Plasma p‐tau217 and cognitive impairment: Evaluating biomarker equity across racial/ethnic groups in HABS‐HD

Abstract

INTRODUCTION

Plasma phosphorylated tau 217 (p‐tau217) is a leading blood‐based biomarker of Alzheimer's disease. Its performance in underrepresented racial/ethnic groups remains insufficiently characterized.

METHODS

We analyzed 2798 participants from the Health and Aging Brain Study–Health Disparities, including non‐Hispanic White, non‐Hispanic Black, and Hispanic adults. Multivariable logistic regression and receiver operating characteristic analyses assessed associations and discriminative accuracy between plasma p‐tau217 and clinical cognitive impairment with racial/ethnic‐specific thresholds.

RESULTS

Across all groups, p‐tau217 levels were higher in cognitively impaired than unimpaired participants. Elevated p‐tau217 was associated with greater odds of cognitive impairment in all racial and ethnic groups. Discriminative accuracy was modest but significant (area under the curve 0.65–0.72), with highest performance in non‐Hispanic Black and lowest in Hispanic participants.

DISCUSSION

Plasma p‐tau217 is robustly associated with cognitive impairment across diverse populations with varying thresholds, highlighting the need for population‐specific calibration to support equitable biomarker implementation.

Highlights

• Non‐Hispanic Black (NHB) and Hispanic participants had a higher prevalence of cognitive impairment than non‐Hispanic White (NHW) participants despite lower mean plasma phosphorylated tau at threonine 217 (p‐tau217) concentrations.

• Plasma p‐tau217 levels and upper reference limits were highest among NHW participants and lowest among Hispanic participants.

• Across all racial/ethnic groups, higher plasma p‐tau217 was strongly associated with greater odds of cognitive impairment.

• Plasma p‐tau217 demonstrated modest discrimination for cognitive impairment, with the highest accuracy among NHB participants and the lowest among Hispanic participants.

• Plasma p‐tau217 showed strong discrimination for amyloid beta positron emission tomography positivity across all groups, with the strongest performance among NHW and Hispanic participants.

1. INTRODUCTION

The high cost and participant burden of positron emission topography (PET) imaging and cerebrospinal fluid (CSF) collection to quantify Alzheimer's disease (AD) pathology underscores the need for scalable, minimally invasive biomarkers suitable for early detection, with plasma assays offering the greatest potential for broad implementation. ¹ , ² Plasma phosphorylated tau at threonine 217 (p‐tau217) has emerged as one of the most promising indicators of AD pathology, amyloid beta (Aβ) plaques, and neurofibrillary tau tangles. ³ , ⁴ , ⁵ Multiple studies demonstrate that plasma p‐tau217 closely predicts AD brain pathology, differentiates AD from other neurodegenerative disorders, and predicts subsequent clinical decline. ⁶ , ⁷ In addition, numerous studies have found that increases in plasma p‐tau217 are associated with cognitive impairment. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² Taken together, these findings position plasma p‐tau217 as a robust, minimally invasive biomarker with potential utility for early detection, prognosis, cognitive impairment detection, and clinical trial enrichment.

Despite this promise, most validation studies of plasma p‐tau217 have been conducted in highly selected, predominantly non‐Hispanic White (NHW) cohorts. ¹³ , ¹⁴ , ¹⁵ The degree to which these findings extend to diverse, community‐based populations remains unclear. This knowledge gap is critical, as non‐Hispanic Black (NHB) and Hispanic older adults experience disproportionately high dementia burden ¹⁶ but remain markedly underrepresented in biomarker research. Recent work highlights this concern: a large multiethnic cohort study demonstrated that underrepresented groups were less likely to meet plasma‐based eligibility criteria for amyloid PET imaging when algorithms incorporating p‐tau217 ratios were applied, even though PET‐confirmed amyloid eligibility did not differ once individuals passed plasma screening. ¹⁷ These findings suggest that plasma‐based algorithms may inadvertently exacerbate disparities in research participation and diagnostic access, underscoring the urgent need to evaluate the performance and generalizability of p‐tau217 cutoffs and discriminatory accuracy across racially/ethnically diverse populations.

The present study examined plasma p‐tau217 in the Health and Aging Brain Study–Health Disparities (HABS‐HD), a large community‐based cohort with three racial/ethnic groups: NHW, NHB, and Hispanic older adults. The objective was to evaluate whether plasma p‐tau217 identifies global cognitive impairment with similar accuracy across the three racial/ethnic groups from a community‐based sample. We hypothesized that the classification accuracy of plasma p‐tau217 would differ across racial/ethnic groups because of variations in comorbidity burden and background prevalence of impairment. We also hypothesized that optimal thresholds for p‐tau217 would differ across racial/ethnic groups. Last, we hypothesized that continuous p‐tau217 levels would be associated with cognitive impairment in each group, although the strength of these associations would vary by racial and ethnic subgroup because of underlying sociodemographic and health‐related factors. This study serves as an essential step toward identifying data‐driven p‐tau217 thresholds within specific racially/ethnically diverse populations by quantifying subgroup‐specific associations and classification performance.

2. METHODS

2.1. HABS‐HD

The HABS‐HD is an ongoing, large, community‐based longitudinal study including NHW, NHB, and Hispanic participants. A detailed description of HABS‐HD recruitment procedures, study protocols, and cohort characteristics has been published previously. ¹⁸ , ¹⁹ The study is approved by the University of North Texas Health Science Center Institutional Review Board, and all participants provide written informed consent in accordance with the Declaration of Helsinki.

2.2. Demographics

Age was recorded in years, and education was quantified as total years of schooling. Sex was coded as male or female. Racial ethnicity was self‐reported as NHW, NHB, or Hispanic.

2.3. Apolipoprotein E

Apolipoprotein E (APOE) genotyping was performed using TaqMan SNP assays (rs429358 and rs7412; ThermoFisher Scientific) using whole blood. APOE status was dichotomized into ε4 carriers (≥ 1 ε4 allele) versus non‐carriers (no ε4 alleles).

2.4. Cardiometabolic risk factors and renal function

Cardiometabolic variables included diabetes, hypertension, dyslipidemia, obesity, and tobacco use, each coded as binary indicators. Complete operational definitions are available in prior HABS‐HD methodological publications. ²⁰ Briefly, hypertension reflected documented diagnosis or medication use; diabetes was defined by diagnosis or treatment; dyslipidemia incorporated clinical history or lipid‐lowering therapy; obesity was defined as body mass index (BMI) ≥ 30 kg/m²; and tobacco use captured current or past dependence. We derived a harmonized estimated glomerular filtration rate (eGFR) variable by assigning each participant the appropriate estimate based on their racial ethnicity. History of kidney disease was also included as a covariate in all statistical models.

RESEARCH IN CONTEXT

1. Systematic Review: Plasma phosphorylated tau at threonine 217 (p‐tau217) is among the most validated blood‐based biomarkers of Alzheimer's disease (AD), with strong correlations to amyloid and tau pathology and clinical progression. However, most validation studies have been conducted in predominantly non‐Hispanic White (NHW) samples, limiting understanding of its performance in racially and ethnically diverse populations. Non‐Hispanic Black (NHB) and Hispanic adults are underrepresented in biomarker research despite experiencing higher dementia burden, and little is known about whether p‐tau217 levels or their associations with cognition differ across groups.

2. Interpretation: In this large, community‐based cohort, plasma p‐tau217 was consistently higher among cognitively impaired individuals across NHW, NHB, and Hispanic participants and was strongly associated with cognitive impairment in all groups. Despite lower mean p‐tau217 concentrations among NHB and Hispanic participants, these groups exhibited a higher prevalence of impairment. Discrimination of cognitive impairment using plasma p‐tau217 was modest overall, with the highest accuracy observed among NHB participants and the lowest among Hispanic participants. In contrast, plasma p‐tau217 showed strong discrimination for amyloid beta positron emission tomography positivity across all groups, with particularly high accuracy among NHW and Hispanic participants.

3. Future Directions: Refinement of reference ranges and thresholds using diverse community‐based cohorts, along with integration of p‐tau217 with complementary neurodegenerative and vascular biomarkers, will be critical to ensuring equitable diagnostic and prognostic utility of blood‐based AD biomarkers.

2.5. Plasma p‐tau217

Plasma p‐tau217 was quantified with Quanterix Simoa immunoassay kits and reported in pg/mL. Detailed HABS‐HD biospecimen collection and assay protocols have been published previously. ²¹ Plasma p‐tau217 was modeled as a continuous variable in its original measurement scale.

2.6. Clinical cognitive status

Cognitive status (cognitively unimpaired, mild cognitive impairment [MCI], or dementia) was assigned through a structured consensus process previously described in the HABS‐HD. ¹⁸ , ¹⁹ The review committee considered self‐ and informant‐reported changes in daily functioning together with performance across the neuropsychological battery. Neuroimaging data did not contribute to diagnostic decisions. For the outcome, we classified participants as cognitively impaired if they met diagnostic criteria for MCI or dementia, consistent with prior biomarker‐validation studies that distinguish unimpaired from clinically impaired groups. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹²

2.7. Neuroimaging

The HABS‐HD amyloid PET protocol follows procedures established by the Alzheimer's Disease Neuroimaging Initiative 3. Amyloid PET was acquired using Siemens Biograph Vision 450 PET/CT scanners after administration of florbetaben (FBB; 8.1 mCi), with four 5 minute frames collected 90 minutes post‐injection. ¹⁹ , ²² Standardized uptake value ratios (SUVR) were computed at the University of Southern California Stevens Neuroimaging and Informatics Institute using FreeSurfer‐derived regions of interest and the cerebellum as the reference region. Global amyloid positivity was defined as SUVR > 1.08, consistent with prior work in HABS‐HD. ²³

2.8. Statistical analysis

Plasma p‐tau217 concentrations (pg/mL) were analyzed as continuous predictors. Cognitive status was dichotomized as cognitively unimpaired (CU) or cognitively impaired (MCI or dementia) based on adjudicated clinical diagnoses. All analyses were stratified by racial and ethnic group (NHW, NHB, Hispanic). Unless otherwise specified, multivariable models included the same covariates: age, sex, years of education, APOE ε4 carrier status, eGFR, kidney disease, history of cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, and tobacco use.

2.9. Group differences in p‐tau217

Within each racial/ethnic group, we compared p‐tau217 distributions between CU and impaired participants using the Wilcoxon rank‐sum test. Jittered scatterplots with boxplots were generated for visualization. Statistical significance was defined as two‐sided P < 0.05.

2.10. Age associations with p‐tau217

To evaluate age‐related variation in p‐tau217, we fit racial/ethnic‐ and cognition‐stratified linear regression models with age modeled per 5‐year increase. Models were adjusted for the covariates listed above. We reported regression coefficients (β) and 95% confidence intervals (CI) and displayed fitted lines with confidence bands.

2.11. Racial/ethnic‐specific reference intervals

We derived racial/ethnic‐specific p‐tau217 reference intervals using the Clinical and Laboratory Standards Institute (CLSI) parametric approach (mean ± 1.96 standard deviation), constraining the lower bound to zero. ²⁴ Resulting upper limits were 1.17 pg/mL (NHW), 0.94 pg/mL (NHB), and 0.90 pg/mL (Hispanic). These were plotted against age.

2.12. Association between elevated p‐tau217 and cognitive impairment

Participants were classified as having “high” p‐tau217 if their value exceeded their race‐specific CLSI upper limit. We fit logistic regression models within each racial/ethnic group to estimate the association between high p‐tau217 and cognitive impairment. Models included the full covariate set.

2.13. Discriminative accuracy for cognitive impairment

To quantify how well plasma p‐tau217 distinguished CU from impaired individuals, we fit fully adjusted logistic regression models for the full cohort and separately for each racial/ethnic subgroup. Predicted probabilities were used to construct receiver operating characteristic (ROC) curves (pROC package), and we calculated area under the curve (AUC) with 95% CI and optimal thresholds using the Youden J statistic. Sensitivity and specificity at the optimal threshold were also estimated. Pairwise racial/ethnic differences in AUC were tested using the DeLong method.

2.14. Discriminative accuracy for Aβ PET positivity

Among participants with Aβ PET imaging, we evaluated the ability of plasma p‐tau217 to predict Aβ positivity (SUVR > 1.08). Fully adjusted logistic regression models were fit for the overall sample and within each racial/ethnic group, adding PET scanner as a covariate. ROC analyses paralleled those described above, and group differences in AUC were compared using DeLong tests.

All analyses were conducted in R (v4.2.3) with two‐sided P values < 0.05 considered statistically significant. Data are available at: https://loni.usc.edu/

3. RESULTS

The analytic cohort included 1086 NHW, 588 NHB, and 1124 Hispanic participants (Table 1). NHW participants had a mean age of 68.5 years, 57% were female, 29% were APOE ε4 carriers, mean p‐tau217 was 0.48 pg/mL, and 17% were cognitively impaired. NHB participants were younger (mean age 63.2), 65% female, 40% APOE ε4 carriers, had a mean p‐tau217 of 0.38 pg/mL, and 41% were cognitively impaired. Hispanic participants had a mean age of 63.3, 65% were female, 20% were APOE ε4 carriers, mean p‐tau217 was 0.36 pg/mL, and 28% were cognitively impaired. Comorbidities including kidney disease, cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, and tobacco use varied across groups and are reported in Table 1.

TABLE 1.

Baseline demographic, clinical, and biomarker characteristics stratified by racial/ethnic group.

Variable	NHW (N = 1086)	NHB (N = 588)	Hispanic (N = 1124)	Overall (N = 2798)
Age (years), mean (SD)	68.5 (8.7)	63.2 (8.0)	63.3 (8.0)	65.3 (8.7)
Female, n (%)	623 (57.4)	380 (64.6)	733 (65.2)	1736 (62.0)
Education (years), mean (SD)	15.6 (2.6)	14.8 (2.7)	10.0 (4.5)	13.2 (4.4)
APOE ε4, n (%)	319 (29.4)	234 (39.8)	222 (19.8)	775 (27.7)
APOE missing, n (%)	12 (1.1)	3 (0.5)	21 (1.9)	36 (1.3)
Kidney disease, n (%)	46 (4.2)	13 (2.2)	41 (3.6)	100 (3.6)
Cardiovascular disease, n (%)	108 (9.9)	42 (7.1)	60 (5.3)	210 (7.5)
Diabetes, n (%)	145 (13.4)	151 (25.7)	404 (35.9)	700 (25.0)
Hypertension, n (%)	621 (57.2)	468 (79.6)	709 (63.1)	1798 (64.3)
Dyslipidemia, n (%)	749 (69.0)	363 (61.7)	815 (72.5)	1927 (68.9)
Obesity, n (%)	428 (39.4)	339 (57.7)	579 (51.5)	1346 (48.1)
Missing, n (%)	10 (0.9)	7 (1.2)	5 (0.4)	22 (0.8)
Tobacco use, n (%)	39 (3.6)	80 (13.6)	82 (7.3)	201 (7.2)
eGFR, mean (SD)	75.7 (15.5)	84.3 (18.6)	86.2 (16.4)	81.6 (17.3)
Missing, n (%)	109 (10.0)	76 (12.9)	143 (12.7)	328 (11.7)
pTau217 (pg/mL), mean (SD)	0.48 (0.36)	0.38 (0.29)	0.36 (0.28)	0.41 (0.32)
pTau217 median [min, max]	0.35 [0.09, 2.41]	0.29 [0.07, 2.49]	0.27 [0.08, 2.31]	0.30 [0.07, 2.49]
Cognitively impaired, n (%)	189 (17.4)	240 (40.8)	310 (27.6)	739 (26.4)
Clinical cognitive stage, n (%)
Unimpaired	897 (82.6)	348 (59.2)	814 (72.4)	2059 (73.6)
Mild cognitive impairment	143 (13.2)	191 (32.5)	241 (21.4)	575 (20.6)
Dementia	46 (4.2)	49 (8.3)	69 (6.1)	164 (5.9)

Notes: Values are presented as mean (SD) for continuous variables and number (percentage) for categorical variables. Cognitive impairment includes individuals with mild cognitive impairment or dementia based on adjudicated consensus diagnoses. APOE ε4 reflects carrier status (heterozygous or homozygous). eGFR values are derived from the updated Chronic Kidney Disease Epidemiology Collaboration equation. Percentages reflect the proportion within each racial/ethnic subgroup (NHW, NHB, and Hispanic). Missing values are shown where applicable. Plasma p‐tau217 is reported in pg/mL.

Abbreviations: APOE, apolipoprotein E; eGFR, estimated glomerular filtration rate; NHB, non‐Hispanic Black; NHW, non‐Hispanic White; p‐tau, phosphorylated tau at threonine 217; SD, standard deviation.

Across all racial and ethnic groups, plasma p‐tau217 levels were higher in cognitively impaired than unimpaired participants (Figure 1A). Age was positively associated with p‐tau217 in both cognitive groups, with variation by racial/ethnic strata (Figure 1B). Among CU adults, p‐tau217 increased with age in NHW (β = 0.04 per 5 years; 95% CI 0.03–0.05) and Hispanic participants (β = 0.03; 95% CI 0.02–0.04), with smaller effects in NHB participants (β = 0.02; 95% CI 0.01–0.04); patterns were similar among cognitively impaired adults. CLSI‐derived reference intervals (Figure 1C) revealed racial/ethnic differences in upper limits of normal p‐tau217, highest in NHW (1.17 pg/mL), followed by NHB (0.94 pg/mL) and Hispanic participants (0.90 pg/mL). MCI cases were more common in NHB and Hispanic groups.

FIGURE 1.

Plasma p‐tau217 distribution, age associations, and CLSI reference intervals across racial/ethnic groups. A, Plasma p‐tau217 concentrations (pg/mL) in CU and cognitively impaired participants, stratified by race/ethnicity (NHW, NHB, Hispanic). Points represent individual values with jitter for visibility; boxplots show medians and interquartile ranges. P values are from Wilcoxon rank‐sum tests comparing CU versus impaired within each racial/ethnic group. B, Associations between age and plasma p‐tau217 among CU and impaired participants, stratified by race/ethnicity. Lines represent multivariable‐adjusted linear regression slopes with 95% CI. Corresponding β estimates (per 5‐year increase in age) and 95% CI are displayed alongside the panels. C, Plasma p‐tau217 values plotted across age with race‐specific CLSI reference intervals (mean ± 1.96 standard deviation; lower bound constrained to 0). Shaded regions denote expected reference ranges; black lines indicate upper reference limits. Points are colored by cognitive diagnosis (CU, MCI, dementia). CI, confidence interval; CLSI, Clinical and Laboratory Standards Institute; CU, cognitively unimpaired; MCI, mild cognitive impairment; NHB, non‐Hispanic Black; NHW, non‐Hispanic White; p‐tau, phosphorylated tau at threonine 217; RI, reference interval

High p‐tau217, defined by racial/ethnic‐specific CLSI limits, was strongly associated with cognitive impairment across all groups (Table S1 in supporting information). Adjusted odds ratios were 2.91 (95% CI 1.50–5.50) in NHW, 13.60 (95% CI 4.07–67.06) in NHB, and 6.37 (95% CI 3.32–12.75) in Hispanic participants.

After accounting for demographic, genetic, renal, and cardiometabolic factors, plasma p‐tau217 demonstrated modest but statistically reliable discrimination between CU and cognitively impaired individuals (Figure 2A). Among NHW participants, p‐tau217 yielded an AUC of 0.70 (95% CI 0.65–0.74; Figure 2B). NHB participants showed the highest discriminative accuracy (AUC 0.72, 95% CI 0.68–0.77; Figure 2C), whereas Hispanic participants had the lowest (AUC 0.65, 95% CI 0.61–0.69; Figure 2D). Optimal Youden thresholds varied by group: NHB 0.47 (specificity 0.76, sensitivity 0.57), NHW 0.20 (0.80, 0.54), and Hispanic 0.25 (0.64, 0.59). The full‐cohort threshold was 0.30. DeLong tests showed no AUC differences between NHW and NHB or NHW and Hispanic, but NHB outperformed Hispanic participants (P = 0.02).

FIGURE 2.

Discriminative accuracy of plasma p‐tau217 for cognitive impairment across racial/ethnic groups. ROC curves derived from fully adjusted logistic regression models predicting cognitive impairment (CI vs. CU) using plasma p‐tau217. Models were run in the full cohort and stratified by race/ethnicity (NHW, NHB, and Hispanic). Each model adjusted for age, sex, education, APOE ε4 status, eGFR, kidney disease, cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, and tobacco use. AUC values with 95% confidence intervals are displayed for each group. Optimal classification thresholds were identified using the Youden index, with corresponding sensitivity and specificity. DeLong tests compared AUCs across groups to assess differences in discriminative performance. APOE, apolipoprotein E; AUC, area under the curve; CI, cognitively impaired; CU, cognitively unimpaired; eGFR, estimated glomerular filtration rate; NHB, non‐Hispanic Black; NHW, non‐Hispanic White; p‐tau, phosphorylated tau at threonine 217; ROC, receiver operating characteristic

In the full model, plasma p‐tau217 showed strong discrimination for Aβ PET positivity (AUC = 0.87, 95% CI 0.85–0.89). The optimal threshold was 0.22, with sensitivity = 0.74 and specificity = 0.86 (Figure 3A). Among NHW participants, p‐tau217 showed excellent discrimination (AUC = 0.90, 95% CI 0.88–0.93) with an optimal threshold of 0.31 (sensitivity 0.80, specificity 0.88; Figure 3B). NHB participants demonstrated good performance (AUC = 0.84, 95% CI 0.78–0.89; threshold 0.19; sensitivity 0.67; specificity 0.87; Figure 3C). Hispanic participants showed similarly strong accuracy (AUC = 0.88, 95% CI 0.84–0.91; threshold 0.14; sensitivity 0.85; specificity 0.76; Figure 3D). AUCs differed only between NHW and NHB groups (P = 0.04).

FIGURE 3.

Race/ethnicity‐stratified ROC curves evaluating plasma p‐tau217 for predicting Aβ PET SUVR positivity. ROC curves for the association between plasma p‐tau217 concentration and amyloid PET SUVR positivity (Global_FBB_SUVR ≥ 1.08) in the overall sample (A) and stratified by race/ethnicity: NHW (B), NHB (C), and Hispanic (D). ROC curves were derived from fully adjusted logistic regression models including age, sex, education, APOE ε4 status, PET scanner type, eGFR, kidney disease, cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, and tobacco use. Shaded diagonal lines indicate reference performance. AUC values with 95% CI and Youden‐optimized thresholds, sensitivities, and specificities are shown within each panel. Aβ, amyloid beta; APOE, apolipoprotein E; CI, confidence interval; eGFR, estimated glomerular filtration rate; FBB, florbetaben; NHB, non‐Hispanic Black; NHW, non‐Hispanic White; PET, positron emission tomography; p‐tau, phosphorylated tau at threonine 217; ROC, receiver operating characteristic; SUVR, standardized uptake value ratio

4. DISCUSSION

Plasma p‐tau217 is among the most promising blood‐based biomarkers for AD, with strong evidence supporting its association with amyloid pathology and cognitive decline. ¹⁵ , ²⁵ , ²⁶ In this large, community‐based cohort intentionally designed to include NHW, NHB, and Hispanic older adults, plasma p‐tau217 was strongly associated with global cognitive impairment across all racial and ethnic groups. These associations were robust across multiple operationalizations of p‐tau217, including continuous concentrations and race‐specific abnormality thresholds, underscoring the biological relevance of p‐tau217 to cognitive impairment in diverse populations.

At the same time, differences in discriminatory accuracy were observed across racial and ethnic groups, with the highest performance observed among NHW participants. These findings highlight the limitations of assuming uniform biomarker performance across populations and emphasize the necessity of evaluating biomarkers in cohorts that reflect the heterogeneity of real‐world clinical and community settings. Importantly, such differences should not be interpreted as evidence of biologically distinct disease mechanisms across racial/ethnic categories.

Germaine to these findings, race and ethnicity are historical and sociopolitical constructs rather than biological classifications, reflecting lived social environments shaped by structural inequities rather than inherent biological variation. ²⁷ Accordingly, variation in biomarker distributions and classification performance is more plausibly attributable to contextual and structural factors including differences in comorbidity burden, access to health care, diagnostic pathways, cumulative exposure to stressors, and historical exclusion from biomarker discovery cohorts. ²⁸

Interpretation of discriminatory accuracy must therefore consider selection bias and outcome definition. Plasma p‐tau217 assays and thresholds have largely been developed and optimized in predominantly NHW, clinic‐based samples enriched for AD pathology. ¹⁵ , ²⁶ , ²⁹ , ³⁰ , ³¹ As a result, higher classification accuracy for amyloid PET positivity among NHW participants likely reflects closer alignment between assay development populations and the biological and clinical profiles of this group rather than superior biomarker validity. In contrast, for global cognitive impairment, a community‐derived outcome adjudicated independently of biomarker status, NHB participants exhibited comparable or slightly higher discrimination than NHW participants. This pattern likely reflects differences in impairment prevalence, etiologic heterogeneity, and cognitive aging trajectories within community‐based populations. ¹⁸ , ¹⁹ Together, these findings underscore how selection bias, cohort composition, and outcome definition shape apparent test performance and caution against extrapolation of thresholds derived from clinic‐based samples to more diverse community settings. ¹⁹ , ³² , ³³ , ³⁴

Enrollment and recruitment processes further shape observed group differences. Although HABS‐HD was intentionally designed to achieve racial/ethnic diversity, the strategies required to do so including targeted community outreach, clinic‐based referrals, bilingual engagement, and site‐specific eligibility criteria which introduce systematic differences across groups. ¹⁹ These mechanisms influence those who are reached, who volunteers, and who ultimately enrolls, thereby shaping distributions of comorbidities, prior diagnostic exposure, health‐care access, and baseline cognitive status. For example, the Hispanic population has a substantially lower education attainment than the NHW and NHB participants; however, NHW and NHB participants have similar education representativeness, which is rarely achieved in diverse community cohorts. As a result, racially/ethnically representative enrollment does not necessarily equate to population representativeness with respect to disease etiology or clinical trajectories especially given that this sample is only recruited from the Dallas/Fort Worth, Texas area. ³⁵ , ³⁶ , ³⁷

These enrollment mechanisms are particularly relevant for interpreting baseline differences in APOE ε4 carrier frequency. The higher prevalence of APOE ε4 among NHB participants and lower prevalence among Hispanic participants in this cohort likely reflect recruitment pathways and referral contexts rather than population‐representative genetic distributions. Differences in APOE ε4 frequency across racial and ethnic groups have been reported previously and vary substantially across study designs and recruitment strategies. ³⁸ , ³⁹ In this context, stratified analytic approaches, rather than direct cross‐group comparisons, provide a more appropriate framework for evaluating biomarker performance. Stratification allows characterization of p‐tau217 associations within each subgroup's distinct sociodemographic, genetic, clinical, and enrollment context while avoiding inappropriate contrasts shaped by differential recruitment mechanisms.

Medical comorbidities known to influence circulating biomarker concentrations also warrant careful consideration. Renal dysfunction and cardiovascular disease, in particular, have been shown to affect peripheral metabolism and clearance of plasma‐based AD biomarkers. ⁴⁰ , ⁴¹ In the present study, models explicitly adjusted for eGFR, self‐reported kidney disease, and cardiovascular disease. Incorporation of these factors that influence plasma biomarkers strengthens confidence that observed associations between p‐tau217 and cognitive impairment reflect AD‐related processes and are independent of comorbid confounding. ²¹ Nonetheless, residual confounding from unmeasured or incompletely captured health conditions including subclinical vascular disease, medication use, inflammatory burden, and lifetime health exposures remains possible. ⁴² , ⁴³

There have been prior reports of lower plasma p‐tau217 concentrations among NHB participants, ⁴⁴ , ⁴⁵ but these studies relied largely on convenience sampling and clinic‐based recruitment. Such designs may preferentially enroll individuals earlier in the disease course or with different comorbidity and health‐care access profiles, thereby contributing to observed biomarker differences independent of underlying AD biology. Similarly, patterns observed in the present study parallel previously reported findings in NHW participants whom were more likely to screen positive for plasma biomarkers, yet biomarker–outcome associations did not differ across ethnoracialized groups once screening thresholds were met. ³³ Overall, these findings suggest that differential entry into biomarker screening, rather than biological differences in biomarker–disease coupling, may drive apparent racial/ethnic differences in biomarker positivity.

Although population‐specific thresholds derived within racial or ethnic strata may help address heterogeneity in biomarker distributions, such approaches are not the only or necessarily the most practical solution for improving equity in blood‐based biomarkers. Alternative strategies include deriving cutpoints from large, population‐based cohorts that are broadly representative across multiple dimensions, including racial/ethnic identity, socioeconomic position, health status, and comorbidity burden. ⁴⁶ Thresholds derived from aforementioned “representative” samples will be more translational and applicable across settings and less dependent on the characteristics of any single subgroup. In addition, analytical approaches that avoid discrete cutpoints like continuous‐risk models, probabilistic frameworks, or multivariable algorithms integrating p‐tau217 with complementary biomarkers will further enhance equity and clinical utility in heterogeneous populations. ⁴⁷

Beyond methodological considerations, inclusion of racially and ethnically diverse participants in biomarker research is an ethical and public health imperative. ¹³ , ⁴⁸ , ⁴⁹ NHB and Hispanic communities experience a disproportionate burden of AD and related dementias alongside longstanding inequities in access to diagnostic evaluation, specialty care, and emerging disease‐modifying therapies. ³⁴ , ⁵⁰ Exclusion of these populations from biomarker development increases the risks and likelihood of perpetuating disparities by producing tools optimized for populations already advantaged within health‐care and research systems. Ensuring that plasma biomarkers such as p‐tau217 are evaluated in high‐risk, historically excluded populations is essential if these tools are to attenuate inequities in diagnosis and care.

Several limitations warrant consideration. The current study is cross‐sectional; hence, we do not have the ability to determine the temporality between plasma p‐tau217 and cognitive impairment. Longitudinal follow‐up will be essential to determine whether group differences in biomarker distribution or accuracy affect prediction of incident impairment and dementia. Although the sample size of the cohort was large and racially diverse, the subgroup sizes especially for the NHB participants and even more for those with available PET imaging limited power to detect interaction effects. Despite extensive covariate adjustment, unmeasured social, clinical, and structural factors likely continue to influence biomarker levels but more so the heterogeneity of underlying contributions to cognitive impairment. Finally, although the p‐tau217 assays used in this study are analytically validated, the calibration and threshold development was completed in predominantly NHW clinic‐based cohorts.

In summary, plasma p‐tau217 demonstrated strong and consistent associations with clinical cognitive impairment across NHW, NHB, and Hispanic older adults in a large, community‐based cohort. Its biological relevance was evident across groups, yet differences in discriminatory performance reflect selection processes, and comorbidity patterns, instead of biologically distinct disease mechanisms. These findings underscore the promise of p‐tau217 as a scalable biomarker for AD while emphasizing the necessity of equitable calibration, population‐inclusive validation, and analytic frameworks that account for social and structural determinants of biomarker performance. Ensuring reliable performance across the populations most affected by AD and related dementias will be critical for achieving fair and effective clinical translation.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the supporting information.

ETHICS APPROVAL

All participants were required to provide appropriate and informed consent to participate in the HABS‐HD research study.

Supporting information

Supporting Information

Supporting Information

APPENDIX 1.

1.2.

*HABS‐HD MPIs: Sid E. O'Bryant, Kristine Yaffe, Arthur Toga, Robert Rissman, and Leigh Johnson; and the HABS‐HD Investigators: Meredith Braskie, Kevin King, James R. Hall, Melissa Petersen, Raymond Palmer, Robert Barber, Yonggang Shi, Fan Zhang, Rajesh Nandy, Roderick McColl, David Mason, Bradley Christian, Nicole Phillips, Stephanie Large, Joe Lee, Badri Vardarajan, Monica Rivera Mindt, Amrita Cheema, Lisa Barnes, Mark Mapstone, Annie Cohen, Amy Kind, Ozioma Okonkwo, Raul Vintimilla, Zhengyang Zhou, Michael Donohue, Rema Raman, Matthew Borzage, Michelle Mielke, Beau Ances, Ganesh Babulal, Jorge Llibre‐Guerra, Carl Hill, and Rocky Vig.