Trends in incidence and epidemiologic characteristics of cerebral venous thrombosis in the United States

Abstract

Objective

To test the hypothesis that race-, age-, and sex-specific incidence of cerebral venous thrombosis (CVT) has increased in the United States over the last decade.

Methods

In this retrospective cohort study, validated ICD codes were used to identify all new cases of CVT (n = 5,567) in the State Inpatients Databases (SIDs) of New York and Florida (2006–2016). A new CVT case was defined as first hospitalization for CVT in the SID without prior CVT hospitalization. CVT counts were combined with annual Census data to compute incidence. Joinpoint regression was used to evaluate trends in incidence over time.

Results

From 2006 to 2016, annual age- and sex-standardized incidence of CVT in cases per 1 million population ranged from 13.9 to 20.2, but incidence varied significantly by sex (women 20.3–26.9, men 6.8–16.8) and by age/sex (women 18–44 years of age 24.0–32.6, men 18–44 years of age 5.3–12.8). Incidence also differed by race (Blacks: 18.6–27.2; Whites: 14.3–18.5; Asians: 5.1–13.8). On joinpoint regression, incidence increased across 2006 to 2016, but most of this increase was driven by an increase in all age groups of men (combined annualized percentage change [APC] 9.2%, p < 0.001), women 45 to 64 years of age (APC 7.8%, p < 0.001), and women ≥65 years of age (APC 7.4%, p < 0.001). Incidence in women 18 to 44 years of age remained unchanged over time.

Conclusion

CVT incidence is disproportionately higher in Blacks compared to other races. New CVT hospitalizations increased significantly over the last decade mainly in men and older women. Further studies are needed to determine whether this increase represents a true increase from changing risk factors or an artifactual increase from improved detection.

Cerebral venous thrombosis (CVT) is a relatively uncommon but important cause of stroke that may be associated with significant morbidity and mortality.^1,2 Older studies estimate the annual incidence of CVT as between 2 and 5 cases per 1 million population,^1,3 but studies providing these estimates relied on data that predated the widespread use of modern noninvasive diagnostic technology for CVT such as CT and magnetic resonance venography and are therefore likely limited by underdetection.^4,5 More recent studies from various parts of the world^6,7 suggest that the incidence may be higher than previously thought, but population-based data on trends in CVT incidence and hospitalizations, particularly from multiethnic populations, are still lacking.^4,5

It has been well established that CVT is a form of stroke that affects predominantly young and middle-aged adults, especially women.⁸ Specific risk factors for CVT such as cancer, thrombophilia, oral contraceptive use, local infections, and head trauma have long been well described,^3,8,9 but the epidemiology of stroke in the United States and other parts of the world is changing. The incidence of stroke in young adults is on the rise and paralleling the observation of increasing vascular risk factors in young adults.^10,11 How the demographic and risk factor profile of patients with CVT has changed over time is unknown. Accurate understanding of the epidemiologic characteristics of CVT is of public health importance.

The primary aims of this study are to quantify current race-, age-, and sex-specific incidence of CVT in the United States; to investigate trends in the incidence and the overall burden of CVT hospitalizations in men and women of the United States; and to describe trends in the prevalence of CVT risk factors in men and women of the United States over the last decade.

Methods

Standard protocol approvals, registrations, and patient consents

According to the Health Care Utilization Project (HCUP), the datasets used in this study are limited datasets in which 16 direct patient identifiers specified in the Privacy Rule have been removed, so the use of such limited data sets does not require review by an institutional review board. The incidence portion of this project conducted with the State Inpatients Databases (SIDs) was approved by Agency for Healthcare Research and Quality after the agency reviewed and determined this project to be consistent with the HCUP Data Use Agreement.

Design and data sources for incident CVT hospitalizations

We used administrative claims data from acute care hospitals contained in the SIDs of Florida (2005–2016) and New York (2005–2016) to retrospectively identify all new cases of CVT over the period from 2006 to 2016. SIDs are part of a family of databases designed by the HCUP of the Agency for Healthcare Research and Quality. They encompass the entire universe of inpatient discharges in participating states and combined account for up to 97% of all US community hospital discharges.

The 2 selected states in this study are large, demographically diverse states that combined account for >10% of the US population. The SIDs of these states also contain verified visitlink variables that allow tracking individual patients longitudinally across numerous hospitalizations over multiple years. Between 10 and 31 discharge diagnoses are coded at each encounter with ICD-9-CM codes before October 2015 or ICD-10-CM afterward.

National burden of CVT hospitalizations

To evaluate the generalizability of findings in these states to the entire United States, we obtained data on national trends in the burden of CVT hospitalizations and the epidemiologic characteristics of CVT from the 2005–2016 National Inpatient Sample (NIS), another HCUP database. The NIS is a 20% stratified sample of all US hospital discharges. Further details on the NIS and SID designs are available at hcup-us.ahrq.gov/.

US population data

Yearly estimates of the mid-year population of the selected states and the entire United States by age and sex for the period of the study were obtained from tables available at the US Census Bureau website (census.gov/). Data on Hispanic ethnicity for all races by states were readily available in the US census website for 2010 to 2016. Race and ethnicity data were obtained for non-Hispanic Whites, non-Hispanic Blacks or African Americans, Hispanics of any race, and Asians to allow consistency with the HCUP race and ethnic categories.

Study population

We identified all adult hospitalizations (age ≥18 years) with a diagnosis of CVT by querying the database using ICD-9 CM codes 437.6, 325, and 671.5x or ICD-10-CM codes I63.6, I67.6, O22.5X, O87.3, and G08. This combination of ICD-9 codes has been validated previously and found to have a positive predictive value of 75.7% and a specificity of 92.7% for CVT,¹² while the combination of ICD-10 codes was found to have a combined positive predictive value of 92.3% for CVT.¹³ We used all primary and secondary discharge codes to identify patients because the use of codes in all positions has been shown to have better accuracy for CVT than primary discharge codes alone.¹² All hospitalizations with missing visitlink variables and patients who were nonresidents of the corresponding state of admission at the time of hospitalization were excluded to allow longitudinal follow-up of all patients over time. When studying burden of CVT hospitalization using the NIS in which there is no unique patient identifier, we excluded hospitalizations in which patients were transferred to other acute care hospitals within 24 hours of admission to minimize the potential for double counting. We also excluded all hospitalizations with missing information on age or sex. Using criteria similar to that described above, we identified all adult stroke hospitalizations (both ischemic and hemorrhagic) in the NIS over the study period using the constellation of ICD-9/10 codes for stroke contained in the HCUP Clinical Classification Software (DXCCS 109).

Definition of incident CVT

Incident CVT in any patient was defined as first hospitalization containing a diagnostic code for CVT and with no prior hospitalizations for CVT in that patient in the preceding years. We used a 1-year washout period (2005) to minimize the influence of old CVT hospitalizations on study estimates after this year. Use of a longer washout period (3 years) did not significantly change estimates.

Definition of CVT risk factors and other variables

Patient-level risk factors for CVT that have previously been established in the literature, including pregnancy and puerperium, prothrombotic conditions, inflammatory diseases, CNS trauma, CNS and ear infection, any cancer, any brain tumor (benign or malignant), dehydration, and contraceptive use, were identified with the HCUP CCS software or the ICD-9/10 codes corresponding to these conditions (data available from Dryad, table e-1, 10.5061/dryad.zw3r2285g).

Hypertension, diabetes mellitus, and obesity were defined with the HCUP variables corresponding to these conditions. Age was stratified at the time of CVT diagnosis into 18 to 44 , 45 to 64, and elderly ≥65 years of age to allow easy correlation with census data groups. Sex was defined with the HCUP variable female (0 = men, 1 = women). The nature of the dataset does not allow other gender categories.

Statistical analysis

Overall annual incidence of CVT per 1,000,000 population was calculated from the number of CVT cases and the total adult population of the participating states for that year. Age-stratified and crude estimates by race and sex were also obtained from stratified counts of CVT and of the subpopulation in each category. Crude estimates were age- and/or sex-standardized, as appropriate, to the 2000 US Census population to make data comparable between groups and across years.

Similarly, we used overall weighted counts of total CVT hospitalizations obtained from the NIS to calculate the burden of CVT hospitalizations per 1,000,000 persons in the overall population and in various sex groups. NIS trend weights were used to make data comparable before and after the NIS redesign in 2012 as recommended by HCUP.

We used joinpoint regression models with the permutation model selection method to evaluate trends in the incidence of CVT cases or trends in burden of CVT hospitalization over time. Joinpoint regression uses a series of Monte Carlo permutation-based tests to identify points of change in trends (joinpoints) in a dataset.¹⁴ The annualized percentage change (APC) is then computed for each of the identified trends by fitting a regression line to the natural logarithm of the rates with calendar year used as a regressor variable.

We evaluated for possible linear trend in the prevalence of each CVT risk factor over time by constructing a logistic regression model with each risk factor as the dependent variable and year of discharge as the independent variable, evaluated continuously, with significance of differences in trend over time assessed with the Wald test.¹⁵

All primary analyses were performed with Stata 14 (StataCorp, LP, College Station, TX). A 2-tailed value of α < 0.05 was required for statistical significance. Because this is an exploratory analysis with no specific hypothesis being tested, adjustment for multiple comparison was not necessary.^16,17 We considered the weighting, clustering, and stratification needed in the complex NIS survey design in all analysis by using the Stata SVY suite of commands with hospital as the primary sampling unit and applying relevant probability sampling weights for robust variance estimation to all models. Joinpoint regression was done with Joinpoint software version 4.7.0.1 (Bethesda, MD), while computation of age-adjusted incidence proportions and standard errors was done with Microsoft Excel (Microsoft Corp, Redmond, WA).

Sensitivity analysis

To better understand any potential influence that transition from an ICD-9– to an ICD-10–based system during the study period may have had on trend estimates, we restricted joinpoint regression of CVT incidence to the period just before ICD-10 introduction, i.e., 2006 to 2014, and qualitatively compared the estimates to that of our primary analysis.

Missing data

After inclusions and exclusions enumerated above, data were missing for race in 0.8% of patients in the SIDs of New York and Florida across the period of 2006 to 2016. Patients with missing race were categorized in an unknown category.

Data availability

The SID and NIS are both publicly accessible datasets that can be obtained easily after completion of the HCUP data use agreement.

Results

Across the period from 2005 to 2016, there were 57,315 weighted CVT admissions in the United States, representing 0.66% of all adult hospitalizations for any cerebrovascular disease over this time. The proportion of stroke admissions that were CVT increased by 70.4% from 0.47% in 2005 to 0.80% in 2016 (p for trend <0.001) (table 1).

Table 1.

Temporal trends in age, sex, and risk factor characteristics of CVT hospitalizations in the United States, 2005 to 2016

Sex and age distributions of CVT hospitalizations in the entire United States

About two-thirds (66.7%) of all CVT hospitalizations were in women, but this proportion declined over time, just as the proportion of hospitalizations in men increased by 88.6% over time (22.9% in 2005 and 43.1% in 2016, p for trend <0.001). The mean age of men (49.6 years) was slightly higher compared to that of women (42.1 years) (p for comparison <0.001), but the mean age of women increased steadily across the entire study period (p for trend <0.001) (table 1).

On further stratification by age and sex, the proportion of all hospitalizations in the United States in young women declined significantly over time, while the proportion of hospitalizations in all other age groups increased over time. In 2005, more than half of all CVT hospitalizations (56.5%) in the entire United States were in young women 18 to 44 years of age, but this proportion declined steadily over time, and by 2016, only 27.6% were in this sex and age group. In contrast, the combined proportions of hospitalizations in elderly adults more than doubled from 11.2% (7.4% in women + 3.8% in men) in 2005 to >24% in 2016 (table 1).

Demographic characteristics of incident CVT cases in Florida and New York alone

The age and sex distributions of incident cases in the states of Florida and New York were very similar to that of the entire United States (table 2). A total of 68.9% of all incident CVT cases in these states were in women, but this proportion also declined over time (table 2); 54.3% of all incident CVT hospitalizations were in Whites, 19.2% were in Blacks, and 14.9% were in Hispanics (table 2).

Table 2.

Age and sex distributions and risk factor characteristics of 5,567 incident cases of CVT admitted to hospitals in Florida and New York, 2006 to 2016

Prevalence and trend in CVT risk factors in the entire United States

We found that 57.4% of hospitalizations in men and 63.7% of hospitalizations in women had comorbid codes for at least 1 CVT risk factor (table 1). Pregnancy and puerperium (21.7%), cancer (11.8%), and inflammatory conditions (11.4%) were the most common associated conditions coded in women, while cancer (19.5%), CNS trauma (11.3%), and CNS infection (11.2%) were the most common conditions in hospitalizations in men (table 1). The proportion of hospitalizations in women with comorbid codes for pregnancy and the puerperium declined by >50% over time (41.3% in 2005 and 16.7% in 2014) in all women (table 1) and even among young women (63.7% in 2005 to 21.6 in 2016) (data available from Dryad, figure e-1, 10.5061/dryad.zw3r2285g). In comparison, the proportion of hospitalizations with comorbid cancer and comorbid CNS trauma increased significantly over time in both sexes (all p for trends <0.001) (table 1). The risk factor characteristics of the incident cases in New York and Florida were very similar to that of the entire United States.

Population incidence and burden of CVT hospitalizations over time

There were 5,567 new cases of CVT in Florida and New York over the period of 2006 to 2016. Age- and sex-standardized incidence of CVT varied across the study period from 13.9 cases per 1 million in 2006 to 20.2 cases per 1 million in 2014, but the age-standardized incidence was consistently greater in women compared to men (table 3) (p < 0.001).

Table 3.

Temporal trends in the incidence of CVT in New York and Florida and burden of hospitalizations for CVT in the entire United States, 2005 to 2016

On joinpoint regression, the incidence of CVT increased across the study period in both sexes, but the rate of increase was >4 times more rapid in men (APC 9.2%, 95% confidence interval [CI] 6.9%–11.6%) compared to women (APC 2.1%, 95% CI 0.8%–3.7%) (figure 1 and table 3). Pairwise comparison accessing whether the 2 regression mean functions are not parallel (test of parallelism) was significant (p < 0.001), indicating that the incidence gap between men and women closed significantly over time (figure 1).

Figure 1. Joinpoint regression of trends in age-standardized incidence of CVT in New York and Florida from 2006 to 2016, according to sex.

APC = annualized percentage change; CI = confidence interval; CVT = cerebral venous thrombosis.

Among women, most of the increase was driven by increased incidence in middle-aged women 45 to 64 years of age (APC 7.8%, 95% CI 5.9%–9.7%) and elderly women ≥65 years of age (APC 7.4%, 95% CI 4.7%–10.2%); the incidence in young women 18 to 44 years of age remained virtually unchanged at 29.2 to 32.6 cases per 1 million patients across most of the study period (APC −0.5%, 95% CI −2.2 to 1.2, p = 0.495) (table 3 and figure 2). In contrast, the incidence increased in all age groups of men (figure 2). A similar pattern of stable hospitalization in young women and increasing new hospitalizations in men and older women was also noted in each state when results were stratified by state, sex, and age (results not shown).

Figure 2. Joinpoint regression of trends in incidence of CVT in New York and Florida from 2006 to 2016, according to sex and age groups.

In men, yearly annualized percentage change (APC) estimates for the age groups of 18 to 44 and 45 to 64 years are displayed underneath their regression lines, and the estimate for ≥65 years of age is displayed above the regression line. In women, APC estimates for those 18 to 44 and ≥65 years of age are depicted above the regression line, and that for women 45 to 64 years of age is depicted underneath the regression line.

Consistent with the increased incidence of CVT in these states, the national burden of hospitalizations with codes for CVT (whether new or recurrent hospitalizations) increased significantly across the study period from 13.7 hospitalizations per 1 million population in 2005 to 26.2 hospitalizations per 1 million in 2016 (table 3). The annual rate of change in hospitalizations was also greater in men (APC 10.7%, 95% CI 9.3%–12.2%) compared to women (APC 3.1%, 95% CI 2.0%–4.2%) (table 3).

CVT incidence by race

When aggregated across the entire period, the average age- and sex-standardized incidence of CVT was highest in Blacks (23.1 cases per 1 million), followed by non-Hispanic Whites (16.5 cases per 1 million) and Hispanics (13.7 cases per 1 million) (p for pairwise mean comparison to Blacks <0.001). The average incidence in Asians (8.56 cases per 1 million) was just over half that of non-Hispanic Whites (table 4).

Table 4.

Age- and sex-standardized incidence of CVT in the New York and Florida from 2010 to 2017 according to race and ethnic groups

Sensitivity analysis

Just as in our primary analysis, the incidence in young women 18 to 44 years of age did not change significantly when restricted to the period of 2006 to 2014 (APC 0.74%, 95% CI −033% to 1.82%) (data available from Dryad, table e-2, 10.5061/dryad.zw3r2285g). The APCs in women 45 to 64 years of age (8.82%) and women ≥65 years of age (9.04%) were marginally higher but within the range of the primary estimates. Similarly, APCs in the various age groups of men were also within the range of those of our primary estimates.

Discussion

In this contemporary population-based study, we provide new information highlighting current trends in the national burden of CVT in the United States. We show that 0.66% of all stroke admissions over the period of 2005 to 2016 were CVTs and that this proportion increased by 70% over time. The annual age- and sex-standardized incidence of CVT in Florida and New York ranged from 13.9 to 20.2 cases per 1 million across the period of 2006 to 2016. Incidence differed by sex, age, and race. Incidence increased over time, but the pace of this increase was significantly faster in men compared to women. Incidence increased in all age groups of men and in middle-aged and elderly women, but that of young women (18–44 years of age) remained unchanged over time. These findings indicate a change in the demographic characteristics of detected CVT cases in the United States.

The rising CVT incidence is likely due in part to better CVT ascertainment. Advances in noninvasive imaging and improved recognition of the diverse presentation of CVT by clinicians have resulted in increased detection.¹ Increased use of neuroimaging is likely a strong contributor to the increased diagnosis. For example, the use of neuroimaging for headache increased by up to 3-fold in recent years.¹⁸ An alternative or contributory explanation is that changes in the prevalence of known, emerging, or unknown CVT risk factors also may have partly led to increased incidence. That the rising incidence differentially spared young women and that the overall pace of increase was more rapid in men compared to women give credence to the latter hypothesis, because improvement in clinical and diagnostic recognition is more likely to have occurred concurrently in all age and sex groups over time unless the true incidence peak in certain age groups have been approached. In this study, we documented a marked increase in the proportion of CVT hospitalizations with comorbid codes for brain tumor, cancer, and inflammatory diseases. The prevalence of obesity, a possible emerging risk factor for CVT when combined with oral contraceptive use,¹⁹ also more than doubled in CVT hospitalizations over time. The rising incidence may therefore be a reflection of these changes in the risk factor profile of patients with CVT. Increased survival after a cancer diagnosis could be a factor in the increased trends in the subgroup of patients with this condition.

Whether real or artifactual from improved detection, the rising incidence implies that CVT is either a growing or previously unrecognized problem in men and older women and no longer just a disease of young women alone. Consistent with this changing incidence, we noted a >50% decline in the proportion of CVT hospitalizations in young women in the entire United States over time due to increasing proportion of hospitalizations in men and older women. These findings contradict that of a recent systematic review that reported a shift in the sex ratio of CVT hospitalizations toward women in the 5 decades from 1966 to 2014.²⁰ Although not directly comparable to ours, this review had included predominantly small retrospective single-center studies that are likely to have been constrained by their small sample size, limited generalizability ,and the potential for referral and publication bias.²⁰

The findings from this study corroborate those of other recent studies from Adelaide, South Australia, from 2005 to 2011 (15.7 cases per 1 million)⁶ and 2 Dutch provinces from 2008 to 2010 (13.2 cases per 1 million)⁷ reporting higher annual CVT incidence than older studies. Our finding of higher incidence in young women (24.0–32.6 cases per 1 million) is also consistent with that reported in the Netherlands (27.8 per 1 million person-years),³ but we provide new information evaluating trends and demonstrate that incidence has not changed in this age group of women over time. This finding is particularly noteworthy given the remarkable increase in CVT incidence in other sex/age groups. Reasons for these findings are not clear, but again, changes in the risk factor characteristics of young women may be contributory. In this study, we noted a precipitous decline in the prevalence of comorbid pregnancy and postpartum, which was the most common risk factor coded in incident cases and in the entire United States (tables 1 and 2 and figure e-1, available from Dryad, 10.5061/dryad.zw3r2285g).While we also have no clear explanation for this intriguing finding, it is possible that changes in obstetric practices over the last decade, including recommendations for the use of aspirin to prevent preeclampsia in at-risk patients,²¹ may also have inadvertently led to a decrease in pregnancy- and postpartum-associated CVTs in some otherwise at-risk patients. Aspirin may prevent venous thromboembolic disease.²² Any potential increase in CVT from other risk factors may have been balanced by a decrease in pregnancy- and postpartum-associated CVTs, but no associations or conclusions can be drawn from this retrospective study. The increased proportion of head and neck infections and cancer-associated CVT in women but not in men in this study also may be related in part to the declining proportion of pregnancy-associated cases. As the proportion related to pregnancy markedly declined, cancer and head and neck infections emerged as more prominent risk factors in women over time. Prospective studies are needed to better understand this and other potential explanatory factors. Men are more predisposed to deep venous thrombosis (DVT) compared to women,²³ and with improved pregnancy care, the trends could be showing a more normal distribution of CVT.

Another salient finding in this study is the racial disparity in CVT incidence. To the best of our knowledge, racial differences in CVT incidence have not been described previously. Multiple studies from Iran over the last decade report annual CVT incidence of 12.3 to 13.5 cases per 1 million in this population,^24,25 but these are based on experiences from single or dual hospital centers and likely to have been limited by referral bias. Another single-center Caribbean study reported an annual incidence of 15 to 20 cases per 1 million in the predominantly Black population of Guadeloupe,²⁶ but data on CVT incidence from representative non-White populations in Africa, Asia,²⁷ Latin America, or even the United States are sparse.⁴ We found that the age- and sex-standardized incidence of CVT in Blacks was higher compared to that of non-Hispanic Whites and more than twice the incidence of CVT in Asians.

The exact reasons for the racial disparity in CVT incidence noted in this study are unknown. Prior studies have reported a higher incidence of arterial ischemic stroke, DVT, and pulmonary embolism in Blacks compared to non-Hispanic Whites and Hispanics.^28–30 DVT and pulmonary embolism incidence in Blacks may be up to 6 times that in Asians.²⁸ However, ischemic stroke, systemic venous thromboembolism, and CVT are distinct clinical entities. Blacks may possess an underlying genetic predisposition to thrombosis.³⁰ Levels of fibrinogen (an essential component of blood clot and established marker of coronary artery disease and stroke) are known to be higher in individuals of African descent compared to European Americans.³⁰ Systemic lupus erythematosus, sarcoidosis, and other inflammatory conditions associated with increased CVT incidence are more prevalent and more severe in Blacks compared to Whites.^31–33 In addition to biological differences, race may be a proxy for socioeconomic and lifestyle risk factors that may possibly increase CVT risk. Obesity, a condition that involves interplay between genetic and environmental factors, is more prevalent in Blacks compared to Whites across the entire age spectrum.³⁰ Studies evaluating the role of access to healthcare, neighborhood factors, and other social determinants of health that separate across racial lines and may possibly influence CVT risk are lacking. Future studies are needed to address these and other potential explanatory factors.

Major strengths of our study include the use of a population-based database covering >97% of the population of the states of Florida and New York. The large number of cases identified from this multiethnic population provided an adequate number of cases for the estimation of CVT incidence in various population subgroups. We provide national estimates of the burden of risk factors that are generalizable to the entire United States.

This study has limitations. Despite entailing some of the highest annual estimates of CVT incidence ever reported, this study likely underestimates the overall true incidence of CVT in the community for several reasons. First, this study was done using a hospital database and thus may not have captured asymptomatic or mild cases of CVT not presenting for hospitalization. Moreover, up to 3.3% of cases of CVT in Florida and New York may be misdiagnosed at initial emergency room presentation.³⁴ Second, although the codes used for CVT diagnosis have previously been validated and have high specificity (92.7%),¹⁹ the relatively modest sensitivity (77.8%)¹² means that a significant proportion of CVT cases may not have been captured.

Cumulative incidence calculation as done in this study is based on the assumption of a closed population with no competing risk such as death and no loss to follow-up. However, some residents of these states die daily, while others emigrate. This can potentially lead to underestimation or overestimation of true incidence, depending on the baseline risk of patients dying or migrating. Nevertheless, any potential influence this may have had on our study estimates is likely to have been small.³⁵ Our study of trends is based on the implicit assumption that coding practices remained unchanged over time, but in our study period, use of ICD-9 and ICD-10 codes overlapped. A recent study showed no substantial impact on the overall accuracy of identification of patients with stroke by transitioning from an ICD-9– to an ICD-10–based coding system in the United States.³⁶ Furthermore, we conducted a sensitivity analysis showing that trend in CVT incidence remained significantly high when limited to the period before ICD-10 introduction. We cannot evaluate the influence of increased CVT ascertainment over time on study findings.

The demonstrated trends in CVT risk factor prevalence should be viewed with caution because we were able to show only a cross-sectional association between these factors and CVT. For example, improved care of patients with cancer may have allowed them to survive long enough to develop CVTs, or a diagnosis of CVT may have prompted search for an underlying cancer or an underlying inflammatory state, but we are unable to show a temporal association between these conditions and CVT. We cannot provide information on individual prothrombotic conditions such as factor V Leiden or prothrombin gene mutation because this information is unavailable in our predominantly ICD-9 database. There is great variability in how physicians diagnose hypercoagulable conditions, but the likely impact of this on study estimates is random misclassification toward the null. Although we took measures such as excluding transfers within 24 hours to minimize the potential for double-counting hospitalizations when evaluating the national burden of CVT, some patients may have still been counted more than once. When using the deidentified NIS, we are also uncertain what proportion of CVT diagnoses may have been carried forward to subsequent unrelated hospitalizations. This may potentially lead to an artifactual increase in CVT burden over time.

As seen in recent studies from other countries, the incidence of CVT in the United States is >3-fold higher than previously reported, but the demographic characteristics of detected cases in the United States are changing. The incidence of CVT is disproportionately higher in Blacks compared to other races. CVT incidence increased over time, but how much of this increase represents an artifactual increase from improved detection vs a true increase is yet to be determined. Our findings need to be replicated in other multiethnic studies to better elucidate the etiologic reasons for this changing trend.

Glossary

APC

annualized percentage change

CI

confidence interval

CVT

cerebral venous thrombosis

DVT

deep venous thrombosis

HCUP

Health Care Utilization Project

ICD-9/10-CM

International Classification of Diseases, 9th/10th revision, clinical modification

NIS

National Inpatient Sample

SID

State Inpatients Database

Appendix. Authors

Study funding

No targeted funding reported.

Disclosure

F.O. Otite, S. Patel, R. Sharma, P. Khandwala, D. Desai, J.G. Latorre, S. Izzy, E.O. Akano, N. Anikpezie, A.M. Malik, D. Yavagal, and P. Khandelwal report no conflicts of interest relevant to this manuscript. S. Chaturvedi is an assistant editor for Stroke. Go to Neurology.org/N for full disclosures.