Functional, molecular, and digital measurements of biological age

The reality of an aging population demands a deeper understanding of aging as a biological process, rather than as a chronological descriptor. Chronological age poorly captures interindividual heterogeneity in physiological and functional decline, disease susceptibility, and mortality risk. In contrast, biological age encompasses deterioration at the molecular, cellular, tissue, organ, functional, and organismal levels and provides insight into why two individuals with the same chronological age exhibit differences in physiological function, disease susceptibility, and mortality risk. While early models of biological age relied on functional markers or composite scores derived largely from longitudinal cohort studies, more recent models integrate molecular profiling with machine learning to ascertain biological aging trajectories. In parallel, new artificial intelligence tools have been applied to various imaging modalities and other forms of complex data to elucidate latent patterns and estimate biological age. In this state-of-the-art Review, we explore historical and modern approaches to estimating biological age and highlight key conceptual, technical, and translational challenges that remain unresolved. As geroscience-guided interventions are incorporated into clinical evaluations, robust and accurate interpretable measures of biological aging are crucial to ascertain treatment effects in clinical trials.

Early BA estimates relied on simple clinical measures drawn from longitudinal aging cohorts: investigators compared individuals with age- and sex-matched reference groups and interpreted deviations as accelerated or delayed aging, sometimes converting these differences into “biological years.” Some groups also combined these clinical variables into composite indices to summarize physiological aging (6, 15, 37–40), often overlapping with constructs of frailty and physiological reserve (41). These tools are easy to collect, clinically intuitive, and capture meaningful variation in physiological reserve and vulnerability to stressors, but they provide limited insight into the cellular and molecular drivers of aging.

Physical performance biomarkers. Physical function was one of the earliest and most pragmatic approaches to estimating BA, in part because declines in mobility and strength are widely recognized features of aging and closely track individuals’ lived experience of functional loss. Gait speed is among the best-studied metrics. In a pooled analysis across nine cohort studies (1986–2000) including more than 34,000 adults aged 65 and over, each 0.1 m/s higher gait speed was associated with a 10%–12% lower mortality risk (42). Its relevance extends into midlife: in the Dunedin birth cohort, slower gait at age 45 tracked with weaker physical performance, older facial appearance, and structural brain changes (e.g., smaller brain volume, greater cortical thinning), suggesting that gait captures early, system-level aging rather than late functional decline alone (43). Because of its reproducibility across cohorts and substantial prognostic value, some have referred to gait speed as a “sixth vital sign,” although its routine clinical adoption has been limited (44).

Grip strength, measured by dynamometer squeeze, is another robust marker of systemic aging (45). In the large, multinational PURE study of 139,691 adults, each 5 kg decrease in grip strength was associated with a 16% higher risk of all-cause mortality (46). A meta-analysis of more than three million participants across 42 cohorts showed similar patterns, linking weaker grip to higher mortality as well as CVD, cancer, and stroke risk (47). Other performance measures, including the timed-up-and-go test, the short physical performance battery, and the sit-to-stand test, also show that reduced strength, balance, and mobility are tightly coupled to adverse health outcomes (48–50). For example, unipedal stance time, a measure of balance and stability, declines more rapidly with aging than gait speed or grip strength (51). Together, these findings illustrate that mobility measures do not simply quantify functional status but may act as integrative readouts of longevity. Newer AI-based tools now utilized in research settings, including video-derived gait analysis and computer vision–based methods, may improve the precision and scalability of these established functional measures (52, 53), offering a promising future direction toward defining a more granular “movement age.”

Sensory function, including vision, hearing, and olfaction, offers a scalable organ-level readout that may reflect systemic aging beyond isolated end-organ loss. In large population-based cohorts, poorer multisensory function is associated with worse physical and cognitive performance and higher short-term mortality, even after adjustment for major confounders (54). Across longitudinal analyses, mortality risk increases with both the number and severity of sensory impairments, suggesting a cumulative multisensory aging signal (55, 56).

Organ function biomarkers. Endothelial function declines with age, with impaired vasodilation reflecting reduced nitric oxide bioavailability and accumulated oxidative stress (57). Flow-mediated dilation, an ultrasound-based measure of conduit artery function, is strongly linked to cardiovascular (CV) events (58) and provides a noninvasive window into vascular aging (59). Because endothelial health integrates metabolic, inflammatory, and hemodynamic inputs, it can be seen as a physiologic interface of systemic aging rather than reflecting vessel-specific pathology alone.

Arterial stiffness reflects age-related alterations in elastin-collagen balance, cross-linking, inflammation, and smooth muscle stiffening. It contributes to isolated systolic hypertension and elevated pulse pressure (PP) (60) and meets criteria for a prognostic CV biomarker (61). Aortic pulse-wave velocity is the most established measure: across longitudinal cohorts, higher pulse-wave velocity is associated with roughly twofold increases in CV and all-cause mortality after adjustment (62). This suggests that vascular stiffening is not merely a CV change but a proxy for global loss of tissue elasticity and repair capacity. Central PP, which better reflects central organ load than brachial BP, also correlates more strongly with subclinical organ damage; each 10 mmHg increase in central PP is associated with an approximately 14% higher CVD risk (63).

Cardiac physiological measures capture cumulative effects of aging on autonomic regulation, myocardial relaxation, and CV reserve (64). Heart rate variability (HRV), derived from beat-to-beat (RR interval) fluctuations, captures autonomic control of cardiac rhythm. Mendelian randomization analyses indicate that lower genetically predicted HRV causally increases CV risk, positioning HRV as a marker of declining autonomic resilience with age (65). The echocardiographic ratio (E/e′), an estimate of left ventricular filling pressure, independently predicts mortality; CV events including myocardial infarction, stroke, and heart failure; or hospitalizations for any cause, with a HR of approximately 1.05–1.06 per 1-unit increase in E/e′ (66). Reduced early-diastolic velocity (e′), indicative of impaired left ventricular relaxation, likewise predicts adverse outcomes and captures age-related diastolic dysfunction even in the absence of overt disease (67). Maximal oxygen uptake, providing an integrated measure of CV, pulmonary, and muscular performance, is one of the strongest indicators of whole-body physiological reserve. In the HUNT3 Fitness Study cohort (healthy adults), maximal oxygen uptake declined progressively approximately 6%–8% per decade, with both submaximal and peak exercise models capturing this trajectory across adulthood (68). Higher aerobic capacity is strongly associated with higher long-term survival; in the 46-year Copenhagen cohort, midlife cardiorespiratory fitness was closely tied to improved longevity (69).

Pulmonary function declines with CA as lung tissue loses elasticity, respiratory muscles weaken, and small airways narrow, reducing ventilatory capacity and gas-exchange efficiency (70). Peak expiratory flow and forced vital capacity show consistent age-related reductions, and lower forced vital capacity is associated with higher cardiometabolic risk even in individuals without overt lung disease (39). Forced expiratory volume in 1-second declines by approximately 20–50 mL per year in longitudinal studies (71, 72) and correlates with CV and mortality risk, indicating that respiratory aging may mirror systemic biological aging rather than isolated organ decline (39).

Blood-based biomarkers. Natriuretic peptides (BNP and NT-proBNP) are established markers of cardiac strain and long-term CV risk, even at levels far below heart failure diagnostic thresholds (73). Data from several aging-focused cohorts support the notion that NT-proBNP rises steadily with age and serves as a BA indicator. In a cohort of 1,079 older adults, NT-proBNP was the single most predictive age-related biomarker, outperforming routine labs for predicting 7-year mortality, enabling derivation of a “proBNPage” that tracked BA more closely than CA (74). Later work showed that higher proBNPage maps onto previously discussed broader aging phenotypes (e.g., slower gait, weaker grip, cognitive complaints) often more strongly than CA (75), and the Rugao Longevity and Ageing Study similarly linked higher BNP levels to these phenotypes in CVD-free older adults (76). A similar pattern emerges with cystatin C, which although viewed as a renal marker, behaves more like a systemic health parameter across aging cohorts. Even in healthy individuals without kidney impairment, cystatin C rises nonlinearly with age, indicating that it tracks intrinsic renal aging and broader systemic decline (77). Meta-analytic evidence shows that elevated cystatin C predicts long-term all-cause (HR ~1.7–1.9) and CV mortality (HR ~2.0), consistently outperforming creatinine-based indices (78). Higher levels are also linked to fewer disease-free years and a greater risk of transitioning to physical or cognitive disability (79).

Plasminogen activator inhibitor-1 (PAI-1) reflects a related but distinct aspect of aging biology. PAI-1 is a serine protease inhibitor central to fibrinolysis that increases in both chronological and stress-induced aging and is a core component of the senescence-associated secretory phenotype (80). In a Berne Amish kindred, carriers of a rare SERPINE1 (the gene encoding PAI-1) loss-of-function allele had longer telomeres, lower fasting insulin, less type 2 diabetes mellitus, and a longer lifespan (81). Larger cohort data align with this pattern: midlife women in the SWAN study with higher PAI-1 showed adverse metabolic and CV profiles (82).

These senescence-linked changes converge with a second central axis of biological aging: chronic, low-grade inflammation (inflammaging) (83). Inflammaging is marked by elevated circulating cytokines, chemokines, and acute-phase reactants related to both immune system dysfunction and accumulated molecular damage over time and is consistently associated with higher and earlier onset morbidity and mortality in older adults (84, 85). IL-6, in particular, correlates with endothelial dysfunction and is linked to atherogenic processes independent of serum lipid levels (86). Complementing IL-6, high-sensitivity C-reactive protein reflects downstream inflammatory signaling and, in contemporary statin-treated populations, has been a stronger predictor of future CV events and mortality than LDL-C, underscoring the concept of residual inflammatory risk (87). Within the TNF axis, circulating soluble TNF receptors (sTNFR1 and sTNFR2) are often used as stable biomarkers of chronic TNF pathway activation and have been associated with CV events and mortality in prospective studies (88–90). Multimarker approaches suggest that coordinated elevation across IL-6, high-sensitivity C-reactive protein, and TNF pathway readouts identifies a higher-risk inflammatory phenotype than any single analyte alone (91, 92). Metabolic markers capture complementary dimensions of aging biology: growth differentiation factor 15, a key senescence-associated secretory phenotype component, rises with metabolic dysregulation and declining renal function, mirroring multisystem physiologic burden (93). By contrast, the IGF-1 axis has been linked to vascular and endothelial function (94). HbA1c also increases with age even in metabolically healthy individuals and predicts mortality and functional decline across cohorts, highlighting its role as a broad marker of metabolic aging rather than merely insulin resistance in type 2 diabetes mellitus or glycemic control (95–98). One explanation for this observation is that HbA1c partly reflects erythrocyte turnover and glycation kinetics, both of which shift with aging independent of glucose exposure (97, 99, 100).

Although individual biomarkers provide valuable information about specific physiological systems, they cannot capture the multisystem decline that characterizes biological aging. One of the earliest composite BA indices was the Klemera–Doubal method, which shows a strong correlation with CA as well as a 7%–9% higher risk of mortality during follow-up, independent of traditional risk factors (6, 16). Approaches such as PhenoAge incorporated clinical biomarkers selected for their associations with mortality, improving predictive accuracy; each 1-year increase in PhenoAge was associated with a 9% higher mortality risk (HR 1.09) in the derivation cohort. For individuals in the highest quintile, mortality rates were approximately 3.3-fold higher than in the lowest quintile despite similar CA (101). Complementing these cross-sectional composites, longitudinal measures such as the Dunedin Pace of Aging metric quantified the slope of decline across 18 biomarkers spanning metabolic, CV, renal, hepatic, immune, and pulmonary systems, measured repeatedly over 20 years in the Dunedin cohort. Individuals aging “one biological year per chronological year” showed a mean Pace of Aging of 1.0, whereas persons aging faster exhibited values ≥ 1.2, consistent with approximately20% faster organ system deterioration and substantially poorer physical and cognitive function by midlife (102).

Closely related to composite measures of biological aging, resilience and intrinsic capacity have emerged as central functional endpoints in geroscience. Resilience reflects an individual’s ability to withstand and recover from physiological and psychosocial stressors, while intrinsic capacity, introduced by WHO, captures the integrated reserve of physical and mental function across five key domains: cognition, locomotion, vitality, psychological well-being, and sensory function (103, 104). Both constructs prioritize functional status over disease burden, reflecting a broader shift from disease-centered toward function-centered assessment focused on maintaining independence, health span, and recovery capacity. As multidomain and dynamic constructs that focus on adaptive reserve rather than accumulated pathology, resilience and intrinsic capacity are increasingly recognized as relevant outcome measures for aging research and interventional trials (105).

Finally, recent work has furthered the notion of composite aging metrics of inflammaging. One example is the inflammatory aging clock (iAge), derived using deep learning (DL) on circulating inflammatory profiles, which tracks multimorbidity, immunosenescence, frailty, and CV aging and highlights CXCL9 as a major contributor linked to vascular dysfunction and arterial stiffness (106).

Advances in molecular biology have enabled aging to be studied not merely as a clinical phenotype but as a quantifiable biological process driven by specific cellular mechanisms (21, 107, 108). Following the original description of the nine cellular and molecular hallmarks of aging, later expanded in subsequent revisions and summarized in Table 2, researchers have increasingly sought to measure these processes with reliability and precision (83, 109, 110). Advances in high-throughput molecular profiling and computational modeling have made it possible to interrogate thousands of molecular features simultaneously, revealing reproducible signatures of biological aging across tissues and populations. Among these, DNA methylation–based “epigenetic clocks” have emerged as some of the most robust predictors of morbidity and mortality, linking changes in CpG methylation to age-related decline (111–116). Parallel efforts in proteomics, metabolomics, glycomics, and inflammatory profiling have expanded this landscape, generating integrative frameworks that capture the multidimensional biology of aging.

Table 2

Features of the cellular and molecular hallmarks of aging, including the nine original hallmarks and selected extensions

Epigenetic clocks. Some of the earliest epigenetic clocks, including those proposed by Horvath (multitissue composite age estimation) and Hannum (single-tissue age estimation), were based mainly on supervised ML prediction models incorporating specific CpG methylation sites (21, 117–119). These first-generation epigenetic clocks employed a linear model that predicted CA from DNA methylation patterns. The Horvath and Hannum clocks were highly correlated with CA (Pearson’s r = 0.96 and 0.72, respectively), and most importantly age gap, the difference in estimated age and one’s CA, predicted all-cause mortality despite not including more relevant risk factors in the prediction model such as BMI, smoking, alcohol consumption, activity level, education, and other comorbidities (120, 121). These first-generation clocks were not without disadvantages. Technical variation (sample preparation and processing) and biological variation (interindividual variability inherent to certain CpG methylation sites) often led to inconsistency with repeat measurements (122, 123). One study reported that technical replication of the Horvath clock yielded high deviation, with a median of 3.6 years (124). Both of these clocks demonstrated weak associations with clinical biomarkers (e.g., BP, glucose metabolism) and diminished correspondence with tissue-specific ages, limitations prompting the development of newer clocks better equipped to predict morbidity and mortality (125).

Second-generation epigenetic clocks, such as DNAm PhenoAge and GrimAge, used linear models based around DNA methylation patterns to predict biological aging phenotypes rather than just CA (101, 125, 126). DNAm PhenoAge was trained using a weighted composite of 10 clinical characteristics to first estimate PhenoAge, as mentioned previously. This phenotype-driven weighted composite was then used to construct a prediction model based on CpG methylation sites and showed meaningful differences in disease and mortality among individuals of the same CA (101, 125). GrimAge was trained on CpG markers acting as surrogates for smoking history and plasma protein levels to predict time to death (125, 126). One study comparing all four clocks in a single cohort (n = 490) with a mean age of 62.2 years (SD = 8.3), found that PhenoAge (r = 0.85, SD = 9.6) and GrimAge (r = 0.81, SD = 7.6) had stronger age correlations than the Horvath and Hannum clocks (r = 0.74, SD = 11) (125). Another study showed that while first-generation clocks had no association with specific measures of functional health and cognitive performance, PhenoAge and GrimAge were associated with outcomes like weaker grip strength, worse lung function, and slower mental speed (107). With this more phenotype-driven approach, both second-generation clocks have outperformed previous clocks in predicting age-related health outcomes and lifespan, with GrimAge being the most predictive epigenetic clock for all-cause mortality (125, 127).

The novel DunedinPACE clock, termed a third-generation clock, utilizes the Pace of Aging metric mentioned earlier alongside DNA methylation patterns to derive a single, point-in-time blood test accounting for the biological patterns differentiating individuals with faster versus slower rates of aging (15, 128). Individuals with a higher DunedinPACE score exhibit more rapid aging-related decline in both physical and cognitive functions. While its predictions are consistent with other clocks, its focus on the rate of aging gives it distinct advantages in avoiding the survival bias inherent in clocks trained on older cohorts, and it can better distinguish ongoing aging-related decline from preexisting deficits developed in childhood (128). Although it gives less weight to mortality than second-generation clocks like GrimAge, it still provides improved incremental predictive value for morbidity and mortality (128, 129). This specific focus makes DunedinPACE a particularly valuable tool for investigating the effects of behavioral or pharmaceutical interventions on the aging process itself.

Epigenetic aging biomarkers (Figure 1B) continue to be refined, including the recently developed organ systems–based CpG methylation clock, Systems Age, which quantifies ages in 11 distinct organ systems using data from a single blood draw while providing a composite age score reflecting overall multisystem aging (130). Biomarkers are also being developed to leverage the multifactorial nature of aging by moving beyond CpG methylation. Another clock, OMICmAge, implements a regression model integrating multiple molecular profiling frameworks — proteomics, glycomics, and metabolomics — to capture aging hallmarks more directly influenced by biological and environmental phenomena. Preliminary results from its preprint publication suggest OMICmAge is precise, reliable over time, and accounts for changes in immune cell subsets that can influence age estimation (131).

Proteomic clocks. As the end products of epigenetic and transcriptional regulation, proteins directly influence phenotype by integrating system-wide biological processes, making them major contributors to the physiological and pathological outcomes of aging (132, 133). Advancements in assay technologies now enable the measurement of thousands of circulating proteins from small blood samples in a highly reproducible manner, increasing the potential of biomarker discovery with a proteomics focus (132, 134). Proteomic clocks leverage these capabilities, with some regression models trained to predict CA or mortality as a surrogate for BA, though they are usually limited by training with smaller sets of proteins known to be associated with specific biological aging processes (106, 132, 133, 135). Recent work aimed to develop a proteomic aging clock (PAC), based on various blood-borne proteins (n = 2,920), to predict all-cause mortality as an estimate of BA similar to second-generation epigenetic clocks (136, 137). The PAC provided age-adjusted predictions of all-cause mortality and the onset of various diseases with a relatively robust correlation with CA (r = 0.77), demonstrating stronger associations than other biomarkers such as DNAm PhenoAge, BioAge, and short leukocyte telomere length (136). However, PAC was not established to have predictive power for disease- or organ-specific BA estimation, though this has been explored in other work with a proteomics focus that may provide greater insight (137, 138).

Additional molecular clocks. A large population study leveraged IgG Fc N-glycosylation profiles to derive a glycan-based estimator of age, commonly referred to as GlycanAge (139). A minimal three-variable index dominated by glycan structures, including FA2B, FA2G2, and FA2BG2, explained approximately 58% of the variance in CA, with a correlation between chronological and predicted age of about r = 0.76 and a residual SD of approximately 9.7 years. Performance was lower but directionally consistent in independent cohorts (139). Because IgG Fc glycosylation is a functional regulator of antibody effector activity and inflammatory potential, age-associated shifts in these glycans may provide mechanistic links between immune remodeling and systemic inflammaging. Importantly, the residual variance in GlycanAge after accounting for CA correlated with physiological and biochemical traits, supporting its interpretation as a proxy for aspects of biological aging, although the model itself was trained primarily on CA (139).

Additional work has been done investigating how quantifying end products of metabolism could inform an association with CA and health outcomes. One such clock, MileAge, was developed by training multiple ML algorithms on plasma metabolites (n = 168) to predict mortality and health span (140). Performance differed based on the algorithm used (nonlinear vs. linear models), yielding moderate correlations with CA (r = 0.36–0.59 and a mean absolute error of 5.31–6.36 years). Though MileAge proved to be an adequate predictor of CA and mortality, other metabolomic clocks have demonstrated improved predictive power, further highlighting the potential of metabolomic aging measurement (140–142).

The exponential growth of digital health data combined with increases in computational power and advanced AI algorithms specifically utilizing DL has led to new opportunities to quantify BA (143). Whereas traditional and molecular biomarkers of disease rely on predefined variables, AI systems can learn latent representations directly from complex, multimodal data, ranging from imaging and electrocardiography to genomics and clinical records (144, 145). In many cases, these models are trained to predict CA, and the residual age gap is interpreted as a proxy for accelerated or decelerated biological aging. These methods extend beyond human-interpretable features to capture nonlinear and multidimensional signatures and may be particularly well-suited to studying the cumulative physiological decline across all organ systems with age. Foundation or large AI models trained on broad datasets are increasingly available publicly, enabling task adaptation and helping to democratize access to AI while supporting robust performance across external cohorts (146, 147). These AI systems span multiple data modalities, including imaging, electrophysiology, wearable sensors, and clinical text, each offering a distinct window into BA (Figure 2).

Figure 2

AI-derived BA from multimodal digital data. Models estimate age from imaging, electrophysiology, wearable signals, and clinical text to capture organ-level physiological change. MI, myocardial infarction.

ECG-age. An ECG captures dynamic voltage changes across the chest and remains one of the most informative, widely used tools for assessing cardiac structure, electrophysiology, and overall CV health. Although clinical interpretation relies on review of a compressed, standardized visual representation of the raw waveform, DL systems can analyze the full, high-dimensional signal and detect subtle, temporally distributed features that are inaccessible to human readers and can reveal previously unrecognized associations with cardiac disease (148). In aging research, deep neural networks trained to predict CA from ECG data have been used to estimate “ECG-age” (149). The gap between the ECG-age and CA has been shown to correlate with hard CV outcomes, including stroke, myocardial infarction, and death, in several independent studies (150, 151). In the largest ECG-age study to date, a deep residual network was trained on 1,558,415 patients from the Brazilian CODE cohort to predict age from 12-lead ECGs. Patients whose ECG-age exceeded CA by >8 years had a higher risk of all-cause mortality during a median follow-up of 3.4 years (HR 1.79, 95% CI 1.69–1.90), whereas those whose ECG-age was >8 years younger had lower mortality (HR 0.78, 95% CI 0.74–0.83), even after adjustment for age and sex. Similar HRs were replicated in two external cohorts (ELSA-Brasil, n = 14,236; SaMi-Trop, n = 1,631), supporting the interpretation of ECG-age as a digital composite measure of CV-BA rather than a simple age regressor (150). These findings suggest that AI-derived ECG-age reflects the cumulative physiological wear of the CV system and may serve as a readily accessible digital biomarker of biological aging, especially with new foundation models being developed (147).

Retinal-age. In a similar approach to deriving ECG-age, high-resolution color fundus photographic images of the retinal microvasculature have been analyzed with DL models to estimate a “retinal-age” (152). The difference in retinal-age and CA has been linked to important health outcomes including mortality, CVD, chronic kidney disease, and dementia (153–160). For example, a DL model trained on retinal fundus images from UK Biobank participants estimated a 2% higher hazard of all-cause mortality per 1-year increase in retinal age gap over follow-up (HR 1.02, 95% CI 1.00–1.03), after multivariable adjustment, again illustrating that a single image-derived age gap can function as a composite health and aging score (153). While these studies have been limited by potential selection bias from image-quality filtering and have shown worse model performance in external validation cohorts (154, 161), there is the potential to improve on this work by leveraging other retinal imaging techniques, such as optical coherence tomography (OCT), OCT angiography, and ultra-wide-field imaging, and to use foundation models as a starting point for future work (146, 161, 162).

Brain-age, facial aging, and other imaging-based models. Generalizing further, a growing body of research demonstrates that diverse imaging modalities harbor detectable signatures of biological aging. Brain MRI, for example, has been used to derive “brain-age,” which allows for age gap measures that strongly correlate with cognitive decline, frailty, and mortality risk across populations (163, 164). In a cohort of 669 older adults from the Lothian Birth Cohort 1936, a ML model trained on brain volumes produced a mean absolute error of 4.2 years for age prediction; individuals with an older-than-expected brain (positive brain-predicted age difference [brain-PAD]) had higher mortality, with each additional year of brain-PAD associated with an approximately 6% increase in all-cause mortality risk (HR 1.06, 95% CI 1.01–1.10) after adjustment for CA and traditional risk factors (165). Chest radiography (CXR) offers a more accessible clinical alternative, enabling estimation of CXR-age that predicts all-cause and CV mortality independent of established risk factors (166, 167). Facial imaging represents yet another distinct pathway, one that does not rely on medical testing and may allow for opportunistic capture from photographs. FaceAge, a convolutional neural network (DL architecture optimized for image and signal analysis), was trained on 58,851 healthy individuals and validated in 4,906 patients with cancer, quantifying a facial age gap that is strongly prognostic. Across pan-cancer, thoracic radiotherapy, and palliative cohorts, each 10-year increase in FaceAge was associated with worse overall survival (e.g., pan-cancer, HR 1.151, P = 0.013, n = 4,906), and cancer patients on average appeared 4.79 years older than matched noncancer controls (P < 0.0001). Incorporating FaceAge improved clinicians’ short-term survival predictions in palliative radiotherapy (AUC 0.80 vs. 0.74; P < 0.0001), highlighting that even a facial photograph–derived age gap can operate as a composite frailty and BA signal (168).

Large language models of clinical text. Modern large language models (LLMs) trained on clinical text and health records are now demonstrating the ability to estimate BA-related measures without the need for imaging or high-dimensional molecular profiling (169). For example, Li et al. (169) developed a framework in which an LLM was prompted with standard health examination reports to predict both overall and organ-specific ages. They found that the LLM-predicted overall age achieved a concordance index of about 0.757 (95% CI 0.752–0.761) for all-cause mortality, outperforming traditional aging proxies such as telomere length, frailty index, epigenetic age scores, and models using ML. The LLM-derived age gap predicted all-cause death, and organ-specific age gaps also predicted corresponding disease categories better than classic ML models (169). This work demonstrates that LLMs are scalable for population-level phenotypic aging estimation because they rely on routinely collected clinical text from existing electronic health data.

Importantly, LLM-based age estimates should not be interpreted as mechanistic models of aging. Instead, they function as high-dimensional summary measures of clinical phenotype, integrating information across diagnoses, laboratory values, and documented health trajectories. It is not surprising that such models show strong associations and discrimination in mortality and disease outcomes, often exceeding those observed for single-modality aging metrics. Unlike proteomic or molecular clocks, however, LLM-derived age gaps do not directly identify upstream biological drivers of aging or specific pathways amenable to intervention; these models are best viewed as downstream tools for identifying individuals with “accelerated aging” phenotypes, complementary to mechanistic aging measures rather than substitutes for them.

Despite substantial progress in the development of BA measures across clinical, molecular, and digital domains, several important gaps limit broader adoption and utility. First, the field lacks consensus on a unifying framework for what constitutes an adequate BA metric. At present, no broadly accepted standard exists, and creating one is beyond the scope of this review (28). Additionally, reliability, longitudinal stability, responsiveness to interventions, and feasibility across diverse healthcare settings are variably measured and rarely compared head-to-head. Furthermore, most BA metrics are supported primarily by cross-sectional and prognostic data, but there remains limited evidence that within-person changes over time reliably reflect aging rate, respond to interventions, or qualify as clinically meaningful surrogate endpoints. As a result, there is no consensus on the magnitude or clinical relevance of changes in BA measures in response to health behaviors or clinical interventions aimed at improving health (27). Standardized benchmarking datasets, shared evaluation criteria, and harmonized reporting standards will be essential to move from isolated clocks toward a more coherent approach to aging assessment. Additionally, it is important to distinguish biomarkers that predict outcomes, reflect accumulated disease burden or physiologic decline, and directly illuminate mechanisms of aging biology, as current evidence is strongest for the first two and more limited for the third.

Second, not all BA metrics are anchored in biology, with molecular clocks as examples of well-cited tools for BA assessment that often aggregate thousands of features into composite scores with limited interpretability (20). Digital and AI-derived clocks rely on high-dimensional representations that are even less transparent, complicating efforts to understand which pathways are being captured (170). Systematic efforts to link clock components to specific hallmarks of aging, tissue, or organ-specific processes and causal pathways are needed to move from association toward mechanism-informed measurement.

Third, a related unresolved challenge is distinguishing aging-related biological change from disease-induced organ dysfunction when interpreting BA estimates. Because aging and disease mechanisms are intertwined, elevated BA may reflect accelerated aging, subclinical pathology, established disease, reduced resilience, or a combination thereof, underscoring the need for contextual interpretation across these overlapping phenomena (3, 9, 27, 28).

Fourth, there is a paucity of comparative studies that directly evaluate traditional, molecular, and digital clocks within the same individuals (171). As a result, the degree of redundancy versus complementarity among different BA measures is poorly understood. It is hypothesized that epigenetic, proteomic, imaging-based, and digital clocks each capture distinct dimensions of aging, reflecting different organs, systems, or timescales; however, further research is needed to verify this and outline comprehensive approaches to BA estimation that are not redundant. This will likely require multimodal integration studies that combine these signals and formally assess their unique and shared contributions to risk prediction, functional decline, and resilience.

Finally, there is limited evidence on how BA metrics should be deployed in clinical settings, including (a) their role in clinical assessment and (b) how evidence of accelerated biological aging should inform clinical management decisions. Few studies have rigorously evaluated how BA responds to interventions known to modify aging-related pathways, how much change is biologically meaningful, or how BA measures should be incorporated into trial design. While the geroscience hypothesis posits that changes in BA measures should impact that aging process directly and may slow the progression of chronic disease development, this is a hypothesis that requires rigorous evaluation and testing in prospective clinical trials. These trials must give attention to cost, workflow integration, and health equity to demonstrate real-world value of BA estimates. Closing these gaps will require coordinated efforts across disciplines, shared infrastructure, and deliberate linkage between measurement science and intervention studies aimed at extending health span. Additional pitfalls in BA interpretation and suggested approaches to mitigate fallacies are provided in Table 3.

Table 3

Common interpretive pitfalls for BA metrics and suggested approaches

Best practices for translating BA metrics into clinical practice and identifying interventions that improve the aging process will require detailed, objective, prospective human studies that use a comprehensive approach to BA estimation as a modifiable clinical endpoint alongside metrics of physical function and organ system–specific health that decline with aging. If an intervention truly impacts the aging process, consistent effects across multiple aging biomarkers, rather than isolated changes, are expected. As an initial step, harmonized, small-to-moderate scale signal-finding trials using a unified testing platform across multiple sites should test diverse approaches for enhancing health span and well-being across genetically and culturally diverse populations. This conceptual framework illustrates how repeated BA assessments, paired with targeted interventions, can be used longitudinally to track aging trajectories and treatment responsiveness, framing BA as a dynamic, modifiable construct rather than a static measurement (Figure 3). This approach to aging research can identify which interventions are most likely to result in reproducible, clinically meaningful changes that target aging biology and will inform design of subsequent phase III clinical trials. Throughout this process, attention to financial cost, workflow integration, regulatory acceptance, and equity must be considered to ensure that potential interventions are not only scientifically valid but feasible for clinical integration and generalizable to real-world healthcare systems.

Figure 3

Conceptual framework for BA-guided intervention studies. Repeated BA assessments are used prospectively alongside targeted interventions to evaluate aging trajectories and intervention responsiveness.

BA metrics offer a promising path to quantify individual variation in aging, anticipate health outcomes beyond CA, and accelerate the evaluation of interventions that target underlying aging processes. At present, the field is characterized by methodological heterogeneity, limited comparative and longitudinal data, and sparse evidence from prospective intervention studies. Moving forward, progress will depend on converging toward consensus definitions, shared evaluation standards, and coordinated, multisite clinical observational and interventional trials that embed harmonized BA panels as endpoints. By aligning measurement science with mechanistic insight, rigorous trial design, and attention to implementation, BA can evolve from a research construct into a practical tool to guide prevention, personalize care, and ultimately extend health span.

BSC, BB, MOA, MSH, WBM, FJP, MAM, DML, MKG, RJC, JTW, and DEV contributed meaningfully to conceptualizing the paper, creating an outline, writing the manuscript, drafting/editing tables and figures, and editing the final work. Order of co–first authors reflects their involvement in the project.

BSC is an advisor with equity in Zoe Biosciences. FJP has received consulting fees and speaking honoraria from Gilead Sciences, ViiV Pharmaceuticals, Merck, and EMD Serono. MKG has served as a consultant for Genentech-Roche, Regeneron, and Kriya Therapeutics. RJC holds US patents 11,803,241 and 12,277,272 and US patent applications 16/588,202 and 14/534,163. DEV has served as a consultant for The Medicines Company and is a cofounder of Zoe Biosciences.

This work is the result of NIH funding, in whole or in part, and is subject to the NIH Public Access Policy. Through acceptance of this federal funding, the NIH has been given a right to make the work publicly available in PubMed Central.

• NIH grant R01HD114776 (to RJC).
• Research Accelerator Program of the Shirley Ryan AbilityLab (to RJC).
• National Heart, Lung, and Blood Institute grant R35HL171553 (to DEV).
• NIH grants R01AG092241, R01AI189353, R01AI165079, and R01AA031197 (to MAM).

Copyright: © 2026, Cheema et al. This is an open access article published under the terms of the Creative Commons Attribution 4.0 International License.

Reference information: J Clin Invest. 2026;136(12):e205777. https://doi.org/10.1172/JCI205777.

1. Harper S. Economic and social implications of aging societies. Science. 2014;346(6209):587–591.
   View this article via: CrossRef PubMed Google Scholar
2. Khan SS, et al. Molecular and physiological manifestations and measurement of aging in humans. Aging Cell. 2017;16(4):624–633.
   View this article via: CrossRef PubMed Google Scholar
3. Niccoli T, Partridge L. Ageing as a risk factor for disease. Curr Biol. 2012;22(17):741–752.
   View this article via: CrossRef PubMed Google Scholar
4. Bloom DE, et al. Macroeconomic implications of population ageing and selected policy responses. Lancet. 2015;385(9968):649–657.
   View this article via: CrossRef PubMed Google Scholar
5. Daviglus ML, et al. Relation of body mass index in young adulthood and middle age to Medicare expenditures in older age. JAMA. 2004;292(22):2743–2749.
   View this article via: CrossRef PubMed Google Scholar
6. Levine ME. Modeling the rate of senescence: can estimated biological age predict mortality more accurately than chronological age? J Gerontol A Biol Sci Med Sci. 2013;68(6):667–674.
   View this article via: CrossRef PubMed Google Scholar
7. Yashin AI, et al. Mortality and aging in a heterogeneous population: a stochastic process model with observed and unobserved variables. Theor Popul Biol. 1985;27(2):154–175.
   View this article via: CrossRef PubMed Google Scholar
8. Hayflick L. Biological aging is no longer an unsolved problem. Ann N Y Acad Sci. 2007;1100:1–13.
   View this article via: CrossRef PubMed Google Scholar
9. Kennedy BK, et al. Geroscience: linking aging to chronic disease. Cell. 2014;159(4):709–713.
   View this article via: CrossRef PubMed Google Scholar
10. Kirkwood TBL. Understanding the odd science of aging. Cell. 2005;120(4):437–447.
    View this article via: CrossRef PubMed Google Scholar
11. Strehler BL, Mildvan AS. General theory of mortality and aging. Science. 1960;132(3418):14–21.
    View this article via: CrossRef PubMed Google Scholar
12. Finch C, Kirkwood TBL, eds. Chance, Development, and Aging. Oxford University Press; 2000:278.
13. Fuellen G, et al. Health and aging: unifying concepts, scores, biomarkers and pathways. Aging Dis. 2019;10(4):883–900.
    View this article via: CrossRef PubMed Google Scholar
14. Jia L, et al. Common methods of biological age estimation. Clin Interv Aging. 2017;12:759–772.
    View this article via: CrossRef PubMed Google Scholar
15. Belsky DW, et al. Quantification of biological aging in young adults. Proc Natl Acad Sci U S A. 2015;112(30):4104–4110.
    View this article via: CrossRef PubMed Google Scholar
16. Klemera P, Doubal S. A new approach to the concept and computation of biological age. Mech Ageing Dev. 2006;127(3):240–248.
    View this article via: CrossRef PubMed Google Scholar
17. Comfort A. Test-battery to measure ageing-rate in man. Lancet. 1969;2(7635):1411–1414.
    View this article via: CrossRef PubMed Google Scholar
18. Mooradian AD. Biomarkers of aging: do we know what to look for? J Gerontol. 1990;45(6):B183–B186.
    View this article via: CrossRef PubMed Google Scholar
19. Cho IH, et al. An empirical comparative study on biological age estimation algorithms with an application of Work Ability Index (WAI). Mech Ageing Dev. 2010;131(2):69–78.
    View this article via: CrossRef PubMed Google Scholar
20. Rutledge J, et al. Measuring biological age using omics data. Nat Rev Genet. 2022;23(12):715–727.
    View this article via: CrossRef PubMed Google Scholar
21. Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 2013;14(10):R115.
    View this article via: CrossRef PubMed Google Scholar
22. Salih A, et al. Conceptual overview of biological age estimation. Aging Dis. 2023;14(3):583–588.
    View this article via: CrossRef PubMed Google Scholar
23. Meng D, et al. Application of AI in biological age prediction. Curr Opin Struct Biol. 2024;85:102777.
    View this article via: CrossRef PubMed Google Scholar
24. Zhavoronkov A, et al. Artificial intelligence in longevity medicine. Nat Aging. 2021;1(1):5–7.
    View this article via: CrossRef PubMed Google Scholar
25. Zhavoronkov A, Mamoshina P. Deep aging clocks: the emergence of AI-based biomarkers of aging and longevity. Trends Pharmacol Sci. 2019;40(8):546–549.
    View this article via: CrossRef PubMed Google Scholar
26. Moore JH, et al. Artificial intelligence based approaches to identify molecular determinants of exceptional health and life span-an interdisciplinary workshop at the national institute on aging. Front Artif Intell. 2019;2:12.
    View this article via: CrossRef PubMed Google Scholar
27. Biomarkers of Aging Consortium, et al. Challenges and recommendations for the translation of biomarkers of aging. Nat Aging. 2024;4(10):1372–1383.
    View this article via: CrossRef PubMed Google Scholar
28. Moqri M, et al. Biomarkers of aging for the identification and evaluation of longevity interventions. Cell. 2023;186(18):3758–3775.
    View this article via: CrossRef PubMed Google Scholar
29. Justice JN, et al. A framework for selection of blood-based biomarkers for geroscience-guided clinical trials: report from the TAME Biomarkers Workgroup. Geroscience. 2018;40(5-6):419–436.
    View this article via: CrossRef PubMed Google Scholar
30. Fries JF. The compression of morbidity. 1983. Milbank Q. 2005;83(4):801–823.
    View this article via: CrossRef PubMed Google Scholar
31. Wilkins JT, et al. Lifetime risk and years lived free of total cardiovascular disease. JAMA. 2012;308(17):1795–1801.
    View this article via: CrossRef PubMed Google Scholar
32. Ahmad FS, et al. Hypertension, obesity, diabetes, and heart failure-free survival: the cardiovascular disease lifetime risk pooling project. JACC Heart Fail. 2016;4(12):911–919.
    View this article via: CrossRef PubMed Google Scholar
33. Cotton RJ, et al. A causal framework for precision rehabilitation [preprint]. https://doi.org/10.48550/arXiv.2411.03919 Posted on arXiv November 6, 2024.
34. Nielsen L, et al. New directions in geroscience: integrating social and behavioral drivers of biological aging. Psychosom Med. 2024;86(5):360–365.
    View this article via: CrossRef PubMed Google Scholar
35. Lakatta EG. So! What’s aging? Is cardiovascular aging a disease? J Mol Cell Cardiol. 2015;83:1–13.
    View this article via: CrossRef PubMed Google Scholar
36. Nadon NL, et al. NIA interventions testing program: investigating putative aging intervention agents in a genetically heterogeneous mouse model. EBioMedicine. 2017;21:3–4.
    View this article via: CrossRef PubMed Google Scholar
37. Andersson C, et al. Framingham heart study: JACC Focus Seminar, 1/8. J Am Coll Cardiol. 2021;77(21):2680–2692.
    View this article via: CrossRef PubMed Google Scholar
38. Ferrucci L. The Baltimore Longitudinal Study of Aging (BLSA): a 50-year-long journey and plans for the future. J Gerontol A Biol Sci Med Sci. 2008;63(12):1416–1419.
    View this article via: CrossRef PubMed Google Scholar
39. Wang C, et al. Biomarkers of aging and lung function in the normative aging study. Aging (Albany NY). 2020;12(12):11942–11966.
    View this article via: CrossRef PubMed Google Scholar
40. Mount S, et al. Measuring successful aging: an exploratory factor analysis of the InCHIANTI Study into different health domains. Aging (Albany NY). 2019;11(10):3023–3040.
    View this article via: CrossRef PubMed Google Scholar
41. Mitnitski A, et al. Age-related frailty and its association with biological markers of ageing. BMC Med. 2015;13:161.
    View this article via: CrossRef PubMed Google Scholar
42. Studenski S, et al. Gait speed and survival in older adults. JAMA. 2011;305(1):50–58.
    View this article via: CrossRef PubMed Google Scholar
43. Rasmussen LJH, et al. Association of neurocognitive and physical function with gait speed in midlife. JAMA Netw Open. 2019;2(10):e1913123.
    View this article via: CrossRef PubMed Google Scholar
44. Middleton A, et al. Walking speed: the functional vital sign. J Aging Phys Act. 2015;23(2):314–322.
    View this article via: PubMed CrossRef Google Scholar
45. Bohannon RW. Grip strength: an indispensable biomarker for older adults. Clin Interv Aging. 2019;14:1681–1691.
    View this article via: CrossRef PubMed Google Scholar
46. Leong DP, et al. Prognostic value of grip strength: findings from the Prospective Urban Rural Epidemiology (PURE) study. Lancet. 2015;386(9990):266–273.
    View this article via: CrossRef PubMed Google Scholar
47. Wu Y, et al. Association of grip strength with risk of all-cause mortality, cardiovascular diseases, and cancer in community-dwelling populations: a meta-analysis of prospective cohort studies. J Am Med Dir Assoc. 2017;18(6):551.e17–551.e35.
    View this article via: CrossRef PubMed Google Scholar
48. de Fátima Ribeiro Silva C, et al. Short physical performance battery as a measure of physical performance and mortality predictor in older adults: a comprehensive literature review. Int J Environ Res Public Health. 2021;18(20):10612.
    View this article via: CrossRef PubMed Google Scholar
49. Christopher A, et al. The reliability and validity of the Timed Up and Go as a clinical tool in individuals with and without disabilities across a lifespan: a systematic review. Disabil Rehabil. 2021;43(13):1799–1813.
    View this article via: CrossRef PubMed Google Scholar
50. Brito LBB de, et al. Ability to sit and rise from the floor as a predictor of all-cause mortality. Eur J Prev Cardiol. 2014;21(7):892–898.
    View this article via: CrossRef PubMed Google Scholar
51. Rezaei A, et al. Age-related changes in gait, balance, and strength parameters: a cross-sectional study. PLoS One. 2024;19(10):e0310764.
    View this article via: CrossRef PubMed Google Scholar
52. Cotton RJ, et al. Transforming gait: video-based spatiotemporal gait analysis. Annu Int Conf IEEE Eng Med Biol Soc. 2022;2022:115–120.
    View this article via: PubMed CrossRef Google Scholar
53. Peiffer JD, et al. Portable biomechanics laboratory: clinically accessible movement analysis from a handheld smartphone [preprint]. https://doi.org/10.48550/arXiv.2507.08268 Posted on arXiv December 23, 2025.
54. Pinto JM, et al. Global sensory impairment predicts morbidity and mortality in older U.S. Adults. J Am Geriatr Soc. 2017;65(12):2587–2595.
    View this article via: CrossRef PubMed Google Scholar
55. Schubert CR, et al. Sensory impairments and risk of mortality in older adults. J Gerontol A Biol Sci Med Sci. 2017;72(5):710–715.
    View this article via: PubMed CrossRef Google Scholar
56. Zhou L, et al. Association of visual and hearing impairments with all-cause mortality in older adults in China. BMC Geriatr. 2025;25(1):572.
    View this article via: CrossRef PubMed Google Scholar
57. Seals DR, et al. Aging and vascular endothelial function in humans. Clin Sci. 2011;120(9):357–375.
    View this article via: CrossRef PubMed Google Scholar
58. Cooper LL, et al. Digital peripheral arterial tonometry and cardiovascular disease events: the framingham heart study. Stroke. 2021;52(9):2866–2873.
    View this article via: CrossRef PubMed Google Scholar
59. Yeboah J, et al. Predictive value of brachial flow-mediated dilation for incident cardiovascular events in a population-based study: the multi-ethnic study of atherosclerosis. Circulation. 2009;120(6):502–509.
    View this article via: CrossRef PubMed Google Scholar
60. Lacolley P, et al. Mechanisms of arterial stiffening: from mechanotransduction to epigenetics. Arterioscler Thromb Vasc Biol. 2020;40(5):1055–1062.
    View this article via: CrossRef PubMed Google Scholar
61. Wilkinson IB, et al. Uses of arterial stiffness in clinical practice. Arterioscler Thromb Vasc Biol. 2020;40(5):1063–1067.
    View this article via: CrossRef PubMed Google Scholar
62. Vlachopoulos C, et al. Prediction of cardiovascular events and all-cause mortality with arterial stiffness: a systematic review and meta-analysis. J Am Coll Cardiol. 2010;55(13):1318–1327.
    View this article via: CrossRef PubMed Google Scholar
63. Lakatta EG, Levy D. Arterial and cardiac aging: major shareholders in cardiovascular disease enterprises: Part I: aging arteries: a “set up” for vascular disease. Circulation. 2003;107(1):139–146.
    View this article via: CrossRef PubMed Google Scholar
64. Lakatta EG, Levy D. Arterial and cardiac aging: major shareholders in cardiovascular disease enterprises: part I: aging arteries: a “set up” for vascular disease. Circulation. 2003;107(1):139–146.
    View this article via: CrossRef PubMed Google Scholar
65. Zhao Y, et al. Heart rate variability and cardiovascular diseases: a Mendelian randomization study. Eur J Clin Invest. 2024;54(1):e14085.
    View this article via: CrossRef PubMed Google Scholar
66. Gharacholou SM, et al. Left ventricular diastolic function and long-term outcomes in patients with normal exercise echocardiographic findings. Am J Cardiol. 2013;112(2):200–207.
    View this article via: CrossRef PubMed Google Scholar
67. Ratneswaren A, et al. Tissue Doppler echocardiography predicts long-term cardiovascular mortality: the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT) legacy 20-year follow-up study. Open Heart. 2025;12(1):e002795.
    View this article via: CrossRef PubMed Google Scholar
68. Loe H, et al. Cardio-respiratory reference data in 4631 healthy men and women 20-90 years: the HUNT 3 fitness study. PLoS One. 2014;9(11):e113884.
    View this article via: CrossRef PubMed Google Scholar
69. Clausen JSR, et al. Midlife cardiorespiratory fitness and the long-term risk of mortality: 46 years of follow-up. J Am Coll Cardiol. 2018;72(9):987–995.
    View this article via: CrossRef PubMed Google Scholar
70. Vaz Fragoso CA, Gill TM. Respiratory impairment and the aging lung: a novel paradigm for assessing pulmonary function. J Gerontol A Biol Sci Med Sci. 2012;67(3):264–275.
    View this article via: PubMed CrossRef Google Scholar
71. Vaz Fragoso CA, et al. Aging-related considerations when evaluating the forced expiratory volume in 1 second (FEV1) over time. J Gerontol A Biol Sci Med Sci. 2016;71(7):929–934.
    View this article via: CrossRef PubMed Google Scholar
72. Thomas ET, et al. Rate of normal lung function decline in ageing adults: a systematic review of prospective cohort studies. BMJ Open. 2019;9(6):e028150.
    View this article via: CrossRef PubMed Google Scholar
73. Wang TJ, et al. Plasma natriuretic peptide levels and the risk of cardiovascular events and death. N Engl J Med. 2004;350(7):655–663.
    View this article via: CrossRef PubMed Google Scholar
74. Muscari A, et al. N-terminal pro B-type natriuretic peptide (NT-proBNP): a possible surrogate of biological age in the elderly people. Geroscience. 2021;43(2):845–857.
    View this article via: CrossRef PubMed Google Scholar
75. Muscari A, et al. The association of proBNPage with manifestations of age-related cardiovascular, physical, and psychological impairment in community-dwelling older adults. Geroscience. 2021;43(4):2087–2100.
    View this article via: CrossRef PubMed Google Scholar
76. Yao S, et al. Association of BNP with frailty in elderly population: Rugao longevity and ageing study. J Nutr Health Aging. 2019;23(1):73–78.
    View this article via: CrossRef PubMed Google Scholar
77. Odden MC, et al. Age and cystatin C in healthy adults: a collaborative study. Nephrol Dial Transplant. 2010;25(2):463–469.
    View this article via: CrossRef PubMed Google Scholar
78. Tanto C, et al. Cystatin C as predictor of long-term mortality in elderly: a systematic review and meta-analysis. SN Compr Clin Med. 2022;4(1):171.
    View this article via: CrossRef Google Scholar
79. Sarnak MJ, et al. Cystatin C and aging success. Arch Intern Med. 2008;168(2):147–153.
    View this article via: CrossRef PubMed Google Scholar
80. Vaughan DE, et al. Plasminogen activator inhibitor-1 is a marker and a mediator of senescence. Arterioscler Thromb Vasc Biol. 2017;37(8):1446–1452.
    View this article via: CrossRef PubMed Google Scholar
81. Khan SS, et al. A null mutation in SERPINE1 protects against biological aging in humans. Sci Adv. 2017;3(11):eaao1617.
    View this article via: CrossRef PubMed Google Scholar
82. Xu Z, Huang Y. Blood PAI-1 and cardiovascular and metabolic risk factors among the middle-aged women from SWAN study. Sci Rep. 2024;14(1):21207.
    View this article via: CrossRef PubMed Google Scholar
83. López-Otín C, et al. Hallmarks of aging: an expanding universe. Cell. 2023;186(2):243–278.
    View this article via: CrossRef PubMed Google Scholar
84. Cavero-Redondo I, et al. Glycated haemoglobin A1c as a risk factor of cardiovascular outcomes and all-cause mortality in diabetic and non-diabetic populations: a systematic review and meta-analysis. BMJ Open. 2017;7(7):e015949.
    View this article via: CrossRef PubMed Google Scholar
85. Franceschi C, Campisi J. Chronic inflammation (inflammaging) and its potential contribution to age-associated diseases. J Gerontol A Biol Sci Med Sci. 2014;69(Suppl 1):S4–S9.
    View this article via: CrossRef PubMed Google Scholar
86. Tyrrell DJ, et al. Age-associated mitochondrial dysfunction accelerates atherogenesis. Circ Res. 2020;126(3):298–314.
    View this article via: CrossRef PubMed Google Scholar
87. Ridker PM, et al. Inflammation and cholesterol as predictors of cardiovascular events among patients receiving statin therapy: a collaborative analysis of three randomised trials. Lancet. 2023;401(10384):1293–1301.
    View this article via: CrossRef PubMed Google Scholar
88. Lazar-Poloczek E, et al. Levels of TNF-α and soluble TNF receptors in normal-weight, overweight and obese patients with dilated non-ischemic cardiomyopathy: does anti-TNF therapy still have potential to be used in heart failure depending on BMI? Biomedicines. 2022;10(11):2959.
    View this article via: CrossRef PubMed Google Scholar
89. Carlsson AC, et al. 10-year associations between tumor necrosis factor receptors 1 and 2 and cardiovascular events in patients with stable coronary heart disease: a CLARICOR (effect of clarithromycin on mortality and morbidity in patients with ischemic heart disease) trial substudy. J Am Heart Assoc. 2018;7(9):e008299.
    View this article via: CrossRef PubMed Google Scholar
90. Neirynck N, et al. Soluble tumor necrosis factor receptor 1 and 2 predict outcomes in advanced chronic kidney disease: a prospective cohort study. PLoS One. 2015;10(3):e0122073.
    View this article via: CrossRef PubMed Google Scholar
91. Schnabel RB, et al. Multiple inflammatory biomarkers in relation to cardiovascular events and mortality in the community. Arterioscler Thromb Vasc Biol. 2013;33(7):1728–1733.
    View this article via: CrossRef PubMed Google Scholar
92. Cesari M, et al. Inflammatory markers and onset of cardiovascular events: results from the Health ABC study. Circulation. 2003;108(19):2317–2322.
    View this article via: CrossRef PubMed Google Scholar
93. Evans DS, et al. Proteomic analysis of the senescence-associated secretory phenotype: GDF-15, IGFBP-2, and Cystatin-C Are associated with multiple aging traits. J Gerontol A Biol Sci Med Sci. 2024;79(3):glad265.
    View this article via: CrossRef PubMed Google Scholar
94. Higashi Y, et al. Aging, atherosclerosis, and IGF-1. J Gerontol A Biol Sci Med Sci. 2012;67(6):626–639.
    View this article via: CrossRef PubMed Google Scholar
95. Pani LN, et al. Effect of aging on A1C levels in individuals without diabetes: evidence from the Framingham Offspring Study and the National Health and Nutrition Examination Survey 2001-2004. Diabetes Care. 2008;31(10):1991–1996.
    View this article via: CrossRef PubMed Google Scholar
96. Selvin E, et al. Glycated hemoglobin, diabetes, and cardiovascular risk in nondiabetic adults. N Engl J Med. 2010;362(9):800–811.
    View this article via: CrossRef PubMed Google Scholar
97. Neto NJ, et al. HbA1c and physical performance in older adults from different aging epidemiological contexts: longitudinal findings of the International Mobility in Aging Study (IMIAS). Arch Gerontol Geriatr. 2023;104:104823.
    View this article via: CrossRef PubMed Google Scholar
98. Palta P, et al. Hemoglobin A_1c and mortality in older adults with and without diabetes: results from the national health and nutrition examination surveys (1988-2011). Diabetes Care. 2017;40(4):453–460.
    View this article via: CrossRef PubMed Google Scholar
99. Dubowitz N, et al. Aging is associated with increased HbA1c levels, independently of glucose levels and insulin resistance, and also with decreased HbA1c diagnostic specificity. Diabet Med. 2014;31(8):927–935.
    View this article via: CrossRef PubMed Google Scholar
100. Rooney MR, et al. Glycaemic markers and all-cause mortality in older adults with and without diabetes: the Atherosclerosis Risk in Communities (ARIC) study. Diabetologia. 2021;64(2):339–348.
     View this article via: CrossRef PubMed Google Scholar
101. Levine ME, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY). 2018;10(4):573–591.
     View this article via: CrossRef PubMed Google Scholar
102. Belsky DW, et al. Quantification of biological aging in young adults. Proc Natl Acad Sci U S A. 2015;112(30):4104–4110.
     View this article via: CrossRef PubMed Google Scholar
103. Merchant RA, et al. Editorial: resilience and successful aging. J Nutr Health Aging. 2022;26(7):652–656.
     View this article via: CrossRef PubMed Google Scholar
104. Zhou Y, Ma L. Intrinsic capacity in older adults: recent advances. Aging Dis. 2022;13(2):353–359.
     View this article via: CrossRef PubMed Google Scholar
105. Bernal MC, et al. A functional approach to model intrinsic capacity in ageing trajectories. Sci Rep. 2025;15(1):7878.
     View this article via: CrossRef PubMed Google Scholar
106. Sayed N, et al. An inflammatory aging clock (iAge) based on deep learning tracks multimorbidity, immunosenescence, frailty and cardiovascular aging. Nat Aging. 2021;1(7):598–615.
     View this article via: CrossRef Google Scholar
107. Maddock J, et al. DNA methylation age and physical and cognitive aging. J Gerontol A Biol Sci Med Sci. 2019;75(3):504–511.
     View this article via: PubMed CrossRef Google Scholar
108. Milholland B, et al. Mutation and catastrophe in the aging genome. Exp Gerontol. 2017;94:34–40.
     View this article via: CrossRef PubMed Google Scholar
109. López-Otín C, et al. The hallmarks of aging. Cell. 2013;153(6):1194–1217.
     View this article via: CrossRef PubMed Google Scholar
110. Tartiere AG, et al. The hallmarks of aging as a conceptual framework for health and longevity research. Front Aging. 2024;5:1334261.
     View this article via: CrossRef PubMed Google Scholar
111. Zetterberg H, et al. The Szilard hypothesis on the nature of aging revisited. Genetics. 2009;182(1):3–9.
     View this article via: CrossRef PubMed Google Scholar
112. Berdyshev GD, et al. [Nucleotide composition of DNA and RNA from somatic tissues of humpback and its changes during spawning]. Biokhimiia. 1967;32(5):988–993.
     View this article via: PubMed Google Scholar
113. Vanyushin BF, et al. The 5-Methylcytosine in DNA of rats: tissue and age specificity and the changes induced by hydrocortisone and other agents. Gerontologia. 2009;19(3):138–152.
     View this article via: CrossRef PubMed Google Scholar
114. Xiong J, et al. DNA hydroxymethylation age of human blood determined by capillary hydrophilic-interaction liquid chromatography/mass spectrometry. Clin Epigenetics. 2015;7(1):72.
     View this article via: CrossRef PubMed Google Scholar
115. Bocklandt S, et al. Epigenetic predictor of age. PLoS One. 2011;6(6):e14821.
     View this article via: CrossRef PubMed Google Scholar
116. Higgins-Chen AT, et al. A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking. Nat Aging. 2022;2(7):644–661.
     View this article via: CrossRef PubMed Google Scholar
117. Higgins-Chen AT, et al. A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking. Nat Aging. 2022;2(7):644–661.
     View this article via: CrossRef PubMed Google Scholar
118. Hannum G, et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell. 2013;49(2):359–367.
     View this article via: CrossRef PubMed Google Scholar
119. Bell CG, et al. DNA methylation aging clocks: challenges and recommendations. Genome Biol. 2019;20(1):249.
     View this article via: CrossRef PubMed Google Scholar
120. Marioni RE, et al. DNA methylation age of blood predicts all-cause mortality in later life. Genome Biol. 2015;16(1):25.
     View this article via: CrossRef PubMed Google Scholar
121. Chen BH, et al. DNA methylation-based measures of biological age: meta-analysis predicting time to death. Aging (Albany NY). 2016;8(9):1844–1865.
     View this article via: CrossRef PubMed Google Scholar
122. Sugden K, et al. Patterns of reliability: assessing the reproducibility and integrity of DNA methylation measurement. Patterns (N Y). 2020;1(2):100014.
     View this article via: CrossRef PubMed Google Scholar
123. Bose M, et al. Evaluation of microarray-based DNA methylation measurement using technical replicates: the Atherosclerosis Risk In Communities (ARIC) Study. BMC Bioinformatics. 2014;15(1):312.
     View this article via: CrossRef PubMed Google Scholar
124. McEwen LM, et al. Systematic evaluation of DNA methylation age estimation with common preprocessing methods and the Infinium MethylationEPIC BeadChip array. Clin Epigenetics. 2018;10(1):123.
     View this article via: CrossRef PubMed Google Scholar
125. McCrory C, et al. Grimage outperforms other epigenetic clocks in the prediction of age-related clinical phenotypes and all-cause mortality. J Gerontol A Biol Sci Med Sci. 2020;76(5):741–749.
     View this article via: CrossRef PubMed Google Scholar
126. Lu AT, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019;11(2):303–327.
     View this article via: CrossRef PubMed Google Scholar
127. Hillary RF, et al. An epigenetic predictor of death captures multi-modal measures of brain health. Mol Psychiatry. 2021;26(8):3806–3816.
     View this article via: CrossRef PubMed Google Scholar
128. Belsky DW, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420.
     View this article via: CrossRef PubMed Google Scholar
129. Elliott ML, et al. Disparities in the pace of biological aging among midlife adults of the same chronological age have implications for future frailty risk and policy. Nat Aging. 2021;1(3):295–308.
     View this article via: CrossRef PubMed Google Scholar
130. Sehgal R, et al. Systems Age: a single blood methylation test to quantify aging heterogeneity across 11 physiological systems. Nat Aging. 2025;5(9):1880–1896.
     View this article via: CrossRef PubMed Google Scholar
131. Chen Q, et al. OMICmAge: an integrative multi-omics approach to quantify biological age with electronic medical records [preptint]. https://doi.org/10.1101/2023.10.16.562114 Posted on bioRxiv October 24, 2023.
132. Sathyan S, et al. Plasma proteomic profile of age, health span, and all-cause mortality in older adults. Aging Cell. 2020;19(11):e13250.
     View this article via: CrossRef PubMed Google Scholar
133. Tanaka T, et al. Plasma proteomic signature of age in healthy humans. Aging Cell. 2018;17(5):e12799.
     View this article via: CrossRef PubMed Google Scholar
134. Gold L, et al. Aptamer-based multiplexed proteomic technology for biomarker discovery. PLoS One. 2010;5(12):e15004.
     View this article via: CrossRef PubMed Google Scholar
135. Unterhuber M, et al. Proteomics-enabled deep learning machine algorithms can enhance prediction of mortality. J Am Coll Cardiol. 2021;78(16):1621–1631.
     View this article via: CrossRef PubMed Google Scholar
136. Kuo C, et al. Proteomic aging clock (PAC) predicts age-related outcomes in middle-aged and older adults. Aging Cell. 2024;23(8):e14195.
     View this article via: CrossRef PubMed Google Scholar
137. Oh HSH, et al. Organ aging signatures in the plasma proteome track health and disease. Nature. 2023;624(7990):164–172.
     View this article via: CrossRef PubMed Google Scholar
138. You J, et al. Plasma proteomic profiles predict individual future health risk. Nat Commun. 2023;14(1):7817.
     View this article via: CrossRef PubMed Google Scholar
139. Krištić J, et al. Glycans are a novel biomarker of chronological and biological ages. J Gerontol A Biol Sci Med Sci. 2014;69(7):779–789.
     View this article via: CrossRef PubMed Google Scholar
140. Mutz J, et al. Metabolomic age (MileAge) predicts health and life span: a comparison of multiple machine learning algorithms. Sci Adv. 2024;10(51):eadp3743.
     View this article via: CrossRef PubMed Google Scholar
141. Hwangbo N, et al. A metabolomic aging clock using human cerebrospinal fluid. J Gerontol A Biol Sci Med Sci. 2021;77(4):744–754.
     View this article via: CrossRef PubMed Google Scholar
142. Lau CHE, et al. NMR metabolomic modeling of age and lifespan: a multicohort analysis. Aging Cell. 2024;23(7):e14164.
     View this article via: CrossRef PubMed Google Scholar
143. Haugg F, et al. Imaging biomarkers of ageing: a review of artificial intelligence-based approaches for age estimation. Lancet Healthy Longev. 2025;6(7):100728.
     View this article via: CrossRef PubMed Google Scholar
144. Cheema B, Pandit J. AI and heart failure: present state and future with multimodal large language models. JACC Adv. 2024;3(9):101029.
     View this article via: CrossRef PubMed Google Scholar
145. Cheema B, et al. Artificial intelligence in the management of heart failure. J Card Fail. 2025;31(10):1561–1573.
     View this article via: CrossRef PubMed Google Scholar
146. Zhou Y, et al. A foundation model for generalizable disease detection from retinal images. Nature. 2023;622(7981):156–163.
     View this article via: CrossRef PubMed Google Scholar
147. Li J, et al. An electrocardiogram foundation model built on over 10 million recordings. NEJM AI. 2025;2(7):10.1056/aioa2401033.
     View this article via: PubMed CrossRef Google Scholar
148. Wu Z, Guo C. Deep learning and electrocardiography: systematic review of current techniques in cardiovascular disease diagnosis and management. Biomed Eng Online. 2025;24(1):23.
     View this article via: CrossRef PubMed Google Scholar
149. Lopez-Jimenez F, et al. Assessing biological age: the potential of ECG Evaluation using artificial intelligence: JACC family series. JACC Clin Electrophysiol. 2024;10(4):775–789.
     View this article via: CrossRef PubMed Google Scholar
150. Lima EM, et al. Deep neural network-estimated electrocardiographic age as a mortality predictor. Nat Commun. 2021;12(1):5117.
     View this article via: CrossRef PubMed Google Scholar
151. Attia ZI, et al. Age and sex estimation using artificial intelligence from standard 12-lead ECGs. Circ Arrhythm Electrophysiol. 2019;12(9):e007284.
     View this article via: CrossRef PubMed Google Scholar
152. Grimbly MJ, et al. Estimating biological age from retinal imaging: a scoping review. BMJ Open Ophthalmol. 2024;9(1):e001794.
     View this article via: CrossRef PubMed Google Scholar
153. Zhu Z, et al. Retinal age gap as a predictive biomarker for mortality risk. Br J Ophthalmol. 2023;107(4):547–554.
     View this article via: CrossRef PubMed Google Scholar
154. Chen R, et al. Association between cardiovascular health metrics and retinal ageing. Geroscience. 2023;45(3):1511–1521.
     View this article via: CrossRef PubMed Google Scholar
155. Zhang S, et al. Association of retinal age gap and risk of kidney failure: a UK Biobank Study. Am J Kidney Dis. 2023;81(5):537–544.e1.
     View this article via: CrossRef PubMed Google Scholar
156. Poplin R, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018;2(3):158–164.
     View this article via: CrossRef PubMed Google Scholar
157. Diaz-Pinto A, et al. Predicting myocardial infarction through retinal scans and minimal personal information. Nat Mach Intell. 2022;4(1):55–61.
     View this article via: CrossRef Google Scholar
158. Cheung CY, et al. A deep learning model for detection of Alzheimer’s disease based on retinal photographs: a retrospective, multicentre case-control study. Lancet Digit Health. 2022;4(11):e806–e815.
     View this article via: CrossRef PubMed Google Scholar
159. Nusinovici S, et al. Retinal photograph-based deep learning predicts biological age, and stratifies morbidity and mortality risk. Age Ageing. 2022;51(4):afac065.
     View this article via: CrossRef PubMed Google Scholar
160. Chen R, et al. Accelerated retinal ageing and multimorbidity in middle-aged and older adults. Geroscience. 2025;47(3):4291–4300.
     View this article via: CrossRef PubMed Google Scholar
161. Grzybowski A, et al. Retina fundus photograph-based artificial intelligence algorithms in medicine: a systematic review. Ophthalmol Ther. 2024;13(8):2125–2149.
     View this article via: CrossRef PubMed Google Scholar
162. Wang J, et al. Artificial intelligence-enhanced retinal imaging as a biomarker for systemic diseases. Theranostics. 2025;15(8):3223–3233.
     View this article via: CrossRef PubMed Google Scholar
163. Tanveer M, et al. Ensemble deep learning in speech signal tasks: a review. Neurocomputing. 2023;550:126436.
     View this article via: CrossRef Google Scholar
164. Gaser C, et al. A perspective on brain-age estimation and its clinical promise. Nat Comput Sci. 2024;4(10):744–751.
     View this article via: CrossRef PubMed Google Scholar
165. Cole JH, et al. Brain age predicts mortality. Mol Psychiatry. 2018;23(5):1385–1392.
     View this article via: CrossRef PubMed Google Scholar
166. Raghu VK, et al. Deep learning to estimate biological age from chest radiographs. JACC Cardiovasc Imaging. 2021;14(11):2226–2236.
     View this article via: CrossRef PubMed Google Scholar
167. Ieki H, et al. Deep learning-based age estimation from chest X-rays indicates cardiovascular prognosis. Commun Med (Lond). 2022;2(1):159.
     View this article via: CrossRef PubMed Google Scholar
168. Bontempi D, et al. FaceAge, a deep learning system to estimate biological age from face photographs to improve prognostication: a model development and validation study. Lancet Digit Health. 2025;7(6):100870.
     View this article via: CrossRef PubMed Google Scholar
169. Li Y, et al. Large language model-based biological age prediction in large-scale populations. Nat Med. 2025;31(9):2977–2990.
     View this article via: CrossRef PubMed Google Scholar
170. Rajpurkar P, et al. AI in health and medicine. Nat Med. 2022;28(1):31–38.
     View this article via: CrossRef PubMed Google Scholar
171. Blesky DW, et al. Eleven telomere, epigenetic clock, and biomarker-composite quantifications of biological aging: do they measure the same thing? Am J Epidemiol. 2018;187(6):1220–1230.
     View this article via: PubMed CrossRef Google Scholar