhage rate; GEE‐Poisson, generalized estimating equation Poisson regression with an exchangeable correlation structure and robust standard errors; and SH, symptomatic hemorrhage.

^* P<0.05.

^† Multivariable models included age, prior bleeding, and SH in the prior year, and estimates for each covariate are mutually adjusted for the others.

Post‐Propensity Score‐Matched Analysis

After Propensity score‐matched analysis, the 2 cohorts had similar baseline characteristics (Table S5), and their nonpregnant SH risks were comparable (ASHR 7.9% versus 8.8% per patient‐year; P=0.787). Across all statistical approaches comparing pregnancy with the corresponding nonpregnant periods in the matched cohorts, pregnancy remained associated with a significantly higher SH risk (Table S6). Neurologic outcomes and their changes over time did not differ significantly between the matched cohorts (Table S7).

DISCUSSION

In this particularly high‐risk location, data on the impact of pregnancy on hemorrhage risk in brainstem CMs are scarce, and it has been unclear whether the coexistence of brainstem location and pregnancy exerts an additive effect on risk. In our cohort, the crude proportion of women who symptomatically hemorrhaged during pregnancy (15.9%) was slightly lower than during non‐pregnant periods (20.6%), which could misleadingly suggest that pregnancy has little effect on SH risk. However, these proportions do not account for the much shorter duration of pregnancy‐related exposure. Using a prospective multicenter cohort and a self‐controlled, person‐time‐based analytic design, our primary, subgroup, and sensitivity analyses consistently showed higher SH incidence during pregnancy than during nonpregnant periods. This is likely related to the high baseline vulnerability of our cohort (84.1% with prior hemorrhage, 71.4% with SH within the preceding year, and 63.5% presenting with hemorrhage at enrollment) and to the short pregnancy windows, in which a modest number of SHs can yield substantially higher incidence rates even though absolute numbers and crude proportions remain low. We also identified 2 clinically relevant patterns: SH risk varied across gestational stages, with higher risk in late pregnancy and postpartum, and among women with prior hemorrhage, pregnancy did not significantly increase risk when more than 1 year had elapsed since the last SH.

Literature Review and Summary

Based on 31 reported pregnant women with brainstem CMs with sufficient clinical data (Table S1), the summarized characteristics were as follows. The median age was 31.0 years (IQR 27.0–34.0; range 15.0–38.0). Twenty‐three women (74.2%) had hemorrhage during pregnancy, occurring in 4, 3, 6, and 1 pregnancy in the first, second, and third trimesters and postpartum, respectively; the timing was not specified in 9 women. Approximately 23 women (74.2%) had FNDs. The median gestational age at delivery was 38.0 weeks (IQR 36.0–40.0; range 34.0–41.0), although gestational age was not reported in 18 women and was described only as term birth in 3. Delivery mode was cesarean section in 10 women and vaginal in 3 and was not reported in the remaining 18. Nine women (29.0%) were explicitly reported to have no rebleeding, whereas 10 (32.3%) had subsequent hemorrhage: 5 rebled within the same pregnancy (3 before delivery and 2 postpartum), 2 at 2 and 3 months postpartum, and 3 at unspecified times; outcomes were not clearly documented in the other 12 women. Eight women (25.8%) underwent surgery (4 during pregnancy before delivery, 2 postpartum, and 2 at 3 months and 6 years postpartum), 16 (51.6%) were managed conservatively, and management was not reported in 7. These case reports are subject to substantial publication bias and are not representative of the broader patient population; however, given the rarity of brainstem CMs and the complexity of managing hemorrhage during pregnancy, the accumulated experience from these individual reports remains informative and may help raise awareness and guide clinical decision‐making in this challenging setting.

Comparison With Previous Studies

Our findings differ from prior reports that mainly included supratentorial or mixed‐location cerebral cavernous malformation cohorts, likely reflecting differences in study design, case mix, and outcome definitions rather than simple inconsistency. For example, in the cohort by Joseph et al, ¹¹ 42.9% (9/21) of women had no prior hemorrhage, suggesting a lower baseline‐risk population, and several earlier studies used partially retrospective follow‐up from age 15 years or from diagnosis, which is more susceptible to recall bias. In contrast, our study initiated prospective follow‐up at enrollment and collected detailed, individual obstetric timelines. Outcome definitions and analytic approaches also differed. We required both clinical deterioration and magnetic resonance imaging evidence to define SH, whereas some earlier series relied primarily on symptoms, which may capture neurological events not necessarily attributable to new hemorrhage. ²⁹ , ³² Moreover, our primary analysis used a self‐controlled, segment‐based design with adjustment for time‐varying covariates, and secondary propensity score‐matched analyses compared pregnant women with baseline‐similar nonpregnant controls. Under this framework, pregnancy did not appear to increase overall lifetime SH risk but was associated with higher SH risk within the pregnancy‐related period.

Implication for Clinical Practices

In a subgroup analysis restricted to the 72 pregnancy segments, a simple history of prior hemorrhage was not associated with pregnancy‐related SH (P=0.409), whereas SH within the preceding year (P=0.027) and older age (P=0.006) were both significantly associated with higher SH risk during pregnancy (Table 5). From a clinical management perspective, our findings provide critical guidance for preconception counseling. Specifically, women with brainstem CMs and no prior hemorrhage may generally proceed with pregnancy under standard care. In contrast, for women with a recent SH, advising them to defer conception until achieving more than 1 year of clinical and radiological stability may be a prudent and actionable strategy to mitigate the risk of recurrent SH during pregnancy. This aligns with the proposal by Bektas et al ³⁴ that conceiving within the first year after hemorrhage may increase rebleeding risk and complicate perinatal management. Prior work further suggests that waiting approximately 2.5 years after lesion stabilization may be even safer, ⁵ , ³⁵ provided this does not result in advanced maternal age, which in our cohort appeared to increase pregnancy‐related SH risk. A plausible biological explanation is that estrogen‐mediated endothelial repair and restoration of tight junctions within the blood–brain barrier may diminish with age. ³⁶ , ³⁷

From a practical standpoint, during the waiting period after an SH, careful blood pressure control and lifestyle modification are advisable, including avoidance of excessive fatigue, night shifts, strenuous physical labor, alcohol, smoking, vigorous exercise, and marked emotional fluctuation. If rebleeding or neurologic worsening occurs, surgical intervention should be considered first, with pregnancy deferred until clinical and radiological stability is re‐established to minimize SH risk during pregnancy and protect both maternal and fetal health. For women who become pregnant after a diagnosis of brainstem CM, our data indicate that SH risk is not significantly increased in the first and second trimesters; nevertheless, close monitoring of blood pressure and neurological status remains essential, and any new or worsening symptoms should prompt magnetic resonance imaging to assess for interval changes in the CM.

In our cohort, SHs occurred predominantly in the second and third trimesters and postpartum, and none occurred during labor and delivery. Consequently, our study cannot determine whether vaginal delivery or cesarean section modifies the risk of SH. The optimal mode of delivery remains unclear, requiring further investigation in larger prospective registries, and should currently be individualized by a multidisciplinary team. ¹¹ , ²⁹ , ³² , ³³ , ³⁴ Although pregnancy‐related SH increases management complexity, most cases can be treated conservatively; in our series, only 2 women required resection and both improved neurologically. After propensity score‐matched (Table S7), neurologic outcomes, changes in functional status, and the proportion undergoing surgery did not differ significantly between matched pregnancy and nonpregnancy cohorts. Nonetheless, clinicians must remain vigilant for rare but potentially catastrophic cases, in which a single severe SH can have major consequences for both mother and fetus.

From an obstetric perspective, particular attention is warranted for women who symptomatically hemorrhage in the second trimester. SH appears relatively uncommon in early pregnancy. When it occurs in late pregnancy or postpartum, emergency surgery can usually be preceded by delivery, often via cesarean section, to secure fetal safety before CM resection in accordance with standard practice. In contrast, SH in the second trimester creates a difficult dilemma: clinicians and families must balance the need for continued fetal maturation against maternal safety, in the face of uncertainty about the timing and likelihood of rebleeding. This watchful waiting is highly stressful and demands considerable patience from both physicians and families. As a treatment strategy, conservative management is generally preferred initially to prolong gestation, promote fetal growth, and improve the survival prospects of a potentially preterm infant. However, a minority of women (such as case 48 in our cohort) may experience recurrent SH with clinical deterioration at a gestational age when the fetus is still too immature for safe delivery. In such situations, close multidisciplinary collaboration is required to proceed with lesion resection while allowing the pregnancy to continue, followed by cesarean delivery once an acceptable gestational age is reached. In daily practice, every effort should be made to intervene proactively—through blood pressure control and lifestyle and risk‐factor modification—to reduce the likelihood of these extreme scenarios.

Limitations

This study was conducted in tertiary referral centers, introducing inherent selection bias. Beyond the high proportion of patients with prior bleeding, this selection bias may result in an abnormally high prevalence of other potential, non‐determinable baseline characteristics associated with prior hemorrhage, which could invisibly exacerbate the risk of hemorrhage during pregnancy. As established in previous literature, ⁵ prior bleeding is a strong independent predictor of future hemorrhage risk, meaning our observed pregnancy effect is heavily influenced by this elevated baseline vulnerability. Consequently, our findings are not broadly generalizable. We explicitly caution against extrapolating our results to patients with incidentally discovered or asymptomatic lesions. We enrolled relatively few women without prior hemorrhage, scenarios for which current guidelines already provide widely accepted recommendations, ³³ , ³⁴ and their broader inclusion might have yielded considerably different associations. Rather, our data are most applicable to women with symptomatic hemorrhage (SH) in the preceding year who lack a clear surgical indication or choose conservative management while planning pregnancy. Furthermore, because pregnancy cannot be randomized, evidence must rely on observational cohorts. Due to the natural tendency of these high‐risk patients to avoid pregnancy, the absolute number of bleeding events during pregnancy remained low. Although our subgroup analyses were strictly pre‐specified rather than being post hoc multiple testing, their statistical power is inherently limited. Therefore, these specific findings should be considered exploratory and hypothesis‐generating, requiring careful interpretation because of wide CIs, and warranting confirmation in larger multicenter cohorts.

CONCLUSIONS

In this prospective, self‐controlled study of women with brainstem CMs, pregnancy was associated with an increased risk of SH, particularly during the third trimester and the early postpartum period (≤6 weeks). SH in the preceding year emerged as an independent predictor of pregnancy‐associated SH, supporting the recommendation that women with prior bleeding defer conception until at least 1 year of clinical and radiologic stability has been achieved. However, because our cohort predominantly consisted of patients with prior bleeding, this study carries inherent selection bias. Consequently, the generalizability of our findings is limited, and these results should be extrapolated cautiously, particularly on patients with incidentally discovered or asymptomatic lesions. Given these limitations, the sample size, and the referral‐center setting, these findings should be regarded as hypothesis‐generating and require confirmation in larger, preferably multicenter, studies.

Sources of Funding

This work was supported by (1) National Natural Science Foundation of China (No. 82301470); (2) National High Level Hospital Clinical Research Funding; (3) Elite Medical Professionals Initiative of China‐Japan Friendship Hospital (No. ZRJY2025‐GG10); (4) Central Government Guided Local Science and Technology Development Special Project of Ningxia Hui Autonomous Region (No. 2023FRD05053). The funder had the following involvement with the study: study design, data collection, analysis, interpretation, writing of the manuscript, and decision to submit it for publication.

Disclosures

None.

Supporting information

Data S1

Tables S1–S7

Figures S1–S3

Reporting checklist for case‐control study