Abstract
Background:
Individuals with Alzheimer’s disease (AD) often present with coexisting vascular pathology that is expressed to different degrees and can lead to clinical heterogeneity.
Objective:
To examine the utility of unsupervised statistical clustering approaches in identifying neuropsychological (NP) test performance subtypes that closely correlate with carotid intima-media thickness (cIMT) in midlife.
Methods:
A hierarchical agglomerative and k-means clustering analysis based on NP scores (standardized for age, sex, and race) was conducted among 1,203 participants (age 48±5.3 years) from the Bogalusa Heart Study. Regression models assessed the association between cIMT ≥50th percentile and NP profiles, and global cognitive score (GCS) tertiles for sensitivity analysis.
Results:
Three NP profiles were identified: Mixed-low performance [16%, n = 192], scores ≥1 SD below the mean on immediate, delayed free recall, recognition verbal memory, and information processing; Average [59%, n = 704]; and Optimal [26%, n = 307] NP performance. Participants with greater cIMT were more likely to have a Mixed-low profile [OR = 3.10, 95% CI (2.13, 4.53), p < 0.001] compared to Optimal. After adjusting for education and cardiovascular (CV) risks, results remained. The association with GCS tertiles was more attenuated [lowest (34%, n = 407) versus highest (33%, n = 403) tertile: adjusted OR = 1.66, 95% CI (1.07, 2.60), p = 0.024].
Conclusion:
As early as midlife, individuals with higher subclinical atherosclerosis were more likely to be in the Mixed-low profile, underscoring the potential malignancy of CV risk as related to NP test performance, suggesting that classification approaches may aid in identifying those at risk for AD/vascular dementia spectrum illness.
INTRODUCTION
It is well recognized that more than 50% of individuals with a diagnosis of Alzheimer’s disease (AD) have coexisting brain pathologies, with vascular neuropathology as particularly prevalent [1, 2]. The underlying vascular component concomitant with AD pathology is promoted by long-term exposure to cardiovascular risk factors, including subclinical atherosclerosis, that results in structural and functional disruption of normal brain hemodynamics, blood-brain barrier damage, microvascular flow instability, and neuroinflammation [3, 4].
The exposure to clinical and pathobiological indicators of vascular disease not only promotes but also exists side-by-side and exacerbates AD pathology [2, 6]. Postmortem studies have shown that individuals with AD and additional neuropathology were more likely to transition from mild cognitive impairment (MCI) to dementia [7]. Thus, the coexistence of multiple pathologies are best understood as falling within an AD/vascular dementia (AD/VaD) spectrum [2, 9]. Furthermore, the AD- and vascular-related mechanisms are expressed to different degrees among different individuals, contributing to heterogeneity in the clinical presentation of dementia syndromes [7, 8].
Although it is commonly thought that neuropsychological (NP) presentations of AD dementia syndromes begin with amnestic impairment followed by executive dysfunction, evidence shows that individuals with AD do not always fit this stereotypical amnestic-predominant pattern, particularly during the early stages of the disease [2, 11]. Likewise, VaD has been traditionally associated with executive dysfunction, but evidence suggests the full influence of vascular pathology is clinically masked among individuals with AD pathology [12, 13]. Studies have shown the prediction accuracy of neuropsychological features, including memory and verbal fluency, can distinguish between neuropathologically defined VaD and AD dementia. However, when AD and VaD are combined into mixed dementia, as in the vast majority of dementia cases, the impact of AD dominates the neuropsychological presentation [14]. Neuroimaging and biomarker assessments have shown potential to identify the pathological substrate for cognitive impairment, but these tools are expensive, experimental, or rely on equipment not readily available in a primary care setting[2, 16].
NP assessment is a cornerstone of the diagnostic process for AD; however, traditional consensus diagnosis mostly relies on NP global summary scores which do not always differentiate between specific patterns of deficit [17–19]. Recently, data-driven classification approaches utilizing person-centered statistics, such as cluster analysis, have been successful in identifying disease subtypes that more closely link to pathobiological processes and have demonstrated that participants with mixed NP deficits have higher rates of progression to dementia [8, 19]. This novel NP-based subtyping may offer a more meaningful method to identify those at high risk of progression to AD dementia during prodromal stages, decreasing diagnostic errors and selecting the most appropriate individuals for clinical trials [20–22]. However, these NP classification approaches have mostly been utilized among elderly individuals from memory clinics with established MCI or dementia syndromes, limiting the generalizability of results to a larger population [19, 23].
This study examines the utility of NP classification approaches in a relatively young (mean age 45 years) community-based cohort of African American and White individuals with a well-documented high burden of cardiovascular risk factors from the Bogalusa Heart Study (BHS). Previous work from the BHS showed that a subclinical vascular indicator of atherosclerosis, carotid intima-media thickness (cIMT) above the 50th percentile, was associated with worse midlife global cognitive function [24]. We aim to identify complex patterns of NP test performance that may be more closely associated with this vascular contributor, that has also been shown to be a potential point of intersection between vascular risk factors and neuropathology [2, 26]. Studying individuals as early as mid-life, could provide evidence for the existence of distinct subgroups who may benefit from preventive and therapeutic strategies targeting underlying etiologies within the AD/VaD spectrum [8, 15].
METHODS
Study design and population
The BHS began in 1973 and is located in the rural town of Bogalusa, Louisiana. The BHS is a community-based longitudinal cohort dedicated to the study of the natural history of cardiovascular disease with cross-sectional examinations starting from childhood [27].During the 2013–2016 examination, the first neuropsychological evaluation was implemented as part of the study protocol, the sample of individuals who participated consisted of 1,298 adults (age 48.17±5.27). From this sample, a total of 1,203 participants with a complete NP assessment, cIMT ultrasound testing, and information pertaining to covariates of interest were selected for the cross-sectional analysis (Supplementary Figure 1). Longitudinal blood pressure (BP) measurement data was obtained from previous examinations to calculate cumulative BP. Additional details about the cohort design have been previously published [27, 28]. All participants provided written informed consent, and the study was approved by the Institutional Review Board of Tulane University Health Sciences Center.
Carotid ultrasound measurements
Far-wall carotid intima-media measurements were obtained using a Toshiba Aplio ultrasound system (Toshiba Medical, Tokyo, Japan) and a 7.5-MHz linear-array transducer [29]. The composite-cIMT was defined as the average of the segmental maximum cIMT (right and left-sided maximum far-wall measurements from the common carotid, internal carotid, and carotid bulb). Our previous work shows that c-IMT above the 50th percentile was associated with worse midlife cognitive functioning. On the basis of previous literature identifying higher c-IMT measures as clinically concerning, particularly as a predictor of cardiovascular disease and also in correlation with cognitive outcomes, the composite-cIMT was dichotomized at the 50th percentile (>0.87 mm) for statistical analyses [24, 30–33]. The composite-cIMT as a continuous measure was also analyzed.
Neuropsychological assessment
The BHS neurocognitive protocol includes a series of NP tests assessing executive control, attention/information processing speed, and verbal episodic memory domains, as recommended in the National Institutes of Health (NIH) toolbox [34]. The Digit Span Forward and Digit Span Backwards (WAIS-IV), Trail-Making Test-Part A (TMT-A) and Part B (TMT-B), Logical Memory I (WAIS-IV), Logical Memory II, Delayed Recognition (WMS-IV), and the Digit Symbol Coding (WAIS-IV) were used as clustering variables. The TMT-A and TMT-B were reverse-scored so that higher scores indicated better performance. This was done after clustering analyses to avoid scores with normalized distribution in the clustering analysis. A global cognitive z-score (GCS) was computed by averaging all z-scores and divided into tertiles.
Achieved educational attainment and premorbid cognitive abilities were estimated using tests that assessed word reading (WRAT-4) and knowledge (WAIS-IV Vocabulary subtest). Since NP performance is highly influenced by cultural constructs and may result in the over-representation of lower and higher scores among Black and White individuals, respectively, we demographically corrected NP scores to yield more sensitive indicators of cognitive function [35, 36]. Because of the young age of study participants, without clinical diagnoses of cognitive impairment, the substantial sample size, and the lack of normative data for NP tests among African-Americans in the rural South, NP tests’ raw scores from the whole sample were standardized into z-scores corrected for age, sex, and race, with a mean of zero and a standard deviation (SD) of 1.0 [24, 37–39]. Additional details of this methodology can be found in the Supplementary Material.
Covariates
Sociodemographic, lifestyle, and cardiometabolic data were obtained through validated questionnaires in accordance with BHS protocols [27]. Covariates, including fasting glucose, high-density (HDL) and low-density (LDL) lipoprotein cholesterol, and triglycerides, were measured at a centralized laboratory following standard protocols [40].
To assess BP exposure from childhood to midlife, the cumulative blood pressure (cumBP) per year of follow-up was calculated using the formula: cumBP =∑[(BP1 + BP2) /2 * time 1 - 2] + [(BP2 + BPn)/2 * time 2 - n], where BP1, BP2, and BPn indicate BP measurements at each study examination [41]. The total cumulative BP was then divided by the number of years of follow-up of each participant, and both systolic and diastolic measures were calculated. Body mass index (BMI) was calculated from duplicate measures of height and weight for each study participant. Non-HDL cholesterol was computed based on the lipid profile.
Information on educational level, employment status, smoking, and medication use was obtained from self-reported questionnaires. The Center for Epidemiologic Studies Depression Scale-10 Item (CESD-10) was used to assess depressive symptomology [42]. A language index expressed as a z-score, assessing word knowledge, was used as an indicator for achieved education and premorbid cognitive abilities, given that race, income, geography, and other social determinants affect the quality of education and are not fully captured by education years alone [43, 44]. Because NP tests were corrected based on participants’ age, sex, and race, these variables were not included in regression analyses [24, 46]. Additional details of this methodology can be found in the Supplementary Material.
Analytical methods
Clustering analysis
To ensure robustness and consistency of the NP subtyping, two different clustering methods were used: a hierarchical agglomerative analysis using Ward’s linkage method and a K-means algorithm with Euclidean distance [47]. The hierarchical agglomerative approach has been widely used in previous studies and was found to be the most parsimonious for identifying separable clusters [20, 48]. The number of clusters (groups) was chosen based on the resulting clustering dendrogram and the highest Duda-Hart (DH) Index (for hierarchical algorithm), and the highest Calinski-Harabasz (CH) Index (for hierarchical and K-means algorithms) [49, 50]. To evaluate the stability of the clustering results, a leave-one-out cross-validation procedure was performed for the hierarchical algorithm. To generate dataset variations, resampling was used for the K-means method. A cross-tabulation of the hierarchical and K-means results was done to calculate the percentage of consistency for each cluster, followed by the Adjusted Rand Index (ARI), to compare clustering solutions (a perfect match yields a value of 1.0) [51, 52]. A t-distributed stochastic neighbor embedding (t-SNE) was used to reduce the dataset dimensionality. Both hierarchical and K-means were applied to partition the result of the reduction [53] (Fig. 1). Additional details of the clustering methodology can be found in the Supplementary Material.
Neuropsychological Test-Based Sub-Groups identified via unsupervised learning. A t-SNE analysis to reduce the dataset into a 2-dimensional data space, both, a) hierarchical and b) K-means were applied to partition the result of the reduction. Group 2 (Normal) and 3 (Early MCI-like) aggregated on opposite sides of the plot while Group 1 (Suboptimal) overlapped minimally.
Statistical analysis
Statistical analyses focused on the most parsimonious cluster solution, the hierarchical agglomerative solution. Identity was given to unlabeled subgroups based on each cluster’s NP test performance (z-scores). Differences in participant characteristics by NP profiles were examined using means and SD for continuous variables and compared using analysis of variance (ANOVA) and the Kruskal-Wallis’s test. For categorical variables, frequencies and percentages were compared using Pearson Chi-Squared test. Bonferroni corrections were performed for multiple comparisons. Sensitivity analysis examined differences in participant characteristics and dementia risk factors between GCS tertiles, including education, smoking behavior, blood pressure and glucose levels, and carotid thickness.
Multinomial logistic regression models were used to assess the association between composite-cIMT categories (above and below the median value) and data-driven NP profiles. Multivariable models included covariates based on a priori knowledge of confounding effects (education proxy, mean cumulative SBP per year of follow-up, glucose level, smoking status, CESD-10 depression scale score, non-HDL cholesterol, triglycerides level, and hypertension medication use). For sensitivity analysis, regression models were used to assess the association between composite-cIMT categories and GCS tertiles. Odds ratios (OR) and 95% confidence intervals (95% CI) were reported. Following a complete case approach, participants with missing c-IMT measurements (n = 73) and NP test scores (n = 22) were excluded (Supplementary Figure 1). All analyses were performed using Stata/IC 15.1 (StataCorp LLC, College Station, TX), and RStudio (version 3.0) software. Statistical tests were two-tailed, and p-values were considered significantif < 0.05.
RESULTS
A total of 1,203 participants were included in the analysis. The mean age of participants was 48 years (SD 5.27) years, and 42% (502) were men. From the total sample, 66% (792) self-identified as White and 34% (411) as Black. The mean composite-cIMT was 0.95 mm (0.32) (Table 1).
Participant Characteristics by NP-profile
Data are presented as mean (SD) for continuous measures and n (%) for categorical measures. †Bonferroni adjusted p-values *Center for Epidemiologic Studies Depression Scale 10-Item (CESD-10).
Cluster analysis
The hierarchical algorithm identified a 3-cluster solution, selected based on the resulting dendrogram and higher cluster indices (CH = 332.44, DH = 0.87), indicating dense and well-separated groups, compared to a 4-cluster solution (CH = 272.85; DH = 0.79). The 4-cluster solution yielded the same clustered-derived (Group 3) with lower NP z-scores (n = 192) than the 3-cluster solution and therefore was not considered. A leave-one-out cross-validation procedure showed robust clusters that remained constant with a classification accuracy that changed from 93.75% to 93.47%. The K-means algorithm with k = 3 clusters outperformed a k = 4 solution in the CH index (CH = 379.27 versus 313.58, respectively). Therefore, a 3-cluster solution was confirmed. After resampling input data, individuals in the cluster with lower NP z-scores (Group 3) remained constant (n = 222).
The two algorithms’ solutions were consistent within each cluster. Cross-tabulation showed consistency of participants in Group 2 of 96.74%, Group 3 of 93.23%, and in Group 1 of 77.13%. The ARI index between the two clustering methods for Groups 2 and 3 was 1.0, and 0.60 for the 3 groups. The t-SNE solution shows that Groups 2 and 3 aggregated on opposite sides of the plot while Group 1 overlapped minimally with Groups 2 and 3 (Fig. 1). This is consistent with the results shown by cross-tabulation and the ARI index.
Neuropsychological profiles
Data-driven Group 1 was labeled Average performance and comprised 59% (n = 704) of the study sample. Participants in this group scored approximately 0.25 SD below the mean on memory tests assessing immediate and delayed free recall and above the mean on all other NP tests. Group 2 was labeled Optimal performance and consisted of 26% (n = 307) of the study sample. Participants in this group performed above average on all eight NP tests (mean z-scores ranged from 0.37 to+1.20 with SDs = 0.51–1.02). Group 3 was labeled Mixed-low performance and represented 16% (n = 192) of the study sample. Participants in this group exhibited scores with > 1 SD below the mean on NP tests measuring immediate and delayed free recall (Logical Memory (LM) I & LM II), LM delayed recognition along with tests measuring graphomotor information processing and executive function (Digit Symbol Coding & TMT-B). They also scored > 0.5 SD below the mean across NP tests measuring attention (Digit Span Forward & TMT-A) (Fig. 2, Supplementary Table 1).
Neuropsychological tests Z-scores by NP-profile. Boxplot displaying the distribution of neuropsychological test Z-scores by neuropsychological profiles. Bars represent the spread of each NP test, with the first quartile, median and third quartile, and the mean represented by a rhombus shape. Red y-axes are marked at 0 and –1 SD.
Sociodemographic, lifestyle, and vascular risk
The three NP profiles differed significantly by sex, race, educational level, and employment status (ps < 0.05). Participants in the Mixed-low profile were older [49.13 ± 4.86 versus 47.45 ± 5.40, p= 0.002] compared to their Optimal NP profile counterparts, but not when compared to the Average profile [48.21 ± 5.29, p= 0.098]. Those in the Mixed-low were more often men [54.2% (104), p< 0.001] and Black [56.2% (108), p< 0.001], with lower education [some college and above = 15.1% (29), p< 0.001], and lower employment rate [30.7% (59), p< 0.001] compared to participants in other NP profiles (Table 1).
Compared to those in the Optimal cognitive profile, participants in the Mixed-low profile had higher SBP levels [126.11 ± 17.99 mmHg versus 121.61 ± 16.85, p= 0.01] higher cumSBP/year of follow-up [115.95 ± 9.43 mmHg/year versus 113.03 ± 9.07, p< 0.001], and higher HDL levels [48.80 ± 16.47, p= 0.007]. There were no significant differences in these characteristics when compared to those in the Average performance profile. However, participants with Mixed-low performance had higher rates of treated hypertension 44.8%, and diabetes* 17.2% compared to the Optimal* and the Average performance profile groups (all ps < 0.05, except *p= 0.07) (Table 1).
cIMT and neuropsychological profiles
All far-wall cIMT measurements were significantly higher among Mixed-low participants when compared to the other NP profiles (ps < 0.01) (Table 1). Multinomial logistic regression model results showed that cIMT-composite>50th percentile (greater than 0.87 mm) was associated with a three-fold higher likelihood of having a Mixed-low performance [OR = 3.10, 95% CI (2.13, 4.53), p < 0.001], and increased likelihood of suboptimal performance [OR = 1.37, 95% CI (1.04, 1.80), p = 0.024] compared to Normal performance [ref. OR = 1]. After adjustment for education proxy, cumSBP, glucose level, smoking status, CESD-10 score, non-HDL cholesterol, triglycerides, and hypertension medication use, the point estimates were attenuated but remained significant; the likelihood of having an early MCI-like performance for individuals with cIMT-composite>50th percentile remained statistically significant when compared to Normal performance [OR = 2.32, 95% CI (1.43, 3.79), p < 0.001] (Table 2).
Association between carotid ultrasound measurement and NP-profiles
cIMT, Carotid intima media thickness **<versus 50th percentile (0.87 mm). *Education proxy, mean cumulative SBP per year of follow-up, glucose level, smoking status, CESD-10 score, non-HDL cholesterol, triglycerides level and hypertension medication use.
Additional regression models with composite-cIMT as a continuous variable showed an increased likelihood of having a Mixed-low performance [OR = 2.22, 95% CI (1.08, 4.55), p = 0.030] when compared to an Optimal profile, after adjustment of confounders (Supplementary Material).
Sensitivity analysis
Participants in the lowest GCS tertile (n= 403), indicating worse cognitive performance, were more often women [213 (52.9%), p= 0.002], had lower educational level [some college and above = 73 (18.1%), p< 0.001], and higher SBP [125.85 ± 18.11, p= 0.001] when compared to the highest GCS tertile (better cognitive performance). No significant age differences were found. In contrast to data-driven NP profiles, there were no differences in prevalence of hypertension or diabetes between the GCS tertile groups (Supplementary Table 2). Furthermore, the association between cIMT-composite ≥50th percentile was much more attenuated than that seen with NP profile groups [lowest versus highest tertile: adjusted OR = 1.67, 95% CI (1.07, 2.60), p = 0.024; middle versus highest tertile: adjusted OR = 0.98, 95% CI (0.68, 1.40), p = 0.899] (Table 3). No significant associations were found when using cIMT-composite as a continuous variable (Supplementary Material).
Association between carotid ultrasound measurement and GCS-tertiles
cIMT, Carotid intima media thickness **<versus 50th percentile (0.87 mm). *Education proxy, mean cumulative SBP per year of follow-up, glucose level, smoking status, CESD-10 score, non-HDL cholesterol, triglycerides level, and hypertension medication use.
DISCUSSION
To our knowledge, this is the first study to employ data-driven NP profiling via unsupervised learning in a community-based cohort of Black and White men and women in midlife. We identified a group of participants with Mixed-low performance on NP tests [54, 55]. This group had substantial NP deficits in tests assessing processing speed/executive function and verbal episodic memory, suggestive of an early MCI-like profile concomitant with a higher burden of vascular and dementia risk factors. Such a profile could be indicative of an emerging brain at risk quite early in the life-course compared to other cohorts [19, 23]. Participants with greater cIMT were more likely to display a Mixed-low profile than an Optimal profile, even after adjustment for vascular and dementia risk factors, including educational attainment and smoking status, as well as the cumulative burden of systolic BP over the lifespan. Furthermore, we found stronger point estimates for the association between cIMT and midlife cognitive performance using data-driven profiles than using the more traditional GCS.
The Mixed-low profile resembled a mixed dysexecutive phenotype which has been observed in clinical settings among individuals with a predominance of vascular pathology [8, 58]. Studies utilizing actuarial methods show that individuals identified with a mixed MCI phenotype have higher progression to dementia and higher rates of AD biomarkers and pathology on autopsy [19, 59]. In contrast, GCS tertile groups include decreasing means in all NP tests from the lowest to the highest tertile without distinction due to the method of their construction. These findings illustrate that the use of traditional approaches, such as categorization by quantiles of GCS, may result in a greater degree of misclassification and less specificity with respect to underlying characteristics [18]. Furthermore, the magnitude of point estimates for the association between cIMT and midlife cognitive function was more attenuated when examined using GCS tertile groups than for the cluster-derived NP profiles, providing additional support in using NP profiles to decrease misclassification.
Because of the well-known neuropathologic heterogeneity and difficulties of classifying dementia across the AD/VaD spectrum, cluster-derived NP profiles could be helpful for informing diagnosis and, thereby, treatment [8, 60]. Previous studies that utilized data-driven approaches have been successful in identifying false positive diagnoses of MCI, which have aided in elucidating stronger effects of therapeutic interventions among individuals with MCI [60]. This study expands existing evidence into midlife, a critical time window for the detection of preclinical stages dementia syndromes.
Our results also show that participants with an Mixed-low profile had higher rates of hypertension and diabetes, both of which have been related to cognitive impairment [61, 62]. In addition, participants had increased exposure to elevated levels of systolic BP over the lifespan as indicated by cumulative systolic blood pressure. This group had thicker cIMT measurements, which speaks to the chronicity of these conditions at a relatively young age, and the potential impact that the cumulative burden of such vascular risk factors can already have on cognition by midlife. Moreover, individuals classified in this group had lower education levels, lower rates of employment, and worse depression-screening scale scores, which have also been associated with dementia syndromes [63, 64]. In contrast, categorization based on traditional quantile groups of GCS showed no difference in hypertension or diabetes prevalence. When this traditional categorization is used to define groups based on cognitive function, it may mask the importance of participants’ vascular risk factors and subclinical disease. Indeed, this is the most likely explanation for the lower point estimates characterizing the association between cIMT and midlife global cognitive performance.
Our study findings demonstrate the potential added value of classifying participants using NP test performance and its association with underlying vascular pathology and underscore the putative likelihood of progression to MCI and dementia syndromes. Typically, the majority of AD and dementia research has been conducted in late life among patients from memory clinics or participants in cohorts that fail to capture earlier stages in life. BHS provides a unique opportunity to study the role of vascular contributions to dementia in a diverse, community-based setting. In contrast to other studies of NP phenotyping, the subjects in our study were community-dwelling adults in midlife with a significant representation of Black individuals (34%) and women (58%), making our results a substantial contribution to the current literature [19, 48].
Strengths of this study include leveraging the cumulative blood pressure values from childhood to adulthood, allowing for a clearer representation of the lifetime blood pressure burden in each of the NP profiles, and robust adjustments for potential confounders in models of the relationship between cIMT and NP profiles. In addition, we systematically applied an unsupervised machine learning framework to ensure internal validity, optimal number of clusters, cluster stability, and consistency between different methods. However, the use of such novel approaches to classify and characterize heterogeneity also requires external validation [15].
With respect to limitations, we recognize that given the unique demographic and clinical characteristics of the sample in this study, results might not be entirely applicable to a more general US population. The need for external validation with additional data modalities and cohorts to ensure reproducible results is imperative. However, BHS participants may be more representative of community-dwelling individuals from rural areas that have been historically underrepresented in AD and related dementias research. We also acknowledge that there are no uniformly accepted methods by which cognitive impairment is used to derive a clinical diagnosis of MCI. Moreover, in future research, other psychosocial factors, psychiatric problems, and other comorbidities need to be considered for a proper clinical diagnosis.
Another limitation of the current research is that neuropathological information will not be confirmed by autopsy in the near future. Thus, in order to fully appraise the added value of our results, there is a need for additional clinical information, including prospective neuroimaging, AD-related biomarker data, and longitudinal NP assessment in future BHS examinations. Due to the nature of unsupervised clustering methods, the clinical meaning of these subgroups is unknown. Nevertheless, the NP performance of the Mixed-low profile is similar to the performance seen in participants from other cohorts with a diagnosis of MCI, including the ADNI-2 and the UCSD Shiley Marcos ADRC, that have used this approach [19, 65]. Future examinations of participants in the BHS will provide data to further evaluate whether and how groups might eventually convert to a clinical MCI or dementia diagnosis and/or experience reversion to Optimal or near-normal levels of cognitive performance.
Conclusion
As early as midlife, individuals with a higher burden of subclinical atherosclerosis were more likely to be classified by data-derived methods into a group reflecting a Mixed-low NP test performance profile. These results in a midlife, community-based cohort, suggest the potential for NP profiling to improve identification of individuals with a high burden of risk factors for progression to MCI and dementia syndromes. The ability to identify high risk individuals at preclinical stages within an AD/VaD spectrum could lead to improved allocation of preventive strategies and more timely and tailored interventions for risk reduction.
Footnotes
ACKNOWLEDGMENTS
The authors woud like to thank the BHS participants, staff members, and study personnel. Their effort is crucial for conducting, sustaining, and continuing with the study.
FUNDING
The National Institute on Aging of the NIH (RF1AG041200, R01AG062309, RF1AG041200-06S2, R01AG077497, 5P30AG013846) and the American Heart Association (AHA) award 20SFRN35490098.
CONFLICT OF INTEREST
Rhoda Au received consulting fees from Biogen, Signant Health, GSK, Novo Nordisk, and Davos Alzheimer’s Collaborative (nonprofit). David Libon received royalties from Oxford University Press and Linus Health. Dr. Carmichael received consulting fees from Novo Nordisk. Dr. Kolachalama received consulting fees from Davos Alzheimer’s Collaborative.
Rhoda Au and David Libon are Editorial Board Members of this journal, but were not involved in the peer-review process, nor had access to any information regarding its peer-review.
All other authors have no conflict of interest to report.
DATA AVAILABILITY
Data not provided in the article and additional information on methods and materials can be shared on responsible request.
