L-shaped association of plasma creatinine with frailty in oldest‑old: Evidence from a cross-sectional study of CLHLS

Abstract

This study sought to identify key circulating hematological and biochemical indicators associated with frailty in the oldest-old and to investigate their interrelationships. Data from 2,434 participants aged ≥ 80 years (3,222 observations) were analyzed from the Chinese Longitudinal Healthy Longevity Survey (2008–2014). Frailty was assessed using a 50-item frailty index (FI). A panel of 14 hematological and biochemical markers was measured at each wave. Candidate markers were initially screened using univariate generalized linear mixed models (GLMMs), followed by further selection via a least absolute shrinkage and selection operator (LASSO) logistic regression model. The selected markers were then evaluated using multivariable GLMMs, adjusting for potential confounders. Nonlinear associations were assessed using restricted cubic splines (RCS). Of the 3,222 observations, 65.3% were classified as frail. LASSO identified albumin (ALB), creatinine (CREA) and malondialdehyde (MDA) emerged as the markers associated with frailty. RCS analyses indicated an L-shaped association between CREA and frailty with a statistical inflection point at 122 µmol/L, while ALB and MDA exhibited linear associations. In the adjusted GLMMs, lower CREA levels were associated with a higher likelihood of frailty when CREA levels were below 122 µmol/L (OR = 0.84, 95% CI: 0.76–0.91, P < 0.001), whereas CREA at or above 122 µmol/L showed no significant association with frailty (OR = 1.04, 95% CI: 0.95–1.14, P = 0.376). Meanwhile, lower ALB levels (OR = 0.90, 95% CI: 0.83–0.99, P = 0.015) and higher MDA levels (OR = 1.22, 95% CI: 1.11–1.34, P < 0.001) were associated with a higher likelihood of frailty. Circulating ALB, MDA and CREA levels were associated with frailty in Chinese adults aged ≥ 80 years. A nonlinear, L-shaped association was observed between CREA and frailty.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-50163-4.

Introduction

Globally, populations are aging at an unprecedented rate. According to the World Health Organization, the number of individuals aged ≥ 60 is projected to exceed 2.1 billion, and those aged ≥ 80 years are expected to more than triple to roughly 426 million by 2050¹. These demographic changes, increasingly evident across low-and middle-income regions, herald a dramatic rise in the population of people aged 80 and over². The oldest-old are especially susceptible to health decline. Indeed, the prevalence of frailty escalates markedly with advancing age, increasing from approximately 11% among those aged 50–59 to over 50% in individuals aged 90 and above³. In some cohorts, most individuals ≥ 80 years are frail^4,5. This evidence highlights that the oldest-old represent a rapidly growing and highly vulnerable subgroup in aging societies.

Frailty is a key geriatric syndrome characterized by reduced physiological reserve and resilience. Clinically, frailty is defined as a state of diminished physiological reserve and increased vulnerability to a broad range of adverse health outcomes^3,6,7. Frail older adults are at a significantly higher risk of adverse events than their non-frail peers. For instance, frailty was associated with substantially increased risks of falls, fractures, progressive disability, hospitalization, and even early mortality⁸. Frailty also portended cognitive decline and institutionalization, and it greatly increases the burden on healthcare systems and caregivers⁹. Given the disproportionately poor functional status and quality of life experienced by frail elders, frailty has become a central target for geriatric medicine and public health¹⁰. Early recognition of frailty is seen as critical to deploying interventions that can preserve independence and reduce costly adverse outcomes.

Despite its importance, current frailty screening methods have notable limitations. Widely used tools include the Fried phenotype, the Rockwood deficit index, the Edmonton Frail Scale and the Clinical Frailty Scale. These tools rely primarily on self-reported deficits and simple performance tests^11,12. While these approaches are practical, they omit key dimensions of biological aging, including cognitive, psychosocial and subclinical processes, and depend on rater-dependent questionnaires¹². Crucially, none incorporate detailed molecular or inflammatory markers, limiting their ability to capture the complex, multisystemic nature of frailty¹³. Growing body of evidence is linking frailty to chronic inflammation, hormonal dysregulation, nutritional deficiencies, sarcopenia, and metabolic dysfunction, however, most studies have evaluated these biomarkers in isolation^8,14,15. Experts have advocated multidimensional panels integrating immune, endocrine, and metabolic signals, as well as clinical data, to improve predictive accuracy¹⁶. Indeed, combining multiple inflammatory and metabolic indicators with traditional risk factors has yielded more robust frailty prediction models^13,17, suggesting that a multi-marker signature could provide a more objective and physiologically grounded assessment of vulnerability in older adults.

In summary, the convergence of global population aging and the heightened vulnerability of the oldest-old has made frailty an increasingly important public health concern. Although existing frailty screening tools are widely used, they primarily focus on functional domains and may not fully capture underlying biological alterations. There is therefore a need to characterize a broad range of circulating biomedical indicators and to better understand how they relate to frailty in advanced age. This work will improve the understanding of the biological markers of frailty and provide population-based evidence relevant to its assessment in in high-risk aging population.

Methods

Study population

Participants were drawn from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), which was conducted by the Center for Healthy Aging and Development of Peking University, Data were used from the 2008, 2012 and 2014 waves of the survey. Each individual had biomedical measurements including blood routine and biochemistry, and survey data on health status. Following the flowchart in Figs. 1, 2 and 434 participants aged 80 years or over were included, comprising 3,222 observations. 61 individuals (2.5%) had three observations, 727 (29.9%) individuals had two observations, and 1,646 (67.6%) had one observation. The study was conducted in accordance with the Declaration of Helsinki, and the ethical approval was obtained from the Research Ethics Committee of Peking University (IRB0001052–13074). All participants or their legal representatives provided written informed consent prior to data collection.

Fig. 1.

Flowchart of participant selection.

Fig. 2.

Distributions of biochemical indicators by frailty status. Note: ALB: albumin, GSP: glycated serum protein, CHOL: total cholesterol, TG: triglyceride, HDL-C: high-density lipoprotein cholesterol, GLU: fasting blood glucose, CREA: creatinine, BUN: blood urea nitrogen, hs-CRP: high-sensitivity C-reactive protein, MDA: malondialdehyde.

Detection of hematological and biomedical indicators

The independent variables included a panel of and biochemical markers. Specifically, hematological measures comprised white blood cell count (WBC), red blood cell count (RBC), hemoglobin (HGB), platelet count (PLT), while biochemical markers consisted of albumin (ALB), total cholesterol (CHOL), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), glycated serum protein (GSP), fasting blood glucose (GLU), creatinine (CREA), blood urea nitrogen (BUN), high-sensitivity C-reactive protein (hs-CRP), and malondialdehyde (MDA). Following each survey wave, venous blood samples were collected and centrifuged to separate the plasma, which was then stored at − 20 °C and transported to the Clinical Laboratory of Capital Medical University. All assays were conducted on an automated biochemistry analyzer (Hitachi 7180, Roche) to ensure consistent and reliable measurements.

Frailty assessment

Frailty was measured using a deficit-based frailty index (FI), as described in a previous study¹⁸. This index incorporated 50 items across multiple domains, including physical health (23 items, such as self-rated health and disease status, basic and instrumental activities of daily living, physical performance, sensory function), cognitive function, as assessed by the Mini-Mental State Examination (24 items), mental health (2 items), and observer-rated health (Table S1). Each variable was scored from 0 to 2 according to severity, and the total score was divided by 51 to yield an FI value between 0.00 and 1.00. Higher values indicate greater frailty. Participants with an FI below 0.25 were classified as non-frail, while those with an FI of 0.25 or higher were deemed frail^19,20.

Covariates

Demographic characteristics and health behaviors were ascertained through a structured questionnaire. These characteristics included age categorized as 80–89, 90–99, and 100 years or older, sex, and body mass index (BMI), classified as underweight (< 18.5 kg/m²), normal weight (18.5–23.9 kg/m²), overweight (24.0–27.9 kg/m²), and obesity (≥ 28.0 kg/m²), place of residence, defined as urban, town, or rural, marital status recorded as married, widowed, or divorced/unmarried, and annual household income, grouped into < 10 000, 10 000–29 999, 30 000–49 999, and ≥ 50 000 CNY. Information on smoking, alcohol consumption, and physical activity was captured by distinguishing current or former users from individuals who had never engaged in these behaviors.

Statistical analysis

Descriptive analysis

Categorical variables were described as n (%) and compared using chi-square tests. Continuous variables were tested for normality using the Kolmogorov–Smirnov test. Non-normally distributed variables were summarized using the median (P₂₅, P₇₅) and compared by Mann–Whitney U test. Violin plots of biomedical indicators by frailty status were generated using the ggstatsplot package in RStudio. Spearman correlation was used to assess the associations among these indicators. Analyses were performed using SPSS 26.0 and R 4.5.0 software, and all tests were two-sided with α = 0.05.

Modeling framework

The statistical modeling was performed using a sequential approach: (1) Univariate GLMMs were first used for preliminary screening of candidate markers; (2) A least absolute shrinkage and selection operator (LASSO) logistic regression was then applied to further select the most robust markers while minimizing multicollinearity; (3) Multivariable GLMMs were employed to evaluate the independent associations of the selected markers with frailty, adjusting for potential confounders; (4) Finally, restricted cubic splines (RCS) were used to explore potential nonlinear relationships.

Variable selection and modeling

Generalized linear mixed model (GLMM) was used to conduct univariate screening with individual ID as a random effect and each biomedical indicator as a fixed effect. Candidate indicators with P < 0.05 were entered into a LASSO logistic regression model, employed as a variable selection technique to address potential multicollinearity, rather than to construct a predictive model. 10-fold cross-validation was used to identify the optimal lambda (λ) and assess model stability. The receiver operating characteristic (ROC) curve of the fitted model was generated to describe the discriminative properties of the selected biomarkers within this dataset, not to validate a predictive tool. The selected variables were then entered into a multivariable GLMM adjusting for age, sex, and other confounders. Variance inflation factors (VIF) then were calculated to check multicollinearity of the fitted model.

Nonlinear relationships

Nonlinear associations between key indicators and frailty likelihood were explored using RCS analysis via the rcs() function in the rms package, with knot number selected based on AIC values.

Interaction and stratified analyses

Interaction terms between age, sex and each significant indicator were added to the GLMM. Single interactions were first tested to identify any that were potentially significant, before considering multiple interactions. Stratified analyses were conducted to descriptively examine associations within subgroups.

Sensitivity analyses

Sensitivity analyses were performed to assess the robustness of our findings. First, we compared the parameter estimates obtained from GLMMs fitted to complete-case data with those obtained using multiple imputation by chained equations (MICE). Second, generalized additive mixed models (GAMMs) incorporating penalized splines were employed to reproduce the functional forms identified in the RCS analyses of GLMM. Third, we implemented generalized estimating equations (GEEs) under both exchangeable and autoregressive of order 1 [AR(1)] working correlation structures. Finally, we refitted the final model as a linear mixed model (LMM), treating the frailty index as a continuous dependent variable.

Results

Frailty status and characteristic distribution in oldest-old

As shown in Table 1, of the 3,222 observations included in this analysis, 2,103 (65.3%) were classified as frail. The median age of all observations was 92 years (interquartile range: 86–100). Of these observations, 64.2% were female, 80.0% had never smoked, and 81.8% had never consumed alcohol. When comparing frail and non-frail observations, statistically significant differences were observed in sex (P < 0.001) and age group (P < 0.001), with frailty more common in older age groups. Additionally, frail observations differed significantly from non-frail ones in terms of BMI, smoking status, alcohol consumption, physical activity, and marital status (all P < 0.001). No significant differences were found between the two groups regarding household income or place of residence.

Table 1.

Demographic characteristics and frailty status of oldest‑old participants.

Characteristics	All participants (n = 3222)	Non-frail (n = 1119)	Frail (n = 2103)	P value
Age, median (P25 P75), n (%)	92 (86 100)	95 (90 101)	87 (83 93)	< 0.001ᵃᵇ
80–89 years	1154 (35.8)	652 (58.3)	502 (23.9)
90–99 years	1107 (34.4)	334 (29.8)	773 (36.8)
≥ 100 years	961 (29.8)	133 (11.9)	828 (39.4)
Sex, n (%)				< 0.001ᵃ
Male	1152 (35.8)	563 (50.3)	589 (28.0)
Female	2070 (64.2)	556 (49.7)	1514 (72.0)
BMI (kg/m²), median (P25 P75), n (%)	20.05 (17.86 22.52)	20.54 (18.52 23.33)	19.84 (17.78 22.22)	< 0.001ᵃᵇ
< 18.5	1003 (31.1)	276 (24.7)	723 (34.4)
18.5–23.9	1715 (53.2)	635 (56.7)	1084 (51.5)
24.0–27.9	390 (12.1)	162 (14.5)	228 (10.8)
≥ 28.0	114 (3.5)	46 (4.1)	68 (3.2)
Smoking, n (%)				< 0.001ᵃ
Never	2578 (80.0)	835 (74.6)	1743 (82.9)
Current or former	644 (20.0)	284 (25.4)	360 (17.1)
Alcohol consumption, n (%)				< 0.001ᵃ
Never	2637 (81.8)	859 (76.8)	1778 (84.5)
Current or former	585 (18.2)	260 (23.2)	325 (15.5)
Physical activity, n (%)				< 0.001ᵃ
Never	2667 (82.8)	864 (77.2)	1803 (85.7)
Current or former	555 (17.2)	255 (22.8)	300 (14.3)
Annual household income, n (%)				0.567ᵃ
< 10 000 CNY	1398 (43.4)	470 (42.0)	928 (44.1)
10 000–29 999 CNY	958 (29.7)	341 (30.5)	617 (29.3)
30 000–49 999 CNY	438 (13.6)	150 (13.4)	288 (13.7)
≥ 50 000 CNY	428 (13.3)	158 (14.1)	270 (12.8)
Marital status, n (%)				< 0.001ᵃ
Married	664 (20.6)	357 (31.9)	307 (14.6)
Widowed	2475 (76.8)	726 (64.9)	1749 (83.2)
Divorced or unmarried	83 (2.6)	36 (3.2)	47 (2.2)
Residence, n (%)				0.107ᵃ
Urban	117 (3.6)	41 (3.7)	76 (3.6)
Town	494 (15.4)	192 (17.1)	302 (14.4)
Rural	2611 (81.0)	886 (79.2)	1725 (82.0)

Notes: ᵃχ² test, ᵇMann–Whitney U test, BMI = body mass index.

Hematological and biochemical markers by frailty status and inter-marker correlations

The violin plots in Fig. 2 and Fig. S1 illustrate the distributions of biochemical and hematological indicators by frailty status. Compared with non-frail participants, those classified as frail exhibited significantly higher median values for PLT (191.00 × 10^9/L vs. 185.00 × 10^9/L), GLU (4.81 mmol/L vs. 4.78 mmol/L), BUN (6.79 mmol/L vs. 6.59 mmol/L), hs-CRP (1.76 mg/L vs. 1.30 mg/L), and MDA (4.81 nmol/L vs. 4.62 nmol/L). In contrast, frail individuals had significantly lower median values for WBC (5.40 × 10^9/L vs. 5.60 × 10^9/L), RBC (4.06 × 10^12/L vs. 4.18 × 10^12/L), HGB (121.00 g/L vs. 122.00 g/L), and CREA (76.30 µmol/L vs. 78.00 µmol/L). No significant differences were observed between the two groups in terms of CHOL, TG, or HDL-C levels. Spearman correlation analysis (Fig. 3) revealed that the correlations between markers ranged from − 0.21 to 0.63. The strongest positive correlation was found between RBC and HGB (r = 0.63, P < 0.01), while the most significant negative correlation was observed between HDL-C and TG (r = − 0.21, P < 0.01). These findings guided the evaluation of multicollinearity and subsequent modeling strategies.

Fig. 3.

Spearman correlation analysis among biomedical indicators. Note: The blue color represents positive correlation, and red color represents negative correlation. The darker the color, the greater the correlation coefficient. *P < 0.05, **P < 0.01. 

Variable selection based on univariable GLMM and LASSO logistic regression

Univariable GLMMs with individual ID as a random effect revealed that older age groups, female sex never smoking, never drinking, never exercising, and widowed status were all associated with a higher likelihood of frailty, as were. In contrast, higher BMI was associated with a lower likelihood of frailty (Table S2). Among the biomedical markers examined, higher levels of WBC, RBC, HGB, CREA, and ALB were linked to a lower likelihood of frailty, while higher levels of PLT, HDL-C, GLU, BUN, hs-CRP, and MDA were associated with a higher likelihood of frailty. Variables with P < 0.05 from these models were further included in subsequent LASSO selection. Ten-fold cross-validation of the LASSO regression model yielded a minimum classification error of approximately 0.30 when log(λ)≈–4 (Fig. 4A). Three key markers including CREA, ALB, and MDA showed a slightly higher error but greater simplicity. The S-shaped error curve indicated effective regularization and stable performance. As shown in Fig. 4B, the ROC curve based on the LASSO model achieved an area under the curve (AUC) of 0.746 (95% CI: 0.729–0.764), indicating the moderate discrimination capacity of the selected biomarkers within this study populations.

Fig. 4.

Selection of the optimal λ (A) and ROC curve (B) of the LASSO regression model Associations of key makers with frailty in multivariable GLMM.

The results of multivariable GLMM were shown in Table 2. Higher plasma CREA levels were associated with a lower likelihood of frailty (P = 0.009). Plasma ALB was inversely associated with likelihood of frailty (P = 0.013), whereas higher MDA levels were positively associated with likelihood of frailty (P < 0.001). Values of VIFs were all below 2, indicating no substantial multicollinearity within the final model.

Table 2.

Associations of key markers with frailty in the oldest-old based on multivariable GLMM.

Variables	Model 1			Model 2
	β	OR(95%CI)	P value		β	OR(95%CI)	P value
CREA	−0.12	0.89(0.82, 0.97)	0.009	CREA_low	−0.18	0.84(0.76, 0.91)	< 0.001
				CREA_high	0.04	1.04(0.95, 1.14)	0.376
MDA	0.21	1.23(1.13, 1.35)	< 0.001	MDA	0.20	1.22(1.11, 1.34)	< 0.001
ALB	−0.11	0.90(0.82, 0.98)	0.013	ALB	−0.10	0.90(0.83, 0.99)	0.015

Note: CREA_low: CREA level is below 122 µmol/L. CREA_high: CREA level is greater than or equal 122 µmol/L. Model 1, the association of key markers with frailty in the oldest-old was examined using a multivariable GLMM, adjustment factors included participant ID (random effect), age, sex, BMI, smoking, alcohol consumption, physical activity and marital status. Model 2, Association of key biomedical indicators with frailty in the oldest-old after CREA stratification based on multivariable GLMM, adjustment factors included participant ID (random effect), age, sex, BMI, smoking, alcohol consumption, physical activity and marital status.

Nonlinear relationships between key makers and frailty

RCS analysis revealed an L-shaped relationship between CREA and frailty (nonlinear P < 0.01), with an inflection point occurring at approximately 122 µmol/L. Frailty likelihood decreased as CREA increased up to 122 µmol/L, and then showed a slight increase at higher levels (Fig. 5). By contrast, the spline curves for ALB and MDA showed no evidence of nonlinearity (nonlinear P = 0.50 and P = 0.21, respectively), supporting approximately linear associations.

Fig. 5.

Nonlinear associations between key makers and frailty. Note: Nonlinear associations between CREA, MDA, ALB and frailty likelihood modeled by RCS. Areas between dashed lines represent 95% CIs derived from 1000 bootstrap samples. The vertical dashed line at CREA = 122 µmol/L marks the reference point used in the spline model and corresponds to the location of the observed inflection point in the CREA–frailty association, serving as a statistical reference rather than a clinical cutoff.

Interaction and stratified analyses

Stratified analysis by CREA levels

When CREA was dichotomized at 122 µmol/L and reintroduced into the GLMM, the results are summarized in Table 2. CREA below 122 µmol/L was associated with a lower likelihood of frailty (β = − 0.18, OR = 0.84, 95% CI: 0.76–0.91, P < 0.001), whereas CREA at or above 122 µmol/L showed no statistically significant association with frailty (β = 0.04, OR = 1.04, 95% CI: 0.95–1.14, P = 0.376). In the same model, MDA remained positively associated with frailty (β = 0.20, OR = 1.22, 95% CI: 1.11–1.34, P < 0.001), whereas ALB showed an inverse association (β = − 0.10, OR = 0.90, 95% CI: 0.83–0.99, P = 0.015). These findings were consistent with the nonlinear pattern observed for CREA in the spline analysis. MDA and ALB remained in the same direction as in the main model.

Interaction and stratified analysis by age and sex

No statistically significant interactions were detected in the GLMM (all P > 0.05), the following subgroup analyses are presented descriptively. As depicted in Fig. 6, in age-stratified analyses, CREA < 122 µmol/L was associated with a lower likelihood of frailty in individuals aged 90–99 (β = − 0.012, OR = 0.987, 95% CI: 0.980–0.994, P < 0.001) and ≥ 100 years (β = − 0.016, OR = 0.984, 95% CI: 0.978–0.991, P < 0.001), but not in individuals aged 80–89 years (P = 0.702). CREA ≥ 122 µmol/L was associated with a higher likelihood of frailty in the 90–99 age group (β = 0.014, OR = 1.014, 95% CI: 1.002–1.026, P = 0.025), but non-significant in other groups. MDA was positively associated with frailty across all age groups, whereas an inverse association for ALB was observed only among centenarians. Sex-stratified estimates showed a similar trend in males, but non-significant effects among females.

Fig. 6.

Age- and sex-stratified associations between key markers and frailty in the oldest-old. Note: CREA_low: CREA level is below 122 µmol/L. CREA_high: CREA level is greater than or equal 122 µmol/L. Adjustment factors included participant ID (random effect), age, sex, BMI, smoking, alcohol consumption, physical activity and marital status (excepting stratification factor). *P < 0.05, **P < 0.01, ***P < 0.001.

Sensitivity analyses

The robustness of the main findings was assessed through a series of sensitivity analyses. As shown in Table S3, the results of GLMM based on multiple imputation for missing data were consistent in direction, magnitude, and significance with those from the complete-case analysis. Re-fitting the models using GAMM with smoothing splines yielded curve shapes for the key indicators that closely resembled the nonlinear patterns identified by RCS in the GLMM (Fig. S2). GEEs with exchangeable and AR1 working correlation structure produced estimates of similar direction and significance for ALB, MDA, and CREA (Table S4). In addition, when the FI was analyzed as a continuous outcome using a linear mixed model, the associations of ALB (β = −0.09, P < 0.001), MDA (β = 0.05, P = 0.002), and CREA (β = −0.06, P < 0.001) remained consistent with those observed in the dichotomous frailty analyses (Table S5). Overall, these sensitivity analyses demonstrated the stability of the observed associations between selected indicators and frailty in the oldest-old.

Discussion

In this study, we identified associations between circulating hematological and biochemical indicators and frailty among Chinese adults aged 80 years or older. Higher levels of plasma MDA and lower ALB levels were linearly associated with a higher likelihood of frailty. Notably, CREA exhibited a nonlinear, L-shaped association with frailty, with a statistical inflection point at approximately 122 µmol/L. These findings suggest that nutritional status, muscle-related physiology, and oxidative stress may be linked to frailty in advanced age.

Given the multidimensional nature of frailty, routinely measured biomarkers such as ALB, CREA, and HGB are more likely to reflect overlapping physiological correlates of frailty-related processes than fully independent predictors. ALB is a well-known indicator of nutritional status and systemic inflammation. Consistent with previous literature showing that hypoalbuminemia increases the likelihood of frailty^21–25, we found that higher ALB levels were associated with a lower likelihood of frailty. Notably, this inverse association was evident only among the oldest subgroup (≥ 100 years) in stratified models, which may reflect survival bias or selective retention of more robust individuals with distinct nutritional or metabolic profiles at extreme ages. We also observed higher hs-CRP and lower HGB levels among frail individuals, aligning with existing meta-analytic evidence linking inflammation to frailty²³. Similarly, MDA is a well-recognized end-product of lipid peroxidation and a commonly used indicator of systemic oxidative stress²⁶. In the present study, higher MDA levels were observed among frail individuals across all age brackets and both sexes. This aligns with previous reports²⁷ and supports the broader view that increased oxidative stress is closely linked to the frail phenotype and may reflect accelerated biological aging processes^8,27.

In contrast to the largely linear associations for ALB and MDA, we identified a non-linear, L-shaped relationship between plasma CREA levels and frailty likelihood. Although the CREA inflection point of 122 µmol/L is derived from cross-sectional data and serves as a model-derived statistical reference rather than a definitive clinical cut-off, it provides highly informative insights into the physiological complexities of the extreme elderly. Currently, there is a lack of established age-specific normative CREA values tailored for this population. In this context, this statistical threshold conceptually highlights two distinct health trajectories. On one hand, CREA levels below this inflection point are inversely associated with likelihood of frailty, likely reflecting the profound impact of age-related muscle depletion, sarcopenia, and malnutrition²⁸. On the other hand, levels above this point may be increasingly driven by pathological kidney aging, vascular diseases, or chronic kidney disease (CKD)²⁹.

Stratified analyses indicated that some variations in these biomarker associations by age and sex, such as the significance of CREA levels differing between men and women, consistent with previous literature³⁰. However, due to formal tests for interactions in the mixed models did not reach statistical significance, these subgroup-specific estimates should be interpreted descriptively. These findings highlight the heterogeneity of the oldest-old population while acknowledging limited statistical power for detecting definitive effect modifications.

Several limitations should be acknowledged. First, the predominantly cross-sectional observational design precludes causal inference. Although extensive covariate adjustment and sensitivity analyses were performed, residual confounding and reverse causality cannot be fully excluded. Second, the study population comprised Chinese individuals aged 80 years or over who had survived to advanced age. Therefore, the findings may not be generalizable to younger populations or individuals of other ethnicities. Third, while a broad range of routinely measured blood biomarkers was examined, direct assessments of muscle mass or specific metabolic markers were not available, limiting further mechanistic insights. Future longitudinal studies with repeated biomarker measurements are warranted to validate whether the observed statistical thresholds, particularly for CREA, can be translated into practical clinical benchmarks. Despite these limitations, our findings provide robust population-based evidence that lower ALB and CREA levels, as well as high MDA levels, are intricately associated with frailty among the oldest-old. These results may help guide future research aimed at better characterizing frailty-related biological processes in ageing populations.

Conclusion

In this study of Chinese adults aged ≥ 80 years, higher plasma ALB and CREA (below the statistical inflection point of approximately 122 µmol/L) levels were independently associated with a lower likelihood of frailty, whereas higher MDA concentrations were linked to a higher likelihood of frailty. RCS analyses revealed a nonlinear, L-shaped relationship between CREA and frailty, while ALB and MDA exhibited largely linear associations. Sensitivity analyses yielded consistent results, supporting the robustness of the observed associations. Future longitudinal studies with repeated measurements and broader phenotyping are warranted to further elucidate the temporal relationships and biological underpinnings of these associations across diverse aging populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CLHLS

Chinese Longitudinal Healthy Longevity Survey

FI

frailty index

ALB

albumin

CREA

creatinine

MDA

malondialdehyde

WBC

white blood cell count

RBC

red blood cell count

HGB

hemoglobin

PLT

platelet count

CHOL

total cholesterol

TG

triglyceride

HDL-C

high-density lipoprotein cholesterol

GSP

glycated serum protein

GLU

fasting blood glucose

BUN

blood urea nitrogen

hs-CRP

high-sensitivity C-reactive protein

BMI

body mass index

LMM

linear mixed model

GLMM

generalized linear mixed model

GAMM

generalized additive mixed model

GEE

generalized estimating equation

LASSO

least absolute shrinkage and selection operator

ROC

receiver operating characteristic

VIF

Variance inflation factors

MICE

multiple imputation by chained equation

AUC

area under the curve

eGFR

estimated glomerular filtration rate

Author contributions

LHX and CML designed the study. QHJ, WHG and TTY extracted and checked the data, and evaluated the frailty of the subjects. LHX and QHJ analyzed the data. LHX and CML contributed to the writing of original manuscript. BJZ, LW, PY, XLC and CML interpreted the results and revised the manuscript. All authors read and approved the final version. The authors gratefully acknowledge the research staff for their essential contributions.

Funding

This research was supported by the Jiangxi Provincial Natural Science Foundation (No. 20232BAB216104), National Natural Science Foundation of China (No. 82460636), and Nanchang University Research Start up Fund (No. 9167-28770619).

Data availability

All data utilized in this study were derived from Peking University Open Research Data Platform (https://doi.org/10.18170/DVN/FWVGN5).

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and the ethical approval was obtained from the Research Ethics Committee of Peking University (IRB0001052–13074). Written informed consent was obtained from all participants or their legally authorized representatives.