Biological age threshold is associated with symptomatic knee osteoarthritis risk in chinese adults: Insights from machine learning analysis of a national cohort

Knee osteoarthritis (KOA), a prevalent and debilitating chronic joint disorder characterized by articular cartilage degradation, synovitis, and subchondral bone remodeling, imposes a substantial burden on global healthcare [1]. Established risk factors, including chronological aging, obesity, joint trauma, inflammation, and genetic predisposition, significantly contribute to its pathogenesis and progression [2]. Globally, KOA ranks as the 11th leading cause of disability, affecting approximately 3.8% of the world's population [3]. In China, an estimated 37.35 million adults aged ≥60 years are affected by symptomatic KOA [4], highlighting the urgent need for improved risk stratification strategies to enable early identification and timely intervention.

Although chronological age (CA) is a well-established risk factor for KOA, it inadequately captures the considerable heterogeneity observed in individual aging trajectories and susceptibility to age-related pathologies [5]. Biological age (BA), constructed using panels of accessible blood biomarkers that reflect systemic physiological integrity, including inflammatory, metabolic, and hormonal profiles, provides a more precise quantification of an individual's functional decline and molecular aging processes than CA alone [6,7]. Globally, various methods have been developed to estimate BA, such as the Klemera-Doubal method (KDM), PhenoAge, and others, each incorporating distinct biomarker panels to capture multidimensional aspects of aging [8,9]. Studies across different populations have shown that accelerated BA, constructed using diverse methodologies, is consistently associated with a higher risk of age-related diseases, including osteoarthritis [10,11]. Mechanistically, accelerated BA is linked to KOA pathogenesis through processes such as cellular senescence and the proinflammatory senescence-associated secretory phenotype (SASP) [12]. Therefore, integrating BA into KOA risk assessment frameworks may improve the understanding of heterogeneous aging-related risk.

The China Health and Retirement Longitudinal Study (CHARLS) provides a valuable resource for constructing predictive models [13]. This nationally representative survey encompassed individuals aged ≥ 45 years in China and gathered extensive information on health, social, and economic variables [14]. Despite the theoretical promise of integrating BA into symptomatic KOA risk assessment, evidence from large-scale, nationally representative studies on the Chinese population remains limited. Moreover, comparative insights from international cohorts regarding BA construction methods and their clinical applicability in the context of KOA are scarce. To address this gap, we conducted a cross-sectional study using data from the 2011 and 2015 waves of CHARLS. We employed descriptive statistics, multivariable logistic regression, restricted cubic spline analysis, and subgroup analyses to examine the association between BA and symptomatic KOA. Furthermore, we developed six machine learning models and used SHapley Additive exPlanations (SHAP) to interpret the optimal model. This study aimed to provide robust evidence for incorporating BA into symptomatic KOA risk assessment, facilitating the identification of high-risk individuals and informing effective prevention strategies.

This cross-sectional analysis of the CHARLS dataset revealed a robust association between BA and symptomatic KOA prevalence. Each 1-year increase in BA was associated with 1.23%–1.58% higher odds of KOA (fully adjusted OR = 1.0123, 95% CI: 1.0049–1.0197, p = 0.001), with a graded dose-response relationship. Participants in the highest BA quartile had a 45.19% higher risk of symptomatic KOA than those in the lowest quartile (OR = 1.4519, 95% CI: 1.1755–1.7956). Restricted cubic spline analysis revealed a non-linear association, with symptomatic KOA risk accelerating steeply beyond BA ≥ 66.7 years (non-linearity p = 0.013). These findings are consistent with international evidence from two cross-sectional studies in the U.S. National Health and Nutrition Examination Survey (NHANES) [20,21]. Both studies consistently demonstrated positive associations between KDM-based biological aging indicators (including BA acceleration) and the risk of osteoarthritis. In the fully adjusted models, the highest quartile of BA acceleration was associated with a 36.3% increase in OA risk (OR = 1.363, 95% CI: 1.213–1.532), with evident dose-response relationships. Additionally, sex-specific effects were observed, with stronger associations identified in females than in males. One study further revealed that biological aging mediates the relationship between central obesity and OA, with a mediation proportion ranging from 0% to 40%. Together, these results support the cross-population robustness of KDM-based biological aging measures as reliable markers for OA risk stratification. The XGBoost machine learning model achieved an accuracy of 83.37% and an AUROC of 0.9078, with SHAP values confirming BA as the most influential feature (mean absolute SHAP value >0.6). These findings underscore the importance of BA as a pivotal biomarker for the risk stratification of symptomatic KOA.

Furthermore, the divergence between BA and CA underscores heterogeneous aging trajectories, likely reflecting cumulative exposure to environmental, lifestyle, and genetic factors. The accelerated risk of symptomatic KOA beyond a BA threshold of ~66.7 years suggests that individuals with advanced biological aging may experience a compressed health span, with an earlier onset of age-related disorders. Future longitudinal studies are needed to explore whether BA acceleration is associated with overall lifespan reduction or earlier disability.

Our results align closely with those of existing mechanistic and epidemiological studies, enhancing our understanding of the relationship between BA and symptomatic KOA. Mechanistically, previous studies have established that cellular senescence and the subsequent SASP play pivotal roles in pathogenesis [22,23]. At the cellular level, senescent chondrocytes within the articular cartilage exhibit a decline in matrix synthesis and upregulation of catabolic factors such as matrix metalloproteinases (MMPs) and pro-inflammatory cytokines such as interleukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α) [24 –26]. These factors disrupt the normal balance between cartilage synthesis and degradation, leading to progressive cartilage erosion and joint inflammation and accelerating the process of KOA [27 –29].

In the context of biological aging, systemic factors associated with BA acceleration likely contribute to local joint pathology. For instance, elevated oxidative stress, a common feature of increased BA, can directly damage chondrocytes and extracellular matrix components [30,31]. It can also promote the activation of nuclear factor kappa B (NF-κB), a key transcription factor that upregulates the production of SASP factors, thus creating a pro-inflammatory microenvironment conducive to KOA development [32,33]. Additionally, chronic low-grade inflammation associated with aging, as reflected by elevated BA levels, may dysregulate immune cell function in joints, including macrophages and T lymphocytes [34,35]. These immune cells secrete cytokines and proteases, exacerbating cartilage damage and synovial inflammation [36 –38].

Epidemiologically, our study extends previous work by demonstrating a population-level association between BA and symptomatic KOA using a large, nationally representative Chinese cohort, addressing the limitation that many prior epidemiological investigations predominantly relied on CA as a risk factor for symptomatic KOA while overlooking the potential value of BA, a metric that integrates multiple biomarkers reflecting an individual's overall biological state. By incorporating BA into the analysis, we provided a more nuanced understanding of disease risk, with the identified non-linear BA-symptomatic KOA relationship supporting the concept proposed in recent literature of critical biological aging thresholds beyond which the risk of joint degenerative diseases, such as KOA, may increase exponentially, consistent with earlier research highlighting BA as a powerful feature of various age-related pathologies, including musculoskeletal disorders [39 –41]. In terms of discriminative accuracy, BA's multisystem biomarker panel offers a more comprehensive measure of an individual's functional decline compared to CA, as further validated by our machine learning results using the XGBoost model, wherein BA served as a top feature with high accuracy and AUROC values, aligning with emerging trends advocating composite biomarkers to improve the assessment of age-related disease risk.

Importantly, our findings demonstrated the incremental value of BA beyond established risk factors such as CA and BMI. This study advances the conventional risk prediction model developed by Wang et al., which relies on demographic variables including age and sex (AUC = 0.719) [4], by introducing BA as an innovative indicator. BA not only integrates biomarkers reflective of multi-system biological aging, but also, when incorporated into the XGBoost model, exhibits the potential to enhance predictive performance, with an achieved AUROC of 0.9078, thereby providing a new dimension for more precise stratification of high-risk populations. While CA is well recognized, BA captures biological heterogeneity within age groups and is strongly associated with symptomatic KOA risk, supported by three key observations: BA emerged as the most influential feature in the robust XGBoost model, outperforming traditional factors in SHAP value contributions; its significant association with symptomatic KOA persisted after rigorous adjustment for covariates in logistic regression; and a BA-specific non-linear risk acceleration point was identified, which CA alone may obscure. This enhanced risk stratification has substantial clinical promise; However, the translation of BA assessment into routine clinical practice faces challenges, including the cost and feasibility of measuring multiple biomarkers, the need for population-specific validation of BA algorithms, and the generalizability of the findings to diverse healthcare settings. The strong association between BA and symptomatic KOA, coupled with its superior performance in discriminative models, suggests that BA may hold potential for enhancing risk stratification frameworks in future studies. Integrating BA assessment into clinical practice could refine screening protocols by identifying individuals with elevated symptomatic KOA risk attributable to their biological aging trajectory, even those with "lower-risk" CA or BMI. This facilitates targeted preventive interventions, including lifestyle modifications, enhanced monitoring, and future senotherapeutics, thereby maximizing benefits and ultimately delaying the onset or slowing the progression of symptomatic KOA. The current study had several limitations that merit consideration. First, its cross-sectional nature prevents definitive inference regarding causality between BA and symptomatic KOA, as temporal relationships cannot be established. Second, the diagnosis of symptomatic KOA was based solely on self-reported clinical information without radiographic confirmation (e.g., Kellgren-Lawrence grading), which may have introduced misclassification bias. Future studies incorporating imaging assessments will strengthen diagnostic precision. Third, the BA calculation relied on a reduced set of 8 systemic biomarkers instead of the standard 12 used in the KDM method because of data availability constraints in the CHARLS. Although prior studies have validated similar reduced sets in Chinese populations [18,19], this approach may introduce bias and affect the absolute value of BA estimates, potentially limiting direct comparability with studies using the full biomarker panel. Future research should aim to utilize the complete set of biomarkers where possible. Fourth, the multistage participant selection process (Fig 1) necessary to obtain a sample with complete data for the core variables (BA and symptomatic KOA status) may have introduced selection bias. A direct sensitivity analysis was not feasible because of the lack of core variables in the excluded groups. Although the final analytical sample retained the broad characteristics of the CHARLS cohort, this potential bias may have affected the representativeness of our sample and limited the generalizability of our findings. Fifth, although occupational physical load and activity levels are potential contributors to KOA risk, we did not adjust for them in our primary model. This decision was based on the limitations of the available data (e.g., lack of detailed lifetime occupational history and objective physical activity measures) and the substantial risk of reverse causality, as prevalent KOA severely impacts an individual's current activities and work capacity. Future prospective studies with dedicated pre-symptomatic exposure assessments are needed to disentangle these complex relationships. Furthermore, although we performed rigorous internal validation through cross-validation and subgroup analyses, the lack of an independent external dataset for validation means that the generalizability of the observed associations may be limited to the study population. Future studies in larger, multicenter cohorts and diverse populations are needed to replicate these associations and confirm the robustness of BA as a marker for symptomatic KOA risk assessment. Finally, the generalizability of the findings to other ethnicities, younger populations, or clinical settings requires further validation, as this study focused on middle-aged and older Chinese adults from a community-based cohort. This study provides a basis for subsequent research. Longitudinal investigations are warranted to explore the dynamic interplay between BA trajectories and symptomatic KOA incidence or progression, which may help to elucidate potential causal relationships. Further studies integrating joint-specific aging biomarkers with systemic BA metrics could help refine risk stratification models for symptomatic KOA. Additionally, interventional studies examining whether targeting biological aging pathways (e.g., senolytics and anti-inflammatory agents) might mitigate symptomatic KOA risk could provide valuable insights into preventive strategies. Finally, refining machine learning models with multi-omics data (e.g., genomics and metabolomics) may enhance the discriminative utility of BA for symptomatic KOA, facilitating more personalized approaches to musculoskeletal health management.

This study demonstrated that BA was significantly associated with symptomatic KOA in a large, nationally representative Chinese cohort, with a non-linear relationship characterized by an accelerated risk escalation beyond the critical BA threshold. The strong association between BA and symptomatic KOA, coupled with its superior performance in machine learning models, highlights its strong association with symptomatic KOA and its value for risk assessment. For instance, BA assessment may have the potential for integration into health screenings to better understand heterogeneous aging-related risk profiles for which targeted preventive strategies, such as lifestyle interventions or more frequent monitoring, could be initiated before the onset of severe symptoms. These findings underscore the utility of integrating BA into epidemiological investigations of age-related musculoskeletal disorders, potentially offering new insights into public health strategies aimed at the early identification and targeted prevention of symptomatic KOA. However, the cross-sectional design precludes causal inference, and reliance on self-reported KOA data without imaging confirmation is a limitation. Future longitudinal studies incorporating objective OA measures are needed to further elucidate the interplay between biological aging and symptomatic KOA pathogenesis.