Kidney function and all-cause mortality in the oldest-old population: a three-cohort longitudinal study

Abstract

With increasing global life expectancy, the prognostic significance of kidney function in centenarians and supercentenarians should be clarified. The conventional estimated glomerular filtration rate (eGFR) threshold of < 60 mL/min/1.73 m² may not accurately reflect chronic kidney disease (CKD) in this population, highlighting the need for age-adapted definitions. This study investigated the association of kidney function with all-cause mortality across three age groups. Data from the Tokyo Centenarian Study, Japan Semi-supercentenarian Study, and Kawasaki Ageing and Wellbeing Project were analyzed in 1,918 participants with baseline serum creatinine measurements were included (median follow-up: 1,399 days). eGFR was calculated using creatinine and cystatin C according to the Japanese Society of Nephrology equation. Associations with mortality were assessed using Cox regression models, and standardized 3-year absolute risks were estimated. Lower cystatin C-based eGFR was consistently associated with higher mortality, whereas creatinine-based eGFR showed weaker associations. Prognostic thresholds for eGFR declined with advancing age—approximately 45, 30, and 15 mL/min/1.73 m² at ages 85–89, 100–104, and ≥ 105 years, respectively. The 3-year absolute risk difference (ARD) between the low and high eGFR groups was modest (≈10%) at ages 85–89 years, larger (≈30%) at 100–104 years, and attenuated at ≥ 105 years, where the 3-year mortality exceeded 80%. eGFR remained a prognostically relevant marker in very old individuals; however, the magnitude of ARDs varied by age—modest in the younger-old adults, larger in centenarians, and attenuated in semi- and supercentenarians—supporting the adoption of age-specific CKD thresholds.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-026-02130-0.

Introduction

Age-related renal structural and functional changes and the increasing prevalence of metabolic disorders contribute to kidney function decline [1–4]. In older adults, an estimated glomerular filtration rate (eGFR) cutoff of < 45 mL/min/1.73 m² has been proposed for chronic kidney disease (CKD) diagnosis, rather than the conventional 45–59 mL/min/1.73 m² range, owing to its greater prognostic significance [1, 5]. Cystatin C (Cys)-based eGFR (eGFR_Cys) is recommended, as creatinine (Cre)-based eGFR (eGFR_Cre) is influenced by age-related reductions in muscle mass [6–8]. However, eGFR_Cys may overestimate CKD prevalence owing to the effects of inflammation and elevated cortisol levels [9, 10]. Thus, the optimal biomarker for eGFR assessment in the oldest-old population remains uncertain [11, 12].

Although relationships among eGFR, urinary albumin-to-creatinine ratio (UACR), and mortality have been reported, individuals aged ≥ 85 years were rarely included [13, 14]. While several studies have investigated eGFR levels in centenarian populations, their sample sizes have been small, and the findings varied substantially by country, clinical setting, and estimation equation [15–17]. Moreover, studies evaluating the association between eGFR and mortality for this age range have been mostly small-scale and lacked corresponding eGFR_Cys estimates, limiting the interpretation of their null findings [18, 19].

To date, no study has comprehensively examined the associations of eGFR_Cre and eGFR_Cys with all-cause mortality in the oldest-old population. In particular, the prognostic thresholds of kidney function have not been compared across the old-old population, centenarians, and individuals aged ≥ 105 years, including semi-supercentenarians and supercentenarians. Previous studies have primarily focused on relative risks, which may be less informative in populations with very high baseline mortality. Evaluation of absolute risk differences (ARDs) may provide a more clinically meaningful understanding of the prognostic impact of kidney function.

Therefore, this study aimed to investigate the associations of eGFR_Cre and eGFR_Cys with all-cause mortality across these three age groups and assess 1- and 3-year ARDs to explore age-related shifts in prognostic thresholds and their clinical relevance.

Materials and Methods

Study design

This prognostic study utilized data from three prospective cohort studies of Japanese individuals aged 85–115 years at study enrollment. This study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines [20]. Ethical approval for the use of all three datasets was granted by the Ethics Committee of Keio University School of Medicine (20,021,020, 20,022,020, and 20,160,297). The datasets were registered as observational studies in the University Hospital Medical Information Network Clinical Trials Registry (UMIN000040446, UMIN000040447, and UMIN000026053).

Settings

Data were obtained from the Tokyo Centenarian Study (TCS; September 2000–July 2018) [21], Japan Semi-supercentenarian Study (JSS; September 2002–January 2023) [22], and Kawasaki Aging and Wellbeing Project (KAWP; March 2017–December 2018, with follow-up data collected through December 2021) (Online Resource 1) [23].

Participants

The TCS enrolled 304 centenarians (78.6% women) between July 2000 and May 2002. After excluding 16 participants without serum Cre measurements, only 288 were analyzed. The JSS recruited 663 centenarians (88.1% women) between September 2002 and March 2019, of whom 92.2% were semi-supercentenarians (aged ≥ 105 years). After excluding 54 participants without serum Cre measurements and 1 with insufficient age verification, only 608 participants were analyzed. The KAWP enrolled 1,026 octogenarians (50.0% women) between March 2017 and December 2018. Following the exclusion of four participants without serum creatinine measurements, only 1,022 participants were analyzed. All participants provided written informed consent. For those lacking decision-making capacity, consent was obtained from an authorized representative.

Serum Cre data were available for all included participants. However, because serum Cys measurement was not mandatory at study enrollment, data for the purpose of biobanking could not be obtained for all participants. Consequently, fewer participants were included in eGFR_Cys-based analyses than in eGFR_Cre-based analyses.

Study exposure

The primary exposure variables were baseline eGFR_Cys and eGFR_Cre. Both equations, developed by the Japanese Society of Nephrology (JSN) incorporate age, sex, and filtration markers, but do not include race-based adjustments [24, 25].

The equations are as follows:

$$\text{eGFR}\_\text{JSNCys}\left(\text{mL},/,\text{min},/,1.73,{\text{m}}^2\right)=\text{K}\times 104\times \text{serum Cys}{\left(\text{mg},/,\text{L}\right)}^{-1.019}\times {0.996}^{\text{Age}}-8$$

K = 1 for male participants and 0.929 for female participants

$$\text{eGFR}\_\text{JSNCr}\left(\text{mL},/,\text{min},/,1.73,{\text{m}}^2\right)=\text{K}\times 194\times \text{serum Cre}{\left(\text{mg},/,\text{dL}\right)}^{-1.094}\times {\text{age}}^{-0.287}$$

K = 1 for male participants and 0.739 for female participants.

Cys levels were measured using a colloidal gold immunoassay, whereas Cre levels were determined using an enzymatic method across all cohorts. In participants aged 85–89 years (KAWP cohort), urinary microalbumin was measured using an immunoturbidimetric assay, whereas the UACR was calculated and categorized using a cutoff value of 30 mg/gCr. Urine samples were not collected from individuals aged ≥ 100 years.

Covariates

Age, sex, educational attainment, smoking status, hypertension, dyslipidemia, diabetes mellitus, cardiovascular disease, use of renin–angiotensin–aldosterone system inhibitors, and C-reactive protein levels were included as covariates (Online Resource 2).

Outcomes

All-cause mortality was determined through annual telephone interviews and mailed surveys in the TCS and JSS cohorts, whereas long-term care insurance records were used to determine mortality in the KAWP cohort [26].

Statistical analysis

The participants’ baseline characteristics were summarized according to age group (85–89, 100–104, 105–109, and ≥ 110 years). Descriptive statistics were expressed as absolute counts with corresponding relative frequencies for categorical variables and as arithmetic means with standard deviations for continuous variables. The participants were stratified into two groups based on eGFR_JSNCys or eGFR_JSNCre using a cutoff of 45 mL/min/1.73 m². The prevalence of CKD stages was reported for each age group. Simultaneously, eGFR was examined using both a six-category classification at 15-mL/min/1.73 m² intervals and dichotomous classification based on the 45-mL/min/1.73 m² cutoff. The adjusted hazard ratios (aHRs) with 95% confidence intervals (CIs) were estimated using Cox proportional hazards models (Cox models), adjusted for the aforementioned covariates, to evaluate the association between eGFR and all-cause mortality. Time to event was defined as the interval from study enrollment to death. The 60–75 mL/min/1.73 m² and 30–45 mL/min/1.73 m² eGFR_JSNCys categories served as the references for participants aged 85–89 years and ≥ 100 years, respectively [1, 27, 28]. The UACR was classified using the cutoff values of 30 and 300 mg/gCr, and it was available only for participants aged 85–89 years. eGFR_JSNCys was included as a covariate along with other specified covariates. When categorical eGFR and UACR values were analyzed together, participants were classified into six groups using the cutoff values of 45 and 60 mL/min/1.73 m² for eGFR and 30 mg/gCr for UACR, considering statistical power. Linear and restricted cubic spline (RCS) analyses with three knots placed at the 10th, 50th, and 90th percentiles were conducted to evaluate the associations among continuous eGFR, UACR, and all-cause mortality. Cox models adjusted for relevant covariates were also used to estimate the standardized 1-year and 3-year absolute risks of all-cause mortality for each eGFR_JSNCre and eGFR_JSNCys category within each age group. For each category, the predicted survival probability at the specified time point was calculated by fixing eGFR_JSNCys at the category median and averaging across the observed covariate distribution. Absolute risk was defined as 1 minus the predicted survival probability, and 95% CIs were derived from 1,000 bootstrap resamples. Owing to the limited sample size, participants aged ≥ 110 years were included in the ≥ 105-year-old group for analysis. All statistical analyses were performed using R (version 4.3.3; R Foundation for Statistical Computing), with statistical significance defined as a two-sided p value of < 0.05.

Results

Participant characteristics

Data were collected from 1,918 participants (85–89 years: 1,022; 100–104 years: 288; 105–109 years: 572; and ≥ 110 years: 36). The baseline characteristics of the 1,818 participants, stratified by eGFR_JSNCys, are presented in Table 1. The proportions of participants with an eGFR_JSNCys of < 45 mL/min/1.73 m² increased from 18.9% in the 85–89-year-old group to 80.4% in the 100–104-year-old group, 89.7% in the 105–109-year-old group, and 93.5% in the ≥ 110-year-old group. In the 85–89-year-old group, 34.8% of participants with an eGFR_JSNCys of < 45 mL/min/1.73 m² and 15.7% with an eGFR_JSNCys of ≥ 45 mL/min/1.73 m² had a UACR of ≥ 30 mg/gCr. Meanwhile, 16.9% of the participants in the 85–89-year-old group, 35.1% in the 100–104-year-old group, 37.2% in the 105–109-year-old group, and 30.6% in the ≥ 110-year-old group exhibited an eGFR_JSNCre of < 45 mL/min/1.73 m² (Online Resource 3).

Table 1.

Baseline characteristics of participants stratified by age at enrollment and eGFR_JSNCys

JSN_Cys	85–89 years				100–104 years				105–109 years				≥ 110 years
	N	< 45	N	≥ 45	N	< 45	N	≥ 45	N	< 45	N	≥ 45	N	< 45	N	≥ 45
Age, years (SD)	189	86.9 ± 1.44	811	86.5 ± 1.35	213	101.6 ± 1.60	52	101.1 ± 1.37	468	106.7 ± 1.14	54	106.7 ± 1.10	29	110.8 ± 0.94	2	111.3 ± 0.54
Female, n (%)	189	111 (58.7)	811	418 (51.5)	213	170 (79.8)	52	36 (69.2)	468	413 (88.2)	54	47 (87.0)	29	27 (93.1)	2	2 (100)
Education (high school), n (%)	188	61 (32.4)	808	251 (31.1)	205	42 (20.5)	50	13 (26.0)	442	50 (11.3)	53	3 (5.66)	27	1 (3.70)	2	1 (50.0)
Smoking (current), n (%)	187	8 (4.28)	809	34 (4.20)	212	3 (1.42)	50	0 (0.00)	461	7 (1.52)	54	0 (0.00)	28	1 (3.57)	2	0 (0.00)
Body mass index, kg/m2 (SD)	189	24.1 ± 2.99	811	23.0 ± 3.12	132	19.8 ± 3.12	40	18.6 ± 3.27	298	19.5 ± 3.28	32	18.2 ± 2.76	18	18.6 ± 2.98	2	17.7 ± 4.53
Hypertension, n (%)	189	153 (81.0)	811	621 (76.6)	210	117 (55.7)	52	28 (53.8)	457	178 (38.9)	54	23 (42.6)	29	9 (31.0)	2	1 (50.0)
Diabetes mellitus, n (%)	189	39 (20.6)	811	141 (17.4)	213	14 (6.57)	52	4 (7.69)	468	26 (5.56)	54	4 (7.41)	29	2 (6.90)	2	0 (0.00)
Dyslipidemia, n (%)	189	111 (58.7)	811	424 (52.3)	213	27 (12.7)	52	10 (19.2)	468	73 (15.6)	54	5 (9.26)	29	6 (20.7)	2	1 (50.0)
Cardiovascular disease, n (%)	189	102 (54.0)	811	291 (35.9)	207	77 (37.2)	51	13 (25.5)	461	197 (42.7)	52	14 (26.9)	26	10 (38.5)	2	0 (0.00)
RAAS inhibitors, n (%)	189	103 (54.5)	811	281 (34.6)	207	22 (10.6)	51	2 (3.92)	461	58 (12.6)	52	6 (11.5)	26	6 (23.1)	2	0 (0.00)
C-reactive protein, n (%)	189	41 (21.7)	811	80 (9.86)	213	71 (33.3)	52	18 (34.6)	468	206 (44.0)	54	21 (38.9)	29	16 (55.2)	2	2 (100)
JSN_Cre, mL/min/1.73 m2 (SD)	189	41.3 ± 9.43	811	62.9 ± 122	213	49.5 ± 18.4	52	80.0 ± 23.8	468	51.4 ± 20.7	54	92.9 ± 33.0	29	53.2 ± 23.0	2	74.9 ± 13.9
JSN_Cys, mL/min/1.73 m2 (SD)	189	36.8 ± 6.43	811	61.5 ± 10.6	213	31.2 ± 9.10	52	54.0 ± 7.45	468	27.8 ± 8.77	54	49.9 ± 4.55	29	27.4 ± 9.60	2	47.6 ± 2.36
UACR, n (%)	184	64 (34.8)	802	126 (15.7)

C-reactive protein and UACR were analyzed as categorical variables using cutoff values of 0.3 mg/dL and 30 mg/gCr, respectively

Prevalence of chronic kidney disease by eGFR_JSNCre and eGFR_JSNCys

Figure 1 illustrates the CKD classification based on eGFR_JSNCre and eGFR_JSNCys levels. In the 85–89-year-old group, CKD staging based on eGFR_JSNCre was largely consistent with that based on eGFR_JSNCys. In the 100–104-year-old and ≥ 105-year-old groups, stages 3b and 4 accounted for more than 75% of patients when classified according to eGFR_JSNCys. By contrast, classification based on eGFR_JSNCre demonstrated a more balanced distribution, with stages 2–3b comprising approximately 80% of the participants. Notably, the proportion of participants with stage 1 CKD increased with advancing age when using eGFR_JSNCre for classification.

Fig. 1.

Age-specific distribution of CKD stages according to eGFR calculated using the JSN equation. The blue bars represent eGFR calculated using the JSN_Cre equation, whereas the green bars represent eGFR calculated using the JSN_Cys equation. Abbreviations: CKD chronic kidney disease, eGFR estimated glomerular filtration rate, JSN_Cre estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with creatinine, JSN_Cys estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with cystatin C, JSN equation the Japanese Society of Nephrology equation

Participant follow-up and mortality outcomes

The median follow-up period for all participants was 1,399 days (with subgroup medians of 1,567 days for the 85–89-year-old group, 742 days for the 100–104-year-old group, and 477 days for the ≥ 105-year-old group). During follow-up, a total of 1,048 participants (54.6%) died: 168 (16.4%) in the 85–89-year-old group, 281 (97.6%) in the 100–104-year-old group, and 599 (98.5%) in the ≥ 105-year-old group. When classified by CKD stage based on eGFR_JSNCre, the incidence rates of all-cause mortality per 100 person-years in the 85–89-year-old group were 7.8 for stage 1, 4.3 for stage 2, 3.4 for stage 3a, 6.5 for stage 3b, and 13.5 for stage 4. In the 100–104-year-old group, the corresponding rates were 43.3, 33.7, 32.4, 39.6, 40.0, and 54.2, respectively. Among participants aged ≥ 105 years, the incidence rates were 81.5, 61.3, 54.2, 50.9, 65.7, and 62.9 across stages 1 to 5, respectively. When classified by CKD stage based on eGFR_JSNCys, the incidence rates in the 85–89-year-old group were 7.4, 3.4, 4.2, 7.5, 9.0, and 28.0 across stages 1 to 5, respectively. In the 100–104-year-old group, the incidence rates were 31.6, 32.4, 44.5, and 52.0 for stages 3b, 3a, 4, and 5, respectively. In the ≥ 105-year-old group, the incidence rates were 580.0, 59.3, 52.5, 58.0, and 82.8 for stages 2 to 5, respectively.

Association of estimated glomerular filtration rate and urinary albumin-to-creatinine ratio with all-cause mortality using eGFR_JSNCre and eGFR_JSNCys as categorical variables

Table 2 and Online Resource 4 present the aHRs for categorical eGFR_JSNCre, eGFR_JSNCys, and their combinations with UACR. Using a threshold of 45 mL/min/1.73 m², an eGFR_JSNCre of < 45 mL/min/1.73 m² was associated with increased mortality only in the 85–89-year-old group (aHR: 1.81, 95% CI: 1.30–2.54). By contrast, an eGFR_JSNCys of < 45 mL/min/1.73 m² was associated with increased mortality in the overall cohort (aHR: 1.29, 95% CI: 1.06–1.56]) and 85–89-year-old group (aHR: 1.72, 95% CI: 1.24–2.38). When eGFR_JSNCys was categorized in 15-mL/min/1.73 m² increments, aHRs for all-cause mortality in the 85–89-year-old group were 1.98 (95% CI: 1.25–3.02) for 30–45 mL/min/1.73 m² and 2.60 (95% CI: 1.26–5.38) for 15–30 mL/min/1.73 m². In the 100–104-year-old group, the corresponding aHRs were 1.38 (95% CI: 1.03–1.91) for 15–30 mL/min/1.73 m² and 1.91 (95% CI: 1.00–3.70) for < 15 mL/min/1.73 m². In the ≥ 105-year-old group, the aHR was 1.49 (95% CI: 1.03–2.17) for < 15 mL/min/1.73 m². In the 85–89-year-old group, the association between an eGFR_JSNCre of < 45 mL/min/1.73 m² and all-cause mortality was significant only among participants with a UACR of ≥ 30 mg/gCr (aHR: 2.93, 95% CI: 1.72–5.02) and not among those with lower UACR levels (aHR: 1.46, 95% CI: 0.89–2.39). Conversely, an eGFR_JSNCys of < 45 mL/min/1.73 m² was associated with increased all-cause mortality regardless of the UACR level (UACR < 30 = aHR: 1.83, 95% CI: 1.11–3.01; UACR ≥ 30 = aHR: 2.96, 95% CI: 1.69–5.17). Mortality was also associated with UACR categories of 30 and 300 mg/gCr (Online Resource 5).

Table 2.

Adjusted hazard ratios for mortality according to eGFR, UACR, and combined categories across age groups

Category	Entire	85–89 years	100–104 years	≥ 105 years
JSN_Cre
eGFR
≥ 45	Ref	Ref	Ref	Ref
< 45	1.06 (0.92–1.22)	1.81 (1.30–2.54)	1.17 (0.88–1.53)	0.83 (0.69–1.01)
eGFR
≥ 75	1.32 (1.06–1.64)	1.72 (1.10–2.69)	1.09 (0.73–1.60)	1.45 (1.11–1.91)
60–75	Ref	Ref	0.67 (0.42–1.06)	1.33 (1.02–1.74)
45–60	1.02 (0.84–1.24)	0.88 (0.60–1.29)	0.90 (0.64–1.22)	1.15 (0.90–1.46)
30–45	1.08 (0.88–1.33)	1.66 (1.09–2.54)	Ref	Ref
15–30	1.35 (1.02–1.78)	3.06 (1.44–5.92)	1.15 (0.65–2.04)	1.28 (0.93–1.78)
< 15	1.41 (0.74–2.71)	NA	1.47 (0.55–3.88)	1.34 (0.57–3.12)
eGFR and UACR
eGFR ≥ 60 and UACR < 30	NA	Ref	NA	NA
eGFR ≥ 60 and UACR ≥ 30	NA	1.59 (0.89–2.85)	NA	NA
eGFR45–59 and UACR < 30	NA	0.77 (0.50–1.18)	NA	NA
eGFR45–59 and UACR ≥ 30	NA	0.74 (0.35–1.58)	NA	NA
eGFR < 45 and UACR < 30	NA	1.46 (0.89–2.39)	NA	NA
eGFR < 45 and UACR ≥ 30	NA	2.93 (1.72–5.02)	NA	NA
JSN_Cys
eGFR
≥ 45	Ref	Ref	Ref	Ref
< 45	1.29 (1.06–1.56)	1.72 (1.24–2.38)	1.11 (0.78–1.58)	0.95 (0.68–1.22)
eGFR
≥ 75	1.36 (0.78–2.36)	1.58 (0.89–2.81)	NA	NA
60–75	Ref	Ref	1.80 (0.85–3.81)	NA
45–60	1.26 (0.90–1.75)	1.26 (0.85–1.87)	0.91 (0.61–1.35)	1.09 (0.79–1.50)
30–45	1.51 (1.07–2.13)	1.98 (1.25–3.02)	Ref	Ref
15–30	1.66 (1.16–2.39)	2.60 (1.26–5.38)	1.38 (1.03–1.91)	1.02 (0.82–1.25)
< 15	2.48 (1.58–3.89)	NA	1.91 (1.00–3.70)	1.49 (1.03–2.17)
eGFR and UACR
eGFR of ≥ 60 and UACR of < 30	NA	Ref	NA	NA
eGFR of ≥ 60 and UACR of ≥ 30	NA	1.56 (0.72–3.36)	NA	NA
eGFR of 45–59 and UACR of < 30	NA	1.23 (0.80–1.90)	NA	NA
eGFR of 45–59 and UACR of ≥ 30	NA	1.47 (0.77–2.80)	NA	NA
eGFR of < 45 and UACR < 30	NA	1.83 (1.11–3.01)	NA	NA
eGFR of < 45 and UACR of ≥ 30	NA	2.96 (1.69–5.17)	NA	NA

Associations among JSN_Cre, JSN_Cys, and all-cause mortality are expressed as adjusted hazard ratios (95% confidence intervals) using the following classifications: a two-category eGFR (≥ 45 or < 45 mL/min/1.73 m²), a six-category eGFR (< 15, 15–30, 30–45, 45–60, 60–75, or ≥ 75 mL/min/1.73 m²), and a three-category eGFR (≥ 60, 45–59, or < 45 mL/min/1.73 m²) combined with a two-category UACR (< 30 or ≥ 30 mg/gCr)

Association of estimated glomerular filtration rate and urinary albumin-to-creatinine ratio with all-cause mortality using eGFR_JSNCre and eGFR_JSNCys as continuous variables

Online Resource 6 and Fig. 2 illustrate the associations among continuous eGFR_JSNCre, eGFR_JSNCys, and mortality. In the linear model, each 1-unit increase in eGFR_JSNCre was associated with a higher mortality risk in the ≥ 105-year-old group (aHR: 1.006, 95% CI: 1.002–1.010). Conversely, each 1-unit increase in eGFR_JSNCys was associated with a lower mortality risk in the overall cohort (aHR: 0.987, 95% CI: 0.982–0.993), 85–89-year-old group (aHR: 0.985, 95% CI: 0.974–0.996), and 100–104-year-old group (aHR: 0.983, 95% CI: 0.972–0.995). A significant interaction was observed between eGFR_JSNCre and age in relation to mortality. RCS analysis demonstrated significant nonlinear associations between both eGFR_JSNCre and eGFR_JSNCys and all-cause mortality, with different patterns across age groups. The relationship between eGFR_JSNCre and mortality followed an inverted bell-shaped curve, with a nadir at approximately 50–60 mL/min/1.73 m² in the overall cohort, 85–89-year-old group, and 100–104-year-old group. In the ≥ 105-year-old group, mortality increased nonlinearly at an eGFR_JSNCre level of > 60 mL/min/1.73 m². By contrast, eGFR_JSNCys exhibited a nonlinear association with mortality in the 85–89-year-old and ≥ 105-year-old groups, with elevated mortality risk observed at eGFR_JSNCys levels of < 45 mL/min/1.73 m² and < 30 mL/min/1.73 m², respectively. In the 85–89-year-old group, the linear model indicated that each 10-unit increase in UACR was associated with a higher mortality risk (aHR: 1.03, 95% CI: 1.01–1.05). However, this association was not significant in the nonlinear model (Online Resource 5 and 7).

Fig. 2.

Restricted cubic spline curves of eGFR_JSNCre and eGFR_JSNCys for all-cause mortality across age groups. The restricted cubic spline curves were adjusted for age, sex, educational attainment, smoking status, hypertension, diabetes mellitus, dyslipidemia, cardiovascular disease, use of renin–angiotensin–aldosterone system inhibitors, and C-reactive protein. Knots were placed at the 10th, 50th, and 90th percentiles, with an eGFR of 45 mL/min/1.73 m² used as the reference point. Abbreviations: aHR adjusted hazard ratio, eGFR estimated glomerular filtration rate, eGFR_JSNCre estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with creatinine, eGFR_JSNCys estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with cystatin C, JSN_Creatinine estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with creatinine, JSN_Cystatin C estimated glomerular filtration rate calculated using the Japanese Society of Nephrology equation with cystatin C

Absolute risk of all-cause mortality according to the eGFR_JSNCre and eGFR_JSNCys categories

Figure 3 and Online Resource 8 present the standardized 1-year and 3-year absolute risks of all-cause mortality according to eGFR_JSNCys categories. Across all age groups, lower eGFR_JSNCys values were associated with higher mortality risks. Among participants aged 85–89 years, the 1-year absolute risk ranged from 1.30% (95% CI: 0.71–1.97; eGFR: 60–75) to 3.55% (95% CI: 1.54–7.11; eGFR: 15–30), whereas the 3-year risk ranged from 6.69% (95% CI: 4.88–8.42) to 17.2% (95% CI: 10.1–25.0). In the 100–104-year-old group, the corresponding risks were substantially higher, ranging from 15.5% (95% CI: 8.96–25.2; eGFR: 45–60) to 35.3% (95% CI: 20.1–56.3; eGFR: < 15) at 1 year and from 46.2% (95% CI: 32.4–61.6) to 78.8% (95% CI: 58.2–93.3) at 3 years. In the ≥ 105-year-old group, absolute risks were markedly elevated across all eGFR categories. One-year risks ranged from 37.8% (95% CI: 32.3–42.2; eGFR: 30–45) to 49.7% (95% CI: 38.6–66.8; eGFR: < 15), whereas 3-year risks ranged from 77.9% (95% CI: 73.4–82.8) to 88.5% (95% CI: 79.4–96.5). Across all age groups, the eGFR_JSNCre categories showed patterns similar to those observed for eGFR_JSNCys (Online Resource 9). Relative gradients in mortality risk across eGFR categories were most pronounced among participants aged 100–104 years but were less evident in the ≥ 105-year-old group, in which the absolute risks were uniformly high.

Fig. 3.

Standardized 1-year and 3-year absolute mortality risks by age group and eGFR_JSNCys category. Standardized 1-year and 3-year absolute risks of all-cause mortality are presented by age group and eGFR_JSNCys category, with 95% confidence intervals. Estimates were standardized for age, sex, educational attainment, smoking status, hypertension, diabetes mellitus, dyslipidemia, cardiovascular disease, and renal disease using the direct method, with the mean covariate distribution of the study population as the reference. Abbreviation: eGFR_JSNCys estimated glomerular filtration rate calculated with cystatin C using the Japanese Society of Nephrology equation

Discussion

This study hypothesized that the association between kidney function and mortality risk diminishes with advancing age, particularly weakening among the oldest-old population. It was further assumed that eGFR_JSNCys would demonstrate a stronger association with mortality than eGFR_JSNCre and that evaluating ARDs would provide clinically meaningful insights in this population. The observed attenuation of the eGFR–mortality relationship with advancing age suggests that uniform threshold values may be insufficient for accurate risk stratification in very old adults. Notably, eGFR_JSNCys exhibited a consistent and stronger association with mortality across all age groups compared with eGFR_JSNCre. The ARD analysis demonstrated a modest impact in participants aged 85–89 years, a more pronounced impact in centenarians, and a limited impact among semi-supercentenarians (aged ≥ 105 years), in whom the extremely high background mortality likely reduced the relative contribution of kidney function to overall risk.

A key finding of this study is that the eGFR threshold associated with increased mortality risk declined progressively with advancing age: < 45 mL/min/1.73 m² at ages 85–89 years, < 30 mL/min/1.73 m² at ages 100–104 years, and < 15 mL/min/1.73 m² at age ≥ 105 years. Although a threshold of 45 mL/min/1.73 m² has been proposed for CKD diagnosis in older adults, the present findings suggest that a uniform threshold may be inappropriate for risk application in the oldest-old population.

With regard to biomarker selection, eGFR_JSNCys demonstrated a stronger and more consistent association with mortality than eGFR_JSNCre across all age groups. Among participants aged 85–89 years, this association persisted regardless of albuminuria status. This finding contrasts with the results of previous studies in older adults (aged ≥ 75 years), which reported an association between eGFR_Cre and mortality even in the absence of albuminuria [1]. In the present study, however, the association between eGFR_JSNCre and mortality was restricted to participants with microalbuminuria. Given the potential overestimation of kidney function by Cre in the context of age-related muscle loss, Cys C may represent a more reliable marker for assessing kidney function in very old adults. Nevertheless, as cystatin C can be influenced by non-GFR determinants, such as inflammation and hormonal factors, these findings should be interpreted with caution.

The ARD analyses further underscored the age-related variation in the strength of association between kidney function and mortality. Among individuals aged 85–89 years, the 3-year ARD was approximately 10%, indicating a modest impact. By contrast, the 3-year ARD approached 30 percentage points in centenarians, suggesting that kidney function remains a significant prognostic factor in this age group. Among participants aged ≥ 105 years, the 3-year mortality exceeded 80%, and the additional prognostic contribution of kidney function was minimal, reflecting risk saturation and an increasing influence of non-renal factors [17]. These findings emphasize the need to reconsider how kidney function is assessed and incorporated into risk stratification frameworks for the oldest-old population.

Taken together, these findings indicate the limitations of applying uniform CKD thresholds, relying on a single biomarker, or basing risk assessment solely on kidney function in very old adults. Among individuals aged 85–89 years, early detection of kidney impairment using appropriate biomarkers may provide opportunities for targeted intervention. In centenarians, kidney function remains an important prognostic indicator; however, comprehensive management that accounts for comorbidities and functional vulnerability is essential. Among semi-supercentenarians, the influence of kidney function appears limited, suggesting that care strategies should prioritize functional status and quality of life over CKD-specific interventions.

Our study had some limitations. First, eGFR and UACR were classified based on a single baseline measurement [29, 30]. In addition, the causes of death were not recorded; thus, the extent to which kidney function contributed to mortality cannot be clarified [31, 32]. Moreover, frailty could not be adjusted for; thus, eGFR may have partly reflected overall frailty rather than true kidney function in the oldest-old populations. Finally, potential selection and information biases across the three cohorts should be carefully considered when interpreting these findings.

Conclusion

To our knowledge, this study is the first to comprehensively examine associations among eGFR_Cre, eGFR_Cys, and all-cause mortality across pre-centenarians, centenarians, and semi-supercentenarians. Risk thresholds for kidney function progressively declined with advancing age, and the strength of the association weakened among supercentenarians. eGFR_JSNCys demonstrated a stronger association with mortality compared with eGFR_JSNCre, suggesting its potential utility as a prognostic marker in very old populations; however, the non-GFR determinants unique to each marker should be considered during interpretation. Moreover, ARD analyses provided additional clinically relevant insights beyond relative risk estimates. With the continuing global increase in life expectancy, these findings underscore the importance of moving beyond conventional CKD assessment frameworks by incorporating age-specific risk evaluation and individualized kidney disease management strategies. Future research should focus on longitudinal assessment of kidney function trajectories, evaluation of cause-specific mortality, and the impact of comorbidities to optimize risk stratification and guide therapeutic strategies in the oldest-old population.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Yasumichi Arai, Yukiko Abe, and Nobuyoshi Hirose; Methodology: Yasumichi Arai and Yukiko Abe; Data curation: Hideaki Kurata, Shu Meguro, Yukiko Abe, Yasumichi Arai, and Nobuyoshi Hirose; Formal analysis: Hideaki Kurata and Shu Meguro; Writing – original draft: Hideaki Kurata; Writing – review & editing: Shu Meguro, Yasumichi Arai, and Kaori Hayashi; Resources: Takashi Sasaki and Yasumichi Arai; Supervision: Yasumichi Arai and Kaori Hayashi; Validation: Yasumichi Arai and Hideaki Kurata; both authors had full access to all study data and verified the integrity and accuracy of the data and analysis.

Funding

This study was supported by the Ministry of Health, Labour and Welfare for the Scientific Research Project for Longevity; Keio University Global Research Institute; Japan Society for the Promotion of Science (23H03337); JST Research Complex Program (JP15667051); Japan Agency for Medical Research and Development (17jm0210051h0001, 23zf0127007h0002, and 23ae0121041h0003); and Kanagawa Institute of Industrial Science and Technology. The funding sources had no role in the design, conduct, analysis, or reporting of this study.

Data availability

Individual-level data from the TCS, JSS, and KAWP cohorts are not publicly available due to ethical and privacy considerations. Access requires a formal data use agreement and approval from the Ethics Committee of Keio University School of Medicine. Requests should be directed to the corresponding author. Upon approval, collaborative analyses may be conducted under the supervision of Dr. Yasumichi Arai, the principal investigator, with support from the coordinating center.

Declarations

Ethical approval for the use of the three datasets—the Tokyo Centenarian Study, the Japan Semi-supercentenarian Study, and the Kawasaki Aging and Wellbeing Project—was obtained from the Ethics Committee of Keio University School of Medicine (approval numbers: 20021020, 20022020, and 20160297). All studies were conducted in accordance with the principles of the Declaration of Helsinki. These datasets were registered as observational studies in the University Hospital Medical Information Network Clinical Trials Registry (UMIN000040446, UMIN000040447, and UMIN000026053). Written informed consent to participate was obtained from all participants.

Competing interest

Yasumichi Arai has received research support from Ezaki Glico Co., Ltd.; Johonson & Johonson K.K.; and Tuning Folk Bio Japan. Hideaki Kurata, Shu Meguro, Yukiko Abe, Takashi Sasaki, Nobuyoshi Hirose, and Kaori Hayashi declare no relevant financial interests