This retrospective cohort study used de-identified data from participants who underwent a general health checkup at Samsung Changwon Hospital between January 1, 2002, and December 31, 2018. We included participants aged ≥60 years after excluding those without data on baseline body weight, height, serum albumin, and creatinine levels. Additionally, we excluded participants with a baseline eGFR <15 mL/min/1.73 m² or those without a follow-up examination on kidney function within 5 years.

Baseline information on age, sex, comorbidities (diabetes, hypertension, coronary artery disease, and stroke), current smoking status, body weight, height, systolic blood pressure (SBP), diastolic blood pressure (DBP), and laboratory variables, including hemoglobin, serum albumin, total cholesterol, serum glucose, serum creatinine, and albuminuria, was collected from the database. Albuminuria was defined as a urine protein dipstick test reading of ≥1.

This study was approved by the Institutional Review Board of Samsung Changwon Hospital (IRB No. SCMC 2022-01-008), which waived the requirement for informed consent from the participants because our study only retrospectively accessed a de-identified dataset based on the health screening cohort of the epidemiological research center at Samsung Changwon Hospital for analysis purposes.

This study complied the ethical guidelines for authorship and publishing in the Annals of Geriatric Medicine and Research.²¹⁾

The primary exposure measure in this study was the GNRI, which was divided into quartiles. The primary outcome was a change in kidney function 5 years after baseline, which was assessed using the 5-year slope of eGFR. The secondary outcome was all-cause mortality 5 years after baseline. The GNRI was calculated using the following formula: 14.89 × serum albumin (g/dL) + 41.7 × (actual body weight [kg]) / (ideal body weight [kg]). The ideal body weight was calculated using the following formula: (height [m])² × 22 (kg/m²). The eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation.²²⁾ A negative slope indicates a decline in eGFR.

Continuous and categorical variables for baseline characteristics were expressed as mean±standard deviation or median (interquartile range [IQR]) and count (%), respectively. We determined the significance of the trends in the variables across the GNRI quartiles using linear regression analysis or the Wilcoxon-type nonparametric trend test, as appropriate. For the primary outcome variable, the 5-year slope of eGFR was estimated in participants with at least one follow-up eGFR data point measured between follow-up years 2 and 6, in addition to the baseline eGFR, using a linear mixed-effects model with adjustment for baseline eGFR, allowing for a random intercept and slope using an unstructured covariance matrix. We conducted linear regression analyses to assess the association between the categorized GNRI and the 5-year slope of eGFR, using the highest GNRI quartile as the reference. The analyses were hierarchically adjusted for baseline covariates including age, sex, presence of diabetes or hypertension, systolic blood pressure (SBP), diastolic blood pressure (DBP), and levels of serum hemoglobin, total cholesterol, serum glucose, and serum creatinine as follows: (1) unadjusted model; (2) model 1, which included age, sex, presence of diabetes or hypertension, SBP, DBP, and levels of serum hemoglobin, total cholesterol, and serum glucose; and (3) model 2, which included serum creatinine in addition to all variables included in model 1. The association between all-cause mortality and categorized GNRI was estimated using Cox proportional hazards regression models with the aforementioned adjustments. The survival curve was calculated using the Kaplan–Meier method to compare the cumulative incidence of all-cause mortality across the GNRI quartiles. Differences among groups were determined using the log-rank test.

Missing baseline parameters included diabetes, hypertension, coronary artery disease, stroke, smoking status, hemoglobin, total cholesterol, and albuminuria. The frequency of missing data was <0.7% for hemoglobin, total cholesterol, and albuminuria, whereas those for comorbidities and smoking status were <20.0% and 25.6%, respectively. Multiple imputations were performed using a multivariate normal model. The imputation model included all the variables of the fully adjusted model and an outcome variable using 50 imputed datasets. All statistical analyses were performed using STATA version 14.2 (StataCorp LLC, College Station, TX, USA).

Data from 403,677 participants who received a health check-up between 2002 and 2018 were extracted from a de-identified dataset for analysis, and 22,632 participants aged 60 years and older were included in our study cohort. After excluding participants with missing data on baseline body weight, height, serum albumin, and serum creatinine, those with baseline eGFR less than 15 mL/min/1.73 m², and those without follow-up serum creatinine data, data from 1,599 participants were finally available for analysis. A flowchart of the cohort construction is shown in Supplementary Fig. S1.

Baseline characteristics of the participants in the GNRI quartiles are discussed in Table 1. The median age of the participants was 63 years (IQR, 61–67), and 54% were male. The mean GNRI scores were 113.7±6.9. The median serum creatinine and eGFR levels were 0.9 mg/dL (IQR, 0.7–1.0) and 81 mL/min/1.73 m² (IQR, 72–93), respectively. The group with a higher GNRI was slightly younger, had a higher proportion of patients with diabetes and hypertension, and had a higher SBP and DBP.

The median value of the eGFR slope was −0.44 mL/min/1.73 m² (IQR, −1.34 to 1.17) per year. The median values of the actual eGFR changes over 5 years by group were −3.36 (IQR, −7.48 to 1.12), −2.88 (−4.86 to 5.30), −2.65 (−7.46 to 7.67), and −1.91 (−3.67 to 8.42) mL/min/1.73 m² for the first, second, third, and fourth quartiles of GNRI, respectively (p<0.001). In the linear regression analyses, the lower GNRI groups were associated with steeper 5-year eGFR slopes, as compared with the highest GNRI group (the reference group). The fully adjusted beta coefficients (β) and 95% confidence intervals (CIs) were −0.50 (−0.86, −0.14), −0.29 (−0.63, 0.05), and −0.19 (−0.53, 0.14) for the first, second, and third quartiles of GNRI, respectively (Fig. 1). Each 10-point decrease in GNRI level was associated with a steeper decline in the 5-year slope of eGFR by 0.26 mL/min/1.73 m² per year. The fully adjusted β and 95% CI were –0.26 (–0.45, –0.08).

In subgroup analyses, the association between the GNRI and 5-year eGFR slope was not modified by age, sex, presence of diabetes or hypertension, SBP, DBP, and levels of serum hemoglobin, total cholesterol, and serum creatinine (all for interaction p>0.5).

Among the 1,599 participants, the median follow-up duration was 7.4 years (IQR, 4.6–12.4). During this period, we identified 108 (7.8 per 1,000 person-years) all-cause deaths. The Kaplan–Meier curve for cumulative all-cause mortality stratified by GNRI quartiles is shown in Fig. 2. All-cause mortality significantly increased in the first GNRI quartile (log-rank p<0.001). The Cox regression analyses showed similar results: compared to the highest quartile of GNRI, the fully adjusted hazard ratios and 95% CIs were 2.20 (1.23, 3.95), 0.65 (0.33, 1.27), and 1.06 (0.57, 2.00) for the first, second, and third quartiles of GNRI, respectively (Table 2).