MRI Visible Perivascular Spaces and the Risk of Incident Mild Cognitive Impairment in a Community Sample

Sample

We studied participants from the Framingham Heart Study (FHS), a large community-based prospective cohort that recruited from the community in Framingham, MA, USA. Launched in 1948, the original mission of the FHS was to investigate the epidemiology of cardiovascular disease [18]. Since, the FHS has evolved into multiple cohorts, for which two were utilized for present analysis: the Original cohort [18] and the Offspring cohort [19]. The Original cohort consists of 5209 adults who completed initial assessment between 1948 and 1953 with follow-up approximately every two years thereafter. The Offspring cohort consists of 5,124 adults who were the children of the first-generation cohort or spouses of these children. The participants were enrolled between 1971 and 1975, with follow-up approximately every four years thereafter.

Participants were invited to undergo brain MRI as part of an ancillary study on neuropsychological testing and brain imaging. Brain MRIs were obtained during exams 25–31 in the Original cohort and 7–9 in the Offspring cohort. For the present study, we included participants who attended the clinic exams corresponding to the brain MRIs and had PVS ratings. Participants were excluded if, at baseline (time of MRI), they were aged less than 50 years, had a diagnosis of dementia, MCI, or other significant neurological disease that manifests on MRI scans (e.g., traumatic brain injury, stroke) or if they were not followed up for MCI. Figure 1 details sample selection. All participants provided written informed consent. The Institutional Review Board of Boston University Medical Center approved the study.

OPEN IN VIEWER

Fig. 1

Study flow diagram.

Brain MRI and PVS rating

Participants underwent brain MRI imaging with a 1T (this occurred between 1999–2005) or 1.5T (after 2005) magnetom scanner [20]. The 1T MRI used a T2-weighted double spin echo coronal imaging sequence of 4 mm contiguous slices. The 1.5T MRI used a T2-weighted double spin-echo coronal imaging sequence of 3 mm contiguous slices [20]. Acquisition and imaging processing details have been published previously [21–23].

PVS were determined using the STandards for ReportIng Vascular changes on nEuroimaging Criteria (STRIVE) consensus criteria [9, 15]. Briefly, PVS had to display signal intensity similar to cerebrospinal fluid, follow the course of penetrating vessels, and have a diameter less than 3 mm.

Visible PVS were rated by three study investigators (neuroradiologist, vascular neurologist, and a well-trained research assistant) who were blind to all participant information. We grouped PVS according to their brain topography, involving the CSO and BG [24]. Ratings were performed on T2-weighted axial MRI sequences, to the extent possible. Following our previous approach,[15] PVS burden was categorized into grades based on PVS counts: I (1–10), II (11–20), III (20–40) and IV (>40).

We have previously reported the intra and inter-rater reliability of PVS ratings (ICC ranging from 0.74 to 0.81 for intra-rater reliability, and 0.80 to 0.90 for inter-rater reliability) [6].

MCI case ascertainment

Participants in the FHS are under continuous and uninterrupted surveillance for the incidence of cognitive impairment and dementia. Participants complete the Mini-Mental State Examination (MMSE) at each examination cycle as a cognitive screen [25]. Participants also complete more extensive neuropsychological testing via several ancillary studies. Using the MMSE, participants are flagged with suspected cognitive impairment if 1) performance falls below education-based thresholds, 2) we observe a decline of 3 or more points between consecutive examinations, or 3) we observe a decrease of 5 or more points from the participant’s highest past MMSE score. We also flag participants with suspected cognitive impairment following referrals or concern expressed by the participant, their family, or primary care physician. Once flagged, participants complete annual regular neuropsychological and neurologic evaluations until they develop dementia or are adjudicated to be normal. Evaluations indicative of possible cognitive impairment are followed by referral to our study dementia review committee, comprising a neurologist and neuropsychologist. The dementia review committee adjudicates a diagnosis of MCI in accordance with the criteria defined by Petersen et al. [26], which defined impairment as performance on a single cognitive test within a domain greater than 1.5 SD below normative expectations. Participants are classified into subtypes depending on single or multiple domain impairment and amnestic (only memory domain impaired) or non-amnestic subtypes (memory domain not impaired). We included any MCI in the present study.

Statistical analyses

We used Cox proportional hazard regression models to investigate the association between PVS burden and the incidence of MCI. Kolmogorov-type supremum tests were used to assess the proportional hazards assumption [27]. Follow-up for MCI was from the time of brain MRI to incident MCI through to 2018 (median, 7.4 years; Q1–Q3, 4.9–13.2; max 19.8). For persons without incident MCI, follow-up was censored at the time of death or the date the participant was last known to be MCI-free, also through to 2018. PVS in the CSO and BG (grades I–IV) were modelled separately as categorical predictors with grade I as the reference group. Additionally, we created a categorical mixed PVS score that captured high burden (grade III or IV PVS) in neither region, strictly in the CSO, strictly in the BG, or both regions (0 = none, 1 = BG region only, 2 = CSO region only, or 3 = both regions), with score 0 as the reference group.

All models were adjusted for age at brain MRI, sex, the time interval between the brain MRI and exam, FHS cohort, education, current smoking, diabetes, hypertension, and prevalent cardiovascular disease. These covariates were chosen to control for confounding based on a priori based on knowledge of the literature and cohort. In a sensitivity analysis, we included additional adjustments for common imaging features of small vessel disease. In the first sensitivity analysis, we added an adjustment for white matter hyperintensity volume as a percentage of intracranial volume. In a second sensitivity analysis, we included an additional adjustment for cerebral brain infarcts. The purpose of these analyses was to determine if the associations between PVS and incident MCI were independent of other common CSVD imaging features. Adjustment for image view (coronal versus axial) did not change inferences and thus was not included in the statistical models. In a post-hoc analysis, we stratified associations between PVS and incident MCI by sex. This was because a previous cross-sectional study reported that PVS volume fraction in the CSO was higher in patients with MCI, versus controls, in women only [28].

MRI scanner strength is reflected on the image view used (coronal/axial) in this sample. Adjustment this variable did not change inferences and thus was not included in the statistical models.

Analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). A p-value<0.05 was considered statistically significant.

Sample characteristics

Table 1 displays the sample demographics, stratified by PVS burden severity. Of the 1314 participants, we observed 263 (20%) cases of incident MCI. Across the sample, the mean age was 68 years (SD, 9) and 54% were female. Overall, 4.8% and 0.3% of the sample had the highest grade of PVS severity in the CSO and BG, respectively. Those participants with the lowest PVS severity ratings tended to be younger, more highly educated, and have the lowest prevalence of cardiovascular disease.

Participants with available PVS ratings (n = 2,019) were compared with those excluded due to missing PVS ratings (n = 2,945). Overall, they were of similar age, sex and risk factor profile except for lipid lowering treatment and triglyceride levels; the latter was missing in 19% of the sample which may have skewed the results but considered unlikely to bias substantially the results.

PVS and incident MCI

Supremum tests showed no deviations from the proportional hazards assumption. The risk of MCI tended to increase with increasing PVS severity in the BG (Table 2). However, relative to the lowest severity rating, only those with a severity rating of 2 displayed a statistically significant increase in MCI risk, perhaps owing to the low number of incident cases in the highest severity groups.

The risk of incident MCI also increased with increasing severity of PVS in the CSO (Table 2). Relative to persons with the lowest severity rating, persons with ratings of 3 and 4 both displayed a higher risk of incident MCI. The increase in MCI risk was 2.55-fold in those with the highest severity of PVS in the CSO.

When capturing high burden (grade III or IV PVS) in neither region (reference), strictly in the CSO, strictly in the BG, or both regions (BG and CSO), we observed a higher risk of incident MCI only for those with a high PVS burden in the CSO (Table 2).

We explored whether the associations between PVS and incident MCI were independent of premorbid cognitive function. To investigate this, we included an additional adjustment for Wide Range Achievement Test (WRAT) scores, which is a common neuropsychological tool to assess premorbid cognitive function. Additional inclusion of WRAT scores to the multivariable adjusted model did not alter our results: Compared with the reference grade I CSO, the relation with incident MCI remained unchanged for grades 2 (HR 1.33, 95% CI 0.92, 1.9), grade 3 (HR 1.65, 95% CI 1.09, 2.52) or 4 (HR 2.84, 95% CI 1.68, 4.8); similar results were observed for BG region were grade 2 (HR 1.55, 95% CI 1.16, 2.09), grade 3 (HR 1.53, 95% CI 1.02, 2.27), and grade 4 (HR 2.44, 95% CI 0.96, 6.19) results were largely unchanged.

In addition, we performed a proportional subdistribution hazard regression using death as a competing risk, observing attenuation of associations with high BG PVS burden, while the associations of CSO PVS burden with MCI remained significant and largely unchanged.

Consideration of additional factors such as lipid levels (total cholesterol, low density, and high-density lipoprotein) in multivariable models did not change inferences.

Sensitivity analyses

Further adjustment for white matter hyperintensity volume attenuated the findings for the BG, such that there were no differences in risk of incident MCI between severity groups (Table 3). In contrast, persons with the highest grade of PVS in the CSO continued to display a higher risk of incident MCI, although the magnitude of effect was attenuated. Similarly, in the mixed region analysis, only persons with a high severity of PVS in the CSO displayed a higher risk of incident MCI. Results were similar to the primary models after including an additional adjustment for cerebral brain infarcts.

Post-hoc analysis of sex differences

The association between each grade of PVS severity in the CSO and incident MCI was more pronounced in women than in men (Table 4). For example, women with a PVS rating of IV in the CS, versus a rating of I, had a 3.49-fold increase in MCI risk. In contrast, men with a PVS rating of IV in the CS, versus a rating of I, had a 2.48-fold increase in MCI risk. For the higher grades of PVS severity in the BG, there were too few cases to permit a meaningful comparison between men and women. When capturing high burden (grade III or IV PVS) in neither region (reference), strictly in the CSO, strictly in the BG, or both regions (BG and CSO), results were relatively similar between males and females.