The Brain Health Index: Integrating vulnerability, resilience, and cognitive function into a unified measure of cognitive health and risk of neurodegenerative disease

Abstract

INTRODUCTION

Assessing brain health and identifying cognitive impairment risk remains challenging, with only 11.4% of MCI cases receiving timely diagnoses. We developed the Brain Health Index (BHI), integrating the Vulnerability Index, Resilience Index, and Number Symbol Coding Task into a unified metric.

METHODS

We evaluated 469 participants (258 congnitively normal [CN], 140 mild cognitive impairment [MCI], 49 Alzheimer's disease and related dementias [ADRD]) using comprehensive clinical, cognitive, and biomarker assessments. After empirically‐derived weighting, BHI thresholds were developed. Cross‐sectional associations and longitudinal analyses were performed, with threshold validation for risk stratification.

RESULTS

BHI demonstrated strong correlations with cognitive (Montreal Cognitive Assessment [MoCA] r ² = 0.408), functional (Functional Activities Questionnaire [FAQ] r ² = 0.278), and biomarker (neurofilament light chain [NfL] r ² = 0.073) measures. Complete mediation was observed for NfL and glial fibrillary acidic protein (GFAP) changes over 1 year. Threshold analysis revealed 89.2% of low BHI participants had cognitive impairment, with only one ADRD case in the high BHI group.

DISCUSSION

The BHI provides a brief, validated, comprehensive brain health metric with clinical utility for risk stratification and intervention monitoring.

Highlights

• Assessing brain health status and identifying individuals at risk for cognitive impairment remains a significant challenge, particularly in early prodromal and symptomatic stages when current therapies may be most effective and patients may be eligible for clinical trials.

• We created a unified metric, the Brain Health Index (BHI), combining resilience and vulnerability factors with cognitive performance that divided individuals into high, indeterminant, and low risk groups.

• The BHI demonstrated strong correlations with cognitive, functional, and biomarker measures.

• The BHI had a 16‐fold increase in identifying individuals who were likely to have cognitive impairment.

• The BHI provides a brief, validated, comprehensive brain health metric with clinical utility for risk stratification and intervention monitoring.

1. INTRODUCTION

Assessing brain health status and identifying individuals at risk for cognitive impairment remains a significant challenge in contemporary neurology and geriatric medicine. Addressing this gap is particularly important in early prodromal and symptomatic stages when current therapies may be most effective and patients may be eligible for clinical trials. ¹ Research has helped define early cognitive changes that precede dementia, identifying at‐risk individuals, improving early detection, and refining differential diagnosis. ² , ³ , ⁴ , ⁵ However, despite advances in fluid, imaging, and genetic biomarkers, mild cognitive impairment (MCI) remains substantially underdiagnosed, especially in disadvantaged populations. ⁶ , ⁷ Estimates suggest that only 11.4% of individuals with incident MCI receive a timely diagnosis. ⁸ Further, many cases of Alzheimer's disease (AD) and related dementias (ADRD) are not diagnosed until the moderate stage. The complexity of early detection is compounded by multiple barriers, including time constraints in primary care settings, insufficient clinician training and confidence to perform cognitive assessments, lack of access to and availability of cognitive specialists (e.g., neurology, geriatric psychiatry, geriatric medicine), and challenges in differentiating normal cognitive aging from signs and symptoms of MCI and AD/ADRD. ⁹

The concept of brain health extends beyond simple cognitive screening and can be described as the state of brain functioning across cognitive, sensory, social‐emotional, behavioral, functional, and motor domains. This comprehensive construct represents a complex interplay of modifiable and non‐modifiable factors that collectively influence cognitive resilience and vulnerability throughout the lifespan. ⁵ , ⁶ According to some estimates, up to 80% of brain health is determined by modifiable factors ¹⁰ , including cognitive, physical, nutritional, and sleep habits that can exert positive effects on both objective cognitive performance and subjective well‐being. However, despite growing recognition of the multidimensional nature of brain health, it remains elusive as a functional definition and is difficult to operationalize and measure holistically. ¹¹

To address these challenges in brain health assessment, we previously developed the Brain Health Platform ¹² , a comprehensive approach designed to evaluate global functioning through the systematic measurement of both modifiable and non‐modifiable risk factors associated with cognitive impairment. This platform comprises three complementary components: the Resilience Index (RI) ¹³ , which quantifies protective lifestyle factors including cognitive reserve, physical activity, social engagement, dietary patterns, mindfulness practices, and cognitive leisure activities; the Vulnerability Index (VI) ¹⁴ , which captures cumulative risk through a weighted assessment of demographic and medical factors known to increase dementia risk; and the Number Symbol Coding Task (NSCT) ¹⁵ , a brief measure of executive functioning that provides real‐time assessment of cognitive performance.

In this study, we advanced this foundational work by introducing the Brain Health Index (BHI), a unified numerical score that mathematically integrates these three measures through empirically derived weighting to create a single, interpretable metric with scores ranging from 0 to 100. We hypothesized that the BHI would demonstrate robust correlations with established markers of brain health across multiple domains, including cognitive and functional assessments, neuroimaging and blood‐based biomarkers, and physical performance measures, thereby validating its utility as a comprehensive brain health metric. However, we anticipated that the BHI would not be a stand‐alone diagnostic, as it was designed to provide a real‐time marker of brain health with an emphasis on actionable risk and resilience factors to potentially mitigate future cognitive impairment and/or future cognitive decline. To test these hypotheses and establish the clinical utility of the BHI, this study examines: (1) cross‐sectional associations between the BHI and a comprehensive battery of clinical, cognitive, functional, and biological measures to establish concurrent validity; (2) longitudinal properties of the BHI as a mediator of change in biomarkers and functional outcomes over time, providing insight into its potential as a predictor of brain health trajectories; and (3) the diagnostic utility of empirically‐derived BHI thresholds for risk stratification, enabling practical implementation in clinical settings for identifying individuals who would benefit most from targeted interventions.

2. METHODS

2.1. Study participants

A total of 469 individuals were evaluated at the Comprehensive Center for Brain Health at the University of Miami Miller School of Medicine, completed the required surveys, and were assigned a BHI. Of these individuals, 81 were evaluated in a clinical setting, and 389 were enrolled as participants in the Healthy Brain Initiative (HBI), a longitudinal observational study of adults over 50 in South Florida. Both cohorts had identical clinical, cognitive, behavioral, and functional assessments modelled on the uniform data set (UDS v3.0) from the National Institute on Aging (NIA) Alzheimer's Disease Center program. Clinical cohort members completed one office visit, while HBI participants were seen annually and included genetic, blood‐based biomarker, and neuroimaging data. ¹⁶ Of the 389 participants recruited into the longitudinal study, 166 were seen for a second annual visit at the time of analysis. Written informed consent from prospective research participants was obtained, while data from clinical cases were collected as part of a retrospective chart review. Both data sources were approved by the University of Miami Institutional Review Board.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature using traditional (e.g., PubMed) sources. While there is great interest in the study of brain health, to date, there is no unified method of assessing brain health to compare and contrast brain health between different individuals or tracking brain health longitudinally. We identified recent publications that discuss existing challenges and barriers.

2. Interpretation: The Brain Health Index (BHI) represents a significant advancement in our ability to quantify and monitor brain health through integration of vulnerability, resilience, and performance into a single, interpretable metric.

3. Future directions: The BHI captures the complex interplay between modifiable and non‐modifiable risk factors while remaining sensitive to underlying biological processes. The BHI shows promise as an outcome measure for clinical trials targeting cognitive health and dementia prevention. Further refinement and validation in generalizable populations would be necessary to optimize the BHI's predictive performance for clinical trial enrichment and outcome assessment.

2.2. Clinical assessment

2.2.1. Clinical assessment

Sociodemographic data, medical history, medications, alcohol/tobacco/substance use history, and family history were collected. A detailed clinical and neurological examination was completed. The Depression, Anxiety, and Apathy scale (DA3) ¹⁷ captured a distinct mood rating. The Healthy Brain 9 (HB9) ¹⁸ captures subjective cognitive complaints. The Modified Hachinski Scale ¹⁹ assessed the risk of vascular cognitive impairment. The mini‐Physical Performance Evaluation (mPPT) ²⁰ assessed physical functionality and frailty. Study partners completed a semi‐structured interview with an experienced research clinician to derive the Clinical Dementia Rating (CDR) and its Sum of Boxes (CDR‐SB). ²¹ The study partners also completed the Functional Activities Questionnaire (FAQ) ²² as an assessment of activities of daily living, and the Neuropsychiatric Inventory Questionnaire (NPI‐Q) as an assessment of behavior. ²³

2.2.2. Neuropsychological assessment

Participants completed the patient version of the Quick Dementia Rating System (QDRS) ²⁴ for a global subjective rating of cognitive status, and the Healthy Brain 9 (HB9) ¹⁸ for a rating of subjective cognitive decline (SCD). The Montreal Cognitive Assessment (MoCA) ²⁵ was administered for a global screen. The neuropsychological test battery included items from the Uniform Data Set (UDS) v3.0 ²⁶ covering memory (Craft Story paragraph recall–verbatim and paraphrase), language (Animal Naming and Multilingual Naming Test), executive function (Trail Making A and B), working memory (Numbers Forward and Backward), and visual (Benton Figure Copy and Recall). The UDS v3.0 battery was supplemented with the Hopkins Verbal Learning Task (episodic memory for word lists–immediate, delayed, and recognition). ²⁷ Raw scores for each test were converted into z‐scores based on the published means and standard deviations to create global and domain‐specific z‐scores. Participants also completed Cognivue Clarity, a 10‐minute computerized cognitive battery, providing a global score (range 0–100) with scores less than 69 representing cognitive impairment. ²⁸

2.2.3. Clinical diagnoses

The CDR, CDR‐SB, and the Global Deterioration Scale (GDS) ²⁹ were used for global staging. While both the CDR and GDS rate normal cognition, MCI, and dementia, only the GDS contains a category for subjective cognitive impairment (GDS 2). Clinical and cognitive data (excluding components of the BHI) were consolidated using a clinical consensus conference to assign individuals to the following diagnostic categories: no cognitive impairment (NCI), SCD, MCI, or ADRD according to published diagnostic criteria.

2.2.4. Magnetic resonance imaging

MRI scans were performed on 229 participants using a GE 3T 750 W scanner and included high‐resolution 3D sagittal MPRAGE, axial FLAIR, and T2* sequences. Quantitative morphometry was assessed with Combinostics cMRI (Finland), a United States Food and Drug Administration (FDA) ‐cleared artificial intelligence (AI) pipeline, to provide cortical, ventricular, and subcortical volumes and white matter hyperintensity burden. ³⁰ Cortical atrophy scores (CAS), generated as z‐score estimates, assessed global cortical atrophy derived from the global cortical atrophy four‐point visual rating scale and adjusted for head size, age, and sex based on normative data from a sample of individuals aged 50–90. Higher CAS scores indicate greater atrophy. ³¹

2.2.5. Blood‐based biomarkers

Blood‐based markers were analyzed in 361 participants using two commercially available platforms. PrecivityAD2 (C2N, St Louis, MO) ³² uses mass spectrometry to provide measures of Aβ40, Aβ42, Aβ42/Aβ40, apolipoprotein E (ApoE) ε4 proteotype, ptau217, ptau217% ratio (e.g., the ratio of phosphorylated to non‐phosphorylated tau217), and the Amyloid Probability Score 2 (APS2). Simoa SR‐X (Quanterix Corporation, Billerica, MA) ³³ uses an immunohistochemical approach to provide measures of ptau181, neurofilament light chain (NfL), and glial fibrillary acidic protein (GFAP). General health labs were measured by Cleveland Heart Labs, a CLIA‐cleared clinical laboratory, and included hemoglobin A1C, high‐density lipoprotein (HDL) cholesterol, low‐density lipoprotein (LDL) cholesterol, total cholesterol, Apolipoprotein A1, Homocysteine, Lipoprotein (a), total cortisol, estimated glomerular filtration rate (eGFR), and high‐sensitivity C‐reactive protein (CRP).

2.3. Components of the Brain Health Index

The Brain Health Index is a numerical index based on a previous study, the Brain Health Platform ¹² , derived from the Resilience Index which measures cognitive resilience based on modifiable factors, the Vulnerability Index, which measures vulnerability to developing impairment based on modifiable and non‐modifiable factors, and the Number Symbol Coding Task, which is a measure of executive functioning using task‐switching and symbol matching. The BHI can be completed in‐person or remotely through an online application in ∼15 min.

2.3.1. Number Symbol Coding Task score

The NSCT is a 90‐second assessment used to measure executive function through examining attention, planning, and set switching. ¹⁵ Participants are provided with a key with one symbol matched to each number. After a brief practice session acquainting the participant with the test, 70 boxes are then presented. The final score (possible range of scores 0–70) is calculated as the number of correct matches after the 90‐second time limit expires, with higher scores representing better cognitive performance.

2.3.2. Vulnerability Index

The VI is a weighted sum of 12 risk factors associated with cognitive impairment. The 12 factors include: age, biological sex, race, years of education, obesity, frailty, depression, diabetes, stroke, heart disease, hypercholesterolemia, and hypertension. ¹⁴ Each factor receives differential weighting based on its established association with cognitive impairment risk; age is categorized as 0 points for individuals under 60, one point for ages 60–75, and three points for those over 75; women receive two points compared to one point for men, reflecting higher dementia prevalence, while Black and Hispanic participants receive two points versus one point for non‐Hispanic White participants; educational attainment is inversely weighted, with two points assigned for 12 or fewer years, one point for 13–16 years, and 0 points for more than 16 years of education; medical comorbidities including diabetes and stroke receive enhanced weighting of two points each due to their strong associations with cognitive decline, while other conditions receive one point when present. Higher scores indicate a higher risk, ranging from 2 to 20. Higher scores indicate a higher risk, ranging from 2 to 20. Other ADRD risk factors reported in the literature (e.g., smoking, alcohol use, head injury) were tested but did not reach significance and were not included in the final version of the VI. ¹⁴ In the BHI, the VI is inverted so that higher scores are associated with better brain health, to match the scales of both the NSCT and RI.

2.3.3. Resilience Index

The RI measures six modifiable lifestyle factors associated with cognitive resilience. ¹³ These six factors consist of the Cognitive Reserve Unit Scale (CRUS), the Social Engagement Scale, Quick Physical Activity Rating (QPAR), Cognitive & Leisure Activity Scale (CLAS), Mediterranean‐DASH Intervention for Neurodegenerative Delay (MIND), and Applied Mindfulness Process Scale (AMPS).

The Cognitive Reserve Unit Scale is derived from scaled scores for the highest educational and occupational attainment achieved by the participants, with a range of scores from 0 to 66, with higher scores representing higher cognitive reserve. The social engagement scale measures participation in social activities using a 1–4 Likert scale. The QPAR measures the frequency and duration of physical activity across 10 categories of increasing intensity with a range of scores between 0 and 153, with higher scores representing greater physical activity. The CLAS measures the frequency of participation in 15 categories of cognitively stimulating activities with a range of scores from 0 to 80, with higher scores signifying greater cognitive activity. The Mediterranean‐DASH Intervention for Neurodegenerative Delay (MIND) Diet Score Sheet is a list of 15 categories of foods with frequencies of intake that measures the adherence to the diet with scores ranging from 0 to 15 (higher scores represent healthier dietary choices). The AMPS measures the frequency of practicing mindful techniques such as decentering and emotional regulation with a range of scores from 0 to 60, with higher scores reflecting a greater practice of mindfulness. ¹³

To calculate the RI score, these six factors are totaled without regard to weighting, with a range between 1 and 378; higher scores indicate greater resilience, with a threshold of 143 and below being considered “high risk” or low resilience.

2.4. Calculation of BHI

We aimed to create an index score ranging from 0 to 100, with higher scores representing optimal brain health and lower scores indicating a greater risk for current or future cognitive impairment. As the score is calculated from a combination of both modifiable and non‐modifiable factors, we also aimed for the BHI to change as lifestyle modifications are made, reflecting an improvement of brain health across time points.

Eight weighted components make up the BHI calculation: the NSCT score, the VI score, and the six factors of the RI with individual weights based on importance calculations. ³⁴ Each component is first normalized between 0 and 100 based on scale‐based minimum and maximum values. Component weights were then determined based on feature importance analyses using SHAP (SHapley Additive exPlanations) ³⁴ , factor analysis, and domain expertise. The NSCT was weighted the highest at 4.5x; the VI and MIND Diet were weighted at 2.5x; the QPAR scale was weighted at 2.0x; the Social Engagement scale and CLAS were both weighted at 1.0x; finally, the CRUS and AMPS were both weighted at 0.5x.

The sum of all normalized and weighted components results in a maximum value of 14.5. Instead of dividing this value by 14.5 to maintain normalization, they were instead divided by 13, with values above 1 thresholded to 1, to maximize the variance of the lower‐to‐mid points of the distribution and eliminate the effects of high‐performing outliers. This value was then multiplied by 100 to result in a final score between 0 and 100.

2.4.1. Statistical analyses

Analyses of covariance (ANCOVAs) using age as a covariate were used on continuous data, with Tukey post‐hoc tests, when age was significant between groups; as the BHI and VI build age into their calculations, they were examined using standard analyses of varianc (ANOVAs) to avoid overcorrection. Chi‐squared analyses were used to examine categorical variables (race and biological sex), as well as comparing diagnostic status (CN and MCI) and BHI threshold impairment status. Linear regressions using BHI as a predictor and other cognitive, functional, demographic, and biomarker variables as the predicted variable were calculated using the Python package pingouin ³⁵ 0.5.5. Threshold scores for determining cognitive risk from the BHI were calculated using R version 4.2.1, cutpointr packaged version 1.1.2. Within‐groups comparisons of the top and bottom 20% of BHI scores for each diagnostic group (CN, MCI, ADRD) were compared using ANOVAs. Alpha was set at 0.002.

The ability of BHI to function as a partial or complete mediator between the baseline and year 1 measures of variables was explored. Mediation analyses for calculating longitudinal mediation effects of the BHI on cognitive, functional, neurological, and biomarker variables were performed using pingouin. Complete mediation occurs when the BHI fully explains the relationship between time and the interest variable, with the longitudinal effect becoming non‐significant when BHI is included in the model, indicating that BHI drives any longitudinal change. Partial mediation occurs when the BHI explains only some of that relationship, with the direct relationship with time only reducing in magnitude.

After thresholds for the BHI were created, sensitivity, specificity, positive predictive values (PPV), negative predictive values (NPV), positive and negative likelihood ratios (LR), and diagnostic odds ratios (DOR) were determined. Likelihood ratios range from 0 to infinity, with larger numbers providing more convincing evidence of disease and smaller numbers indicating that disease is less likely. Unlike positive and negative predictive values, likelihood ratios are independent of disease prevalence. The DOR is derived by the ratio of positive LR/negative LR, providing a measure of test effectiveness.

3. RESULTS

3.1. Participant characteristics

Participants (n = 469) had a mean age of 69.4 ± 9.8 years (range 38–93) with a mean education of 15.8 ± 3.1 years (range 9–20). A total of 67.4% were non‐Hispanic White individuals, and 66.3% were female (Table 1). Participants had a mean baseline BHI score of 60.6 ± 11.5 (range 26–86). Final consensus diagnoses indicated 258 participants were cognitively normal controls (156 NCI, 102 SCD), 140 participants were diagnosed with MCI (69 due to AD, 51 due to ADRD), and 49 participants were diagnosed with ADRD. Twenty participants were diagnosed as impaired but not MCI, and 1 participant was diagnosed with mild behavioral impairment; these participants, in addition to 2 ADRD participants with non‐neurodegenerative dementia, were excluded from comparisons of diagnosis, resulting in a sample size of 447 participants: 258 controls, 140 MCI, 49 AD/ADRD (Table 1). When impairment status was not examined (e.g., regressions using BHI as the dependent variable, thresholding analyses), all participants were included.

TABLE 1.

Participant characteristics.

Parameter	Cohort (n = 469)	CN (n = 258)	MCI (n = 140)	ADRD (n = 49)	p‐Value	NCI(n=156)	SCD (n = 102)	p‐Value *
Age, year #	69.4 (9.8)	67.4 (8.9)	71.2 (10.4)	77.3 (7.2)	<0.001 a	67.7 (9.3)	66.9 (8.5)	0.439
% Female #	66.0%	74.0%	55.7%	53.1%	<0.001 b	0.7 (0.4)	0.7 (0.4)	0.883
Race, % non‐Hispanic White #	69.6%	66.7%	67.1%	91.8%	0.055	0.7 (0.5)	0.6 (0.5)	0.184
Education, year #	15.7 (3.0)	16.1 (2.8)	15.5 (3.0)	14.9 (2.4)	0.022	16.3 (2.6)	15.8 (3.1)	0.224
% ApoE carrier	30.0%	28.4%	27.4%	40.0%	0.132	28.7%	28.1%	0.924
FAQ total	2.1 (4.9)	0.5 (1.4)	1.6 (3.1)	11.5 (9.1)	<0.001 d	0.3 (0.8)	0.9 (1.9)	<0.001
QDRS total	1.7 (2.3)	0.9 (1.2)	2.0 (2.2)	5.1 (3.7)	<0.001 a	0.4 (0.6)	1.7 (1.4)	<0.001
AD8 total	1.7 (1.8)	1.2 (1.5)	2.3 (1.9)	3.2 (2.1)	<0.001 a	0.3 (0.6)	2.5 (1.5)	<0.001
DA3 depression #	5.1 (3.4)	4.9 (3.26)	5.4 (3.7)	8.7 (4.9)	0.017	4.3 (2.8)	6.1 (3.4)	<0.001
DA3 anxiety	3.5 (3.1)	3.3 (3.0)	3.6 (3.1)	6.0 (4.0)	0.051	2.7 (2.5)	4.4 (3.4)	<0.001
MoCA total	24.4 (4.2)	26.6 (2.4)	22.9 (2.8)	17.2 (5.1)	<0.001 a	26.9 (2.3)	26.1 (2.4)	0.006
HB9 total	4.3 (5.1)	3.6 (4.1)	5.5 (6.0)	8.7 (9.1)	0.001 b	1.9 (2.0)	6.4 (4.9)	<0.001
CDR Sum of Boxes	0.9 (1.9)	0.1 (0.2)	0.9 (0.6)	5.4 (3.4)	<0.001 a	0.04 (0.1)	0.1 (0.3)	0.004
Cognitive Z‐Score: Memory	−0.24 (0.75)	0.08 (0.58)	−0.90 (0.60)	−2.01 (0.20)	<0.001 a	0.11 (0.57)	0.04 (0.61)	0.372
Cognitive Z‐Score: Executive Function	2.41 (0.88)	2.76 (0.53)	1.86 (0.99)	1.10 (0.80)	<0.001 a	2.78 (0.51)	2.74 (0.56)	0.545
Cognitive Z‐Score: Attention	−0.83 (0.51)	−0.65 (0.47)	−1.07 (0.44)	−1.04 (0.57)	<0.001 b	−0.67 (0.45)	−0.63 (0.51)	0.517
Cognitive Z‐Score: Language	−0.28 (0.97)	0.04 (0.66)	−0.82 (1.24)	−2.24 (0.86)	<0.001 a	0.04 (0.62)	0.03 (0.72)	0.872
Cognitive Z‐Score: Visuospatial	0.22 (0.81)	0.39 (0.72)	−0.18 (0.86)	−1.60 (0.63)	<0.001 a	0.40 (0.70)	0.38 (0.75)	0.805
Cognitive Z‐Score: Global	0.94 (0.54)	0.52 (0.35)	−0.26 (0.48)	−1.08 (0.36)	<0.001 a	0.52 (0.32)	0.51 (0.39)	0.761
Number‐symbol coding task #	38.4 (11.7)	43.8 (8.9)	34.4 (9.5)	21.9 (11.1)	<0.001 a	43.9 (9.3)	43.7 (8.5)	0.827
Vulnerability index #	8.1 (3.0)	7.2 (2.5)	8.7 (3.1)	11.4 (2.8)	<0.001 a	7.0 (2.6)	7.5 (2.3)	0.080
Resilience Index #	159.7 (39.8)	174.8 (33.0)	151.3 (37.6)	104.7 (25.9)	<0.001 a	179.8 (32.1)	167.1 (33.1)	0.003
RI component: CRUS #	46.6 (13.6)	49.5 (12.0)	44.6 (13.6)	38.0 (13.9)	<0.001 a	50.3 (11.7)	48.1 (12.4)	0.155
RI component: Social #	3.2 (0.8)	3.3 (0.6)	3.1 (0.8)	2.5 (0.9)	<0.001 c	3.5 (0.6)	3.2 (0.7)	<0.001
RI component: QPAR #	33.9 (19.6)	38.8 (19.5)	31.3 (18.2)	17.8 (14.3)	<0.001 a	39.1 (19.2)	38.2 (20.2)	0.718
RI component: CLAS #	30.0 (9.5)	31.9 (9.3)	29.1 (9.2)	22.9 (8.0)	<0.001 c	32.6 (9.2)	30.8 (9.3)	0.122
RI component: MIND #	9.51 (2.2)	9.8 (2.2)	9.5 (2.1)	8.3 (2.2)	<0.001 c	10.0 (2.1)	9.5 (2.3)	0.077
RI component: AMPS #	42.5 (9.9)	44.3 (8.6)	41.6 (10.2)	35.8 (12.1)	<0.001 c	46.5 (7.4)	41.0 (9.2)	<0.001
Brain Health Index	60.6 (11.5)	66.0 (8.6)	57.2 (9.2)	42.5 (9.6)	<0.001 a	67.1 (8.8)	64.3 (8.1)	0.010

Abbreviations: ADRD, Alzheimer's disease and related disorders; AMPS, Applied Mindfulness Process Scale; BHI, Brain Health Index; CDR, Clinical Dementia Rating; CLAS, Cognitive Leisure Activity Scale; CN, cognitively normal; CRUS, Cognitive Reserve Unit Scale; DA3, Depression, Anxiety, Apathy Assessment; FAQ, Functional Activities Questionnaire; GFAP, glial fibrillary acidic protein; HB9, healthy brain 9; MCI, mild cognitive impairment; MIND, Mediterranean‐DASH intervention for neurodegenerative delay; MoCA, Montreal Cognitive Assessment; NCI, no cognitive impairment; QDRS, Quick Dementia Rating System, patient‐reported; QPAR, Quick Physical Activity Rating; SCD, subjective cognitive decline.

^*Values in boldface are statistically significant.

^# Included as a measure within the BHI.

^a Significant difference between all groups.

^b Significant between CN and MCI/ADRD.

^c Significant between CN and ADRD.

^d Significant between ADRD and CN/MCI.

3.2. BHI and other measures across diagnostic groups

Age was significantly different between diagnostic groups (p < 0.001). Sex at birth also significantly differed between the controls and AD/ADRD (p < 0.001), with female representation decreasing from CN (74.03%) to MCI (55.71%) and ADRD (53.06%). Years of education showed a decreasing trend across diagnostic categories (16.1 years for CN, 15.5 for MCI, 15.0 for ADRD; p = 0.022). These demographic variables are included in the VI, a component of the BHI.

Differences were observed for most clinical, cognitive, and functional measures (Table 1). The VI, NSCT, the total RI, and the BHI were significantly different between all groups (all p < 0.001). Of the RI components, the QPAR and CRUS were significantly different between all groups (all p < 0.001), while the Social Engagement, CLAS, MIND Diet, and AMPS were only significant between controls and AD/ADRD.

3.3. Comparisons within diagnostic groups

Within‐groups analyses using ANOVA revealed significant differences between the top 20% and bottom 20% of BHI scores for each diagnostic group.

Using an ANOVA, the top 20% of the control group (BHI 77.6 ± 3.3; N = 55) and the bottom 20% (BHI 53.8 ± 4.2; N = 52) were significantly different (p < 0.001). This is reflective of better brain health and differences in demographic, performance, and biomarker variables: age (63.6 ± 8.6 vs. 69.9 ± 9.6, p = 0.001), non‐Hispanic White race (84.0% vs. 48.1%, p < 0.001), years of education (17.0 ± 2.2 vs.14.4 ± 3.0, p < 0.001); HB9 (2.2 ± 3.4 vs. 6.0 ± 5.5, p < 0.001); memory z‐score (0.5 ± 0.6 vs. −0.1 ± 0.6, p < 0.001), executive function z‐score (0.7 ± 0.4 vs. −0.2 ± 0.5, p < 0.001), language z‐score (0.4 ± 0.6 vs. −0.0 ± 0.6, p < 0.001), mPPT (13.5 ± 2.0 vs.11.5 ± 2.9, p < 0.001), ApoE A1 (5.3 ± 0.3 vs. 5.8 ± 0.7, p < 0.001), and eGFR (75.1 ± 14.9 vs. 58.1 ± 20.4, p < 0.001).

The top 20% of the MCI group (N = 35, BHI 68.6 ± 4.4) and the bottom group (N = 29, BHI 44.5 ± 4.9) showed significant differences in BHI scores (p < 0.001). Similar to the control group, this is reflective of better brain health and differences in demographic, performance, and biomarker variables: age (66.9 ± 9.9 vs. 75.5 ± 10.5, p = 0.001), AD8 (1.3 ± 1.3 vs. 2.9 ± 2.0, p = 0.001), HB9 (3.0 ± 3.4 vs. 10.5 ± 9.4, p = 0.001), CDR‐SB (0.7 ± 0.3 vs. 1.4 ± 0.8, p < 0.001), executive function z‐score (−0.1 ± 0.8 vs. −1.7 ± 1.0, p < 0.001), global z‐score (−0.4 ± 0.4 vs. −0.9 ± 0.6, p = 0.001), mPPT (12.2 ± 2.44 vs. 9.3 ± 2.8, p < 0.001), NPI (1.3 ± 2.8 vs. 4.3 ± 4.6, p = 0.002), and white matter hyperintensity (WMH) burden (2.9 ± 1.9 vs. 9.1 ± 6.5, p = 0.001).

Fewer significant differences were found within the ADRD group. In addition to the BHI (p < 0.001), the top 20% (N = 11, BHI 56.82 ± 4.21) differed from the bottom 20% (N = 10, BHI 30.80 ± 2.93) in the MoCA total (21.18 ± 3.25 vs. 14.90 ± 4.91, p = 0.002) and high‐sensitivity CRP (1.02 ± 0.68 vs. 3.65 ± 1.66, p = 0.002).

3.4. BHI and subjective cognitive complaints

Of the 258 control participants, 156 had NCI and 102 had SCD. The total RI and two components of the RI were significantly lower in the SCD group; total RI (NCI:179.8 ± 32.1; SCD: 167.1 ± 33.1, p = 0.003), Social Engagement Scale (NCI: 3.2 ± 0.7; SCD: 3.5 ± 0.6 NCC, p < 0.001) and AMPS (NCI:41.0 ± 9.2 SCD; 46.5 ± 7.4, p < 0.001). The BHI was trending lower in the SCD group (NCI:67.1 ± 8.8; SCD: 64.3 ± 8.1, p = 0.01). There was no difference in VI or NSCT scores.

3.5. Cross‐sectional associations of the BHI and brain health measures

Associations with the BHI were examined using linear regression analyses, with BHI included as the independent predictor variable. Strong associations were observed with functional impairment metrics, including FAQ Total (r ² = 0.278, p < 0.001), QDRS Total (r ² = 0.238, p < 0.001), AD8 Total (r ² = 0.175, p < 0.001), and CDR‐SB (r ² = 0.355, p < 0.001). Cognitive performance measures were reliably predicted by BHI, most notably with the MoCA Total (r ² = 0.408, p < 0.001) and executive function z‐score (r ² = 0.366, p < 0.001). BHI was a significant predictor of NfL (r ² = 0.073, p < 0.001), GFAP (r ² = 0.035, p < 0.001), hippocampal volume (r² = 0.249, p < 0.001), and medial temporal lobe volume (r² = 0.170, p < 0.001). The mPPT (r² = 0.336, p < 0.001), A1C (r ² = −0.122, p < 0.001), and LDL cholesterol (r ² = 0.107, p < 0.001) were also significantly predicted (Table 2).

TABLE 2.

Cross‐sectional associations of BHI with clinical, cognitive, and biomarker measures.

Variable	N	Reg r 2	Reg p‐value
Age, year #	469	0.172	<0.001
% Female #	469	0.013	0.022
Race, % non‐Hispanic white #	469	0.102	<0.001
% Cognitively impaired	469	0.272	<0.001
Education, year #	469	0.105	<0.001
% ApoE carrier	383	0.001	0.077
FAQ total	469	0.278	<0.001
QDRS total	469	0.238	<0.001
AD8 total	469	0.175	<0.001
DA3 depression #	469	0.187	<0.001
DA3 anxiety	469	0.117	<0.001
MoCA total	469	0.408	<0.001
HB9 total	469	0.161	<0.001
CDR Sum of Boxes	469	0.355	<0.001
Cognivue Clarity total score	332	0.068	<0.001
Amyloid beta 42/40 ratio (plasma; C2N)	315	0.262	<0.001
Phosphorylated pTau 181 ratio (plasma; C2N)	213	0.118	0.694
Phosphorylated pTau 217 %ratio (plasma; C2N)	315	0.036	0.456
Neurofilament light (plasma; quanterix)	328	0.073	<0.001
GFAP (plasma; quanterix)	328	0.035	<0.001
mPPT total	469	0.336	<0.001
Cognitive Z‐Score: Memory	469	0.168	<0.001
Cognitive Z‐Score: Executive Function	469	0.366	<0.001
Cognitive Z‐Score: Attention	469	0.084	<0.001
Cognitive Z‐Score: Language	469	0.166	<0.001
Cognitive Z‐Score: Visuospatial	469	0.036	<0.001
Cognitive Z‐Score: Global	469	0.282	<0.001
Blood panel: Hemoglobin A1C	361	0.122	<0.001
Blood panel: HDL cholesterol	361	0.045	0.044
Blood panel: LDL cholesterol	361	0.107	<0.001
Blood panel: Total cholesterol	361	0.084	<0.001
Blood panel: Apolipoprotein A1	361	0.032	<0.001
Blood panel: Homocysteine	361	0.021	0.025
Blood panel: Lipoprotein (a)	361	0.012	0.024
Blood panel: Total cortisol	361	0.007	<0.001
Blood panel: eGFR	361	0.060	<0.001
MRI: Hippocampus total volume	257	0.249	<0.001
MRI: Parahippocampal gyrus total volume	257	0.113	<0.001
MRI: Inferior lateral ventricle total volume	257	0.058	<0.001
MRI: Medial temporal lobe total volume	257	0.170	<0.001
MRI: WMH volume	257	0.129	0.311
MRI: Ventricles total volume	257	0.041	<0.001
MRI: Total cortical volume	257	0.200	0.102
MRI: Cortical atrophy score	257	0.036	<0.001
Tobacco consumption (years)	153	0.036	0.034
Alcohol consumption (years)	215	0.172	0.057

Note: Values in boldface are statistically significant.

Abbreviations: BHI, Brain Health Index; CDR, Clinical Dementia Rating; DA3, Depression, Anxiety, Apathy Assessment; eGFR, Estimated Glomerular Filtration Rate; FAQ, Functional Activities Questionnaire; GFAP, glial fibrillary acidic protein; HB9, healthy brain 9; HDL, high density lipoprotein; LDL, low density lipoprotein; MoCA, Montreal Cognitive Assessment; mPPT, Mini Physical Performance Test; QDRS, Quick Dementia Rating System, patient‐reported; Reg r ², regression analysis, Pearson R‐value; WMH, white matter hyperintensity.

^# Included as a measure within the BHI

3.6. Longitudinal measures of BHI

From our sample, 166 participants returned for a second visit (119 CN, 5 impaired not MCI, 42 MCI; age: 68.9 ± 9.4). BHI did not significantly differ between baseline (M = 65.3, SD = 9.4) and year 1 (M = 65.2, SD = 10.3). Mediation analyses were performed with BHI as the mediator and time as the independent variable. Complete mediation was found with plasma markers of NfL (p = 0.012) and GFAP (p = 0.012), as well as executive function z‐scores (p = 0.008) and the HB9 (p = 0.012). Partial mediation was observed in the ptau217 ratio, QDRS, AD8, MoCA total, CDR‐SB, and mPPT.

Six participants progressed from control to MCI between baseline and the year 1 follow‐up. A repeated measures ANOVA revealed a significant interaction between group (progressed vs stable) and time (F = 6.31, p = 0.013), indicating that the progressed group showed a different pattern of BHI change over time. Pairwise t‐tests confirmed that while both groups had similar BHI scores at baseline (p = 0.33), the progressed group had significantly lower BHI scores at the one‐year follow‐up compared to stable participants (57.8 ± 5.0 vs. 65.5 ± 10.3, p = 0.011).

3.7. Thresholding the BHI

Thresholds for the BHI were established to facilitate clinical interpretation and risk stratification. Cutpoint analysis identified three clinically meaningful threshold categories: low (≤54), indeterminate (55–62), and high (≥63). The diagnostic composition differed significantly across these thresholds (Table 3). The low BHI group (n = 122) comprised 41 ADRD (33.6%), 58 MCI (47.5%), and 23 CN (18.9%) participants, 13 of whom had subjective cognitive complaints. The high BHI group (n = 219) consisted of 46 MCI (21.0%), and 172 CN (78.5%) participants, with 1 ADRD (frontotemporal degeneration). The indeterminate BHI group (n = 126) included 8 ADRD (0.6%), 55 MCI (43.6%), and 63 CN (50.0%) participants and likely represents a mix of cognitive normal individuals at increased risk of cognitive impairment and impaired individuals with better cognitive performance and lower risk of progression (Figure 1).

TABLE 3.

Thresholding of the BHI.

Parameter	Low (BHI ≤ 54)	Indeterminate (BHI 55‐62)	High (BHI ≥ 63)	p‐Value
% Impaired X	81.1%	50.0%	21.5%	<0.001 a
Age, year #	74.4 (10.2)	70.3 (8.8)	66.2 (8.8)	<0.001 a
% Female # , X	60.7%	65.9%	69.9%	0.225
Race, % non‐Hispanic white # , X	63.9%	63.5%	72.1%	0.151
Education, year #	14.3 (2.9)	15.7 (3.0)	16.6 (2.8)	<0.001 a
FAQ total	5.8 (7.9)	1.1 (2.4)	0.4 (1.3)	<0.001 b
QDRS total	3.3 (3.3)	1.5 (1.8)	0.8 (1.1)	<0.001 a
AD8 total	2.8 (2.2)	1.9 (1.8)	1.1 (1.3)	<0.001 a
DA3 depression #	6.7 (3.7)	5.8 (3.7)	4.3 (2.8)	<0.001 b
DA3 anxiety	4.6 (3.2)	3.9 (3.6)	2.9 (2.6)	<0.001 b
MoCA total	20.9 (5.1)	24.5 (3.1)	26.4 (2.5)	<0.001 a
HB9 total	7.4 (6.5)	5.1 (5.7)	2.8 (3.2)	<0.001 a
CDR Sum of Boxes	2.6 (3.2)	0.6 (0.9)	0.2 (0.4)	<0.001 a
Cognitive Z‐Score: Memory	−0.71 (0.75)	−0.39 (0.71)	−0.01 (0.68)	<0.001 a
Cognitive Z‐Score: Executive function	1.54 (1.06)	2.27 (0.72)	2.76 (0.60)	<0.001 a
Cognitive Z‐Score: Attention	−0.99 (0.51)	−0.87 (0.53)	−0.72 (0.48)	<0.001 b
Cognitive Z‐Score: Language	−0.84 (1.20)	−0.43 (0.95)	−0.01 (0.79)	<0.001 a
Cognitive Z‐Score: Visuospatial	−0.07 (0.87)	0.21 (0.79)	0.32 (0.78)	0.003
Cognitive Z‐Score: Global	−0.18 (0.62)	0.15 (0.47)	0.47 (0.42)	<0.001 a
Cognivue Clarity total score	64.9 (14.6)	70.7 (12.7)	77.5 (12.9)	<0.001 b
Amyloid beta 42/40 ratio (plasma; C2N)	0.11 (0.01)	0.10 (0.01)	0.10 (0.01)	0.761
pTau 181 ratio (plasma; C2N)	0.17 (0.05)	0.17 (0.04)	0.17 (0.05)	0.656
pTau 217 %ratio (plasma; C2N)	1.9 (2.3)	2.1 (2.3)	1.5 (1.7)	0.060
Neurofilament light (plasma; quanterix)	16.9 (11.5)	13.4 (7.2)	11.2 (5.4)	<0.001 b
GFAP (plasma; quanterix)	253.3 (212.8)	200.2 (102.2)	186.5 (95.0)	0.002 b
mPPT total	9.6 (3.2)	11.7 (2.4)	12.8 (2.1)	<0.001 a
Blood panel: Hemoglobin A1C	6.0 (1.0)	5.6 (0.5)	5.5 (0.5)	<0.001 a
Blood panel: HDL cholesterol	61.2 (19.8)	63.5 (12.9)	67.4 (18.9)	0.078
Blood panel: LDL cholesterol	96.7 (35.8)	103.5 (37.8)	109.4 (33.4)	0.024
Blood panel: Total cholesterol	178.0 (41.6)	187.8 (42.7)	195.6 (39.8)	0.005
Blood panel: Lipoprotein (a)	89.8 (103.8)	78.8 (74.5)	68.6 (88.2)	0.255
Blood panel: eGFR	60.0 (20.1)	67.4 (19.4)	69.3 (16.0)	0.007
MRI: Hippocampus total volume	6.8 (1.0)	7.2 (1.1)	7.5 (0.8)	<0.001 e
MRI: Parahippocampal gyrus total volume	5.7 (0.9)	5.8 (0.8)	6.0 (0.7)	0.022
MRI: Inferior lateral ventricle total volume	1.5 (0.9)	1.2 (0.7)	1.0 (0.6)	<0.001 e
MRI: Medial temporal lobe total volume	19.3 (2.8)	19.7 (2.5)	20.5 (2.0)	0.004
MRI: WMH volume	9.0 (11.1)	5.1 (6.6)	4.1 (7.1)	0.003
MRI: Ventricles total volume	46.2 (22.0)	35.4 (16.7)	32.9 (17.9)	<0.001 e
MRI: Total cortical volume	469.9 (58.3)	485.0 (44.4)	503.2 (57.0)	0.001 e
MRI: Cortical atrophy score	2.0 (1.1)	1.5 (0.8)	1.4 (0.9)	<0.001 e
Number‐symbol coding task #	26.5 (10.2)	36.7 (6.5)	46.1 (8.3)	<0.001 a
Vulnerability Index #	11.2 (2.9)	8.4 (2.0)	6.3 (2.0)	<0.001 a
Resilience Index #	120.0 (30.6)	155.4 (26.7)	184.5 (30.8)	<0.001 a
RI component: CRUS #	38.5 (14.8)	45.6 (13.2)	51.9 (10.1)	<0.001 a
RI component: Social engagement #	2.8 (0.9)	3.1 (0.7)	3.5 (0.6)	<0.001 a
RI component: QPAR #	20.5 (13.7)	31.4 (14.8)	42.7 (20.2)	<0.001 a
RI component: CLAS #	24.4 (9.5)	30.2 (8.7)	33.1 (8.6)	<0.001 a
RI component: MIND #	7.9 (2.1)	9.4 (1.9)	10.5 (1.9)	<0.001 a
RI component: AMPS #	36.7 (10.4)	42.4 (8.9)	46.0 (8.6)	<0.001 a
BHI	45.5 (7.0)	58.7 (2.4)	70.2 (5.5)	<0.001 a

Abbreviations: AMPS, Applied Mindfulness Process Scale; BHI, Brain Health Index; CDR, Clinical Dementia Rating; CLAS, Cognitive Leisure Activity Scale; CRUS, Cognitive Reserve Unit Scale; DA3, Depression, Anxiety, Apathy Assessment; eGFR, estimated Glomerular Filtration Rate; FAQ, Functional Activities Questionnaire; GFAP, glial fibrillary acidic protein; HB9, healthy brain 9; HDL, high density lipoprotein; LDL, low density lipoprotein; MIND, Mediterranean‐DASH intervention for neurodegenerative delay; MoCA, Montreal cognitive impairment; mPPT, Mini Physical Performance Test; QDRS, Quick Dementia Rating System, patient‐reported; QPAR, Quick Physical Activity Rating; WMH, white matter hyperintensity.

^# Included as a measure within the BHI,

^X Chi‐squared; all other analyses are analysis of variance.

^a Significant difference between all groups.

^b Significant between high and low/indeterminate.

^c Significant between low/high and indeterminate.

^d Significant between low and Indeterminate/high.

^e Significant between low and high.

FIGURE 1.

Histogram displaying the Brain Health Index (BHI) by diagnostic status with thresholds. BHI scores are displayed for cognitively normal (CN, green), mild cognitive impairment (MCI, orange), and Alzheimer's disease and related disorders (ADRD, red) participants. The shaded gray region (BHI 55‐62) represents the indeterminate risk category, separating the low BHI score threshold (≤54), comprised of 81.0% impaired individuals with 11/20 cognitively normal (CN) participants having subjective cognitive complaints, from the high BHI score threshold (≥63), comprised of 79% CN participants with only one ADRD participant with frontotemporal degeneration. The middle (indeterminant) range of BHI scores had a mixture of CN, mild cognitive impairment (MCI), and ADRD cases, representing a mix of cognitive normal individuals at increased risk of cognitive impairment and impaired individuals with better cognitive performance and lower risk of progression.

Clinical and functional measures demonstrated significant differences across threshold groups. FAQ scores were progressively lower, indicating better functioning, in the high BHI group (5.8 ± 7.9 in low, 1.1 ± 2.4 in indeterminate, 0.4 ± 1.3 in high; p < 0.001). Similar patterns were observed for QDRS Total (3.3 ± 3.3 to 0.8 ± 1.1), AD8 Total (2.8 ± 2.2 to 1.1 ± 1.3), and CDR‐SB (2.6 ± 3.2 to 0.2 ± 0.4), all p < 0.001. Anxiety showed a similar pattern (4.6 ± 3.2 to 2.9 ± 2.6, p < 0.001). The MoCA Total scores increased progressively across thresholds (20.9 ± 5.1 in low, 24.5 ± 3.1 in indeterminate, 26.4 ± 2.5 in high; p < 0.001).

Cognitive performance measures differed significantly across threshold groups. Executive function z‐scores showed the strongest differentiation, ranging from 1.54 ± 1.06 in the low group to 2.76 ± 0.60 in the high group (p < 0.001), likely due to the inclusion of the NSCT within the BHI. Global z‐scores also demonstrated significant differences (−0.18 ± 0.62 to 0.47 ± 0.42). Other domain z‐scores (memory, language, attention) demonstrated significant differences (all p < 0.001), while visuospatial z‐scores showed a modest difference which approached but did not reach significance (−0.07 ± 0.87 to 0.32 ± 0.78, p = 0.003).

Biomarker analyses further exhibited patterns of better health in higher groups. NfL concentrations decreased with higher BHI thresholds (16.9 ± 11.5 pg/mL in low, 13.4 ± 7.2 pg/mL in indeterminate, 11.2 ± 5.4 pg/mL in high; p < 0.001). GFAP levels showed a similar pattern (253.3 ± 212.8 pg/mL to 186.5 ± 95.0 pg/mL, p = 0.002). A1C values were significantly higher in the low BHI group (6.0 ± 1.0%) compared to indeterminate (5.6 ± 0.5%) and high (5.5 ± 0.5%) groups (p < 0.001). HDL cholesterol showed a non‐significant trend toward higher values across thresholds (61.2 ± 19.8 mg/dL to 67.4 ± 18.9 mg/dL, p = 0.078). No significant differences were observed in amyloid beta 42/40 ratio or phosphorylated tau measures across threshold groups, while MRI metrics demonstrated significant patterns, including higher hippocampal volume (6.8 ± 1.0 to 7.5 ± 0.8 mL, p < 0.001) in higher BHI groups. 

3.8. Predictive models of health outcomes

A Random Forest Classifier was developed to evaluate the predictive capacity of the BHI for determining cognitive impairment status. The model was trained using a three‐fold three‐repeat repeated stratified cross‐validation approach with hyperparameter optimization to ensure robustness and generalizability. The classifier achieved an overall accuracy of 70% (weighted average precision: 0.70, recall: 0.70, F1‐score: 0.70) in discriminating between cognitively impaired and non‐impaired participants. Sensitivity was calculated at 0.60, indicating moderate ability to correctly identify individuals with cognitive impairment, while specificity was notably higher at 0.79, demonstrating stronger performance in correctly identifying cognitively normal individuals. The positive predictive value (0.70) and negative predictive value (0.71) were balanced, suggesting similar performance in confirmation and exclusion prediction. The diagnostic odds ratio was 5.54, and the area under the receiver operating characteristic curve (AUC) averaged 0.76, indicating good discriminative ability. Confusion matrix analysis revealed that among 209 cognitively impaired participants, 125 were correctly classified (59.8%), while among 258 CN participants, 204 were correctly identified (79.1%).

Examining the performance of the thresholded BHI as a classifier of impairment status reveals a strong ability for high and low thresholds to detect high impairment and low impairment risk, but an inability for the Indeterminate BHI category to effectively discriminate between diagnostic status, reflecting the heterogeneous composition within this category. In the high BHI threshold, 172 out of 219 (78.5%) participants in that group were classified as cognitively normal. In the low BHI group, 99 out of 122 (81.1%) were classified as cognitively impaired. Using baseline evaluations, we examined the ability of the high and low BHI thresholds to distinguish cognitively normal individuals from those with cognitive impairment. This strategy performed well with a high specificity (88.2%) and PPV (81.4%). The sensitivity was lower at 68.2%, but the NPV was acceptable at 78.5%. The positive LR was 5.8 and the negative LR was 0.36, providing a DOR of 16.1, supporting the use of the BHI to predict individuals who were likely to have cognitive impairment.

However, the indeterminate BHI group was evenly split, with 50% of participants being cognitively impaired, mostly with MCI. These results reflect that the BHI is primarily designed to assess overall brain health, with risk of impairment as a downstream effect. CN individuals with lower BHI scores would be potentially at greater risk of future cognitive impairment and appear in the Indeterminate BHI range. Conversely, cognitively impaired individuals with higher BHI scores in the indeterminate HBI range would be predicted to progress more slowly or have a non‐degenerative etiology of their impairment. Few individuals with ADRD were in the indeterminate HBI range. Individuals who score within the indeterminate HBI range would need further evaluation to determine their cognitive status and underlying etiology.

4. DISCUSSION

The BHI represents a comprehensive metric of overall brain health status and risk of cognitive impairment that can be completed in‐person or online in ∼15 min. By integrating the VI, RI, and NSCT into a single weighted score, the BHI captures the complex interplay between risk factors, protective factors, and current cognitive performance in a manner that is greater than the sum of its parts. Building upon our prior work ¹² , which established that multidimensional examination of the three measures could identify cognitive status, the BHI advances this approach through the implementation of differential weighting strategies. This weighted combination allows for the relative contribution of each component to be optimized based on its discriminative power, resulting in superior performance compared to unweighted approaches.

Longitudinal analyses reveal that the BHI functions as a mediator of change in key biomarkers of neurodegeneration and neuroinflammation over a one‐year period. Furthermore, the BHI demonstrated robust cross‐sectional associations across multiple domains with established brain health measures, including functional impairment (i.e., FAQ), metabolic health indicators (e.g., LDL cholesterol, hemoglobin A1C), physical performance (i.e., mPPT), and domain‐specific cognitive z‐scores, even those not directly assessed (e.g., memory, language) by the BHI.

4.1. Performance of the BHI

While the BHI was not designed as a diagnostic instrument, its strong associations with clinical diagnoses and DOR of 16.1 demonstrate its validity as a measure of brain health status and risk of future impairment and progression. The BHI showed strong discriminative ability across the cognitive spectrum, with significant differences observed between controls, MCI, and AD/ADRD groups, especially when thresholds are applied. Notably, the BHI's performance is designed to be independent of etiology; this approach enhances the BHI's utility as a general brain health metric, as it captures common downstream effects of various neurodegenerative and non‐degenerative processes on overall brain function. The progressive decline in BHI scores across diagnostic categories, coupled with its strong correlations with objective markers of neurodegeneration, including plasma biomarkers (e.g., NfL, GFAP) and neuroimaging (e.g., hippocampal volume, WMH), promotes its use as a quantitative indicator of brain health. However, BHI should not be used as a diagnostic in isolation from a comprehensive clinical evaluation.

The BHI achieves this through a calibrated weighting strategy that reflects the relative importance of each component in measuring brain health. Executive functioning and current cognitive performance measured by the NSCT is the most heavily weighted component at 4.5×, reflecting that attention, problem solving, processing speed, and task‐switching abilities represent sensitive indicators of brain health that decline early across multiple neurodegenerative conditions, as opposed to memory tests that are less associated with non‐AD processes ¹⁵ , ³⁶ , ³⁷ . Nutrition, assessed by adherence to the MIND diet, contributes a substantial amount to brain health trajectories ³⁸ , ³⁹ , especially in its association with cardiovascular health; it is the next most heavily weighted component (2.5×). The VI, a measure of cumulative risk factor burden, is also weighted at 2.5×. Physical activity, assessed through the QPAR, reflects the well‐established neuroprotective effects of exercise ⁴⁰ and is weighted at 2×. Social engagement and cognitive leisure activities maintain standard weights, as they measure indirect contributors to brain health. The CRUS capturing the highest educational and occupational attainment and mindfulness contribute at 0.5×, reflecting their life‐long contributions to brain health rather than being directly actionable factors. This hierarchical structure, derived through feature importance analyses, ensures that the BHI captures the multidimensional nature of brain health. The BHI aligns with recent reports that multidomain lifestyle interventions can have beneficial effects on brain health and reduce the risk of ADRD ⁴¹ , ⁴² , and the use of the BHI can provide a unitary score to track brain health in the clinic or in research projects.

Within‐group analyses provide evidence for the BHI's capacity to detect heterogeneity in brain health status. Among controls, the differences observed between the top and bottom 20% of BHI scores, including subjective cognitive decline (i.e., HB9), cognitive performance, physical functioning, and health markers, suggest that the BHI identifies those individuals who may be at risk for future cognitive decline. Similar patterns are found within the MCI group, where top versus bottom BHI scores distinguished participants across CDR‐SB, neuropsychiatric symptoms, WMH burden, and cognitive performance. The more limited differences observed within the AD/ADRD group, restricted primarily to global cognition (i.e., MoCA) and inflammation (i.e., CRP), reflect that the modifiable factors captured by the BHI have diminished influence related to the more progressive underlying pathological processes. These findings underscore the BHI's potential to refine risk stratification within diagnostic categories, assist with early detection, and enable granular tracking of disease risk and progression.

4.2. Associations with biomarkers of brain health

The comprehensive nature of the BHI requires validation against measures beyond those traditionally associated with cognitive health, as many established risk factors, including age, educational attainment, and common medical comorbidities, are already incorporated into its calculation. BHI was a significant predictor of plasma biomarkers of neuronal injury (i.e., NfL) and neuroinflammation (i.e., GFAP), providing narrative for the BHI's ability to capture underlying neuropathological processes. The partial effects observed for other measures, including pTau‐217, functional assessments, and physical performance metrics, further underscore the BHI's capacity to predict longitudinal trajectories across multiple domains of brain health.

Beyond blood‐based biomarkers, the BHI demonstrated associations with neuroimaging and health indicators. Hippocampal volume, cortical atrophy, and WMH burden were each significantly predicted, providing structural validation for the BHI's sensitivity to neurodegenerative changes. Equally important are associations with systemic health markers, including glycemic control and lipid profiles, which reflect the increasingly recognized interconnection between metabolic health and cognitive function. The strong relationship with physical performance measures, particularly the mPPT, highlights how the BHI captures the bidirectional relationship between physical and cognitive health. ⁴³ , ⁴⁴

4.3. Subjective cognitive decline

The relationship between subjective complaints and the BHI reveals important insights into the limitations and complexities of self‐reported cognitive symptoms. Participants with SCD trended toward lower BHI scores compared to those without complaints, supporting that SCD individuals may be at higher risk of AD/ADRD, albeit at a much slower rate than individuals with MCI. ⁴⁵ SCD participants exhibited lower mindfulness scores than NCI participants, with no significant difference compared to MCI participants in mindfulness, diet, social engagement, cognitive leisure activities, and physical activity, which provides further support that SCD represents either a transitional state between normal cognition and MCI or the onset of age‐related cognitive decline.

4.4. Utility in clinical and research contexts

The BHI offers multiple avenues for clinical implementation (Figure 2), including in primary care settings where early detection, risk stratification, clinical decision making, and longitudinal monitoring of brain health represent an unmet need. ⁴⁶ The ∼15‐min BHI could be completed before a visit, to guide clinicians’ next steps and increase the ability to track brain health trajectories over time, with declining scores potentially signaling emerging issues that warrant further evaluation. Conversely, a patient whose BHI scores are improving may indicate successful adoption and adherence to brain‐healthy lifestyle modifications, providing both patients and providers with objective feedback on the effectiveness of preventative strategies. Implementation of early detection of cognitive impairment, for example, with the use of the BHI, could improve patient outcomes. ⁴⁶ Thresholds enhance clinical utility by offering clear risk stratification: the absence of ADRD cases in the high BHI group and the finding that 89.2% of low BHI participants exhibited cognitive impairment provide actionable cutpoints for clinical decision‐making. The stepwise progression of biomarkers across these thresholds, particularly in NfL and GFAP, suggests that the BHI could serve as a cost‐effective estimation of global changes in brain health.

FIGURE 2.

Example of an implementation scheme for incorporating the Brain Health Index (BHI) into clinical practice. This schematic representation demonstrates how the BHI can be implemented into primary and specialty practices. Three new patients present to the clinic for evaluation, each with unique medical histories, cognitive complaints, and risk factors. All three have the same score on a brief cognitive test—in this case 36 (borderline cognitive impairment) on the Number Symbol Coding Test (NSCT). However, each has different scores on the Resilience Index (RI) and Vulnerability Index (VI). BHI scores place each of the three individuals into different risk strata before their clinical visit. A two‐stage evaluation process could then offer an in‐office evaluation (history, physical, medication review) followed by a blood‐based biomarker such as pTau 217. The first patient with a BHI score of 39 falls in the high‐risk group; the office evaluation suggests mild cognitive impairment (MCI), and his biomarker suggests the presence of amyloid, supporting the diagnosis of MCI due to AD. A clinician could then initiate a treatment protocol with an amyloid‐lowering therapy or offer enrollment in a clinical trial. The second patient with a BHI of 69 is in the low‐risk group with only subjective complaints confirmed by negative biomarkers after evaluation. This individual could be reassured, provided health information on risk and resilience factors identified in the BHI assessment, and re‐evaluated at a later time. The third patient with a BHI of 58 falls into the indeterminant group with a diagnosis of MCI of unknown etiology after the formal evaluation. The risk and resilience factors identified in the BHI could assist in further work‐up, including getting a confirmatory biomarker such as an amyloid positron emission tomography (PET) scan.

Beyond primary care applications, the BHI shows promise as an outcome measure for clinical trials targeting cognitive health and dementia prevention. Its unique integration of vulnerability, resilience, and cognitive performance creates the potential for a dynamic metric capable of providing outcomes for intervention trials, where improvements in modifiable factors such as diet, physical activity, and social engagement would be directly reflected in BHI scores, while the NSCT component provides sensitivity to cognitive changes. While the Random Forest classification results demonstrating 0.76 AUC indicate good discriminative ability, further refinement and validation in generalizable populations would be necessary to optimize the BHI's predictive performance for clinical trial enrichment and outcome assessment. As cohorts may come from different sources (e.g., clinic, community, epidemiology sampling), assessing the characteristics of referral resources prior to administering the BHI should be considered to help interpret results.

4.5. Limitations

While a large amount of cross‐sectional data was collected, at the time of writing, only 166 participants completed a longitudinal follow‐up, constraining our understanding of how BHI scores change over time and their predictive validity for cognitive decline across different risk strata. The cohort is a community‐based convenience sample and thus may not be fully generalizable. Although 36.0% had less than a college education, the sample was highly educated (mean 15.8 years). However, the sample had a good distribution of ethnoracial groups and socioeconomic status. The presentation, progression, and risk factors for MCI can be complex and may differ between population‐based cohorts versus clinical referral samples. This can be due to different underlying causes (e.g., neurodegenerative disease, vascular disease, polypharmacy, psychiatric illness) with variable trajectories. The BHI should be examined in these different MCI populations in future studies. The evaluation procedures have evolved over time to reflect advances in knowledge, resulting in mild variations in the data collected; however, the majority of measures were the same, and consensus diagnostic procedures have remained unchanged. External validation in independent, diverse cohorts is essential to confirm the generalizability of the BHI thresholds and establish population‐specific norms that would enhance clinical utility across different demographic groups and healthcare settings.

4.6. Conclusion

The BHI represents a significant advancement in our ability to quantify and monitor brain health through the integration of vulnerability factors, resilience measures, and cognitive performance into a single, interpretable metric. By capturing the complex interplay between modifiable and non‐modifiable risk factors while remaining sensitive to underlying biological processes, the BHI offers a practical tool to identify at‐risk individuals and track intervention effectiveness.

AUTHOR CONTRIBUTIONS

Study conception, funding acquisition, and study design were performed by Dr James E. Galvin. Material preparation, data collection, and analysis were performed by Dr Michael J. Kleiman and Dr James E. Galvin. Data curation was conducted by Mr. Gregory Gibbs and Dr Mahesh S. Joshi. The first draft of the manuscript was written by Dr Michael J. Kleiman, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

Dr. James E. Galvin is the creator of the Resilience Index and the Number Symbol Coding Test. Dr. James E. Galvin and Dr Michael J. Kleiman are co‐creators of the Vulnerability Index and Brain Health Index. The copyrights are held by the University of Miami Miller School of Medicine. Dr. James E. Galvin is an Associate Editor for Alzheimer's and Dementia, but did not participate in any part of the peer‐review process. Dr. James E. Galvin is Chief Scientific Officer for Cognivue, Inc., and receives consulting fees. Mr. Gregory Gibbs and Dr Mahesh S. Joshi declare that they have no competing interests. The authors take full responsibility for the data and have the right to publish all data. Author disclosures are available in the Supporting Information

ETHICAL APPROVAL

The Healthy Brain Initiative, an ongoing longitudinal study, was first approved by the University of Miami's Institutional Review Board (Reference # 20200208) on 05/7/2020, with the most recent continuing review approved on 03/14/2025 and effective through 01/23/2026.

CONSENT STATEMENT

All participants provided written informed consent. This study was performed in accordance with the Helsinki Declaration of 1964.

Supporting information

Supporting information

Kleiman MJ, Gibbs G, Joshi MS, Galvin JE. The Brain Health Index: Integrating vulnerability, resilience, and cognitive function into a unified measure of cognitive health and risk of neurodegenerative disease. Alzheimer's Dement. 2025;21:1‐e70723. 10.1002/alz.70723

DATA AVAILABILITY STATEMENT

A de‐identified dataset for this project is available to all interested parties. Please contact J.E.G. at jeg200@miami.edu.