Acute Ischemic Stroke in Patients With Pulmonary Arteriovenous Malformations: Paradoxical Embolism or Epiphenomenon?

Abstract

Background

Pulmonary arteriovenous malformations (PAVM) are pathological connections between arterial and venous circulations in the lung. The anomalous channel provides a conduit for emboli of venous origin to migrate paradoxically. With access to systemic circulation, thromboembolic processes such as deep vein thrombosis may increase the risk of cerebrovascular accidents such as acute ischemic strokes (AIS). This retrospective cross‐sectional study aims to characterize the contribution of PAVMs to the development of AIS in the setting of thromboembolic processes.

Methods

The 2010 to 2019 National Inpatient Sample was queried for patients with PAVM and AIS using International Classification of Diseases, Ninth/Tenth Revision (ICD‐9; ICD‐10) codes. Baseline demographics and outcomes of interest, including complications and hospitalization metrics, were retrospectively analyzed.

Results

Of 7 465 187 patients treated for an AIS, 1 864 (0.02%) were found to have a concomitant PAVM. Multivariate regression found patients with AIS‐PAVM were more likely to be aged <65 years and less likely to have traditional stroke risk factors such as cardiovascular disease, obesity, or smoking. Furthermore, in propensity‐matched analysis, patients with AIS‐PAVM were more likely to experience thrombosis‐related pathologies and vascular complications. Patients with AIS‐PAVM were also more likely to receive a mechanical thrombectomy or intravenous thrombolysis, yet the mortality between those with PAVMs and those without did not differ significantly. Still, patients with AIS‐PAVM incurred significantly prolonged stays in the hospital and increased total charges.

Conclusion

Given the relative rarity of PAVMs, a high index of suspicion, especially in the setting of deep vein thrombosis, is necessary to identify this condition in a timely manner when evaluating cryptogenic AIS. Abnormal right‐to‐left shunting through pulmonary vasculature may serve as a conduit for the translation of a deep vein thrombosis paradoxically into a distal occlusion of cerebral arteries. Our findings substantiate that AIS may follow a vastly different disease process in patients with PAVM.

Nonstandard Abbreviations and Acronyms

AIS

acute ischemic stroke

DVT

deep vein thrombosis

HHT

hereditary hemorrhagic telangiectasia

PAVM

pulmonary arteriovenous malformation

Clinical Perspective

• Pulmonary arteriovenous malformations may serve as a conduit for deep vein thrombosis to cerebral circulation, thereby causing an acute ischemic stroke.

• Currently, the American Heart Association/American Stroke Association guidelines do not elaborate much on pulmonary arteriovenous malformations as a source of embolic strokes of undetermined significance.

• Pulmonary arteriovenous malformations should be screened for and incorporated into the workup for cryptogenic stroke as they may facilitate the pathological connection between vascular conditions such as deep vein thrombosis and acute ischemic strokes.

Pulmonary arteriovenous malformations (PAVMs) are direct, pathological connections between a pulmonary artery and vein without an intermediary capillary bed. ¹ PAVMs are typically congenital lesions often associated with the autosomal dominant condition hereditary hemorrhagic telangiectasia (HHT), although they can also be acquired from various medical, infectious, and traumatic causes. ² , ³

Although rare, PAVMs should be included in the differential diagnosis for workup of common pulmonary pathologies. Of concern, PAVMs can function as a right‐to‐left intrapulmonary extracardiac shunt, permitting infectious diseases and peripheral venous emboli access to systemic arterial vasculature. Deep vein thrombosis (DVT) and pulmonary embolism may be particularly problematic conditions in patients with a known PAVM. They may transform into paradoxical emboli that lodge in and occlude cerebral vessels, resulting in an acute ischemic stroke (AIS), which carries significant risk of morbidity and mortality. ² , ⁴ Furthermore, PAVMs and AISs may demonstrate a bidirectionally unfavorable relationship; compensatory increases in flow to preserve cerebral function attributable to hypoxic changes in the brain may further stress and compromise the already weak integrity of PAVMs. ⁴ , ⁵

Approximately 25% to 40% of ischemic strokes are considered cryptogenic. ⁶ Therefore, it is critical to investigate whether a thromboembolic process is involved – along with its origin – to quickly apply the most appropriate intervention if available. To delineate nonlacunar ischemic strokes attributable to embolism of unknown origin, the subtype embolic stroke of undetermined significance has emerged in recent literature to focus and improve diagnosis and treatment; however, its utility remains unclear with regard to noncardiac pathology. ⁷ Considering the conduits through which paradoxical emboli could cause AIS, currently only diagnosis and treatment of patent foramen ovale has been extensively investigated relative to less common pathologies such as PAVMs. ⁸ , ⁹ The American Heart Association/American Stroke Association guidelines recommend exploring cardiac etiologies after cryptogenic stroke via echocardiography with or without contrast after embolic stroke of undetermined significance, yet noncardiac causes, such as PAVM, are not discussed. ⁷

In this descriptive study, we aim to characterize risk factors and complications of DVT and AIS in the context of PAVM using data from the National Inpatient Sample database. We hypothesize that patients with DVT are more likely to experience AIS in the context of PAVM.

Methods

Data Source and Patient Selection

While the database must be obtained from the Healthcare Cost and Utilization Project, all supporting data and relevant diagnosis codes are available within the included tables, figures, and Supplemental Materials. The National Inpatient Sample, a service supported by the Agency for Healthcare Research and Quality, is the largest publicly accessible database of all‐payer inpatient data. ¹⁰ With appropriate sampling weights, the National Inpatient Sample represents ≈95% of the United States population. We queried the years 2010 to 2019 using International Classification of Diseases, Ninth and Tenth Revision (ICD‐9; ICD‐10) codes. Patients with AIS were selected using the codes (ICD‐9, 443, 434; ICD‐10, I63). Patients with PAVM (ICD‐9, 41470, 7473; ICD‐10, Q2572, I280) were compared with patients without PAVM. In addition, a subpopulation was created to query for patients with DVT to evaluate the risk of stroke in these patients in the setting of PAVM.

Variables and Outcomes

Baseline demographics including age and sex were extracted. Comorbidities such as diabetes, hypertension, hyperlipidemia, congestive heart failure, chronic obstructive pulmonary disease, chronic renal failure, atrial fibrillation, anticoagulation/antiplatelet treatment, peripheral vascular disease, smoking status, alcohol or substance abuse, obesity, obstructive sleep apnea, frailty, and HHT were compared between those with PAVM and those without. Frailty was quantified using an 11‐factor modified frailty index, which encompasses relevant conditions in a patient's history such as functional status, diabetes, chronic obstructive pulmonary disease, congestive heart failure, myocardial infarction, cardiac interventions, hypertension with medication, peripheral vascular disease, impaired sensorium, transient ischemic attacks without deficit, and cerebrovascular accident with neurological impairments. ¹¹ Stroke‐related indices were used to compare stroke severity. These included mechanical ventilation, coma, cerebral edema, herniation, aphasia, and hemiplegia. A composite stroke severity scale was created as the sum of these indices. This scale has been used in previous studies to quantify stroke severity. ¹²

The occurrences of DVT, pulmonary embolism, red blood cell transfusion, and hemorrhagic transformation were compared between the cohorts. Outcomes of length of stay, inpatient death, and routine discharge were analyzed. Prolonged length of stay was defined as a hospital stay of ≥6 days.

Statistical Analysis

Baseline clinical and demographic characteristics were provided though descriptive statistics. Continuous variables collected were evaluated by the Shapiro–Wilk test for presence of a normal distribution. Student's t‐test and the Mann–Whitney U test were applied to compare normally and nonnormally distributed continuous variables, respectively. Fisher's exact test or Pearson's chi‐squared test was used to compare categorical variables when appropriate. Multivariate regression was applied to assess the associations between the likelihood of a certain outcome or characteristic and the primary outcome, AIS, across populations with and without a PAVM. Propensity score matching was used to control for relevant covariates, such as hyperlipidemia, atrial fibrillation, congestive heart failure, chronic renal failure, smoking history, alcohol abuse, obesity, sleep apnea, and frailty, between PAVM cases and non‐PAVM control cohorts. Randomized populations of patients with AIS and AIS‐PAVM were generated from the observational data by controlling for possible covariates, including demographic, comorbidity, and stroke severity‐related variables. A 1:1 nearest neighbor match evaluated complications and outcomes between the 2 cohorts. Statistical Product and Service Solutions was employed for statistical analysis (IBM SPSS Statistics for Windows, version 28.0, Armonk NY,).

Permissions

The following study did not require formal approval by an ethical standards committee, as no human, vertebrate, or invertebrate experimentation was performed. All data were obtained retrospectively from a national publicly accessible database. Informed consent was not required, as all data were anonymized. No patient information was explicitly identifiable, so no disclosures were necessary. Institutional review board approval was waived, as the National Inpatient Sample database is publicly available. No funding was obtained or necessary.

Results

Baseline Demographics

Of 7 465 187 patients with AIS, 1864 (0.02%) had concomitant PAVM. Patients with PAVM were less likely to be aged >65 years (odds ratio [OR], 0.248 [95% CI, 0.225–0.273]; P<0.001). Patients with AIS‐PAVM were younger, with an average age of 47.0 years, versus 65.9 years for patients with non–PAVM‐AIS (P<0.001) and 56.4 years for the entire AIS population. They were also less likely to be considered frail (OR, 0.669 [95% CI, 0.602–0.743]; P<0.001) and likely to be obese (OR, 0.308 [95% CI, 0.215–0.442]; P<0.001), but they were more likely to be functionally dependent (OR, 2.168 [95% CI, 1.304–3.604]; P=0.005).

Patients with AIS‐PAVM were less likely to carry a diagnosis of diabetes (OR, 0.29 [95% CI, 0.255–0.33]; P<0.001), hypertension (OR, 0.494 [95% CI, 0.45–0.541]; P<0.001), hyperlipidemia (OR, 0.44, [95% CI, 0.4–0.485]; P<0.001), congestive heart failure (OR, 0.85 [95% CI, 0.737–0.979]; P=0.027), chronic obstructive pulmonary disease (OR, 0.861 [95% CI, 0.76–0.975]; P=0.020), atrial fibrillation (OR, 0.438 [95% CI, 0.38–0.504]; P<0.001), and peripheral vascular disease (OR, 0.371 [95% CI, 0.289–0.477]; P<0.001). In concordance with the natural history of HHT, patients with PAVM were more likely to carry an HHT diagnosis (P<0.001).

Patients with PAVM were also less likely to be taking anticoagulant or antiplatelet medications (OR, 0.871 [95% CI, 0.781–0.970]; P<0.001). Alcohol abuse was more prevalent in the PAVM cohort (OR, 1.655 [95% CI, 1.323–2.071]; P<0.001).

In the second data set representing the subpopulation of DVT patients, 1 814 299 patients had a DVT, of whom 57 885 (0.3%) also experienced an AIS. Those with PAVMs and DVTs were more likely to experience an AIS than those with DVTs and no PAVMs (OR, 6.46 [95% CI, 4.76–8.77]; P<0.001) (Table 1).

Table 1.

Baseline Demographics

AIS
	All, n (%)(N=7 465 187)	No PAVM, n (%) (N=7 463 323)	PAVM, n (%) (N=1864)	OR	P value
Age>65	5 032 129 (67.4)	5 031 497 (67.4)	632 (33.9)	0.248 (0.225–0.273)	<0.001
Sex, female	3 743 953 (50.2)	3 743 020 (50.2)	933 (50.1)	0.996 (0.91–1.091)	0.945
Diabetes	2 699 093 (36.2)	2 698 830 (36.2)	263 (14.1)	0.29 (0.255–0.33)	<0.001
Hypertension	4 367 145 (58.5)	4 366 380 (58.5)	765 (41)	0.494 (0.45–0.541)	<0.001
Hyperlipidemia	3 939 328 (52.8)	3 938 713 (52.8)	615 (33)	0.44 (0.4–0.485)	<0.001
CHF	992 658 (13.3)	992 443 (13.3)	215 (11.5)	0.85 (0.737–0.979)	0.027
COPD	1 333 500 (17.9)	1 333 206 (17.9)	294 (15.8)	0.861 (0.76–0.975)	0.020
CRF	1 388 767 (18.6)	1 388 595 (18.6)	172 (9.2)	0.445 (0.38–0.52)	<0.001
AFib	1 754 214 (23.5)	1 753 993 (23.5)	221 (11.9)	0.438 (0.38–0.504)	<0.001
Anticoagulation/Antiplatelet	1 899 468 (25.4)	1 899 041 (25.4)	427 (22.9)	0.871 (0.781–0.970)	0.012
PVD	652 083 (8.7)	652 019 (8.7)	64 (3.4)	0.371 (0.289–0.477)	<0.001
Smoker	1 971 464 (26.4)	1 971 043 (26.4)	421 (22.6)	0.813 (0.729–0.906)	<0.001
Alcohol abuse	196 857 (2.6)	196 777 (2.6)	80 (4.3)	1.655 (1.323–2.071)	<0.001
Substance abuse	859 984 (11.5)	859 792 (11.5)	192 (10.3)	0.882 (0.76–1.024)	0.102
Dependent status	27 821 (0.4)	27 806 (0.4)	15 (0.8)	2.168 (1.304–3.604)	0.005
Obesity	376 097 (5)	376 067 (5)	30 (1.6)	0.308 (0.215–0.442)	<0.001
OSA	424055 (5.7)	423 981 (5.7)	74 (4)	0.686 (0.544–0.866)	0.002
Frailty	2 472 927 (33.1)	2 472 463 (33.1)	464 (24.9)	0.669 (0.602–0.743)	<0.001
HHT	1 026 (0%)	856 (0%)	170 (9.1%)	874.871 (737.12–1038.365)	<0.011

DVT
	All, n (%) (n=1 814 299)	No PAVM, n (%) (n=1 814 014)	PAVM, n (%) (n=285)	OR	P value
AIS	57 885 (0.3)	57 385 (3.2)	50 (17.5)	6.46 (4.76–8.77)	<0.001

AFib indicates atrial fibrillation; AIS, acute ischemic stroke; CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; CRF, chronic renal failure; DVT, deep vein thrombosis; HHT, hereditary hemorrhagic telangiectasia; OR, odds ratio; OSA, obstructive sleep apnea; PAVM, pulmonary arteriovenous malformation; and PVD, peripheral vascular disease.

Stroke Characterization and Management

Patients with AIS‐PAVM were more likely to experience cerebral edema (OR, 1.354 [95% CI, 1.089–1.683]; P=0.007), aphasia (OR, 1.235 [95% CI, 1.093–1.394]; P<0.0010), and hemiplegia (OR, 1.142 [95% CI, 1.032–1.264]; P=0.001).

The PAVM cohort was more likely to register a composite stroke severity scale >1 (OR, 1.636 [95% CI, 1.494–1.792]; P<0.001). Patients with PAVM who received treatment were also more likely to have received intravenous thrombolysis (OR, 1.378 [95% CI, 1.007–1.884]; P=0.05) at an outside hospital, require mechanical ventilation (OR, 2.616 [95% CI, 2.301–2.974]; P<0.001), and undergo mechanical thrombectomy (OR, 1.391 [95% CI, 1.064–1.819]; P=0.0017 (Table 2). After 2015, we noted a significant increase in the rate of thrombectomy (OR, 9.0 [95% CI, 8.883–9.118]; P<0.001) but no change in the rate of PAVM. (Table S1).

Table 2.

Stroke‐Related Indices

	All, n (%) (n=7 465 187)	No PAVM, n (%) (n=7 463 323)	PAVM, n (%) (n=1864)	OR	P value
Mechanical ventilation	461 508 (6.2)	461 234 (6.2)	274 (14.7)	2.616 (2.301–2.974)	<0.001
Coma	404 588 (5.4)	404 493 (5.4)	95 (5.1)	0.937 (0.762–1.151)	0.541
Cerebral edema	254 563 (3.4)	254 478 (3.4)	85 (4.6)	1.354 (1.089–1.683)	0.007
Herniation	83 662 (1.1)	83 646 (1.1)	16 (0.9)	0.763 (0.467–1.249)	0.322
Aphasia	1 045 468 (14)	1 045 156 (14)	312 (16.7)	1.235 (1.093–1.394)	<0.001
Hemiplegia	1 862 697 (25)	1 862 184 (25)	513 (27.5)	1.142 (1.032–1.264)	0.011
Composite stroke severity scale >1	2 938 638 (39.4)	2 937 678 (39.4)	960 (51.5)	1.636 (1.494–1.792)	<0.001
Mechanical thrombectomy	159 689 (2.1)	159 634 (2.1)	55 (3)	1.391 (1.064–1.819)	0.017
tPA outside hospital within 24 h	116 990 (1.6)	116 950 (1.6)	40 (2.1)	1.378 (1.007–1.884)	0.05

Propensity score matching. OR indicates odds ratio; PAVM, pulmonary arteriovenous malformation; and tPA, tissue‐type plasminogen activator.

Outcomes

Propensity score matched analysis revealed that patients with PAVM were more likely to experience a DVT (OR, 10.247 [95% CI, 1.305–80.449]; P<0.011), pulmonary embolism (OR, 4.404 [95% CI, 1.468–13.216]; P=0.006), or intracranial hemorrhage (OR, 3.03 [95% CI, 1.4–6.557]; P<0.005). Patients with PAVM were also more likely to require a red blood cell transfusion (OR, 8.101 [95% CI, 3.147–20.851]; P<0.001) and prolonged length‐of‐stay duration (OR, 2.664 95% CI, 1.898–3.739]; P<0.001). Mortality and routine discharge disposition did not significantly differ across the 2 groups (Tables 3 and 4).

Table 3.

Vascular‐Related Pathologies and Complications

	All, n (%) (N = 750)	No PAVM, n (%) (N = 375)	PAVM, n (%) (N = 375)	OR	P value
DVT	11 (1.5)	1 (0.3)	10 (2.7)	10.247 (1.305–80.449)	0.011
PE	21 (2.8)	4 (1.1)	17 (4.5)	4.404 (1.468–13.216)	0.006
RBC transfusion	42 (5.6)	5 (1.3)	37 (9.9)	8.101 (3.147–20.851)	<0.001
Hemorrhagic transformation	35 (4.7)	9 (2.4)	26 (6.9)	3.03 (1.4–6.557)	0.005

DVT indicates deep vein thrombosis; OR, odds ratio; PE, pulmonary embolism; PAVM, pulmonary arteriovenous malformation; and RBC, red blood cell.

Table 4.

Outcomes

	All (N = 750)	No PAVM (N = 375)	PAVM (N = 375)	OR	P value
Prolonged LOS (>6 d), n (%)	202 (26.9)	121 (32.2)	197 (52.5)	2.323 (1.727–3.125)	<0.001
Inpatient death, n (%)	51 (6.8)	28 (7.5)	23 (6.1)	0.81 (0.457–1.434)	0.562
Routine discharge, n (%)	356 (47.5)	172 (45.9)	184 (49.1)	1.137 (0.853–1.515)	0.421
Total charges (US$)	146 614	75 734	218 842	–	<0.001

LOS indicates length of stay; OR, odds ratio; and PAVM, pulmonary arteriovenous malformation.

Discussion

Our findings suggest that patients with PAVM‐AIS have a significantly higher rate of thrombotic events and vascular complications than patients with non–PAVM‐AIS. Furthermore, in the population of patients with DVT, those with PAVMs are more likely to also experience an AIS than those without PAVMs. The underlying pathophysiology of this phenomenon is likely multifaceted. PAVMs provide a connection between right and left circulations, bypassing protective capillary beds, rendering patients even more vulnerable to venous thromboembolisms that could cause AIS. ¹³ , ¹⁴ Furthermore, PAVMs compromise critical physiological functions of the lungs such as oxygenation, which may increase systemic stroke risk via hypoxemia. The concomitant risk of ischemia in the lungs posed by DVTs may threaten any compensatory physiological changes that have occurred. ¹⁵ It is also likely that the observed differences represent an epiphenomenon rather than a direct causal relationship. DVTs and pulmonary embolisms have a rather acute onset and may occur anywhere in the trajectory of AIS treatment, so it is important to acknowledge that while the patients with AIS‐PAVM may face the risk of a different stroke pathophysiology, more research is necessary to elucidate the exact mechanism and to account for other covariates.

A significant proportion of the PAVM‐AIS cohort also carried the diagnosis of HHT. ¹⁶ Low serum iron has been repeatedly identified as a risk factor for AIS because of impaired oxygen delivery, increased viscosity of blood, and greater ratios of serum fibrinogen and factor VIII. ¹⁷ , ¹⁸ Iron deficiency may be common in patients with PAVM‐AIS and HHT attributable to hemorrhagic loss through arteriovenous malformation rupture. Our findings may substantiate previous suggestions for those with HHT to undergo regular surveillance of their visceral arteriovenous malformations. ¹⁹

Our study found that patients with AIS‐PAVM had longer length of stay, a measure of increased health care resource usage. Comparison of costs incurred during AIS treatment for patients with concurrent PAVMs versus prophylactic imaging and percutaneous closures of PAVMs may shed light on the value of preventative health care measures.

Current indications for the treatment of PAVM include greater shunt grade and size, as they have been associated with increasing severity of clinical manifestations. ¹⁹ Typically, patients with a grade 1 pulmonary shunt, as determined by transthoracic contrast echocardiography, are managed conservatively (ie, observation); grade 2, 3, or 4 pulmonary shunts often call for transcatheter embolotherapy. ¹⁹ Integration of existing regimens such as the Well's criteria and PAVM screening into stroke prevention efforts may allow for more targeted treatment, as we found that patients with DVT are more likely to experience AIS in the context of PAVM. ²⁰ A more comprehensive understanding of this pathology may facilitate earlier interventions such as microvascular resection or endovascular embolization. ¹⁴ , ²¹

Together, our findings bolster the need for greater surveillance of PAVMs and stroke prevention measures, especially since PAVMs may disproportionately affect young patients and those with commonly implicated conditions (eg, HHT). It may be advisable to proactively monitor stroke risk factors, such as iron deficiency, or prescribe antiplatelet medications prophylactically when appropriate, as suggested by some studies. ¹³ This can also be applied to patients with a history of predisposing inherited or acquired DVT risk factors.

We recognize limitations in the methodology and data sources that may preclude further analyses. First, many PAVMs go undetected as they may be microscopic in size and present asymptomatically and are confirmed only via specific screening and imaging studies. Yet studies have documented that PAVM ischemic stroke diagnoses have doubled, suggesting an uptick in detection perhaps associated with increasing awareness of the condition. ² , ¹⁶ Inconsistent screening protocols and diagnostic criteria throughout the study period and across institutions and professional organizations likely confound demographic analyses favoring young, white patients in certain socioeconomic standing with optimal health care access. ¹³ Additionally, we acknowledge the limitations inherent to electronic health record–based databases, including variation in the usage, interpretation, and sensitivity of ICD‐9/10 codes. Exclusion of long‐term acute care hospitals and inconsistent coding for major variables such as race and obesity may not accurately capture the true diversity of patient populations and clinical presentations. Furthermore, inconsistent terminology in the literature to define PAVM (versus pulmonary arteriovenous fistulae, hemangiomas, telangiectasias, and others) and cryptogenic AIS (versus embolic stroke of undetermined significance) may complicate attempts to aggregate all pertinent data and perform more robust analyses. Finally, ICD codes do not provide sufficient information about the temporality or directionality of concurrent conditions; therefore, further investigation beyond our descriptive analysis may be advisable to establish true causal associations.

Conclusion

Patients with DVTs may be more likely to experience an AIS if a PAVM is also present. Patients with AIS‐PAVM are more likely to encounter thrombotic events and vascular complications. The mechanism likely capitalizes upon the lack of a filtration apparatus because of the PAVM, resulting in a paradoxical embolization. These patients tend to be younger and incur greater health care costs. As such, PAVMs merit integration into the traditional workup for cryptogenic stroke, especially if the patient is young or categorically at risk. Imaging, such as computed tomography, magnetic resonance imaging, and angiography, and diagnostic testing, such as transthoracic contrast echocardiography, shunt‐fraction measurement, bubble studies, and genetic screening for HHT, should be obtained so the most appropriate intervention may be applied in a timely manner.

Sources of Funding

No funding was required for this work outside of statistical licenses.

Disclosures

None of the authors report any disclosures or conflicts of interests pertaining to any aspect of this work.

Supporting information

Supporting Materials A: ICD9/ICD10 Codin

Supporting Materials B: Composite Stroke Severity Scal

Table S1: Prevalence Trends of PAVM and Mechanical Thrombectomy