Abstract
Background
Covert cerebrovascular disease (CCD) has been shown to be associated with dementia in population‐based studies with magnetic resonance imaging (MRI) screening, but dementia risk associated with incidentally discovered CCD is not known.
Methods and Results
Individuals aged ≥50 years enrolled in the Kaiser Permanente Southern California health system receiving head computed tomography (CT) or MRI for nonstroke indications from 2009 to 2019, without prior ischemic stroke/transient ischemic attack, dementia/Alzheimer disease, or visit reason/scan indication suggestive of cognitive decline or stroke were included. Natural language processing identified incidentally discovered covert brain infarction (id‐CBI) and white matter disease (id‐WMD) on the neuroimage report; white matter disease was characterized as mild, moderate, severe, or undetermined. We estimated risk of dementia associated with id‐CBI and id‐WMD.
Among 241 050 qualified individuals, natural language processing identified 69 931 (29.0%) with id‐WMD and 11 328 (4.7%) with id‐CBI. Dementia incidence rates (per 1000 person‐years) were 23.5 (95% CI, 22.9–24.0) for patients with id‐WMD, 29.4 (95% CI, 27.9–31.0) with id‐CBI, and 6.0 (95% CI, 5.8–6.2) without id‐CCD. The association of id‐WMD with future dementia was stronger in younger (aged <70 years) versus older (aged ≥70 years) patients and for CT‐ versus MRI‐discovered lesions. For patients with versus without id‐WMD on CT, the adjusted HR was 2.87 (95% CI, 2.58–3.19) for younger and 1.87 (95% CI, 1.79–1.95) for older patients. For patients with versus without id‐WMD on MRI, the adjusted HR for dementia risk was 2.28 (95% CI, 1.99–2.62) for younger and 1.48 (95% CI, 1.32–1.66) for older patients. The adjusted HR for id‐CBI was 2.02 (95% CI, 1.70–2.41) for younger and 1.22 (95% CI, 1.15–1.30) for older patients for either modality. Dementia risk was strongly correlated with id‐WMD severity; adjusted HRs compared with patients who were negative for id‐WMD by MRI ranged from 1.41 (95% CI, 1.25–1.60) for those with mild disease on MRI to 4.11 (95% CI, 3.58–4.72) for those with severe disease on CT.
Conclusions
Incidentally discovered CCD is common and associated with a high risk of dementia, representing an opportunity for prevention. The association is strengthened when discovered at younger age, by increasing id‐WMD severity, and when id‐WMD is detected by CT scan rather than MRI.
Keywords: covert brain infarction, covert cerebrovascular disease, dementia, dementia risk, white matter disease
Subject Categories: Computerized Tomography (CT), Imaging, Magnetic Resonance Imaging (MRI)
Nonstandard Abbreviations and Acronyms
- CBI
covert brain infarction
- CCD
covert cerebrovascular diseases
- id‐CBI
incidentally discovered covert brain infarction
- id‐CCD
incidentally discovered covert cerebrovascular diseases
- id‐WMD
incidentally discovered white matter disease
- KPSC
Kaiser Permanente Southern California
- NLP
natural language processing
- WMD
white matter disease
Clinical Perspective.
What Is New?
This is the first cohort study examining the prognostic significance of incidentally discovered covert cerebrovascular disease, comprising covert brain infarction and white matter disease, on future dementia risk.
Overall, the crude hazard ratio of developing dementia in patients with covert cerebrovascular disease was ≈3‐ to 4‐fold compared with those without.
The adjusted hazard ratios were higher in younger patients with covert cerebrovascular disease, in those whose disease was discovered by computed tomography rather than by magnetic resonance imaging, and in those with more severe disease.
What Are the Clinical Implications?
These findings have important implications for future dementia prevention and risk stratification.
Cardiovascular risk factors contribute to the development of several common forms of dementia and are a focus of efforts to prevent these diseases. 1 , 2 Identifying high‐risk individuals before the onset of severe cognitive decline is a central goal in dementia prevention research. One potentially appealing strategy is to identify individuals with covert cerebrovascular diseases (CCD), comprising both covert brain infarction (CBI) and white matter disease (WMD), which are known to be associated with an increased risk for dementia in cohort studies with participants undergoing screening by protocol‐driven neuroimaging. 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10
However, building upon insights from screened cohort studies and translating their findings into real‐world clinical practice has unique challenges. Findings from population‐based research with protocol‐driven magnetic resonance imaging (MRI) screening may not be generalizable to real‐world cohorts, which include only patients selected for clinically indicated neuroimaging, where imaging may be dominated by computed tomography (CT) scans and where neuroimaging interpretation and reporting are heterogeneous and poorly standardized. 11 , 12 Yet studying incidentally discovered CCD (id‐CCD) is impeded by the poor clinical documentation of these lesions; there are no International Classification of Diseases, Ninth Revision (ICD‐9) codes for CBI or WMD and no Tenth Revision (ICD‐10) codes for CBI. Additionally, these lesions are rarely documented in a patient's problem list even when they are reported on neuroimaging reports. Given these barriers to identifying patients with CCD, it is unsurprising that no current standard of care for the management of these conditions or evidence‐based strategies for prevention of dementia following their discovery exists, despite the fact that they are commonly encountered in routine care. 13 , 14
To facilitate identification of individuals with id‐CCD, we previously developed a natural language processing (NLP) algorithm identifying these conditions with a high degree of accuracy. 15 , 16 In this study, we port the NLP algorithm into Kaiser Permanente Southern California (KPSC), a large integrated health care system, to examine the prognostic significance of NLP‐identified, id‐CCD on risk of dementia.
Methods
Environment
The data that support the findings of this study may be used to support new studies in collaboration with qualified researchers with provision of funding, given the approval of the corresponding author, the senior author, and the KPSC research program. The data sets generated and analyzed during the current study are not publicly available because of ethical standards. The authors do not have permission to share data. Authors Wansu Chen and Yichen Zhou had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. The methods of analysis are described herein and are available in greater detail upon request to the authors. In this cohort study, we used health plan enrollees of KPSC, an integrated health care organization that serves 4.8 million individuals (≈19% of the region's population) at 15 hospitals and 230+ medical offices, broadly representative of the residents in the region. 17 Data were extracted, via a research data warehouse, from KPSC's electronic health record. This system integrates all aspects of care, including inpatient, emergency department, outpatient, pharmacy, and lab services, as well as billing and claims. The study protocol was approved by the Tufts Medical Center and KPSC Institutional Review Boards, which waived the need for informed consent as all data were anonymized.
Population
We included individuals aged ≥50 years enrolled in the KPSC health system who received head neuroimaging (CT, MRI) for a nonstroke indication between 2009 and 2019 and who had no history of ischemic stroke or dementia/Alzheimer disease. Patients with transient ischemic attack, hemiplegia, and hemiparesis and other medical conditions were also excluded to ensure a stroke‐ and dementia‐free cohort.
Data S1 includes the complete ICD‐9/ICD‐10/Current Procedural Terminology codes used to exclude patients. For the current analysis, we additionally excluded patients with a visit reason or scan indication suggestive of cognitive symptoms or decline (eg, confusion, disorientation, altered mental status, or dementia evaluation). The top 5 scan indications were headache or facial pain; dizziness, vertigo, or syncope; ear or hearing issue; stroke risk factor; and head trauma or fall (Table S1).
If there were multiple neuroimaging studies, the first was considered the index scan. For neuroimaging evidence of cerebral infarction to be considered “covert,” individuals were included in the study only if they did not acquire a new ICD code for a diagnosis of cerebral infarction or dementia within 60 days after the index scan. Patients who were not actively enrolled in the KPSC health plan on the index date or not continuously enrolled in the prior 12 months were also excluded (a gap of ≤45 days was allowed).
Identification of Patients With CCD
We aimed to identify patients with CCD, comprising either CBI or WMD. Prominent perivascular spaces and microinfarcts were not included, as CBI is defined as a brain lesion presumed to be attributable to vascular occlusion found by neuroimaging (MRI or CT) in patients without clinical manifestations of stroke. WMD implies diminished density of white matter seen on brain CT, which in turn is hyperintense on T2‐weighted, proton‐density, and fluid‐attenuated inversion recovery brain MRI sequences. 18 , 19
An NLP algorithm developed at Mayo Clinic and Tufts Medical Center was applied to neuroimaging reports associated with these index scans to identify individuals with incidentally discovered CBI (id‐CBI) or incidentally discovered WMD (id‐WMD). 15 As described in prior work and Data S2, these algorithms adopted the open‐source NLP pipeline MedTagger for generic NLP processing and task‐specific knowledge engineering (coding of specific words and phrases referencing CBI and WMD) to yield identification of CBI and WMD that was on par with human readers of the neuroimaging reports. 15 After 2 rounds of training on data from Kaiser, this NLP algorithm achieved F‐scores of 0.90 and 0.91 in id‐CBI and id‐WMD on a separate test sample (n=490), with a prevalence of 29 of 490 and 170 of 490 for id‐CBI and id‐WMD, respectively (Data S3). NLP algorithms also captured CBI location (Data S4) and WMD severity (Data S5).
Follow‐Up
Follow‐up started 60 days after the index scan and ended with the earliest of the following events: disenrollment from the health plan, end of the study (December 31, 2019), death, or dementia (outcome). The 60‐day window after the index scan was selected to further exclude patients with symptoms of cognitive decline that might have prompted the neuroimaging scan.
Outcome Definition
The primary outcome of this study was dementia. Based on shared vascular risk factors and symptomatology, we use a broad definition of dementia including Alzheimer disease, vascular dementia, and mixed dementia 20 , 21 (ICD‐9 codes 290.x, 294.20, or 331.0 and corresponding ICD‐10 codes). Positive predictive value has been reported to be >95%. 22 , 23 Only a single code was required if accompanied by a cognitive enhancement medication dispensation, or any 2 codes on different dates (with the second as the event date) if cognitive enhancement medication was not dispensed. Equivalent ICD‐9 and ICD‐10 codes used and qualifying cognitive enhancement medications are shown in Data S6.
Statistical Analysis
Kaplan–Meier plots were used to present dementia‐free survival in patients with and without id‐CBI and with and without id‐WMD, with differences assessed by the log‐rank test. The overall and risk‐factor–stratified crude incidence rates and the 95% CIs were calculated using Poisson regression and reported as per 1000 person‐years of follow‐up time. We examined the crude and adjusted associations of id‐CBI and of id‐WMD with dementia, using Cox proportional hazards regression models. For adjusted effects, we included known cardiovascular risk factors for stroke based on prediction models in the literature, including the following covariates: age, sex, race and ethnicity (non‐Hispanic White; Asian/Pacific Islander; Black; Hispanic; multiple/other/unknown), diabetes, hypercholesterolemia, history of smoking, mean systolic blood pressure (averaged over the prior year, excluding extreme values <70 or >200 to avoid including measurements from periods of critical illness), atrial fibrillation, carotid disease, congestive heart failure, peripheral arterial disease, and use of antiplatelet or statin therapy. 24 , 25 We also included dementia risk factors, including presence of depression, body mass index, and exercise. 26 , 27 , 28 , 29
For each risk factor, the proportional hazards assumption was examined by the Schoenfeld residuals test. Interaction terms were selected on the basis of clinical judgment. In particular, we hypothesized that the effects of id‐WMD and id‐CBI would vary on the basis of imaging modality. We thus compared the effect of the presence versus the absence of id‐CBI or id‐WMD separately in those examined by CT and by MRI. We also anticipated that these lesions might have greater prognostic importance in younger versus older patients and so included interactions with age (age <70 years versus age ≥70 years).
Sensitivity Analysis
Three sensitivity analyses were performed: (1) defining the outcome on the basis of a single dementia diagnostic code (to examine the stability of results when a more sensitive outcome definition is used); (2) excluding patients who were either on an antithrombotic at baseline or who had a clinical indication for antithrombotic therapy (to examine effects among an antithrombotic‐free population who might be considered appropriate for a trial testing “secondary prevention” with an antiplatelet agent); (3) starting follow‐up 1 year after the index scan (instead of 60 days after the scan) (to rule out with greater certainty that the index scan was not ordered because of a suspected diagnosis of cognitive decline or dementia).
Analysis of WMD Severity
To further stratify risk of dementia, we classified patients according to WMD severity using NLP, as discussed in Data S2. Using the description in neuroradiology reports, patients with id‐WMD were classified into 3 severity grades: mild, moderate, or severe. Scan reports with insufficient information on severity were classified as “undetermined.” Analyses were performed in a similar fashion to the main analysis described above. However, id‐WMD severity was interacted with imaging modality and thus treated as 10 different classes (ie, no WMD, mild WMD, moderate WMD, severe WMD, and undetermined WMD severity for CT and for MRI). The reference class for all hazards was those patients who underwent MRI and were found to be free of id‐WMD (ie, the lowest risk group). We underscore that this approach, using a common reference class, is distinct from the main analysis, which contrasted dementia hazards for patients with and without id‐CCD for each modality separately (eg, comparing hazards for dementia among patients with versus without id‐WMD among those undergoing CT and separately among those undergoing MRI, using different reference classes for these contrasts).
Analyses were performed using SAS version 9.4 for Unix (SAS Institute, Cary, NC) and R version 3.6.0 (R Foundation, Vienna, Austria).
Results
A total of 241 050 individuals receiving brain neuroimaging, with a total of 1 049 777 person‐years of follow‐up time, were included in our analysis cohort (Figure S1).
The median follow‐up time was 3.73 years (range, 61 days to 10.83 years; interquartile range, 1.61–6.85 years); 62 479 (25.9%) patients received MRI and 178 571 (74.1%) received a CT scan. CCD was identified in 74 975 (31.1%) including 11 328 (4.7%) with CBI and 69 931 (29.0%) with WMD. There were 11 554 cases of dementia identified in follow‐up, with a median time to event among those with dementia of 2.95 (interquartile range, 1.40–5.16) years. Table 1 describes patient characteristics in the total cohort and in those with CBI (regardless of WMD) and those with WMD (regardless of CBI).
Table 1.
Patient Demographic and Clinical Characteristics at Baseline
| Patient characteristics | Entire cohort | Subset: patients with id‐CBI | Subset: patients with id‐WMD | Subset: patients without id‐CCD |
|---|---|---|---|---|
| N | 241 050 | 11 328 | 69 931 | 166 075 |
| Demographics | ||||
| Age, mean (SD) | 64.9 (10.42) | 72.2 (10.79) | 70.7 (10.77) | 62.3 (9.17) |
| Female, n (%) | 147 812 (61.3) | 6447 (56.9) | 42 659 (61.0) | 102 261 (61.6) |
| Race and ethnicity, n (%) | ||||
| Asian and Pacific Islander | 28 353 (11.8) | 1256 (11.1) | 8030 (11.5) | 19 755 (11.9) |
| Black | 27 305 (11.3) | 1643 (14.5) | 7821 (11.2) | 18 745 (11.3) |
| Hispanic | 78 146 (32.4) | 2869 (25.3) | 17 419 (24.9) | 59 253 (35.7) |
| Multiple/Other/Unknown | 3999 (1.7) | 134 (1.2) | 990 (1.4) | 2951 (1.8) |
| Non‐Hispanic White | 103 247 (42.8) | 5426 (47.9) | 35 671 (51.0) | 65 371 (39.4) |
| Stroke risk factors, n (%) | ||||
| Atrial fibrillation | 14 073 (5.8) | 1314 (11.6) | 6372 (9.1) | 7247 (4.4) |
| Carotid atherosclerosis | 2312 (1.0) | 206 (1.8) | 1125 (1.6) | 1120 (0.7) |
| Congestive heart failure | 11 450 (4.8) | 1214 (10.7) | 5228 (7.5) | 5784 (3.5) |
| Coronary artery disease | 25 793 (10.7) | 2162 (19.1) | 10 324 (14.8) | 14 667 (8.8) |
| Diabetes | 62 537 (25.9) | 3995 (35.3) | 20 385 (29.2) | 40 479 (24.4) |
| Hypercholesterolemia | 163 808 (68.0) | 8505 (75.1) | 51 027 (73.0) | 109 089 (65.7) |
| Hypertension | 145 536 (60.4) | 8941 (78.9) | 49 759 (71.2) | 92 061 (55.4) |
| Peripheral arterial disease | 10 539 (4.4) | 1067 (9.4) | 4749 (6.8) | 5434 (3.3) |
| Tobacco use (ever) | ||||
| Yes | 111 272 (46.2) | 5998 (53.0) | 35 124 (50.2) | 73 552 (44.3) |
| No | 129 005 (53.5) | 5294 (46.7) | 34 672 (49.6) | 91 910 (55.3) |
| Unknown | 773 (0.3) | 36 (0.3) | 135 (0.2) | 613 (0.4) |
| Number of stroke risk factors | 2.3 (1.47) | 2.9 (1.59) | 2.6 (1.52) | 2.1 (1.42) |
| Systolic blood pressure, mean (SD) | 129.3 (13.17) | 132.7 (13.85) | 131.4 (13.09) | 128.4 (13.07) |
| Modality | ||||
| CT | 178 571 (74.1) | 9041 (79.8) | 36 903 (52.8) | 137 121 (82.6) |
| MRI | 62 479 (25.9) | 2287 (20.2) | 33 028 (47.2) | 28 954 (17.4) |
| Antiplatelet use | 10 986 (4.6) | 753 (6.7) | 3928 (5.6) | 6763 (4.1) |
| Statin use | 103 572 (43.0) | 6022 (53.2) | 34 998 (50.1) | 66 023 (39.8) |
| Depression | 47 057 (19.5) | 2189 (19.3) | 13 713 (19.6) | 32 322 (19.5) |
| Exercise (h/wk), mean (SD) | 1.7 (2.53) | 1.5 (2.39) | 1.7 (2.50) | 1.8 (2.54) |
| BMI (kg/m2), mean (SD) | 28.7 (6.04) | 27.9 (5.94) | 27.9 (5.83) | 29.0 (6.09) |
| id‐CBI, n (%) | 11 328 (4.7) | NA | 6284 (9.0) | NA |
| id‐WMD, n (%) | 69 931 (29.0) | 6284 (55.5) | NA | NA |
| id‐CCD,* n (%) | 74 975 (31.1) | NA | NA | NA |
BMI indicates body mass index; CT, computed tomography; id‐CBI, incidentally discovered covert brain infarction; id‐CCD, incidentally discovered covert cerebrovascular disease; id‐WMD, incidentally discovered white matter disease; and MRI, magnetic resonance imaging.
id‐CBI or id‐WMD.
The crude dementia incidence rates (per 1000 person‐years) were 29.4 (95% CI, 27.9–31.0) in patients with id‐CBI and 23.5 (95% CI, 22.90–24.0) for patients with id‐WMD. This compares to an incidence rate of only 6.0 (95% CI, 5.8–6.2) in patients free of cerebrovascular disease (Table 2).
Table 2.
Dementia* Incidence Rate Overall and in Subgroups (n=241 050)
| Patient characteristics | Patients with id=CBI (n=11 328) | Patient with id‐WMD (n=69 931) | Patients with both id‐CBI and id‐WMD disease (n=6284) | Patients without id‐CBI or id‐WMD (n=166 075) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Average follow‐up, y | No. events | Incidence rate (95% CI) | Average follow‐up, y | No. events | Incidence rate (95% CI) | Average follow‐up, y | No. events | Incidence rate (95% CI) | Average follow‐up, y | No. events | Incidence rate (95% CI) | |
| All | 4.28 | 1425 | 29.4 (27.9–31.0) | 4.08 | 6697 | 23.5 (22.9–24.0) | 3.89 | 1016 | 41.5 (39.0–44.1) | 4.46 | 4448 | 6.0 (5.8–6.2) |
| Age group, y | ||||||||||||
| 50–59 | 4.61 | 24 | 3.0 (2.0–4.4) | 4.40 | 57 | 1.0 (0.8–1.3) | 4.40 | 10 | 4.3 (2.1–7.9) | 4.43 | 167 | 0.5 (0.4–0.6) |
| 60–69 | 4.68 | 115 | 8.2 (6.8–9.8) | 4.33 | 559 | 6.4 (5.9–6.9) | 4.29 | 70 | 11.9 (9.4–14.9) | 4.55 | 849 | 3.5 (3.3–3.7) |
| 70–79 | 4.50 | 484 | 29.3 (26.8–32.0) | 4.16 | 2436 | 27.1 (26.0–28.1) | 4.16 | 325 | 34.9 (31.3–38.9) | 4.57 | 1882 | 16.3 (15.5–17.0) |
| 80+ | 3.39 | 802 | 80.3 (74.9–86.0) | 3.33 | 3645 | 73.1 (70.7–75.5) | 3.24 | 611 | 87.6 (80.8–94.7) | 3.87 | 1550 | 49.4 (47.0–51.9) |
| Sex | ||||||||||||
| Female | 4.49 | 876 | 30.3 (28.3–32.3) | 4.19 | 4339 | 24.3 (23.6–25.0) | 4.08 | 619 | 42.7 (39.5–46.2) | 4.56 | 2868 | 6.1 (5.9–6.4) |
| Male | 4.00 | 549 | 28.1 (25.9–30.6) | 3.91 | 2358 | 22.1 (21.2–23.0) | 3.66 | 397 | 39.7 (36.0–43.8) | 4.30 | 1580 | 5.8 (5.5–6.1) |
| Race and ethnicity | ||||||||||||
| Non‐Hispanic white | 4.19 | 752 | 33.0 (30.7–35.5) | 4.09 | 3778 | 25.9 (25.1–26.7) | 3.85 | 567 | 45.8 (42.1–49.7) | 4.53 | 2176 | 7.3 (7.0–7.7) |
| Asian/Pacific Islander | 4.59 | 114 | 19.8 (16.4–23.6) | 4.26 | 555 | 16.2 (14.9–17.6) | 4.10 | 74 | 26.2 (20.7–32.7) | 4.70 | 382 | 4.1 (3.7–4.5) |
| Black | 4.39 | 255 | 35.4 (31.2–39.9) | 4.13 | 928 | 28.8 (26.9–30.6) | 3.90 | 175 | 49.7 (42.7–57.5) | 4.73 | 634 | 7.2 (6.6–7.7) |
| Hispanic | 4.28 | 293 | 23.9 (21.3–26.7) | 3.99 | 1393 | 20.0 (19.0–21.1) | 3.97 | 195 | 35.2 (30.5–40.4) | 4.26 | 1218 | 4.8 (4.6–5.1) |
| Other† | 3.20 | 11 | 25.7 (13.6–44.4) | 3.31 | 43 | 13.1 (9.6–17.5) | 2.66 | 5 | 24.7 (8.0–57.6) | 3.55 | 38 | 3.6 (2.6–4.9) |
| Imaging modality | ||||||||||||
| CT | 4.25 | 1258 | 32.8 (31.0–34.6) | 3.81 | 5307 | 37.7 (36.7–38.7) | 3.76 | 871 | 51.6 (48.3–55.1) | 4.46 | 4067 | 6.7 (6.5–6.9) |
| MRI | 4.40 | 167 | 16.6 (14.2–19.3) | 4.37 | 1390 | 9.6 (9.1–10.1) | 4.24 | 145 | 19.1 (16.2–22.4) | 4.47 | 381 | 2.9 (2.7–3.3) |
CT indicates computed tomography; ICD, International Classification of Diseases; id‐CBI, incidentally discovered covert brain infarction; id‐WMD, incidentally discovered white matter disease; and MRI, magnetic resonance imaging.
Dementia was defined by 2 ICD diagnostic codes or 1 diagnostic code and 1 medication dispensing.
Multiple, other, or unknown race or ethnicity.
Dementia‐free survival in those with and without id‐CBI, and with and without id‐WMD, stratified by modality (CT versus MRI), is shown in Figure 1A and 1B, respectively. These graphs display clearly the higher incidence rates in patients with id‐CCD detected by CT compared with MRI. In patients with id‐CBI, when detected by CT, the annualized incidence rate was 32.8 (95% CI, 31.0–34.5), about twice that when detected by MRI, 16.6 (95% CI, 14.2–19.3). In patients with id‐WMD, when detected by CT, the annualized incidence rate was 37.7 (95% CI, 36.7–38.7), almost 4‐fold than when detected with MRI, 9.6 (95% CI, 9.1–10.1). Overall, the crude hazard ratio (HR) for dementia associated with id‐WMD was 3.72 (95% CI, 3.58–3.86); and with id‐CBI was 2.91 (95% CI, 2.75–3.07) (Table 3), comparing patients with versus without these findings, within each modality.
Figure 1. Kaplan–Meier plot of dementia‐free survival with and without covert brain infarct (A) and with and without white matter disease (B) stratified by imaging modality (CT vs MRI).
Differences in dementia‐free survival were assessed by the log‐rank test. CT indicates computed tomography; id‐CBI, incidentally discovered covert brain infarct; id‐WMD, incidentally discovered white matter disease; and MRI, magnetic resonance imaging.
Table 3.
Crude and Adjusted HRs for Dementia by CBI and WMD
| HR | 95% CI | |
|---|---|---|
| Crude | ||
| id‐CBI | 2.91 | 2.75–3.07 |
| id‐WMD | 3.72 | 3.58–3.86 |
| Adjusted | ||
| id‐CBI | 1.28 | 1.21–1.35 |
| id‐WMD | 1.93 | 1.85–2.00 |
| Adjusted (varied by age and modality) | ||
| id‐CBI | ||
| Age <70 y | 2.02 | 1.70–2.41 |
| Age ≥70 y | 1.22 | 1.15–1.30 |
| id‐WMD | ||
| MRI, age <70 | 2.28 | 1.99–2.62 |
| MRI, age ≥70 | 1.48 | 1.32–1.66 |
| CT, age <70 | 2.87 | 2.58–3.19 |
| CT, age ≥70 | 1.87 | 1.79–1.95 |
CBI indicates covert brain infarction; CT, computed tomography; HR, hazard ratio; id‐CBI, incidentally discovered covert brain infarction; id‐WMD, incidentally discovered white matter disease; MRI, magnetic resonance imaging; and WMD, white matter disease.
The multivariable model included age, sex, race and ethnicity, atrial fibrillation, carotid atherosclerosis, congestive heart failure, coronary artery disease, diabetes, hypercholesterolemia, hypertension, peripheral arterial disease, ever tobacco use, mean systolic blood pressure in 1 year, antiplatelet use, stain use, depression, exercise (h/wk), body mass index (kg/m2), modality, age*CBI, age*id‐WMD, and modality*id‐WMD. Refer to Table S2 for HRs of all the covariates.
In a multivariable model controlling for major cardiovascular and dementia risk factors, we examined the effect of id‐WMD on dementia risk. Among patients undergoing CT scan, the adjusted HR for the presence versus the absence of id‐WMD was 2.87 (95% CI, 2.58–3.19) for those aged <70 and 1.87 (95% CI, 1.79–1.95) for those aged ≥70 years. Among patients undergoing MRI, the adjusted HR for the presence versus the absence of id‐WMD was 2.28 (95% CI, 1.99–2.62) for those aged <70 years and 1.48 (95% CI, 1.32–1.66) for those aged ≥70. The adjusted HR associated with id‐CBI was 2.02 (95% CI, 1.70–2.41) for patients aged <70 years and 1.22 (95% CI, 1.15–1.30) for patients aged ≥70 years regardless of modality.
Table S2 shows the crude and adjusted effects of other risk factors included in our model. In general, the predictive effects of all cardiovascular and dementia risk factors were in the expected direction in the multivariable model. Patients on statins or antiplatelet agents at the time of the index scan had crudely higher risks of dementia, presumably because of confounding by indication, but these effects were nullified after full adjustment.
Sensitivity Analysis
Dementia incidence patterns were largely the same across sensitivity analyses with a roughly uniform relative increase in outcome rates when dementia was established with only a single diagnostic code (Table S3): 35.2 (95% CI, 33.5–36.9) per 1000 person‐years for patients with id‐CBI, 27.9 (95% CI, 27.3–28.6) for patients with WMD, and 7.3 (95% CI, 7.1–7.5) for patients without id‐CBI or id‐WMD. Patients without any indications for antithrombotic therapy had overall lower dementia incidence than the full cohort: 23.8 (95% CI, 22.1–25.5) per 1000 person‐years in those with id‐CBI, 18.4 (95% CI, 17.8–19.0) for those with id‐WMD, and 4.6 (95% CI, 4.5–4.8) for those without id‐CCD (Table S4). Neither incidence rates (Table S5) nor HRs (Table S6) changed substantially when follow‐up started from 1 year after the index scan rather than 60 days.
Analysis of WMD Severity
As shown in Table S7, among patients imaged by CT scan, 21% (36 903/178 571) were found to have id‐WMD. Of these, 55% had mild disease, 10% had moderate disease, and 5% had severe disease; severity was “undetermined” for the remaining 30% of patients. Among patients imaged by MRI, 53% (33 028/62 479) were found to have id‐WMD. Of these, 62% had mild disease, 15% had moderate disease, and 5% had severe disease; severity was underdetermined in the remaining 17% of patients. Within each imaging modality, the incidence of dementia increased monotonically in the expected direction among those with no, mild, moderate, and severe disease. However, as shown in Figure 2 and Table S7, the incidence was substantially higher among those imaged by CT. Incidence ranged from 3.1 (95% CI, 2.8–3.4) per 1000 person‐years among those who were id‐WMD negative by MRI to 64.1 (95% CI, 58.3–70.4) per 1000 person‐years for those with severe id‐WMD as detected by CT scan. Patients with mild disease by MRI had a similar incidence of dementia to patients with no disease on CT, and patients with severe disease on MRI had a similar incidence to patients with mild disease on CT. Adjusted HRs compared with patients who were negative for id‐WMD by MRI ranged from 1.41 (95% CI, 1.25–1.60) for those with mild disease on MRI to 4.11 (95% CI, 3.58–4.72) for those with severe disease on CT scan (Table S7).
Figure 2. Kaplan–Meier plot of dementia‐free survival by modality and white matter disease severity grade.
CT indicates computed tomography; and MR, magnetic resonance.
Discussion
In this large observational cohort with ≈250 000 subjects and over 1 million person‐years of follow‐up, we found that the presence of incidentally discovered, NLP‐identified CCD was strongly associated with an increased risk of future dementia. In patients aged between 50 and 70 years, id‐CBI conferred a 2‐fold increased risk, after adjustment for other risk factors. On CT scan, patients with id‐WMD had an adjusted risk almost 3‐fold that of those without id‐WMD on CT scan; with MRI, the adjusted risk associated with id‐WMD was ≈2‐fold. These risk estimates are comparable to or greater than those for other major risk factors for dementia, including symptomatic ischemic stroke. 30 , 31 , 32 , 33 The size of these effects was reduced in patients aged ≥70 years, emphasizing the importance of early identification and the need for effective midlife prevention strategies. These findings establish that routinely obtained neuroimaging reports carry substantial information that can be used to identify patients at high risk of progressing to dementia.
To our knowledge, this is the first cohort on which the prognostic significance of id‐CCD with regard to future dementia has been studied. A prior study in this cohort found that id‐CBI was a “stroke equivalent” in terms of the risk of future clinically evident stroke. 34 While the findings here are similar to some studies of CCD in screened cohorts, 3 , 35 there are additional insights not previously known. In particular, we found a dramatic increase in the incidence of dementia following CT‐detected id‐CCD compared with MRI‐detected CCD. The incidence of dementia was 2‐fold higher when id‐CBI was detected by CT scan versus MRI and 4‐fold higher when id‐WMD was detected by CT scan versus MRI (Table 2). The difference in risk is presumably attributable to the lower sensitivity of a CT scan, particularly for id‐WMD. Thus, when changes are detected with a CT scan, id‐CCD is generally more advanced. This is consistent with our prior findings in this cohort of a much lower age‐related incidence of id‐WMD in patients imaged by CT scanning compared with MRI. 36
Importantly, in 77% of patients with id‐WMD, we were able to extract prognostically informative disease severity information from routinely obtained neuroimaging reports, which underscored the difference between CT‐ and MRI‐detected id‐WMD. The cohort of patients without WMD by CT scan appears to have a similar risk of dementia as compared with patients with mild WMD on MRI, presumably attributable in part to the presence of undetected WMD within the CT scan cohort, and patients with mild id‐WMD on CT scan have a risk of dementia similar to or greater than those with severe id‐WMD by MRI. Patients with severe WMD on CT have a dementia incidence ≈20‐fold higher than patients free of WMD on MRI, with an adjusted HR of 4, indicating that neuroimaging reports contain substantial information useful for risk stratification extractable by the NLP algorithm.
Patients with id‐CCD may be an attractive target for dementia prevention for several reasons. First, the overall incidence of dementia in these patients is high, well over 2% per year in the current study, which makes it a feasible outcome for a prevention trial. Second, it identifies patients with an elevated risk of dementias attributed to vascular risk factors for which there are already many known effective therapies. Third, the number of patients with id‐CCD appears to be abundant and now identifiable: We found >11 000 patients aged ≥50 years with CBI and ≈70 000 with WMD without prior symptomatic stroke or cognitive decline, including dementia, in a single health system, suggesting that neuroimaging reports might be a repository of a large number of high‐risk patients suitable for trial enrollment. Because of the abundance of these patients, treating id‐CCD with aggressive vascular risk factor reduction may have substantial impact on the population risk of dementia and stroke.
While prevention strategies for dementia are lacking, therapies used for secondary stroke prevention are attractive candidate interventions in this population, given the vascular contributions to the development of many dementias. Antiplatelet medications have not previously been shown to be beneficial for prevention of dementia in a general population but, in theory, may be helpful for patients with vascular brain injury (eg, CCD). 37 In this study, few patients were on prescribed antiplatelets (≈5%), an unknown percentage were on over‐the‐counter aspirin, and about one‐third had other indications for antithrombotic therapy. Given the size of this cohort, it is likely that a large population of patients may be candidates for randomization to antiplatelets in a clinical trial. Similarly, only half of the patients in this cohort were on statins. Finally, besides age, hypertension remains the most consistently shared risk factor between CCD, symptomatic stroke, and dementia, making it a compelling target for intervention among patients with id‐CCD. 38 , 39 , 40 , 41 , 42 Novel agents currently being studied in stroke prevention may have unique benefits or a different risk–benefit balance with id‐CCD. 43 , 44
There are limitations to this study and the proposed approach to patient identification. We focused on neuroimaging findings of id‐CBI and id‐WMD and did not consider other neuroimaging findings that may indicate CCD (eg, microbleeds or perivascular spaces) or increased dementia risk (eg, prominent ventricles and sulci). Patients were selected into this cohort because they had clinically indicated neuroimaging scans. Thus, the patients are not representative of the full spectrum of patients with WMD and CBI. Also, while we were able to control for many of the major risk factors for dementia, the variables within our data set are not exhaustive, and other risk factors may have further attenuated the adjusted effects we estimated. Additionally, a previous study by our team showed that agreement between the NLP algorithm and direct reading of images by expert readers was imperfect. Nevertheless, the agreement was similar or better than that achieved by 2 expert readers on direct image review. 16 We acknowledge that, as these reports were obtained from routine care across multiple, heterogeneous clinical settings, there may be substantial limitations to the consistency of and the degree of details recoverable from these reports.
On the other hand, the use of routinely obtained neuroimaging reports as our source of data may also be viewed as the chief strength of our study. First, in the absence of population screening for CCD, incidentally discovered CCD represents the most clinically relevant population. Further, by using reports obtained as part of routine care, we were able to estimate the prognostic value of imaging findings across diverse settings using heterogeneous imaging modalities without precise and standardized definitions of CBI and WMD, demonstrating the usefulness of routinely obtained neuroimage reports. Interestingly, while MRI would be considered the gold standard to screen for CCD, CT scans comprised the majority of neuroimages included in our study; lesions discovered by CT scan appear to be substantially more prognostic than those discovered by MRI.
Conclusions
id‐CCD including id‐CBI and id‐WMD are common, identifiable with an NLP algorithm, and associated with a high risk for future dementia. Algorithm‐supported identification of patients with these conditions on routinely obtained neuroimaging reports may facilitate dementia prevention studies and risk stratification.
Sources of Funding
This work was funded by a National Institutes of Health grant (R01‐NS102233). The funder had no role in the design/conduct of the study, manuscript preparation, or decision to publish.
Disclosures
The authors have no conflicts of interest to report.
Supporting information
Yichen Zhou is currently located at the AIML, Fractal Analytics, Inc.
Preprint posted on medRxiv February 10, 2022. https://doi.org/10.1101/2022.02.09.22270682
Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.122.027672
For Sources of Funding and Disclosures, see page 10.
REFERENCES
- 1. Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, Brayne C, Burns A, Cohen‐Mansfield J, Cooper C, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet (London, England). 2020;396:413–446. doi: 10.1016/S0140-6736(20)30367-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Langa KM, Foster NL, Larson EB. Mixed dementia: emerging concepts and therapeutic implications. JAMA. 2004;292:2901–2908. doi: 10.1001/jama.292.23.2901 [DOI] [PubMed] [Google Scholar]
- 3. Vermeer SE, Prins ND, den Heijer T, Hofman A, Koudstaal PJ, Breteler MM. Silent brain infarcts and the risk of dementia and cognitive decline. N Engl J Med. 2003;348:1215–1222. doi: 10.1056/NEJMoa022066 [DOI] [PubMed] [Google Scholar]
- 4. Sigurdsson S, Aspelund T, Kjartansson O, Gudmundsson EF, Jonsdottir MK, Eiriksdottir G, Jonsson PV, van Buchem MA, Gudnason V, Launer LJ. Incidence of brain infarcts, cognitive change, and risk of dementia in the general population: the AGES‐Reykjavik study (age gene/environment susceptibility‐Reykjavik study). Stroke. 2017;48:2353–2360. doi: 10.1161/STROKEAHA.117.017357 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Longstreth WT, Manolio TA, Arnold A, Burke GL, Bryan N, Jungreis CA, Enright PL, O'Leary D, Fried L. Clinical correlates of white matter findings on cranial magnetic resonance imaging of 3301 elderly people. The Cardiovascular Health Study. Stroke. 1996;27:1274–1282. doi: 10.1161/01.str.27.8.1274 [DOI] [PubMed] [Google Scholar]
- 6. Carmichael O, Schwarz C, Drucker D, Fletcher E, Harvey D, Beckett L, Jack CR, Weiner M, DeCarli C. Alzheimer's disease neuroimaging initiative. Longitudinal changes in white matter disease and cognition in the first year of the Alzheimer disease neuroimaging initiative. Arch Neurol. 2010;67:1370–1378. doi: 10.1001/archneurol.2010.284 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Moghekar A, Kraut M, Elkins W, Troncoso J, Zonderman AB, Resnick SM, O'Brien RJ. Cerebral white matter disease is associated with Alzheimer pathology in a prospective cohort. Alzheimers Dement. 2012;8:S71–S77. doi: 10.1016/j.jalz.2012.04.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Song IU, Kim JS, Kim YI, Eah KY, Lee KS. Clinical significance of silent cerebral infarctions in patients with Alzheimer disease. Cogn Behav Neurol. 2007;20:93–98. doi: 10.1097/WNN.0b013e31805d859e [DOI] [PubMed] [Google Scholar]
- 9. Wright CB, Festa JR, Paik MC, Schmiedigen A, Brown TR, Yoshita M, DeCarli C, Sacco R, Stern Y. White matter hyperintensities and subclinical infarction: associations with psychomotor speed and cognitive flexibility. Stroke. 2008;39:800–805. doi: 10.1161/STROKEAHA.107.484147 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Liebetrau M, Steen B, Hamann GF, Skoog I. Silent and symptomatic infarcts on cranial computerized tomography in relation to dementia and mortality: a population‐based study in 85‐year‐old subjects. Stroke. 2004;35:1816–1820. doi: 10.1161/01.STR.0000131928.47478.44 [DOI] [PubMed] [Google Scholar]
- 11. Leung LY, Han PKJ, Lundquist C, Weinstein G, Thaler DE, Kent DM. Clinicians' perspectives on incidentally discovered silent brain infarcts ‐ a qualitative study. PLoS One. 2018;13:e0194971. doi: 10.1371/journal.pone.0194971 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Leung LY, Han PKJ, Lundquist C, Weinstein G, Thaler DE, Kent DM. Patients' responses to incidentally discovered silent brain infarcts–a qualitative study. J Patient Rep Outcomes. 2019;3:23. doi: 10.1186/s41687-019-0112-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Smith EE, Saposnik G, Biessels GJ, Doubal FN, Fornage M, Gorelick PB, Greenberg SM, Higashida RT, Kasner SE, Seshadri S. Prevention of stroke in patients with silent cerebrovascular disease: a scientific statement for healthcare professionals from the American Heart Association/American Stroke Association. Stroke. 2017;48:e44–e71. doi: 10.1161/STR.0000000000000116 [DOI] [PubMed] [Google Scholar]
- 14. Wardlaw JM, Debette S, Jokinen H, De Leeuw FE, Pantoni L, Chabriat H, Staals J, Doubal F, Rudilosso S, Eppinger S, et al. ESO guideline on covert cerebral small vessel disease. Eur Stroke J. 2021;6:Iv. doi: 10.1177/23969873211027002 [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
- 15. Fu S, Leung LY, Wang Y, Raulli AO, Kallmes DF, Kinsman KA, Nelson KB, Clark MS, Luetmer PH, Kingsbury PR, et al. Natural language processing for the identification of silent brain infarcts from neuroimaging reports. JMIR Med Inform. 2019;7:e12109. doi: 10.2196/12109 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Leung LY, Fu S, Luetmer PH, Kallmes DF, Madan N, Weinstein G, Lehman VT, Rydberg CH, Nelson J, Liu H, et al. Agreement between neuroimages and reports for natural language processing‐based detection of silent brain infarcts and white matter disease. BMC Neurol. 2021;21:189. doi: 10.1186/s12883-021-02221-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Koebnick C, Langer‐Gould AM, Gould MK, Chao CR, Iyer RL, Smith N, Chen W, Jacobsen SJ. Sociodemographic characteristics of members of a large, integrated health care system: comparison with US Census Bureau data. Perm J. 2012;16:37–41. doi: 10.7812/tpp/12-031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Hachinski VC, Potter P, Merskey H. Leuko‐araiosis: an ancient term for a new problem. Can J Neurol Sci. 1986;13:533–534. doi: 10.1017/s0317167100037264 [DOI] [PubMed] [Google Scholar]
- 19. Hachinski VC, Potter P, Merskey H. Leuko‐araiosis. Arch Neurol. 1987;44:21–23. doi: 10.1001/archneur.1987.00520130013009 [DOI] [PubMed] [Google Scholar]
- 20. Taylor DH Jr, Fillenbaum GG, Ezell ME. The accuracy of medicare claims data in identifying Alzheimer's disease. J Clin Epidemiol. 2002;55:929–937. doi: 10.1016/s0895-4356(02)00452-3 [DOI] [PubMed] [Google Scholar]
- 21. Taylor DH Jr, Østbye T, Langa KM, Weir D, Plassman BL. The accuracy of Medicare claims as an epidemiological tool: the case of dementia revisited. J Alzheimers Dis. 2009;17:807–815. doi: 10.3233/jad-2009-1099 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Quan H, Li B, Saunders LD, Parsons GA, Nilsson CI, Alibhai A, Ghali WA. Assessing validity of ICD‐9‐CM and ICD‐10 administrative data in recording clinical conditions in a unique dually coded database. Health Serv Res. 2008;43:1424–1441. doi: 10.1111/j.1475-6773.2007.00822.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. van de Vorst IE, Vaartjes I, Sinnecker LF, Beks LJ, Bots ML, Koek HL. The validity of national hospital discharge register data on dementia: a comparative analysis using clinical data from a university medical centre. Neth J Med. 2015;73:69–75. [PubMed] [Google Scholar]
- 24. Flueckiger P, Longstreth W, Herrington D, Yeboah J. Revised Framingham stroke risk score, nontraditional risk markers, and incident stroke in a multiethnic cohort. Stroke. 2018;49:363–369. doi: 10.1161/STROKEAHA.117.018928 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Wolf PA, D'Agostino RB, Belanger AJ, Kannel WB. Probability of stroke: a risk profile from the Framingham study. Stroke. 1991;22:312–318. doi: 10.1161/01.str.22.3.312 [DOI] [PubMed] [Google Scholar]
- 26. Yan S, Fu W, Wang C, Mao J, Liu B, Zou L, Lv C. Association between sedentary behavior and the risk of dementia: a systematic review and meta‐analysis. Transl Psychiatry. 2020;10:112. doi: 10.1038/s41398-020-0799-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Chan YE, Chen MH, Tsai SJ, Bai YM, Tsai CF, Cheng CM, Su TP, Chang WH, Chen TJ, Li CT. Treatment‐resistant depression enhances risks of dementia and alzheimer's disease: a nationwide longitudinal study. J Affect Disord. 2020;274:806–812. doi: 10.1016/j.jad.2020.05.150 [DOI] [PubMed] [Google Scholar]
- 28. Barnes DE, Covinsky KE, Whitmer RA, Kuller LH, Lopez OL, Yaffe K. Dementia risk indices: a framework for identifying individuals with a high dementia risk. Alzheimers Dement. 2010;6:138–141. doi: 10.1016/j.jalz.2010.01.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Qizilbash N, Gregson J, Johnson ME, Pearce N, Douglas I, Wing K, Evans SJW, Pocock SJ. BMI and risk of dementia in two million people over two decades: a retrospective cohort study. Lancet Diabetes Endocrinol. 2015;3:431–436. doi: 10.1016/S2213-8587(15)00033-9 [DOI] [PubMed] [Google Scholar]
- 30. Zhou J, Yu JT, Wang HF, Meng XF, Tan CC, Wang J, Wang C, Tan L. Association between stroke and Alzheimer's disease: systematic review and meta‐analysis. Journal of Alzheimer's disease. 2015;43:479–489. doi: 10.3233/JAD-140666 [DOI] [PubMed] [Google Scholar]
- 31. Georgakis MK, Duering M, Wardlaw JM, Dichgans M. WMH and long‐term outcomes in ischemic stroke: a systematic review and meta‐analysis. Neurology. 2019;92:e1298–e1308. doi: 10.1212/WNL.0000000000007142 [DOI] [PubMed] [Google Scholar]
- 32. Kuźma E, Lourida I, Moore SF, Levine DA, Ukoumunne OC, Llewellyn DJ. Stroke and dementia risk: a systematic review and meta‐analysis. Alzheimers Dement. 2018;14:1416–1426. doi: 10.1016/j.jalz.2018.06.3061 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Pendlebury ST, Rothwell PM. Prevalence, incidence, and factors associated with pre‐stroke and post‐stroke dementia: a systematic review and meta‐analysis. Lancet Neurol. 2009;3:1006–1018. doi: 10.1016/S1474-4422(09)70236-4 [DOI] [PubMed] [Google Scholar]
- 34. Kent DM, Leung LY, Zhou Y, Luetmer PH, Kallmes DF, Fu S, Zheng CX, Liu H, Chen W. Association of silent cerebrovascular disease identified using natural language processing and future ischemic stroke. Neurology. 2021;97:e1313–e1321. doi: 10.1212/WNL.0000000000012602 Epub ahead of print [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Debette S, Markus HS. The clinical importance of white matter hyperintensities on brain magnetic resonance imaging: systematic review and meta‐analysis. BMJ. 2010;341:c3666. doi: 10.1136/bmj.c3666 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Leung LY, Zhou Y, Fu S, Zheng CX, Liu H, Chen W, Kent DM. Risk factors for incidentally discovered silent brain infarcts and white matter disease in a real world cohort identified by artificial intelligence. BMC Neurol. 2021;52:AP57. doi: 10.1161/str.52.suppl_1.P57 [DOI] [Google Scholar]
- 37. Jordan F, Quinn TJ, McGuinness B, Passmore P, Kelly JP, Tudur Smith C, Murphy K, Devane D. Aspirin and other non‐steroidal anti‐inflammatory drugs for the prevention of dementia. Cochrane Database Syst Rev. 2020;4:CD011459. doi: 10.1002/14651858.CD011459.pub2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. SPRINT MIND Investigators for the SPRINT Research Group , Williamson JD, Pajewski NM, Auchus AP, Bryan RN, Chelune G, Cheung AK, Cleveland ML, Coker LH, Crowe MG, et al. Effect of intensive vs standard blood pressure control on probable dementia: a randomized clinical trial. JAMA. 2019;321:553–561. doi: 10.1001/jama.2018.21442 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Alosco ML, Sugarman MA, Besser LM, Tripodis Y, Martin B, Palmisano JN, Kowall NW, Au R, Mez J, DeCarli C, et al. A clinicopathological investigation of white matter hyperintensities and Alzheimer's disease neuropathology. J Alzheimers Dis. 2018;63:1347–1360. doi: 10.3233/JAD-180017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Gupta A, Giambrone AE, Gialdini G, Finn C, Delgado D, Gutierrez J, Wright C, Beiser AS, Seshadri S, Pandya A, et al. Silent brain infarction and risk of future stroke: a systematic review and meta‐analysis. Stroke. 2016;47:719–725. doi: 10.1161/STROKEAHA.115.011889 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Fanning JP, Wong AA, Fraser JF. The epidemiology of silent brain infarction: a systematic review of population‐based cohorts. BMC Med. 2014;12:119. doi: 10.1186/s12916-014-0119-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Kuller LH, Longstreth WT, Arnold AM, Bernick C, Bryan RN, Beauchamp NJ. Cardiovascular health study collaborative research group. White matter hyperintensity on cranial magnetic resonance imaging: a predictor of stroke. Stroke. 2004;35:1821–1825. doi: 10.1161/01.STR.0000132193.35955.69 [DOI] [PubMed] [Google Scholar]
- 43. Smith EE, Markus HS. New treatment approaches to modify the course of cerebral small vessel diseases. Stroke. 2020;51:38–46. doi: 10.1161/STROKEAHA.119.024150 [DOI] [PubMed] [Google Scholar]
- 44. Bath PM, Wardlaw JM. Pharmacological treatment and prevention of cerebral small vessel disease: a review of potential interventions. Int J Stroke. 2015;10:469–478. doi: 10.1111/ijs.12466 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Liu H, Bielinski SJ, Sohn S, Murphy S, Wagholikar KB, Jonnalagadda SR, KE KER, Wu ST, Kullo IJ, Chute CG. An information extraction framework for cohort identification using electronic health records. AMIA Summits Transl Sci Proc. 2013;2013:149–153. [PMC free article] [PubMed] [Google Scholar]
- 46. Ferrucci D, Lally A. UIMA: an architectural approach to unstructured information processing in the corporate research environment. Nat Lang Eng. 2004;10:327–348. [Google Scholar]
- 47. Fu S, Leung LY, Raulli A‐O, Kallmes DF, Kinsman KA, Nelson KB, Clark MS, Luetmer PH, Kingsbury PR, Kent DM, et al. Assessment of the impact of EHR heterogeneity for clinical research through a case study of silent brain infarction. BMC Med Informatics Decis Mak. 2020;20:1–12. doi: 10.1186/s12911-020-1072-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Fu S, Chen D, He H, Liu S, Moon S, Peterson KJ, Shen F, Wang L, Wang Y, Wen A, et al. Clinical concept extraction: a methodology review. J Biomed Inform. 2020;109:103526. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Van Rijsbergen CJ. The Geometry of Information Retrieval. Cambridge University Press; 2004. doi: 10.1017/CBO9780511543333 [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.