Predictors for Late Post-Intracerebral Hemorrhage Dementia in Patients with Probable Cerebral Amyloid Angiopathy

Abstract

Background and Objective:

Cerebral amyloid angiopathy (CAA) accounts for the majority of lobar intracerebral hemorrhage (ICH); however, the risk factors for dementia conversion after ICH occurrence in CAA patients are unknown, especially in the long-term period after ICH. Therefore, we aimed to unravel the predictors for late post-ICH dementia (6 months after ICH event) in probable CAA patients.

Methods:

From a large consecutive MRI prospective cohort of spontaneous ICH (2006–2017), we identified probable CAA patients (modified Boston criteria) without dementia 6 months post-ICH. Cognitive outcome during follow-up was determined based on the information from standardized clinical visit notes. We used Cox regression analysis to investigate the association between baseline demographic characteristics, past medical history, MRI biomarkers, and late post-ICH dementia conversion (dementia occurred after 6 months).

Results:

Among 97 non-demented lobar ICH patients with probable CAA, 25 patients (25.8%) developed dementia during a median follow-up time of 2.5 years (IQR 1.5–3.8 years). Pre-existing mild cognitive impairment, increased white matter hyperintensities (WMH) burden, the presence of disseminated cortical superficial siderosis (cSS), and higher total small vessel disease score for CAA were all independent predictors for late dementia conversion.

Conclusion:

In probable CAA patients presenting with lobar ICH, high WMH burden and presence of disseminated cSS are useful neuroimaging biomarkers for dementia risk stratification. These findings have implications for clinical practice and future trial design.

INTRODUCTION

Cerebral amyloid angiopathy (CAA) is a cerebral small vessel disease (SVD) characterized by progressive deposition of amyloid-β (Aβ) in the walls of cortical and leptomeningeal arterioles. Lobar intracerebral hemorrhage (ICH) and cognitive impairment are the two most common clinical presentations which allow for the identification of patients with CAA [1]. The contribution of CAA to the causation of dementia has been well established in patients with hereditary cerebral hemorrhage with amyloidosis-Dutch type, supported by the association between the amount of quantified CAA burden and the presence of dementia independent of neurofibrillary pathology, plaque density, or age [2]. Increasing evidence has demonstrated the association between ICH at all locations and cognitive impairment/dementia [3]. Moreover, patients with lobar ICH have a higher risk of post-ICH dementia compared to non-lobar ICH patients [4, 5]. Although CAA accounts for the majority of lobar ICH [6], clinical and/or neuroimaging markers that predict post-ICH dementia in patients with probable CAA have not been fully defined.

Previous studies have separated the follow-up phases after ICH incidence into early (within 6 months after ICH event) and late stages (6 months after ICH event) [4, 5]. ICH characteristics of location and volume were revealed to predict early dementia conversion. By contrast, the underlying SVD burden was predictive for late dementia after ICH [4]. The discrepant risk factors for dementia conversion at different post-ICH stages suggested heterogeneous biological mechanisms leading to post-ICH dementia at different disease phases. For patients who preserved cognitive functioning in the early post-ICH phase, the risk for developing late dementia was still high (5.8% per year) [4]. Therefore, it is of great clinical importance to identify these ICH survivors with high risk of developing late post-ICH dementia. However, it is still unknown which risk factors have predictive value for late post-ICH dementia in probable CAA patients.

Therefore, we aimed to 1) investigate the cumulative rate of late post-ICH dementia in patients with probable CAA; and 2) identify clinical/neuroimaging biomarkers predicting late post-ICH dementia in patients with probable CAA.

METHODS

Study population

Figure 1 shows the flow chart of enrollment. A total of 204 patients with probable CAA according to the Modified Boston Criteria were identified from an established large prospective cohort of consecutive patients admitted to the Massachusetts General Hospital with spontaneous symptomatic ICH between January 2006 and July 2017 [7]. The inclusion criteria of the cohort were 1) age ≥55 years old; 2) lobar location of ICH; 3) MRI results demonstrating a) microbleed(s) restricted to lobar, cortical or corticosubcortical regions (cerebellar microbleeds allowed) and/or b) focal or disseminated cortical superficial siderosis (cSS); 4) absence of other causes of hemorrhage or cSS. Patients with deep locations of ICH/microbleeds (except cerebellar microbleeds) and patients with inadequate quality of MRI scans were excluded. Since we focused on the delayed post-ICH dementia conversion, we excluded patients with baseline dementia before ICH event (n = 23), patients who died (n = 45) or who became demented within 6 months after ICH (n = 8), as well as patients without follow-up data on cognitive status (n = 31). A total of 97 patients were included into the final analysis. The demographic characteristics, medical history, and cognitive status of patients were prospectively collected at the hospital visit through in-person interview with the patients or surrogates. The status of cognition was categorized into normal, mild cognitive impairment (MCI), and dementia. The diagnosis of MCI was determined when patients had impaired cognition but preserved independence in functional abilities [8, 9]. The diagnosis of dementia was established if the cognitive decline significantly interfered with the function at work or in usual daily activities for the patients. These diagnostic criteria were in accordance with the core clinical criteria for all-cause MCI and dementia as part of the National Institute on Aging and Alzheimer’s Association workgroup criteria [10, 11]. Educational level was categorized according to years of education, ranging from 0 to 4 (0, none; 1, 1–6 years; 2, 7–9 years; 3, 10–13 years; 4, 14+ years). This study was approved by the institutional review boards from our institution.

Fig. 1.

Enrollment flowchart. CAA, cerebral amyloid angiopathy; ICH, intracerebral hemorrhage.

Neuroimage acquisition and analysis

ICH volume was calculated from the CT scan taken at the day of acute hospital admission for the index ICH as previously described [12]. Clinical 1.5 Tesla MRI (Siemens Healthcare, Magnetom Avanto, Erlange, Germany) scans within 2 weeks to the index ICH were used for neuroimage analysis. The MR sequences included whole brain T1-, T2-, and T2*-weighted gradient echo and T2-fluid attenuated inversion recovery (FLAIR; TR/TE 10000/140 ms, inversion time 2,200 ms, 5 mm slice thickness, 1 mm inter-slice gap).

Trained raters blinded to all clinical information evaluated the neuroimaging markers, including the presence of lacunes, location and number of microbleeds, cSS, white matter hyperintensity (WMH) volume and severity, presence and severity of enlarged perivascular spaces (EPVS), and cortical atrophy according to STRIVE criteria [13]. Lacunes were defined in FLAIR sequences as round or ovoid fluid-filled cavities between 3 to 15 mm in diameter. GRE sequences were used to evaluate microbleeds and for cSS measurement. cSS status was categorized into absence, focal (restricted to ≤3 sulci), or disseminated (affecting 4 or more sulci) [7]. WMH volumes were evaluated on T2 axial FLAIR sequences using a semi-automated segmentation method with MRIcron software (University of South Carolina, USA) as previously described [14]. Total brain volumes (TBV) were obtained based on FLAIR images, by non-linearly registering it to an age-appropriate FLAIR template [15] using Advanced Normalization ToolS (ANTS) [16]. A brain mask was manually created on the template, transformed into subject space, and manually assessed and corrected where necessary. The WMH volume were then normalized by TBV before further analysis to account for differences in brain volume. Periventricular and deep WMH were also classified using the 0–3 Fazekas scale [17], and the two scores were summed to determine total severity of WMH of each subject [18]. EPVS in the centrum semiovale (CSO-EPVS) was measured using a validated 4-point visual rating scale [19]. The individual scores (0–3) of cortical atrophy evaluated using T1 sequences from five individual regions (frontal, temporal, parietal, occipital lobes, and insular cortex) were summed to obtain a total score for cortical atrophy [20]. The pathologically validated total SVD score for CAA was calculated based on the four individual SVD markers (CMBs, cSS, WMH, and CSO-EPVS) [21]. Specifically, for the total score, 2–4 CMBs, focal cSS, CSO-EPVS ≥20, Fazekas ≥2 in deep WMH or Fazekas 3 in periventricular WMH accounted for 1 point respectively, whereas >4 CMBs and disseminated cSS accounted for 2 points.

Cognitive status on follow-up

Our main outcome event was conversion to dementia during follow-up. Standardized notes from clinical follow-up visits were systematically reviewed to determine the cognitive status during follow-up using standardized data collection forms. The information collected in the forms includes whether 1) there was cognitive decline from the patient’s previous level of cognitive function, and 2) whether patients still worked or conducted activities they did prior to presentation. If there was a change in cognitive function, the reasons for the change and the resources of how this information was obtained were noted. The diagnosis of dementia was made if patients demonstrated incapacity to perform in their work or in their usual daily activities [11]. Patients with normal cognition or MCI at follow-up were classified as non-converters, while those who developed dementia were classified as converters. We also recorded the occurrence of ischemic or hemorrhagic stroke before the date of the diagnosis of dementia for converters or during follow-up period for non-converters.

Statistics

Time to event was calculated from date of index ICH until the date of dementia diagnosis. Data were censored at the last reliable follow-up time for non-convertors in the survival analysis. We conducted univariable Cox proportional hazards analysis to calculate unadjusted hazard ratios (HR) for the risk of dementia conversion of each potential clinical and neuroimaging predictor. The potential neuroimaging predictors included the presence of lacunes, WMH volume and Fazekas score, lobar CMB, cSS, CSO-EPVS, and total SVD scores for CAA. The potential clinical predictors included age, sex, education, history of hypertension, hypercholesterolemia, diabetes mellitus, and history of ischemic/hemorrhagic stroke. Variables with p < 0.1 in univariable analysis were included in multivariable Cox proportional hazards models. Since the total SVD score for CAA is composed of four individual neuroimaging markers, separated multivariable models were performed including either the individual score component or the total score. We implemented the stepwise backward selection model to generate the best minimal model by eliminating non-significant variables (p > 0.05). Multicollinearity was assessed using variance inflation factors (VIF) and predictors with VIF >2.5 were removed from the model.

The SPSS 21 statistical package was used for statistical analysis (IBM Corp., Armonk, NY). Significance level was set at 0.05 for all analyses.

RESULTS

Table 1 summarizes the baseline characteristics of our study population. Among 97 non-demented post-ICH patients with probable CAA, 25 patients developed dementia (25.8%) during a median follow-up time of 2.5 (IQR 1.5–3.8) years, leading to a cumulative incidence rate of 37.4% (95% CI: 22.9%–51.9%) at 5 years. Eleven patients had pre-existing MCI before the index ICH. Figure 2 shows the comparison of cumulative incidence of dementia between patients with pre-existing MCI and patients with normal cognition before the index baseline ICH (HR 5.854, 95% CI: 2.376–14.423, p < 0.001).

Table 1.

Baseline characteristics of non-demented probable CAA patients with lobar ICH

Variables	Value (n = 97)
Demographic characteristics
Age (y)	73.92±8.73
Sex (female)	43 (44.3%)
Education (score)	4 (3, 4)
Previous medical history
Hemorrhagic stroke	9 (9.3%)
Hypertension	69 (71.1%)
Diabetes	12 (12.4%)
Hypercholesterolemia	52 (53.6%)
Pre-existing MCI	11 (11.3%)
ICH characteristics
ICH volume (cc)	23.18±21.86
IVH involvement	18 (18.6%)
Neuroimaging biomarkers
Lacunes (presence)	25 (25.8%)
CMBs (n)	2 (1, 6.5)
CMBs≥5	27 (27.8%)
WMH volume (cc)	17.88 (9.24, 29.82)
TBV volume (cc)	1447.25±155.77
WMH* (Fazekas score)	3 (2, 4)
CSO-EPVS (score)	3 (2, 3)
cSS (presence)	25 (25.8%)
cSS burden
Focal	19 (19.6%)
Disseminated	13 (13.4%)
Global cortical atrophy	6 (4, 9)
SVD score for CAA	2 (1, 3)
SVD score for CAA (≥3 versus <3)	41 (42.3%)

Data with normal distribution were shown as Mean ± SD. Nominal data were displayed with the number and percentage. Ordinal data were shown as median [25%, 75%]. ICH, intracerebral hemorrhage; MCI, mild cognitive impairment; IVH, intravetricular hemorrhage; CMBs, cerebral microbleeds; WMH, white matter hyperintensity; TBV, total brain volume; cSS, cortical superficial siderosis; CSO-EPVS, centrum semiovale enlarged perivascular space; SVD, small vessel disease; CAA, cerebral amyloid angiopathy.

^*Fazekas scores of periventricular and deep WMH were summed up.

Fig. 2.

Kaplan-Meier plot of cumulative incidence of delayed dementia in post-ICH patients with probable CAA. ICH, intracerebral hemorrhage; MCI, mild cognitive impairment.

In the univariable analyses, older age, lower education level, pre-existing MCI, lobar CMBs ≥5, higher WMH burden (determined by Fazekas score or WMH/TBV ratio), disseminated cSS, and higher total SVD score for CAA were associated with higher unadjusted hazard ratios of dementia conversion (p < 0.1) (Table 2). Due to the small number of dementia conversion outcome events, we implemented two separate multivariable models to determine independent predictors, including either clinical or neuroimaging factors identified in the univariable analysis. In the model including only clinical features (age, education, and MCI), only pre-existing MCI independently predicted dementia conversion (HR 4.617, 95% CI:1.848–11.536, p = 0.001). In the model including only neuroimaging biomarkers, only the presence of disseminated cSS and higher WMH burden independently predicted dementia conversion. In a sensitivity analysis, combining MCI status together with the presence of disseminated cSS and WMH burden in a single model, all three factors have independent predictive value for dementia conversion (p < 0.05). Furthermore, we included the total SVD score for CAA (representing the cumulative SVD burden) together with MCI status before ICH, in a backward stepwise model. Both MCI and total SVD score for CAA ≥3 demonstrated significant independent predictive value for dementia conversion (Table 3).

Table 2.

Univariable analysis of risk factors for delayed post-ICH dementia conversion in patients with probable CAA

Variables	HR	95% CI	p
Age (per year)	1.052	1.001–1.106	0.047
Sex (female versus male)	1.582	0.705–3.548	0.266
Education (score)	0.563	0.328–0.968	0.038
Pre-existing MCI (yes versus no)	5.854	2.376–14.423	<0.001
Hemorrhagic stroke (yes versus no)	2.151	0.729–6.353	0.166
Hypertension (yes versus no)	1.082	0.465–2.518	0.854
Diabetes (yes versus no)	0.94	0.279–3.175	0.921
Hypercholesterolemia (yes versus no)	1.907	0.820–4.436	0.134
Event (follow-up, yes versus no)	0.745	0.174–3.186	0.691
ICH volume (cc) (per cc)	0.987	0.963–1.011	0.272
IVH involvement (yes versus no)	1.062	0.398–2.833	0.905
Lacunes (yes versus no)	0.676	0.252–1.816	0.437
CMBs (number)	0.997	0.990–1.005	0.521
CMBs≥5 (yes versus no)	2.051	0.908–4.631	0.084
WMH/TBV (%)	1.480	1.175–1.864	0.001
WMH (Fazekas score)	1.509	1.104–2.063	0.010
CSO-EPVS (score)	0.862	0.524–1.418	0.558
cSS (yes versus no)	2.300	1.045–5.064	0.039
cSS burden
Focal cSS versus no cSS	1.763	0.675–4.609	0.247
Disseminated cSS (yes versus no)	3.590	1.271–10.135	0.016
Global cortical atrophy	1.049	0.945–1.164	0.368
SVD score for CAA (≥3 versus <3)	2.858	1.263–6.468	0.012

Age, pre-existing MCI, CMB ≥5, WMH volume, WMH (Fazekas score), presence of disseminated cSS, and SVD score for CAA (≥3 versus <3) were potential risk factors for dementia conversion (p < 0.1). HR, hazard ratio; ICH, intracerebral hemorrhage; MCI, mild cognitive impairment; IVH, intravetricular hemorrhage; CMBs, cerebral microbleeds; WMH, white matter hyperintensity; cSS, cortical superficial siderosis; CSO-EPVS, centrum semiovale enlarged perivascular space; SVD, small vessel disease; CAA, cerebral amyloid angiopathy.

^*Fazekas scores of periventricular and deep WMH were summed up.

Table 3.

Multivariable analysis of risk factors for delayed post-ICH dementia conversion in patients with probable CAA

Variables	HR	95% CI	p
Model 1 (demographic factors)
Age (per year)	1.042	0.989–1.098	0.123
Education (score)	0.66	0.361–1.208	0.178
MCI (yes versus no)	4.617	1.848–11.536	0.001*
Model 2 (neuroimaging factors)
Disseminated cSS (yes versus no)	3.115	1.030–9.424	0.044*
WMH (Fazekas score)	1.439	1.036–2.001	0.03*
Lobar CMBs≥5	1.105	0.431–2.834	0.835
WMH/TBV (%)#	1.480	1.157–1.891	0.002*
Model 3 (sensitivity analysis: factors combined)
MCI (yes versus no)	5.880	2.250–15.367	<0.001*
Disseminated cSS (yes versus no)	3.275	1.129–9.499	0.029*
WMH (Fazekas score)	1.427	1.057–1.927	0.02*
WMH/TBV (%)#	1.426	1.112–1.830	0.005*
Model 4 (sensitivity analysis: factors combined)
MCI (yes versus no)	6.043	2.401–15.212	<0.001*
SVD score for CAA≥3	2.961	1.278–6.861	0.011*

Model 1 and model 2 included demographic factors and neuroimaging factors separately, and factors survived from model 1 and model 2 were then included in model 3. Model 4 used SVD score for CAA to replace individual neuroimaging markers.

^#WHM/TBV (%) was included in the model to replace WHM Fazekas score.

^*with statistical significance.

HR, hazard ratio; MCI, mild cognitive impairment; CMBs, cerebral microbleeds; WMH, white matter hyperintensity; cSS, cortical superficial siderosis; SVD, small vessel disease; CAA, cerebral amyloid angiopathy.

In a further sensitivity analysis restricting the cohort to probable CAA patients with normal cognition at baseline (n = 86), we found similar results. The presence of disseminated cSS (HR 4.443, 95% CI: 1.309, 15.075, p = 0.017), WMH/TBV (%; HR 1.500, 1.104–2.038, p = 0.017), and SVD score for CAA ≥3 (HR 3.223, 95% CI:1.226–8.475, p = 0.018) independently predicted dementia conversion.

DISCUSSION

Our study found that increased WMH burden, presence of disseminated cSS, and higher SVD score for CAA predicted high risk of developing late dementia 6 months after index ICH in patients with probable CAA. The predictive value of these markers was consistent in both probable CAA patients with or without pre-existing MCI. Meanwhile, the characteristics of the index ICH were not associated with late post-ICH dementia in probable CAA.

It is notable that higher WMH burden, either measured by volume or Fazekas scores, was associated with new-onset dementia conversion in post-ICH patients with probable CAA, 6 months after index event. The underlying pathology of WMH appears mostly to reflect demyelination and axonal loss caused by SVD [22, 23]. Previous studies have consistently reported the association between white matter damage and cognitive impairment, mostly based on data from the general population [24], patients with ischemic stroke [25] or patients with Alzheimer’s disease pathology [26]. Recently, a few studies have investigated the predictive value of WMH on dementia conversion in patients with ICH [4, 5, 27], but have shown inconsistent results. WMH burden on MRI, 6 months after index ICH, had no predictive value for delayed onset of dementia in a cohort of patients with lobar ICH [6]. Similarly, neither CT nor MRI based WMH burden were predictive for new onset of cognitive impairment 30 days after lobar ICH event [27]. However, another study showed WMH burden measured by CT scans independently predicts delayed onset of dementia in patients with lobar ICH [4]. The different types of imaging tools and varied time of scanning after ICH may explain the discrepant results from these studies. Compared with these previous studies, our study focused on a more homogeneous cohort, including only lobar ICH due to probable CAA. Our findings support the hypothesis that in patients with probable CAA, the severity of the underlying microangiopathy (using the surrogate measure of WMH burden) is the driving pathology for dementia conversion in these patients.

Our study also found disseminated cSS was predictive of dementia conversion in lobar-ICH patients with probable CAA. cSS is hypothesized to result from repeated episodes of bleeding in the subarachnoid space [28]. The presence of cSS is reported to be a highly specific neuroimaging marker for CAA and is one of the diagnostic hallmarks of the disease [7]. The association between the presence of cSS and ICH recurrence in CAA is well known [29, 30]. While a few studies have reported the association between the presence of cSS and cognitive impairment, these studies were not specifically focused on patients with probable CAA [5, 31]. The associations identified in these previous studies are likely due to the use of general cohorts, in which cSS identifies patients with CAA and that CAA itself strongly correlates with cognitive impairment [7, 32]. Within probable CAA, our current data suggest that the presence of disseminated cSS reflects a more aggressive phenotype of the disease, which predisposes individuals to dementia.

Considering that two components of the total SVD score for CAA, WMH and disseminated cSS, were predictors for dementia conversion, it is not surprising that the score itself also provided significant predictive value. The predictive value of the total SVD score for CAA on dementia is consistent with our previous study including only probable CAA patients without ICH [33]. However, in the probable CAA without ICH cohort, only the total score, instead of any of the four individual components, could predict the dementia conversion. This discrepancy may suggest that the effect of specific individual neuroimaging markers on cognition differs between different phenotypes of probable CAA.

This study has limitations. Since all patients were recruited from clinical care, the lack of pathological confirmation restricts our capacity to identify the proportional contribution of CAA and Alzheimer’s disease pathology on dementia conversion in this cohort, though the co-existing pathologies of these two disorders have been commonly observed in the elderly population [34, 35]. Additionally, other non-CAA microvascular pathology, commonly reported in the elderly, may also contribute to the cognitive decline in this cohort [36]. Moreover, we lack the detailed neuropsychological testing that could detect subtle cognitive change over time. Pre-ICH cognitive status was tracked retrospectively after the onset occurrence. This limitation may lead to the reduced insensitivity of detecting more risk factors. However, detailed clinical notes written by treating neurologists were systematically reviewed to assess whether cognitive impairment impacted the subjects’ daily living ability and social function. The post-ICH cognitive status was cut into dementia and non-dementia categories as dementia conversion could serve as a practical and reliable outcome measurement commonly used in clinical studies [37, 38]. Additionally, the small sample size may limit the power to unravel the value of clinical features in predicting dementia conversion. However, our study represents one of the largest cohorts of ICH patients with probable CAA so far. The increased utilization of MRI in patients with ICH may facilitate the ability to identify more probable CAA patients in the future [39]. Last but not the least, CAA includes a heterogeneous group of disorders. Clinically, the sporadic CAA we included in our study is the most commonly form in the elderly, while there are also rare familial forms occurring in younger patients [40]. Therefore, results of this study can only be interpreted within the sporadic CAA.

In summary, the total burden of SVD, measured with a composite SVD score for CAA, is associated with late dementia conversion in lobar ICH patients with probable CAA. This effect appears to be driven by high WMH burden and presence of disseminated cSS.