Peripheral immunosenescence biomarkers and longitudinal cognitive decline: a large population-based study

Abstract

Peripheral immunosenescence has been increasingly implicated in dementia pathogenesis. However, the contributions of specific lymphocyte subsets to cognitive outcomes remain poorly defined, and evidence from large-scale longitudinal studies is scarce. We leveraged data from the US Health and Retirement Study (n = 7444; mean age: 67.7 ± 9.7 years). Our primary analysis calculated three ratio metrics to quantify T-cell immunosenescence: the CD4:CD8 ratio, CD4 + EMRA: Naïve and CD8 + EMRA: Naïve ratios. Their longitudinal associations with 6-year cognitive decline were evaluated by linear mixed model adjusted for multiple covariates. We found that a higher CD8 + EMRA: Naïve ratio was significantly associated with accelerated cognitive decline (-0.050 point/year per SD; 95% CI: −0.066 to −0.033, P < 0.001). A nonlinear association was observed for the CD4:CD8 ratio (P for nonlinearity = 0.033), with only an intermediate level (the third quartile) associated with a slower cognitive decline. No significant association was found for the CD4 + EMRA: Naïve ratio. This study reveals a compartmentalized pattern in the associations of immunosenescent metrics with cognitive aging. The findings indicate a specific detrimental pathway characterized by an elevated CD8 + EMRA:Naïve ratio, whereas balanced CD4:CD8 homeostasis may play a beneficial role.

Introduction

Dementia exerts an enormous global public health challenge. In 2019, over 57.4 million people worldwide suffered from dementia, with corresponding economic costs of USD $2.8 trillion^1,2. By 2050, dementia is projected to affect 152.8 million people and incur economic costs of USD $16.9 trillion^1,2. Effective treatments for dementia remain scarce; a primary factor for this therapeutic stagnation lies in dementia’s high heterogeneity, in which multiple pathological processes interact over an extended preclinical phase, leading to progressive cognitive decline^3,4.

Accumulating evidence supports the critical contribution of peripheral immunosenescence in the development of dementia^5–8. Peripheral immunosenescence refers to the age-related deterioration of peripheral immune system, characterized by a constellation of changes including thymic involution, a shrinking naïve T cell pool, an expansion of terminally differentiated cells, and systematic inflammaging^9,10. While the brain was historically considered a strictly immune-privileged site, current evidence has redefined this concept, recognizing that there is substantial neuroimmune communication¹¹. Under physiological conditions, peripheral immune cells, particularly T cells, routinely patrol the central nervous system (CNS) to maintain homeostatic surveillance^12,13. These cells predominantly populate the CNS border sites, including the perivascular and subarachnoid spaces, choroid plexus, and meninges, with the cerebrospinal fluid (CSF) acting as an essential conduit for immune surveillance^11,14. However, this regulated surveillance can be disrupted when the blood-brain barrier (BBB) becomes compromised. The BBB is not an impermeable shield but rather a dynamic structure vulnerable to dysfunction driven by aging-related stressors, particularly inflammaging^15,16. The compromised BBB permits the infiltration of peripheral immune cells into the CNS parenchyma, which then engage in complex crosstalk with CNS-resident microglia, triggering a self-perpetuating inflammatory cycle and accelerating neurodegeneration⁵.

Previous epidemiological studies have extensively reported associations between inflammatory immune biomarkers (such as C-reactive protein and Interleukin-6) and cognitive impairment⁸. However, the findings have been mixed and their translational potential are limited by the context-dependent pleiotropy of these inflammatory biomarkers; for instance, classically pro-inflammatory cytokines under specific conditions, can exhibit anti-inflammatory or even neuroprotective properties as well^17,18. While recent evidence has suggested that lower peripheral lymphocyte count, as a general measure of adaptive immunosenescence, was more consistently linked to an increased risk of dementia, the contributions of specific lymphocyte subsets to cognitive outcomes remain poorly defined^6–8. This critical gap has precluded the early identification of precise immunosenescent signatures and the development of targeted immunomodulatory strategies to prevent cognitive deterioration. Several studies have provided valuable insights by indicating the differences in peripheral T-cell profiles between cognitively impaired and intact individuals, while they were predominantly constrained by small clinical samples lacking representativeness and generalizability, or by cross-sectional designs that were susceptible to reverse causation^19–22. Moreover, a growing body of research has moved beyond absolute cell counts to employ ratio metrics that capture the composition of lymphocyte compartments, providing a more integrated proxy of immunosenescence^23–26. These metrics have been linked to systemic aging-related outcomes such as mortality and disability^23,24, while their specific association with longitudinal cognitive outcome remained unexplored. Therefore, we aimed to investigate the associations between peripheral metrics of immunosenescence and rate of change in cognitive function among general older adults, leveraging a nationally representative US aging cohort with comprehensive immunophenotypic and cognitive measurements.

Results

Baseline characteristics

We included 7444 participants (mean age: 67.7 ± 9.7 years, 59.6% female, 73.8% White ethnicity) in the current analysis, with median follow-up duration of 6 (IQR: 4–6) years (Table 1). A total of 5120 (68.8%) participants were CMV seropositive. The median values (IQR) for the CD4:CD8 ratio, CD4 + EMRA: Naïve ratio, and CD8 + EMRA: Naïve ratio were 3.11 (1.96–4.80), 0.04 (0.01–0.13) and 2.16 (0.85–5.34), respectively. The baseline characteristics across quartiles of ratio metrics are shown in Supplemental Tables 3 to 5.

Table 1.

Baseline characteristics of included participants

Characteristics	Statistics a
Age (years)	67.7 ± 9.7
Female (%)	4436 (59.6)
White ethnicity (%)	5493 (73.8)
High school or above (%)	6209 (83.4)
Living alone (%)	2490 (33.4)
Current smoking (%)	819 (11.0)
Drinking at least once per week (%)	3112 (41.8)
Physical active b (%)	5409 (72.7)
Body mass index (kg/m2)	29.2 ± 6.1
Hypertension (%)	4509 (60.6)
Diabetes (%)	1998 (26.8)
Stroke (%)	535 (7.2)
Heart diseases (%)	1768 (23.8)
Cancer (%)	1086 (14.6)
Chronic lung diseases (%)	750 (10.1)
Depressive symptoms (%)	969 (13.0)
Cytomegalovirus seropositive (%)	5120 (68.8)
CD4:CD8 ratio	3.11 (1.96, 4.80)
CD4 + EMRA: Naïve ratio	0.04 (0.01, 0.13)
CD8 + EMRA: Naïve ratio	2.16 (0.85, 5.34)
B cell Memory: Naïve ratio	0.16 (0.08, 0.30)
NK cell CD56 Low: CD56 High ratio	157.83 (87.87, 279.32)
Global cognitive score	15.8 ± 3.9
Memory score	10.2 ± 3.1
Executive function score	5.6 ± 1.7
Median follow-up duration (years)	6.0 (4.0, 6.0)

^a Values are presented as number (percentage), mean ± standard deviation, or median (interquartile ranges).

^b Physical active was defined as participation in moderate or vigorous physical activity at least once per week.

Association between immune metrics and cognitive decline

Table 2 summarizes the longitudinal associations between the different ratio metrics and rates of change in global cognitive function. For the CD4:CD8 ratio, compared with the lowest quartile, a significantly slower rate of cognitive decline was only observed in the third quartiles (0.051 point/year; 95% confidence interval [CI]: 0.005 to 0.097, P = 0.029), but not in other quartiles. The continuous CD4:CD8 ratio was not significantly associated with cognitive decline (P = 0.543). For the CD8 + EMRA: Naïve ratio, this metric was consistently associated with accelerated cognitive decline, both when analyzed as a continuous variable (−0.050 point/year per SD; 95% CI: −0.066 to −0.033, P < 0.001) and across quartiles (P for trend < 0.001). Specifically, compared to the lowest quartile, participants in the third and the highest quartiles presented significantly faster rates of cognitive decline, with estimates of -0.053 point/year (95% CI: −0.098 to −0.007, P = 0.023) and −0.144 point/year (95% CI: −0.190 to −0.098, P < 0.001), respectively. No significant associations between CD4 + EMRA: Naïve ratio and cognitive decline were observed.

Table 2.

Associations between peripheral immune metrics and rate of change in global cognitive score (point/year), using linear mixed models

Immune metrics	Global cognitive decline
	Beta (95% CI)	P
CD4:CD8 ratio
Q1	Reference	/
Q2	0.010 (−0.036, 0.056)	0.669
Q3	0.051 (0.005, 0.097)	0.029
Q4	−0.002 (−0.048, 0.044)	0.929
Test for linear trend	0.004 (−0.011, 0.018)	0.638
Per SD increment	0.005 (−0.011, 0.022)	0.543
CD4 + EMRA: Naïve ratio
Q1
Q2	−0.002 (−0.048, 0.044)	0.932
Q3	−0.011 (−0.056, 0.035)	0.649
Q4	−0.011 (−0.058, 0.035)	0.626
Test for linear trend	−0.004 (−0.019, 0.010)	0.566
Per SD increment	−0.006 (−0.022, 0.010)	0.478
CD8 + EMRA: Naïve ratio
Q1
Q2	−0.041 (−0.086, 0.004)	0.071
Q3	−0.053 (−0.098, −0.007)	0.023
Q4	−0.144 (−0.190, −0.098)	<0.001
Test for linear trend	−0.044 (−0.058, −0.029)	<0.001
Per SD increment	−0.050 (−0.066, −0.033)	<0.001

Models are adjusted for sociodemographic factors (baseline age, sex, ethnicity, cohabitation status, and educational attainment), lifestyle factors (smoking status, drinking status, and physical activity), clinical factors (body mass index, depressive symptoms, self-reported history of hypertension, diabetes, heart diseases, stroke, chronic lung disease, and cancer, cytomegalovirus serostatus).

Test for linear trend was performed by treating quartiles as a numeric variable.

The bold values indicate statistical significance at the level of P < 0.05.

Figure 1 provides a further visualization of the observed associations across quartiles of the immune metrics. As shown by the RCS curves in Fig. 2, a nonlinear dose-response pattern was identified only for the CD4:CD8 ratio (P for nonlinearity = 0.033).

Fig. 1. Mean differences in global cognitive decline rates (point/year) across quartiles of immune metrics.

For each immune metric, the first quartile was the reference group. Solid lines represent the estimates in Table 2 using linear mixed models, and shadows represent the 95% confident intervals.

Fig. 2. Dose-response curves of immune metrics and cognitive decline.

The ratio metrics were log-transformed and standardized to Z-scores for modelling. For each immune metric, a restricted cubic spline was fitted with 4 knots at the 5%, 35%, 65% and 95% percentiles in the linear mixed model. Solid lines represent the estimates of rate in global cognitive decline (point/year), and shadows represent the 95% confident intervals.

Sensitivity analyses

The results remained consistent when the ratio metrics were calculated without adding the small constant (Supplemental Table 6). Figure 3 visualizes findings of the subgroup analyses. No significant interactions were detected. The associations between CD8 + EMRA: Naïve ratio and accelerated cognitive decline showed directional consistency across each subgroup, while a trend of modifying effects of CMV serostatus (P for interaction = 0.067) was observed, suggesting that the effect size was more pronounced among CMV-seropositive individuals.

Fig. 3. Subgroup analyses of associations between immune metrics and global cognitive decline, using linear mixed models.

The ratio metrics were log-transformed and standardized to Z-scores for modelling. A three-way interaction term (immune metric × stratified factor × time) was tested to identify potential modifying effects.

Associations of the different immune metrics with decline in the subdomains of cognitive function are summarized in Supplemental Table 7. We found significant associations between executive function decline and the three immune metrics. Participants in the third quartiles of CD4:CD8 ratio showed a significantly slower decline rate compared with the lowest quartile (0.021 point/year; 95% CI: 0.003 to 0.039, P = 0.021), while higher ratios of CD4 + EMRA: Naïve and CD8 + EMRA: Naïve were associated with accelerated decline (P = 0.006 and P < 0.001, respectively). A significant association with episodic memory decline was only observed for CD8 + EMRA: Naïve ratio (−0.028 point/year per SD; 95% CI: −0.042 to −0.014, P < 0.001).

Exploratory analyses of B cells and NK cells

As shown in Supplemental Table 8, for B cell Memory: Naïve ratio, only participants in the second quartile showed a significantly slower rate of cognitive decline compared with those in the lowest quartile (0.050 point/year; 95% CI: 0.005 to 0.096, P = 0.031). We also observed a trend of higher NK cell CD56 Low: CD56 High ratio and accelerated cognitive decline (−0.016 point/year per SD; 95% CI: −0.032 to 0.0001, P = 0.052).

Discussion

Leveraging data of over 7000 US adults from a large, nationally representative aging cohort, our study reveals a nuanced and compartment-specific relationship between peripheral immunosenescence and longitudinal cognitive decline, providing a more refined understanding of the specific immune pathways contributing to cognitive aging. Our primary analysis found that a higher ratio of CD8 + EMRA: Naïve, but not the CD4 + EMRA: Naïve ratio, was associated with accelerated global cognitive decline, indicating a critical functional divergence in the roles of these two major T-cell lineages in cognitive aging. Moreover, we observed a non-linear relationship for the overall CD4:CD8 ratio, suggesting that a balanced T-cell ecosystem, rather than simply higher or lower ratios, may be beneficial for preserving cognitive function.

Previous population-based studies have demonstrated associations between general biomarkers of immune aging (such as lymphocyte counts) and dementia risk^6–8. Our study pinpoints that the signal for cognitive decline may be particularly strong within the CD8+ compartment. Interestingly, prior studies based on the HRS have primarily identified the CD4+ compartment as a stronger predictor of systemic adverse outcomes such as mortality, biological age, and comorbidity^23,26,27. This discrepancy suggests that distinct T cell differentiation pathways may exert compartmentalized effects on different facets of the aging process. Accumulating evidence has demonstrated that in aging brain, the infiltrated CD8 + T cells interact with microglia, initiating a self-perpetuating neuroinflammatory cascade that critically drives neurodegeneration⁵. Furthermore, previous research has revealed that in people with mild cognitive impairment or AD, an adaptive immune signature was an increase of peripheral CD8 + EMRA T cells, which correlated to worse cognitive function²⁰. Notably, the clonal expansion of cytotoxic and proinflammatory CD8 + EMRA T cells has also been detected in the cerebrospinal fluid of AD patients²⁰. Our subgroup analyses reinforced the robust association between a higher CD8 + EMRA: Naïve ratio and accelerated cognitive decline across demographic and genetic subgroups defined by sex, race, and APOE ε4 carrier status. The attenuated association observed in participants aged < 65 years aligned with prior evidence that the impact of adaptive immunity on dementia risk increased with age and was most pronounced after 65 years old⁶. Notably, we observed a trend of modifying effect of CMV serostatus (P for interaction = 0.067), suggesting that people with CMV seropositive were more vulnerable to the detrimental effects of CD8 + T-cell immunosenescence. This is highly relevant as over two-thirds participants of our cohort were seropositive, highlighting a critical public health concern. Collectively, these findings advocate for future work on targeted immunotherapies against senescent CD8 + T cells and populational-level policies mitigating the social drivers of immunosenescence to prevent cognitive decline.

In contrast to the predominantly cytotoxic nature of CD8 + T cells, CD4 + T cells exhibit considerable functional diversity and plasticity, serving central roles in immune signaling and regulation. Pro-inflammatory subsets, such as T helper type 1 (Th1) and Th17 cells, have been implicated in BBB disruption and microglial activation, thereby aggravating neuropathology^5,28. Conversely, anti-inflammatory subsets, including Th2 cells and regulatory T cells (Tregs), display neuroprotective properties and help mitigate neuroinflammation^5,28. However, the HRS did not directly assess these specific functional subsets. Our null finding regarding the CD4 + EMRA: Naive ratio and global cognitive decline implies that the mere accumulation of senescent CD4+ cells may not be a primary driver of cognitive deteriorations. Interestingly, our domain-specific sensitivity analyses identified that a higher ratio was associated with accelerated decline in executive function, but not in memory. Given that executive dysfunction is more reflective of vascular cognitive impairment, whereas memory impairment is more characteristic of Alzheimer’s disease pathology²⁹, our findings suggest that CD4+ immunosenescence may be preferentially linked to vascular pathological processes underlying cognitive decline. Therefore, future studies should integrate both T-cell differentiation stage and functional lineage data to fully elucidate these pathway-specific links.

A low CD4:CD8 ratio has been an established marker of profound immunosenescence, particularly in the context of chronic immune activation such as HIV infection³⁰. However, findings on the association between CD4:CD8 ratio and health outcomes in the general population have been inconsistent^24,26,31. For example, one HRS-based study found that a higher ratio was linked to greater physical disability²⁴, while another study reported null association with mortality²⁶. Despite the differences in outcomes and study designs, these studies commonly modeled the CD4:CD8 ratio as a continuous variable, potentially masking non-linear relationships. In contrast, our study categorized the ratio into quartiles and observed that only the third quartile, which represented an intermediate level, was associated with slower cognitive decline compared to the lowest quartile. This pattern was further clarified by the RCS analysis, which revealed a significant non-linear relationship. The nonlinearity suggests that both extremes of the ratio represent distinct forms of immune dysregulation detrimental to cognitive function. A low ratio reflects pronounced immunosenescence and inflammaging, which can fuel a self-perpetuating neuroinflammatory cascade⁵. Conversely, an excessively high ratio may indicate a collapse in CD8+ immune surveillance or an aberrant CD4+ expansion predisposing to autoimmunity³⁰, which may lead to cognitive deterioration through pathways such as subclinical viral reactivation and autoimmune-mediated neuronal damage^32,33. Thus, the intermediate level may represent a favorable homeostatic state that preserves cognitive function. Our findings caution against the simplistic view that a higher CD4:CD8 ratio is invariably beneficial. Therefore, therapeutic strategies should prioritize restoring immune homeostasis, rather than indiscriminately boosting either compartment.

Beyond our primary focus on T-cell senescence, the exploratory analyses on B cells and NK cells suggest that the immunologic landscape of cognitive aging may be even broader. For the B cell Memory: Naïve ratio, we found that individuals in the second quartile exhibited a slower rate of cognitive decline compared to those in the lowest quartile. This result preliminarily suggested that a moderate level of antigen experience and memory formation within the humoral immune system may be protective against cognitive decline³⁴. Additionally, we observed a trend of association between a higher NK cell CD56 Low: CD56 High ratio (indicating a shift towards less differentiated hypofunctional NK cells) and faster cognitive decline. This observation is consistent with the concept that decline in NK cell cytotoxicity compromises the clearance of senescent or stressed cells in the brain, thereby allowing for the accumulation of damaged cells and aggregated proteins^35,36. Nevertheless, these hypothesis-generating findings should be interpreted with caution and warrant further validation in dedicated studies interrogating the roles of innate and adaptive lymphocytes in cognitive aging.

This study has several notable strengths. First, we leveraged comprehensive immunosenescence measures within a large population-based cohort. In contrast to prior cross-sectional analyses using the HRS data¹⁹, our longitudinal framework with 6 years of follow-up establishes temporal precedence and enables dynamic assessment of cognitive trajectories, thereby mitigating the risk of reverse causation. Second, we employed composite ratio metrics that more accurately reflect the systemic homeostatic shifts underlying immunosenescence than individual cell compartments. Furthermore, complementary analytical approaches enhanced the depth of our findings: RCS models accounted for non-linear associations overlooked in previous studies, while subgroup analyses supported the robustness of our findings and explored potential modifying effects. However, there are also some limitations. First, causality cannot be established given the observational nature. Although we have carefully adjusted for a variety of covariates, residual confounding from unmeasured factors, such as medication use and immunological treatments, could not be ruled out. Second, immunophenotyping in large-scale cohort studies is logistically challenging, though robust protocols and stringent quality controls were applied in the HRS to minimize measurement errors. A previous study has indicated that delayed processing and cryopreservation of PBMCs, may lead to slight deviations in the percentages of several cell subsets compared to fresh samples³⁷. Besides, a single-timepoint measurement of immune profiles, as available in the current HRS data, may not fully capture the complex temporal evolution of immunosenescence across the lifespan. While our analytical panel of immune cell subset ratios provided an integrated perspective, it remains constrained by the scope of the HRS immunophenotyping data. As such, the potential contributions of finer-grained immunological features, such as the functional polarization of CD4 + T cells and the expression of markers associated with T-cell activation and exhaustion, remain to be investigated. Future longitudinal studies incorporating repeated, multi-dimensional immune profiling are essential to fully delineate the dynamic trajectory of immunosenescence and precisely quantify its impact on cognitive function. Third, the cognitive assessments in the HRS, while appropriate for screening in a large-scale population-based setting, may have limited sensitivity compared to specialized neuropsychological batteries for detecting subtle cognitive changes. Finally, as our study sample was derived from a US population predominantly of White ethnicity, the generalizability of our findings to other countries or ethnic groups requires further validation.

In conclusion, our study reveals that the associations between various immunosenescent metrics and cognitive decline follow a compartmentalized pattern. The robust association of CD8 + EMRA: Naïve ratio with accelerated decline underscores a specific detrimental pathway, while the non-linear association of the CD4:CD8 ratio suggests a beneficial role of balanced immune homeostasis. Our exploratory analyses further suggest that the immune-brain axis may extend beyond T-cell senescence to encompass broader innate and adaptive lymphocytes. Collectively, these findings underscore the multifaceted impacts of peripheral immunity on cognitive aging. Future research should validate these immune metrics in diverse populations and elucidate the underlying neuropathological mechanisms by incorporating neuroimaging and core AD biomarkers.

Methods

Study design and population

This is a longitudinal cohort study based on the Health and Retirement Study (HRS), an ongoing, nationally representative longitudinal survey of community-dwelling adults aged over 50 years in the United States. The HRS has been administered biennially since 1992 (Wave 1)³⁸. It was approved by the Institutional Review Board at the University of Michigan and the National Institute on Aging (HUM00061128), and written informed consents were obtained from all participants. This study was performed in accordance with the Declaration of Helsinki and reported following the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) guidelines³⁹.

The current study included participants from the HRS core 2016 survey (Wave 13) who had completed the Venous Blood Study (VBS). All respondents to the HRS 2016 survey were invited to provide consent for venous blood collection, except for proxy respondents and nursing home residents. The final VBS sample comprised 9934 participants (completion rate = 65%). Immunophenotyping was conducted for the entire VBS cohort. The cognitive assessments were administered biennially in the subsequent surveys of 2018 (Wave 14), 2020 (Wave 15) and 2022 (Wave 16). As outlined in the flow chart (Supplemental Fig. 1), the analytical sample for the present study included 7444 participants who completed the cognitive assessment at baseline HRS Wave 13, had no dementia at baseline (defined as neither a self-reported diagnosis of dementia, Alzheimer’s disease [AD], senility, or other serious memory impairment, nor a global cognitive score < 7 points^40,41), had valid data on peripheral immune biomarkers, and had at least one follow-up cognitive assessment between Waves 14 and 16.

Immunophenotyping

Details on collection, processing, storage of cells, quality control, and analysis of flow cytometry results, have been well-documented elsewhere^37,42,43. In brief, peripheral blood mononuclear cells (PBMCs) were isolated from whole blood and cryopreserved. Prior to immunophenotyping, frozen PBMCs were thawed in batches of ten samples. Immune cell subsets were defined according to the standardized protocol for the Human Immunology Project⁴⁴, with an addition of CD95 antibody to distinguish effector cytotoxic T-cell subsets⁴². All flow cytometry measurements were performed on an LSRII or a Fortessa X20 flow cytometer (BD Biosciences, San Diego, CA). A validated hybrid gating strategy incorporating automated hierarchical gating by OpenCyto and manual refinement by FlowAnnotator was employed⁴². Representative gating plots are publicly available in Supplementary Fig. 1 of the original study⁴². Extensive quality control procedures were implemented to monitor laboratory shifts and drifts and to ensure data reliability and reproducibility⁴³.

A total of 24 immune cell subsets were determined using flow cytometry, encompassing five major populations: T cells, B cells, natural killer (NK) cells, monocytes, and dendritic cells. T cells and B cells were reported as a percentage of total lymphocytes, while other populations were reported as a percentage of total PBMCs. Each subset was reported as a percentage of its parent population.

Our primary analysis focused on T cells, as the age-related alterations in T cell subsets are well-established hallmarks of immunosenescence⁹. Consistent with previous studies^23–25, we calculated the following ratio metrics to quantify T-cell immunosenescence: the CD4:CD8 ratio; and the ratio of terminally differentiated effector memory cells re-expressing CD45RA (EMRA, CD45RA + CCR7−CD28−) to naïve cells (CD45RA + CCR7 + CD28+) within the CD4+ and CD8+ compartments, denoted as the CD4 + EMRA: Naïve and CD8 + EMRA: Naïve ratios, respectively. A lower CD4:CD8 ratio and higher EMRA: Naïve ratios were interpreted as indicative of a more aged immune profile. In exploratory analysis, we expanded our scope by additionally including two less-investigated ratios^23,25: B cell Memory: Naïve and NK cell CD56 Low: CD56 High ratio, for which higher values also reflect greater immunosenescence. The specific cell subsets used for ratio measures, along with their defining surface markers and observed frequencies (expressed as percentages) in our analytical sample, are summarized in Supplemental Table 1. Detailed information on the antibodies used for immunophenotyping, including clone, fluorochrome, provider (with catalog number), and volume per sample, is provided in Supplemental Table 2.

Consistent with a previous study, prior to calculation, we added a small constant (0.001) to both the numerator and denominator of all ratios to avoid undefined values and permit log-transformation²⁵. All ratio values were then natural-log transformed and standardized to Z-scores for regression modeling to approximate normal distributions and ensure comparability.

Cognitive assessment

The cognitive function assessment in the HRS was adapted from the Telephone Interview for Cognitive Status⁴⁵. This instrument has demonstrated good consistency and validity, with previous validation studies reporting an accuracy of 81%–87% in identifying dementia^40,46,47. The HRS cognitive assessment has been extensively used in numerous studies on cognitive aging^41,48,49.

The HRS cognitive assessment covers two key domains: (1) episodic memory, assessed via immediate and delayed recall of a 10‑word list (score range: 0–20 points), and (2) executive function, using one serial-sevens subtraction test to evaluate working memory (0–5 points) and one counting backward test to evaluate processing speed and attention (0–2 points). The total score thus ranges from 0 to 27 points, with higher scores reflecting better global cognitive performance.

Covariates

Several factors that may confound the association between immunosenescence metrics and cognitive decline were considered^{3,23,24,41,48,49}. The covariates included sociodemographic (age, sex, ethnicity, cohabitation status, educational attainment), lifestyle (smoking status, drinking status, physical activity), and clinical factors (body mass index, depressive symptoms, self-reported history of chronic diseases including hypertension, diabetes, heart diseases, stroke, lung disease and cancer, and cytomegalovirus (CMV) serostatus). Specifically, depressive symptoms were defined as a score ≥ 4 on the eight-item Center for Epidemiologic Studies Depression Scale^41,48. CMV seroprevalence was measured using IgG antibodies to CMV in serum using the Roche e411 immunoassay analyzer, with results classified as non-reactive (<0.5 COI), borderline (0.5 to <1.0 COI) or reactive (≥1.0 COI). We defined CMV serostatus as seronegative (<1 COI) or seropositive (≥1.0 COI)^25,26.

Statistical analysis

For descriptive statistics, the mean ± standard deviation (SD) and median (interquartile ranges [IQR]) were applied for numerical characteristics, while the number (percentage) was applied for categorical characteristics. The longitudinal associations between immune metrics and change in global cognitive function were evaluated by the linear mixed model. This model has been extensively adopted for evaluating repeated measurements of continuous outcomes, such as cognitive score^41,48–51, and is able to utilize all available data during follow-up to estimate the rate of change in the outcome⁵². For each immune metric, we treated it as a categorical variable (according to quartiles) and a continuous variable, respectively. The model included the immune metric, time (years since baseline), the interaction term of immune metric × time, and all covariates listed above. Both the intercept and the slope of time were fitted as random effects to account for between-individual variations in baseline cognitive score and the rate of cognitive change during follow-up, respectively. Therefore, the regression coefficient of the interaction term reflected the difference in the rate of change in the cognitive score (point/year) across different levels of the immune metric, with a negative value representing an accelerated cognitive decline.

In addition, we used ordinary least squares regression with restricted cubic spline (RCS) to explore potential dose-response relationships between immune metrics and cognitive decline rate^41,53. Four spline knots at 5%, 35%, 65%, and 95% percentiles were adopted for all smooth curves. Likelihood ratio test was used for testing potential non-linearity.

We performed several sensitivity analyses to examine the robustness of our findings. First, we calculated the ratio metrics without adding the small constant, thus excluding a small proportion of participants (0.1%) who had a zero value in either the numerator or denominator. We then repeated the analyses. Second, we performed subgroup analyses to explore the potential modifying effects of sex, ethnicity, age (< 65 or ≥65 years), CMV serostatus (seronegative versus seropositive), and genetic susceptibility to cognitive decline. Genetic susceptibility was evaluated by APOE ε4 status (non-carrier versus carrier), and this subgroup analysis was performed among 6596 participants (88.6%) with available genetic data. Aligning with previous studies, we performed a three-way interaction test^50,51, in which we simultaneously included three two-way interaction terms (immune metric × time, stratified factor × time, and immune metric × stratified factor) and one three-way interaction term (immune metric × stratified factor × time) in the linear mixed models. Third, we further evaluated the associations of immune metrics with different cognitive domains (i.e., episodic memory, executive function).

Beyond our primary focus on T-cell subsets, we conducted exploratory analyses to examine other immune cell compartments. Specifically, we evaluated associations of cognitive decline with the B cell Memory: Naïve ratio, as well as the NK cell CD56 Low: CD56 High ratio.

All analyses were conducted by R 4.2.1 (R Foundation, Vienna, Austria). A two-sided P value < 0.05 was considered statistically significant.

Supplementary information

Supplementary information ^{(478.1KB, pdf)}

Author contributions

RH, CL and YW conceived and designed the study. RH and CL performed data analysis. RH drafted the manuscript. All authors contributed to critical revision of the manuscript for important intellectual content.

Data availability

The data used for the current study are publicly available (https://hrs.isr.umich.edu/data-products).

Code availability

The R codes for the current analyses are available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41514-026-00423-4.