Traumatic Brain Injury and Early Onset Dementia in Post 9–11 Veterans

Abstract

Objectives:

The objective of this work was to assess TBI-related risks factors for early-onset dementia (EOD).

Background:

Young Post-9/11 Veterans may be at elevated risk for EOD because they experience high rates of traumatic brain injury (TBI) in early/mid adulthood. However, few studies have explored the longitudinal relationship between traumatic brain injury (TBI) and the emergence of EOD subtypes.

Methods:

This matched case-control study used data from the Veterans Health Administration (VHA) to identify Veterans with EOD. To address the low positive predictive value (PPV=0.27) of dementia algorithms in VHA records, we identified Alzheimer’s disease (AD) and frontotemporal dementia (FTD) with PPV of 0.88 and 0.96 as our outcomes. Logistic regression was employed to assess the adjusted risks for dementia subtypes.

Results:

The EOD cohort was comprised of Veterans with AD (n=689) and FTD (n=284). There were no significant demographic differences between the EOD cohort and their matched controls. After adjustment, EOD odds ratio increased given prior history of TBI (OR: 3.05, 2.42–3.83). Other significant prior exposures included epilepsy (OR: 4.8, 3.3–6.97), other neurological conditions (OR: 2.0, 1.35–2.97), depression (OR: 1.35, 1.12–1.63) and cardiac disease (OR: 1.36, 1.1–1.67).

Conclusion:

Post-9/11 Veterans have higher odds of EOD following TBI. A sensitivity analysis across TBI severity confirmed this trend, indicating that the odds for both AD and FTD increased after more severe TBIs.

Introduction:

Early-onset dementia (EOD) is a rare form of dementia defined by progressive mental deterioration prior to age 65, with a prevalence estimated in the civilian population of <1% (1,2). EOD can have genetic underpinnings but has also been linked to preventable causes like traumatic brain injury (TBI) (3–11). There are several potential mechanistic links between TBI and the development of dementia (12,13). Both animal models and human postmortem studies have found that elevated abnormal tau, amyloid beta and TDP-43 can persist for years after TBI, and these are potential markers of neurodegeneration (14,15,16).

However, few studies have explored the risk for EOD across the gradient of TBI severity, although Barnes, et al. found the risk for dementia (all ages) was elevated even after mild TBI without loss of consciousness (17). While mTBI is well documented as a risk factor for chronic traumatic encephalopathy (CTE) (18), there is mixed evidence linking mTBI exposure to subsequent dementia diagnosis (19). Even less is known about the association between TBI of varying severity and specific EOD subtypes, including frontotemporal dementia (FTD), Alzheimer’s disease (AD), vascular dementia (VaD), and Lewy body dementia (LBD). The recent NIH summit on Alzheimer’s Disease and Related Disorders had a specific emphasis on AD and FTD, both of which have significant lifelong consequences that are magnified in those with EOD (20). In this work, we specifically focus on early onset FTD and AD due to their frequency and excellent positive predictive value (PPV) in Veterans Health Administration (VHA) records (21). Less common forms of EOD such as early onset VaD were not considered due to low PPV.

Our prior work demonstrated that the International Classification of Diseases (ICD) codes used by VA to identify dementia in older patients were not accurate when used in patients under the age of 65 (22). Marceaux et al. (21) demonstrated that VHA administrative data has an EOD positive predictive value (PPV) of just 27.3%, which means most Veterans identified by these algorithms did not have EOD confirmed by clinicians evaluating the patient. However, the identification of some specific subtypes was more favorable, and the PPV for codes specific to FTD (PPV=0.96) and AD (PPV=0.88) were highly accurate. Thus, identification of risk factors for these specific types of EOD is feasible using national health system data.

In this work, we explored the longitudinal relationship between TBI-related exposures and downstream risk of EOD in a Post 9/11 Veteran cohort using a matched case-control study. We hypothesized that any history of TBI (regardless of severity) represents a significant risk factor for EOD, and that this relationship would remain statistically significant after adjusting for sociodemographic factors and associated medical comorbidities. Consistent with prior research that found TBI was associated with increased risk of FTD (6,23), we hypothesized that Veterans with TBI would have higher odds of FTD specifically.

Methods

After obtaining approval by institutional review boards (IRB) with a waiver of informed consent from the University of Utah and the Department of Defense (DoD) Human Research Protection Office, we linked national VA and DoD health system data (inpatient, outpatient, pharmacy) from the DoD and VA Infrastructure for Clinical Intelligence using an encrypted identifier. We also used data from the VA comprehensive TBI evaluation (CTBIE) and screening assessments to assist in the identification of TBI. Measures of cognitive reserve were not considered in this study because they were unavailable in the administrative record.

Cohort Development

From the cohort of Veterans who served after September 11, 2001, we identified a subset of Veterans with at least 2 years of care within the VA during fiscal years (FY) 2002–2018, who also received DoD care for at least two years between FY2000-FY2019 (N=1,226,253). To develop the EOD cohort, we used algorithms to identify all Veterans within the full cohort with AD (PPV=0.88) or FTD (PPV=0.96) who were diagnosed at age <=65 (n=973). We then matched each member of the EOD group to similar controls using a nearest neighbor algorithm that paired individuals based on sociodemographic factors and military history. Each sociodemographic variable (see Table 1) was identified first using DoD data, followed by VA data if DoD data were not available or missing.

Table 1:

Comparison of sociodemographic factors and medical history across study groups.

	Percentage	Full cohort N=1.22×106	MC n=973	EOD n=973	EOAD n=689	EOFTD n=284	p value MC v. EOD
Demographics	Age, years	41.7	54.7	54.0	56.3	51.6	>0.5
	Female	17	15.6	15.6	17.4	11.3	>0.5
	Married	48.9	65.3	65.1	66.6	61.6	>0.5
	Retired	5.3	10.3	21.3	22.9	17.6	>0.5
	SVC Disability	46.5	52.2	69.4	58.6	52.9	>0.5
Race/Ethnicity	White	59.8	62.3	62.8	58.9	72.2	>0.5
	Black	18.0	18.1	18.3	19.7	14.8	>0.5
	Hispanic	10.0	9.9	10.3	12.3	5.3	>0.5
	Other	14.0	9.0	8.6	8.9	7.7	>0.5
Military	Deployed	69.9	51.5	51.6	48.3	59.5	>0.5
TBI	TBI, any	26.5	11.8	36.3	28.8	54.9	<0.001 ⋆
Cardiovascular	Stroke	4.1	7.5	16.9	14.8	22.2	<0.001 ⋆
	Cardiac	11.1	20.2	32.1	32.1	32.0	<0.001 ⋆
	Smoking	47.1	43.7	48.6	49.6	46.1	0.03
	Diabetes	8.9	11.7	14.4	14.8	13.4	0.06
	Hypertension	28.9	46.2	48.8	50.2	45.4	0.24
	Cholesterol	38.5	55.0	56.5	56.7	56.0	>0.5
	Obesity	31.7	35.9	34.3	31.6	40.8	>0.5
Mental Health	Anxiety	35.3	27.0	39.2	35.7	47.5	<0.001 ⋆
	Depression	42.6	30.8	48.6	44.7	58.1	<0.001 ⋆
	Attention	7.7	3.9	8.1	6.5	12.0	<0.001 ⋆
	PTSD	37.6	19.0	29.5	24.9	40.8	<0.001 ⋆
	Subst. abuse	35.6	19.1	22.9	20.6	28.5	0.037
	Suicide attempt	4.0	1.0	2.1	1.5	3.5	0.07
Neurological	Epilepsy	2.5	1.6	11.3	8.9	17.3	<0.001 ⋆
	Other neural	2.3	2.1	17.3	14.5	24.3	<0.001 ⋆
Chronic Disease	Kidney disease	2.4	3.2	6.3	6.1	7.0	<0.001 ⋆
	Liver disease	9.9	1.5	6.3	6.1	7.0	0.02
Post-concussive	Headache	30.2	31.9	45.4	42.4	52.8	<0.001 ⋆
symptoms	Memory loss	3.9	3.9	24.8	23.1	29.2	<0.001 ⋆
	Insomnia	26.2	27.3	33.5	28.9	44.7	0.004

(MC: Matched Control, EOD: All early-onset dementia, EO AD: Early-onset Alzheimer’s disease, EO FTD: Early-onset Frontotemporal dementia, SVC disability: Service-connected disability, Cholesterol: Hypercholesterolemia. While the full cohort differed significantly all dementia groups, there were no significant difference in demographic characteristics, deployment, or race/ethnicity after matching (MC vs. EOD).

(⋆ indicates significant difference between the EOD group and matched control group, testing for paired proportions at p<0.05 with Bonferroni familywise correction.)

Measures and Index Date

Early-Onset Dementia

Early-Onset Dementia was identified using diagnoses from inpatient and outpatient data. Cases included Veterans with two or more dementia codes documented at least 7 days apart (12), that included at least one code for AD (ICD-9: 331.0; ICD10: G30.9) or FTD (ICD-9: 331.1; ICD10: G31.0). We identified the index date as the date of the first recorded dementia diagnosis for those with EOD. For those without EOD, a date of dementia diagnosis is not available. Instead, we simulated index dates using a Monte Carlo method to model age-correlated dates of first diagnosis in the EOD cohort and applied the model to simulate index dates for matched controls, with excellent agreement (see appendix 1). To minimize temporal confounding, where definite dates were available, only medical history preceding each patient’s index date by >1 year were considered.

Traumatic Brain Injury

Traumatic Brain Injury was identified using ICD-9/10 codes from DoD and VA data, as well as self-report on the CTBIE using a process implemented previously for the Chronic Effects of Neurotrauma Consortium (24), (self-reported loss/alteration of consciousness [LOC/AOC] and post-traumatic amnesia [PTA], self-reported exposure to penetrating head injury followed by ICD 9/10 codes where self-report data was not available). As the timing between the actual date of TBI and EOD diagnosis is subject to self-report and noise, we performed a sensitivity analysis to determine the impact of index date shifting on TBI odds ratio, which found no significant variations (see Limitations).

For the primary analysis predicting EOD, TBI was classified as present or not present. For sub-analyses comparing AD and FTD, TBI was also categorized as mild, moderate/severe, or penetrating in accordance with the DoD-VA mTBI guidelines. Moderate and severe TBI were collapsed due to small numbers of severe TBI (25). TBI severity was derived from the CTBIE reported durations for 1. LOC (< 30 minutes → mild; > 30 minutes → moderate/severe). 2. AOC (up to 24 hours → mild; >1 day → moderate/severe), and 3. PTA (up to 24 hours → mild; >1 day → moderate/severe). Penetrating TBI was determined by self-report of an injury that penetrated the skull. For those without CTBIE data, we used ICD 9/10 code algorithms to identify TBI severity that were outlined by the Armed Forces Health Surveillance System (26), which also included unclassified TBI and history of TBI (data not shown).

Comorbid Conditions

Comorbid Conditions commonly associated with dementia or EOD were identified using ICD-9/ICD-10 Codes (see Appendix 2) based on algorithms used in prior research, one year or more prior to the index date (24,27–29). Conditions includes cardiovascular/cerebrovascular disease (smoking history, obesity, hypertension, diabetes mellitus, hypercholesterolemia, stroke/transient ischemic attack), cardiac disease (congestive heart failure, cardiac arrhythmia, valvular heart disease), mental health (anxiety, attention/concentration, post-traumatic stress disorder, suicide attempt, depression), neurological conditions/symptoms (epilepsy, other neurological conditions [atrophic lateral sclerosis, multiple sclerosis, Parkinson’s disease, anoxic brain injury, encephalopathy]), other chronic diseases (renal, pulmonary, and hepatic), and post-concussion symptoms (headache, insomnia, memory loss). Consistent with prior studies of chronic conditions, we required two documented diagnoses at least 7 days apart to reduce identification of false positive cases where a diagnosis was used to “rule out” a condition (30,31).

Statistical analysis

Matching

Matching was performed using a one nearest neighbor (1NN) search based on ball tree partitioning that matched demographic (age, sex, race/ethnicity, employment, marital status) and military (deployment, service-connected disability, retirement from service) characteristics. We declined to approximate a fully randomized experiment for reasons outlined elsewhere (32). McNemar’s test for paired proportions was used to test incidence differences between the comparison and EOD groups at significance p<0.05 with familywise Bonferroni correction.

Simulation

Simulation of diagnosis dates was performed for matched controls. Estimating control diagnosis dates using EOD date-covariate relationships is inadvisable, because the results will depend on the EOD marginal distribution, which will be shifted for those without EOD. Instead, a Monte Carlo simulation generated age-correlated index dates by sampling observed distributions (see Appendix 1).

Logistic Regression

Logistic Regression was implemented to calculate adjusted odds ratios (OR). L2 ridge regression regularization was implemented to encourage conservative results. In the analysis across TBI severity, those with no evidence of TBI were the comparator group. P values and 95% confidence intervals (CI95%) of adjusted odds ratios were calculated using the stats models python package. All analyses were scripted in Python 3.8.

Results

Cohort Analysis

Table 1 presents the demographic characteristics and incidence of medical conditions for five groups: 1. full cohort (n=1,226,253), 2. matched control group (n=973), 3. EOD positive cohort (n=973), 4. Veterans with early-onset AD (n=689), and 5) Veterans with early-onset FTD (n=284). Compared to the full cohort, Veterans with EOD tended to be older, married, white race/ethnicity, retired, and have higher rates of service-connected disability and TBI (Table 1). The EOD cohort also showed higher incidence of most cardiovascular diseases, mental health diagnoses, neurological conditions, and chronic disease compared to the full cohort.

After matching, McNemar tests for paired proportions indicated the EOD cohort did not have significant demographic differences compared the MC group (p>0.5). After matching, several unmatched conditions no longer showed significant differences, including hypercholesterolemia (p=0.51), hypertension (p=0.24), obesity (p = 0.5), and suicide attempt (p=0.07).

To provide a broad visual survey of AD/FTD correlates, the crude odds for AD and FTD compared to MCs were calculated separately and are presented as a scatterplot in Figure 1 for various conditions. Color is used to group the conditions into related categories, and the distance of each point from the 1:1 line indicates the extent to which a condition is more closely associated with FTD or AD. The highest odds for both FTD and AD were epilepsy, other neurological conditions, memory loss (see discussion), and TBI (any severity). Of the four dominant factors, TBI was the most frequently documented condition in the EOD cohort, as represented by its circle size.

Figure 1:

Potential prior indicators of early-onset Alzheimer’s disease and frontotemporal dementia in Veterans. A log-log scatterplot shows the crude odds ratio for Veterans with EO AD vs. matched controls (x-axis) against the crude odds ratio for Veterans with EO FTD vs. matched controls (y-axis). Each prior indicator of EOD is color coded by its category (inset). The size of each data point conveys incidence in the overall EOD positive cohort (100% incidence shown in legend).

Mental health conditions displayed slightly higher odds of FTD than AD (Figure 1: Light blue points, including anxiety, depression, schizophrenia, suicidal attempt, PTSD, and bipolar disorder). Chronic kidney disease showed the highest crude odds of all chronic conditions for both AD and FTD. Prior stroke presented the highest odds of FTD and AD among the metabolic/vascular conditions tested.

Adjusted Odds for Early Onset Dementia

To explore adjusted relationships between TBI and EOD (vs. matched controls), a larger logistic model was created, accounting for a wide range of sociodemographic factors and associated medical comorbidities (Figure 2). Sociodemographic adjustments were undertaken to mitigate correlations between sociodemographic characteristics and cognitive decline. The adjusted odds ratios for EOD are shown in Figure 2 sorted by rank. TBI of any specified severity was significantly associated with EOD (OR: 3.05, CI95%: 2.42 – 3.83, p<0.001) with respect to matched controls. Consistent with crude findings from Figure 1, the adjusted model found epilepsy (OR: 4.8, CI95%: 3.3 – 6.97), other neurological diagnoses (OR 2.0, CI95%: 1.35 – 2.97), and cardiac disease (OR: 1.36, CI95%:1.1 – 1.67) were also significant. Although memory loss is not considered an independent risk factor for dementia (see discussion), it was a signal of concern if present >1 year prior to first EOD diagnosis (OR: 4.75, CI95%: 3.33 – 2.42).

Figure 2:

Odds ratios for EOD regressed on sociodemographic and prior clinical exposures. Point estimates and 95% confidence bounds of the adjusted odds ratios for EOD (n=937) are shown compared to the reference group of 1:1 matched controls. For example, the odds ratio of ‘other neurological’ conditions is 2.0, indicating that the prior presence of this diagnosis is associated with a 2:1 increased odds of EOD relative to matched controls. Epilepsy, memory loss, TBI, other neurological conditions, cardiac disease, and depression had the highest odds. ★ indicates significance at p<0.05 and ★★ indicates significance at p<0.002 (Bonferroni correction given = 25 tests).

In an additional analysis, we interrogated the odds of FTD and AD after TBI exposure. The adjusted odds ratios for FTD and AD were calculated using a logistic regression controlling for age, sex, race/ethnicity, employment, marital status, epilepsy, and other neurological conditions (Figure 3), using no TBI as a reference. FTD odds (5.96, CI95%: 3.96 – 8.99) and AD odds (2.26, CI95%: 1.59 – 3.21) were elevated for Veterans with any history of TBI at least 1 year prior to dementia diagnosis. FTD was consistently higher than AD odds after equivalent severity TBI.

Figure 3:

Adjusted odds of FTD and AD after TBI. Point estimates and 95% confidence bounds of the odds ratios for FTD and AD are shown after adjusting for age, sex, race/ethnicity, employment, marital status, epilepsy, and other neurological conditions using logistic regression (left). More severe TBIs were associated with higher odds of both FTD and AD, compared to the baseline of no TBI. The difference in odds ratios across subtypes increased for more severe TBIs (right). Apart from AD after penetrating TBI (p=0.21), all odds were highly significant (p<0.001).

Discussion

This analysis is the first to our knowledge to examine the association between TBI and EOD in a population of Post-9/11 Veterans using dementia diagnoses that are known to have high specificity. Findings generally supported our hypotheses. We found that TBI was significantly associated with both AD and FTD. Epilepsy also showed a strong relationship with EOD, which may be related to the comorbid nature of epilepsy and TBI in young post-9/11 Veterans (33,34). Chronic illnesses including cardiac disease, depression, and other neurological conditions were all associated with elevated risk for EOD, and it may be valuable for future work to explore whether this cluster suggests a vulnerable phenotype of increased neurologic disorder in early life.

Memory loss is not considered an independent risk factor for EOD because it is typically an early symptom of the dementia syndrome. Nonetheless, forgetfulness and recollection difficulties, while nonspecific, are frequent symptoms of many conditions, including post-concussive syndrome (35). Both FTD and TBI frequently involve the frontal lobes, and both are vulnerable to age-related change. Additionally, the documentation of symptom onset after injury may be inaccurately reported by the informant, or incompletely documented. These important aspects are often considered in establishing FTD diagnosis. Therefore, some conflation of FTD with the direct sequelae of TBI may occur, although the high PPV of FTD suggests this is a rare occurrence. Furthermore, these symptom-related diagnoses are typically documented when they are the primary reason for receiving care; thus, individuals with this diagnosis may have more focused issues related to memory. Further analyses that account for the relative proximity of memory loss diagnoses to dementia and TBI diagnosis dates may provide insight.

We also found that the odds of FTD were higher than AD following more severe TBI. This builds on work exploring FTD as a tautopathy linked to TBI, and the work of Kalkonde et al (6), which was a small study that included 63 older Veterans with FTD that revealed 4.4 greater odds of FTD in individuals with a self-reported history of any TBI, compared with those without. Our findings also extend reports of TBI as a risk factor for FTD in a population-based sample of older adults that reported head trauma as an independent risk factor for FTD (3.3 adjusted Odds Ratio) when compared to age and sex-matched individuals. A larger convenience sample analysis found a 1.6 increased OR of a history of remote TBI with extended LOC (≥ 5 minutes) in patients with FTD compared to control (36), although factors such as comorbid mental health problems can delay FTD diagnosis (2). Although the age of our cohort was significantly younger than in previous studies linking remote history of TBI to FTD, the magnitude of the associations we found was within the range of prior reports. Interestingly, other mechanisms common to both conditions, including insufficiency of progranulin and TDP-43 proteinopathy have also been postulated as possible links, whereby TBI may initiate a series of changes that may manifest or exacerbate symptoms of neurodegenerative diseases (37–43).

Although the association between TBI and FTD was stronger than that between TBI and AD in our sample, mechanisms for increased vulnerability to other neurodegenerative disease including amyotrophic lateral sclerosis (ALS), Parkinson’s disease (PD), and chronic traumatic encephalopathy (CTE) following TBI have also been proposed. These include neuroinflammation, nucleocytoplasmic transport deficits mediating TDP-43 pathology, vascular impairment, tau pathology and synaptic dysfunction (44–47).

Overall, our findings suggest the risk profiles for different EOD subtypes show broad similarities but can also exhibit important distinctions. One natural progression of this work is to question whether other dementia subtypes (such as LBD) also display specific associations and risks. In addressing these questions, it is critical to develop and validate algorithms that have high PPV for specific dementia subtypes. Alongside high PPV, our analysis was predicated on the appropriate matching of controls and simulation of matched index dates. Future work exploring subtypes may benefit from these methods, and we provide our codes for matching and index date simulation in appendix 3.

Limitations

This population-based study provides a broad view of the association of TBI, TBI-related sequela, and other chronic disease with EOD. However, genetic information was unavailable in the administrative record, and genetics was not controlled for in this study. Case-control designs using retrospective databases are subject to several potential sources of error, including ascertainment bias, confounding by indication, and time-related biases including the choice of exposure risk window (48).

Ascertainment.

First, the ascertainment of TBI is subject to recall bias and the limitations of self and informant-based report. Additionally, the majority of TBI’s (especially mTBIs) are identified by screening and CTBIE which can occur after a diagnosis of FTD or AD. Second, we have limited information on TBI, including lifetime TBI history (e.g., number, type, mechanism of injury). This study was constrained to DoD and VA data sources, and treatments outside these systems are not included. Assuming that there is no difference in care outside the networks for those with and without EOD, this misclassification bias should result in an underestimate of the true association.

Sensitivity.

Clinically, subtypes such as the FTD behavioral variant might not always be detected by screening for EOD diagnoses. This is because many of the symptoms (e.g., disinhibition, apathy, and compulsive behaviors) are also attributable to comorbid behavioral health issues, which are also more common than EOD (1). Therefore, our EOD cohort may represent a conservative lower bound. Although the PPV of the detection algorithm for early onset AD and FTD were both high, some individuals may have been incorrectly diagnosed, or incorrectly classified. However, both false negatives and false positives would both reduce the effect sizes and differences in odds observed between the control and EOD groups. Therefore, the ORs presented here are likely to be a lower bound on the true values.

Additionally, in the absence of confirmatory biomarkers, such as CSF amyloid beta protein, tau, and phosphotau, or positron emission tomography (PET) studies with amyloid or tau ligands, diagnosis is less accurate than the gold-standard pathologic diagnosis (59–51). Thus, it is likely that some patients diagnosed with FTD or AD may have other neurodegenerative pathologies, such as VaD, and LBD.

Time windowing.

Encouragingly, a sensitivity analysis to determine the impact of windowing with respect to the index date on TBI odds ratios found no differences in our conclusions, regardless of whether TBI dates were strictly time filtered, or if temporal filtering was not performed at all. This is because both the control and exposure TBI rates drop as the included time window narrows, leaving the odds ratio largely unchanged.

Conclusion

Veterans with a history of traumatic brain injury are at increased risk for multiple types of EOD, including AD and FTD, which have distinct clinical presentations and needs. These data recapitulate previous studies linking TBI with EOD, supporting a need for further study to identify additional biomarkers for early identification. TBI severity may play an important role in EOD etiology and diagnosis, and these findings suggest that it is important for clinicians to consider the risk of EOD following TBI among Post 9–11 Veterans. This work lays the foundation for targeted identification, prevention, and interventions for specific EOD subtypes.