Long-Term Risk of Stroke after Traumatic Brain Injury: A Population-Based Medical Record Review Study

Abstract

Objective:

To reliably inform secondary prevention strategies and reduce morbidity and mortality after traumatic brain injury (TBI), we sought to understand the long-term risk of stroke after TBI in patients aged 40 years and older in comparison to age- and sex-matched referents from a population-based cohort.

Materials and Methods:

TBI cases in Olmsted County, Minnesota from January 1, 1985, to December 31, 1999, were confirmed by manual review, classified by injury severity and mechanism, and nonhead trauma was quantified. Each TBI case was matched to 2 sex- and age-matched population-based referents without TBI and with similar severity nonhead trauma. Records of cases and referents were manually abstracted to confirm stroke diagnosis. Stroke events during initial hospitalization for TBI were excluded.

Results:

In total, 1,410 TBI cases were confirmed, 61% classified as possible TBI (least severe, consistent with concussive), with the most common mechanism being falls. There were 162 stroke events among those with TBI (11.5%) and 269 among referents (9.5%). Median time to stroke from the index date for those with TBI was 10.2 years (Q1–Q3 5.2–17.8), and for referents 12.1 years (Q1–Q3 6.2–17.3), p = 0.215. All-severity TBI was associated with increased risk of stroke (HR: 1.32, 95% CI: 1.06–1.63, p = 0.011), but only definite TBI (consistent with moderate-severe) was associated with significant risk (HR: 2.20, 95% CI: 1.04–4.64, p = 0.038) when stratified by severity.

Discussion/Conclusion:

By confirming TBI cases, stroke diagnoses, and injury severity classification using manual review with levels of accuracy not previously reported, these results indicate moderate-severe TBI increases long-term risk for stroke. These findings confirm the need to regularly assess long-term vascular risk after TBI to implement disease prevention strategies.

Introduction

Traumatic brain injury (TBI) annually accounts for up to 283,680 hospitalizations and 52,844 deaths in the USA [1, 2] and has been associated with an increased risk for developing subsequent medical conditions that may further increase its morbidity and mortality [3–5]. TBI has been associated with an increased risk of stroke [6–12], which has important implications because stroke is the fifth leading cause of death and is also commonly associated with long-term functionally limiting neurological impairment [9, 13–16]. Accurate estimates of stroke risk after TBI are essential to reliably inform clinical surveillance strategies and reduce the morbidity associated with TBI.

Existing literature examining the relationship between TBI and stroke has predominantly identified TBI and stroke cases using administrative data designed for billing, rather than research purposes [6–8, 11, 17], which can limit accuracy [16, 18, 19]. Further, although there is consistent evidence of an increased risk of stroke in the acute phase after TBI [9, 15], less is known of the long-term risk and how this risk changes across the spectrum of TBI severity [20–25].

The purpose of this study was to confirm cases of TBI in a population-based cohort of adults by medical record review, stratify them by injury severity, match each case to a referent in the population, and determine the long-term risk for stroke in each group. We further sought to determine whether TBI was differentially associated with ischemic stroke as compared to hemorrhagic stroke, to what extent stroke risk was affected by TBI severity, and the effect of nonhead trauma on stroke risk. We hypothesized that stroke risk would be increased with severity of TBI in comparison with age- and sex-matched referents after adjusting for nonhead trauma and comorbidity.

Methods

This cohort study was approved by Institutional Review Boards at Mayo Clinic and Olmsted Medical Center.

Study Setting

Rochester, Minnesota, is in Olmsted County (2018 census population 156,277) and home to Mayo Clinic, a large private medical center. Beginning in 1907, comprehensive data have been collected and linked to a unique identifier for each patient registered, a linkage that developed into the Rochester Epidemiology Project (REP) in 1966 [26]. The REP is a powerful tool for population-based epidemiological studies and allows for a unique assessment of the natural history of TBI and related sequelae [18, 27–29]. Data from the REP encompass more than 6,239,353 person-years of follow-up and include demographic information, drug prescriptions, surgical procedure codes, and diagnostic codes assigned at every medical contact [30–32]. These data are then screened using a coding system designed specifically for clinical and research purposes, using a modification of the Hospital Adaptation of ICDA – International Classification of Diseases, Ninth Revision (ICD-9), and the International Classification of Diseases, Tenth Revision (ICD-10) [33].

Identification of TBI

The methods for identification of the TBI sample have previously been described [18, 29]. TBI was defined as a “traumatically induced injury that contributed to the physiological disruption of brain function” [18, 29, 34, 35]. Using the REP, a list of potential Olmsted County residents was constructed with codes suggestive of TBI for the period January 1, 1985, to December 31, 1999. Because our goal was to estimate the long-term risk for stroke, we limited the population to those who had at least 5 years of follow-up and those >40 years of age as of the index date because stroke incidence is highest in older individuals [13, 14]. Residents with potential TBI who were screened based on these criteria were then confirmed by manual review. Because additional body injury co-occurring with a TBI may affect stroke risk, all confirmed TBI events were categorized as having either an isolated TBI (case) or TBI associated with nonhead traumatic injury (special case) [29]. For TBI events with nonhead trauma (special case), each of the accompanying nonhead injuries was assigned a diagnosis code based on an empiric measure of injury severity. The trauma mortality prediction model was applied to assign an overall measure of nonhead injury severity to each individual [36].

TBI severity was characterized using the Mayo Classification System for TBI Severity [37]. The Mayo Classification System stratifies each case by injury severity based on the strength of available information in the medical record. It is particularly valuable in retrospective studies where data can be limited by variably missing indicators of injury severity [18, 32]. TBI was classified into definite TBI (consistent with moderate-severe TBI); probable TBI (consistent with mild TBI); and possible TBI (consistent with concussive TBI) (online suppl. Fig. 1; for all online suppl. material, see www.karger.com/doi/10.1159/000525111). All TBI events were further characterized by mechanism of injury according to methods developed by the National Center for Injury Prevention and Control, Division of Injury Disability, Outcomes, and Programs.

Selection of Age- and Sex-Matched Referents

Date of first incident TBI was considered the index date for matching unexposed referents. Each person with a confirmed TBI was matched to 2 sex- and age-matched referents without TBI who were county residents seen in any Olmsted County medical facility in the year (±1) of the exposed individual’s TBI event and were chosen at random from those eligible. Special referents were defined as those who were matched to residents with confirmed TBI events who also experienced co-occurring nonhead trauma (special case). Special referents were also to have been hospitalized or seen in the Emergency Department for nonhead traumatic injuries with a comparable trauma mortality prediction model score within ±2 years of the exposed TBI index date.

Identification of Outcomes

Stroke outcomes were assessed in two phases. First, electronic REP resources were used to identify all instances of stroke based on diagnostic codes, with a broad definition of stroke utilized to ensure capture of all events (online suppl. 2). Data were manually abstracted from paper and electronic health records to confirm the diagnosis of stroke via clinical notes and review of imaging (M.S.), with abstractor blinded to the status of TBI exposure. Diagnostic code dates were reviewed individually from oldest to newest until the true first stroke date was identified, and stroke type (ischemic or hemorrhagic) was recorded. Diagnostic codes for stroke occurring during the same hospitalization for TBI were excluded to focus analysis on long-term effects of TBI rather than sequelae of acute trauma. Codes that were identified via review as transient neurologic symptoms without imaging findings and those that were secondary to potentially repeat trauma were not classified as a stroke.

Statistical Analyses

Cox proportional hazard models were used to examine the relationship between TBI and stroke with time to stroke as the timescale. The hazard ratio (HR) and 95% confidence intervals (CIs) are reported for prediction of occurrence of stroke for those with TBI compared to their matched referents. All individuals with confirmed TBI and all sex- and age-matched referents were followed in the REP census from the TBI event or corresponding index date until the earliest date of (a) the first clinical diagnosis of a stroke; (b) the last medical visit in Olmsted County prior to data abstraction on May 31, 2018; or (c) death. If referents developed a confirmed TBI event after their index date, they were censored at that time. Multivariable models adjusted for the Charlson comorbidity index [38]. We also determined whether the association between TBI and stroke was modified by sex by adding an interaction term between TBI and sex. Individuals with TBI were analyzed as a total sample and stratified by injury severity. All statistical tests were 2-sided, and p values less than or equal to 0.05 were considered statistically significant. Statistical analyses were performed using SAS version 9.4 (SAS Inc. Cary, NC, USA).

Results

A total of 45,791 records were identified as potential TBI from January 1, 1985, through December 31, 1999, of which 14,114 met criteria for abstraction. Manual review for these individuals confirmed 1,410 TBI cases including 1,216 with isolated TBI and 194 with co-occurring nonhead trauma (Fig. 1). The median age of the sample at index date was 54 years (range, 40–97); 43% were male and 91% white (Table 1). The majority of those with TBI were classified as possible (least severe) injuries (61%), and the most common mechanism was falls (48%). 2,820 sex- and age-matched referents without TBI were utilized and were divided between regular and special referents (Fig. 1).

Fig. 1.

Flow diagram representing acquisition and stratification of TBI cases and matched referents.

Table 1.

Sample characteristics of TBI cases

	Total (n = 1,410)	Age 40–64 (n = 1,000)	Age ≥65 (n = 410)
Median age (range), years	54 (40–97)	49 (40–64)	75 (65–97)
Male, n (%)	600 (42.6)	468 (46.8)	132 (32.2)
White, n (%)	1,281 (90.9)	888 (89.0)	393 (95.9)
TBI classification, n (%)
Definite	109 (7.7)	67 (6.7)	42 (10.2)
Probable	447 (31.7)	309 (30.9)	138 (33.7)
Possible	854 (60.6)	624 (62.4)	230 (56.1)
Mechanism of injury, n (%)
Fall	681 (48.3)	395 (39.5)	286 (69.7)
Motor vehicle collision	456 (32.3)	363 (36.3)	93 (22.7)
Hit by object	178 (12.6)	153 (15.3)	25 (6.1)
Assault or gunshot	47 (3.3)	44 (4.4)	3 (0.7)
Sports or recreation	35 (2.5)	34 (3.4)	1 (0.2)
Other	13 (0.9)	11 (1.1)	2 (0.5)

There were 162 stroke events among those with TBI (11.5%) and 269 among referents (9.5%) (Table 2). Age at stroke and time from TBI to stroke were similar between groups. The median follow-up for those with TBI was 21.0 years (range Q1–Q3 14.1–25.7) and for referents 20.7 years (Q1–Q3 14.5–24.9), p = 0.168. The median time to stroke for those with TBI was 10.2 years (Q1–Q3 5.2–17.8) and for referents 12.1 years (Q1–Q3 6.2–17.3), p = 0.215. Stroke type was predominantly ischemic for both groups, and the percentage of stroke events among those with TBI and referents was similar when stratified by TBI severity.

Table 2.

Stroke events among TBI cases and referents

	TBI cases (n = 1,410)	Referents (n = 2,820)
Stroke event (%)	162 (11.5)	269 (9.5)
Age at stroke (Q1–Q3), median years	79 (71–84)	78 (71–85)
TBI to stroke in years median (Q1–Q3)	10.2 (5.2–17.8)	12.1 (6.2–17.3)
Median years to TBI follow-up (Q1–Q3)	21.0 (14.1–25.7)	20.7 (14.5–24.9)
Stroke type
Ischemic, n (%)	152 (93.8)	231 (85.9)
Hemorrhagic	10 (6.2)	37 (13.8)
Both	0	1 (0.4)
By TBI severity classification
Definite, n (%)	15 (9.3)	21 (7.8)
Probable	55 (34.0)	94 (34.9)
Possible	92 (56.8)	154 (57.2)

Table 3 and Figure 2 show that when considering TBI as a group, including all severity categories, TBI was associated with an increased risk of stroke (HR: 1.32, 95% CI: 1.06–1.63, p = 0.011). When stratified by TBI severity, only definite (HR: 2.20, 95% CI: 1.04–4.64, p = 0.038) was associated with significant risk of stroke, but not probable (HR: 1.21, 95% CI: 0.84–1.76, p = 0.306) or possible (HR: 1.29, 95% CI: 0.97–1.70, p = 0.079) TBI. After adjusting for Charlson comorbidity index, overall risk of stroke remained similar (HR: 1.32, 95% CI: 1.10–1.60, p = 0.014). We also assessed whether sex modified the association between TBI and stroke, but the interaction term was not significant (HR: 0.96, 95% CI: 0.83–1.10, p = 0.531). In addition, nonhead trauma did not significantly increase risk for stroke beyond that of TBI (HR: 1.50, 95% CI: 0.79–2.86, p = 0.217).

Table 3.

Risk for stroke after TBI

	HR (95% CI)	p value
Total (1,410 cases, 2,820 referents)	1.32 (1.06–1.63)	0.011
By TBI severity classification
Definite (109 cases, 218 referents)	2.20 (1.04–4.64)	0.038
Probable (447, 894)	1.21 (0.84–1.76)	0.306
Possible (854, 1,708)	1.29 (0.97–1.70)	0.079

Fig. 2.

Kaplan Meier curves representing cumulative incidence of stroke over time after TBI as a whole, and when strafied by injury severity.

Discussion

Analyses of a population-based sample of TBI, confirmed by medical record review, showed an increased long-term risk of stroke when compared to age- and sex-matched referents. Stroke occurred a median of 10.2 years after TBI. When stratified by TBI severity, stroke risk was significantly increased for definite, but not possible or probable TBI. The relationship between TBI and stroke did not differ by sex, and the risk remained after adjusting for comorbidities.

Our results are most consistent with Burke et al. [7], who reported TBI patients had an increased risk of subsequent hospitalization for an ischemic stroke using diagnostic codes and a follow-up of 28 months. In that study, stroke within 1 year of TBI was excluded. Multiple analyses of the association between TBI and long-term risk for stroke have been conducted in a Taiwanese database utilizing ICD coding for the identification of both stroke and TBI. Over 5 years of follow-up, Chen et al. [8] found an increased risk of post-stroke TBI at 1 year and 5 years compared to an age- and sex-matched referent group. Notably, when TBI severity was inferred by the presence or absence of skull fracture, the risk was even greater [8]. Eric Nyam et al. [11] also found an increased risk of stroke after TBI over a 5-year follow-up period but did not stratify by TBI severity. In a cohort of adults with mild TBI determined by grouping diagnostic codes and surviving greater than a year, Lee et al. [20] found increased risk of ischemic stroke over 6 years of follow-up after injury.

Together, the existing literature has consistently reported that TBI is associated with an increased risk for stroke, but that the risk of stroke diminishes over time after the TBI and persists regardless of injury severity. The results presented here suggest that the risk of stroke remains increased several years after TBI but only in those with definite TBI (consistent with moderate-severe). The difference in our results may be accounted for by confirming TBI and stroke events by manual record review, which has long been the standard for accuracy in epidemiology [18]. We additionally classified results by TBI severity. Multiple investigators have shown that the accuracy of identifying TBI using diagnostic codes is substantially limited [18, 39, 40]. In the REP, when using CDC-recommended coding for TBI, the CDC approach identified only 40% of cases that were confirmed by manual review. The majority of missed cases were probable and possible in severity, which have the highest incidence [18]. Further, although the use of administrative data to classify stroke has been validated overall [41], Jenkins et al. [16] found that ICD codes for stroke were not sensitive for finding new cerebral infarcts, and coding has been shown to misclassify TBI-associated hemorrhage as a new hemorrhagic stroke. While our findings of an increased risk for stroke after TBI are consistent with what is reported in existing literature, its magnitude and association with increased injury severity specifically after the acute phase after injury are likely to be more accurate given manual data abstraction and confirmation of both TBI and stroke.

The long-term effects of TBI and methods by which TBI influences risk for subsequent stroke become more undefined as time since injury increases, and remain a current topic of heightened interest and debate [4, 42]. TBI is known to be associated with reduced physical activity levels [43, 44], a strong predictor of stroke risk [45]. TBI has also been associated with an increased risk for mood disorders, metabolic syndrome, diabetes mellitus, and substance use [13, 46], all of which increase vascular risk. While the possibility that persistent trauma-associated changes in vascular anatomy, cerebrovascular regulation, coagulation, and metabolism may contribute to this risk, other factors beyond unmodifiable risk associated with aging require further research, as an accurate understanding of modifiable risk factors can help with stroke prevention for those who sustain a TBI [47, 48].

Our study has limitations. Given a higher risk of stroke with increasing age, the age range in this study was restricted to those aged over 40 years. Although the incidence of stroke is generally higher among older age groups, several studies have reported a higher risk of stroke after TBI in those less than 30 [10, 11]. Thus, our results may not be generalizable for those with TBI younger than age 40. While we adjusted for nonhead injury along with a comorbidity index, there may be other factors that were un-accounted for such as level of physical activity and smoking/alcohol use in cases compared to referents. Additionally, while individuals were matched by age and sex, additional confounding factors may be present. Finally, St Sauver et al. [49] found that the age, sex, and ethnic characteristics of Olmsted County residents were similar to those of the state of Minnesota and the Upper Midwest. However, Olmsted County residents were found to be less ethnically diverse and to have a higher average socioeconomic status than the overall US population [49]. Therefore, our results may not be generalizable to more ethnically or socioeconomically diverse populations [30].

Conclusions

In this population-based study, TBI was associated with a long-term risk of stroke using medical record-confirmed exposure and outcome data, which provides a previously unreported high level of accuracy. Further, the association between TBI and stroke increased with greater injury severity. These findings highlight the need to understand the long-term risk factors for stroke after TBI, regularly assess vascular risk after TBI, and implement prevention strategies unique to individuals. Additionally, further research can consider using data of history of TBI in additional stroke prediction algorithms which may help define those patients at highest long-term stroke risk that require more thorough prevention strategies and surveillance.

Data Availability Statement

Datasets generated or analyzed during this study were obtained via the REP database and are available upon reasonable request from the corresponding author.