This was an observational cohort study that enrolled 127 KTRs under routine ambulatory care at the Nephrology and Transplantology Department of the University Hospital in Kraków, Poland, between 1 September 2016 and 30 June 2019. The median (IQR) follow-up duration was 29 [25–32] months. All patients provided written informed consent prior to study participation. This study was conducted in accordance with Declaration of Helsinki and ICH/GCP guidelines and represents an extension of our prior work and institutional research into circulating biomarkers in KTRs. Approval from the local Bioethics Committee was granted in extension of the original biomarker assessment study and updated for patient follow-up (Bioethics Committee Decision No. 1072.6120.202.2022).

Demographic and clinical data were collected at the time of patient enrolment and over follow-up based on manual chart review. Longitudinal assessments were obtained through screening of electronic and paper-based medical records from consecutive ambulatory clinic visits by physicians, at pre-defined 3-to-6-month intervals. The main outcomes assessed over follow-up included: (i) hospitalization at the University Hospital departments for any cause; (ii) all-cause mortality; and (iii) death-censored graft loss, defined as graft dysfunction requiring permanent return to dialysis or re-transplantation listing. Study outcome data was gathered based on direct patient, family or dialysis centre contact, when available. Individual record censoring occurred at the date of death, dialysis transfer or last ambulatory care visit assessment.

Hypertension was defined as outpatient visit blood pressure measurement ≥140/90 mmHg or recorded use of antihypertensive medications. Dyslipidaemia was defined based on existing medical record or use of lipid-lowering agents. Similarly, diabetes was identified based on pre-existing records or use of glucose-lowering therapy and/or biochemical data according to Polish Diabetes Society guidelines. Coronary artery disease (CAD) was defined broadly as a composite outcome of either history of myocardial infarction, stroke/transient ischemic attack, percutaneous coronary intervention or coronary artery bypass grafting, or existing record of chronic coronary syndrome. We did not include heart failure within this definition. Body mass index (BMI) and waist-to-hip ratio (WHR) were calculated based on standard formula. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) 2021 race-free equation via nephro package [31]. CKD was classified according to KDIGO guidelines [32].

We collected peripheral venous blood samples from patients presenting for first study visit following overnight fast. Serum was obtained and stored in serum separator tubes, allowed to clot at room temperature, and then centrifuged. Aliquots were frozen at −70 °C and stored. Assays were conducted later, in batch and under standardized conditions. Procedures were performed by experienced laboratory staff blinded to patient outcomes.

Concentrations of study markers were measured using commercially available enzyme-linked immunosorbent assay (ELISA) kits, in accordance with manufacturer instructions. For α-Klotho (Immuno-Biological Laboratories Co., Ltd., Fujioka, Japan; Catalog No. 27998) sensitivity was 6.15 pg/mL; measurement range 93.75–6,000 pg/mL; intra-assay coefficient of variation (CV) < 3.5% and inter-assay CV was <11.4%. For sirtuin-1 (Wuhan EIAab Science Co., Ltd., Wuhan, China; Catalog No. E3949h) sensitivity <32 pg/mL; detection range 78–5,000 pg/mL; intra-assay CV was <4.3% and inter-assay CV was <7.2%. For high-sensitivity Interleukin-6 (hsIL-6; Quantikine HS ELISA Human IL-6 Immunoassay; R&D Systems, Inc., Minneapolis, MN, USA; Catalog No. HS600B) the mean minimum detectable dose was 0.039 pg/mL; reference range: 0.447–9.96 pg/mL; intra-assay CV was <7.8% and inter-assay CV was <9.6%.

Routine biochemical tests, including serum creatinine, were measured using standardized methods on automated analysers, as per standard laboratory protocol (Hitachi 917, Hitachi, Tokyo, Japan; Modular P, Roche Diagnostics, Mannheim, Germany).

Analyses were performed in R 4.5.3 (R Core Team, 2026; R Foundation for Statistical Computing, Vienna, Austria). Continuous and categorical variables were summarized as median with interquartile range (IQR) and counts with proportion (N, %), respectively. Variable distributions were assessed with visual inspection of density plots and Shapiro-Wilk test. For between-group comparisons we utilized exact Wilcoxon rank-sum tests for continuous variables, whereas Fisher’s test was employed for categorical variable comparison. To assess relationships from a robust perspective, we utilized Spearman’s rank correlation coefficient (rho), with 95% confidence intervals (CI) derived using percentile bootstrap over 2000 replicates.

The relationship between ln (α-Klotho) and final eGFR was modelled using hierarchical linear regression approach. CIs for coefficients were derived using bootstrap methods. We performed influence diagnostics and Cohen’s f² was utilized to quantify effect size. For additional sensitivity, we confirmed linearity of relationships based on alternative restricted cubic spline (3 knot; P = 0.07 for non-linearity).

Due to overdispersion of hospitalization counts, we considered different modelling approaches. Overall, we utilized quasi-Poisson regression, with log-transformed follow-up time as offset variable. To assess whether ln (α-Klotho) was independently associated with outcome after adjustment for all candidate predictors, we fitted a fully-adjusted bias-reduced Poisson model via brglm2 package. Additionally, non-parametric bootstrap and a negative binomial sensitivity model were considered. The Firth model was refitted with baseline CKD-EPI eGFR added as a covariate to further test if serum α-Klotho provides additional information over eGFR. Multicollinearity was assessed with variance inflation factors. Statistical tests were two-tailed and we treated P < 0.05 as statistically significant.

This study included 127 KTRs who were followed for a median (IQR) time of 29 [25–32] months. Most patients were of male gender (N = 86, 67.7%) and middle-aged, with median (IQR) age of 54 [33–52] years. We compared serum α-Klotho concentrations between KTRs and 32 healthy controls (16 male [50.0%]; mean age 50 years, age range 29–74 years). We observed lower circulating α-Klotho levels in serum of KTRs, with median (IQR) values of 616 (514–788) pg/mL versus 1,042 (922–1,472) pg/mL (P < 0.001), which correspond to an estimated difference of 463 (95% CI 346–605) pg/mL (see Figure 1).

For descriptive comparison of clinical characteristics, we stratified patients using a 75th-percentile cut-off (788 pg/mL) into “low” (n = 95) and “high” (n = 32) α-Klotho groups (see Table 1). However, for inferential analyses, serum α-Klotho was modelled as a continuous variable with natural log transformation due to distributional considerations. Both “low” and “high” α-Klotho groups were comparable in age, sex, body mass indices, metabolic parameters, smoking status, and baseline eGFR. Interestingly, hypertension was more frequently recorded in the low α-Klotho group (P = 0.016). The high α-Klotho group also showed lower log-transformed hsIL-6 levels, with median (IQR) values of 0.81 (0.37–1.20) versus 1.31 (0.94–1.73) (P < 0.001). Respectively, higher SIRT1 levels, with median (IQR) values of 11.46 (9.78–14.48) versus 9.82 (8.11–11.52) ng/mL were recorded in the “high” α-Klotho groyp, as compared to “low” α-Klotho group (P = 0.010). Of note, this difference in SIRT1 concentrations was not robust to the cut-off, as detailed in Section Relationships between clinical characteristics and α-Klotho and SIRT1 concentrations in serum of KTRs.

For monotonic relationship analyses, we followed Cohen’s approach to interpretation of Spearman’s correlation strength (rho = 0.10, weak; rho = 0.30, moderate; rho = 0.50, strong). We observed that serum α-Klotho was consistently associated with allograft function across all timepoints: baseline eGFR (rho = 0.304), follow-up T1 (rho = 0.315), and follow-up T2 (rho = 0.29). Circulating α-Klotho in serum was inversely correlated with hsIL-6 levels (rho = −0.39, 95% CI -0.60–0.15, P = 0.002). Importantly, this association was also consistent in alternative analysis with α-Klotho categorization approaches (P = 0.001 for the 75th-percentile cut-off, P = 0.026 for the median cut-off).

By contrast, the α-Klotho-SIRT1 relationship was weaker and cut-off dependent. In continuous analysis, Spearman rho was estimated at 0.21 (95% CI -0.04–0.44, P = 0.098). Group differences are reported descriptively in Table 1 (P = 0.010 for the 75th-percentile cut-off), though were not preserved using a median split (P = 0.31). The SIRT1 finding should be interpreted with caution and requires validation. Other monotonic relationships between α-Klotho and clinical features were weak: HbA1c (rho = −0.140), BMI (rho = −0.087), mean arterial pressure (rho = 0.099), waist-to-hip ratio (rho = 0.020), and age (rho = 0.006).

At baseline, median (IQR) eGFR was 57.4 (41.7–72.9) mL/min/1.73 m², with a median change of 0 (−10.4–7.1) mL/min/1.73 m² over follow-up. Most KTRs (N = 103, 80%) were characterized with stable graft function, which we defined as absolute change within 15 mL/min/1.73 m². At the last available follow-up visit, KDIGO stage distribution was as follows: G1 (11.8%), G2 (35.4%), G3a (20.5%), G3b (12.6%), G4 (11.8%), and G5 (7.9%).

We observed that when patients were stratified by serum concentrations of α-Klotho categorized into tertiles, they differed by eGFR (P = 0.010; Figure 2A). Specifically, patients in the lowest tertile had a median (IQR) final eGFR of 44.2 (25.6–65.9) mL/min/1.73 m², which is markedly lower than the second tertile (63.9 (52.7–76.5); P = 0.048) and the third tertile (61.4 (49.6–89.0); P = 0.024). Patients with eGFR decline also showed numerically lower median ln (α-Klotho) (P = 0.14; Figure 2B). Further, the continuous monotonic relationship between ln (α-Klotho) and final eGFR was formally significant (rho = 0.28, P = 0.002; Figure 2C), and the distribution of α-Klotho tertiles across KDIGO stages also supported a significant trend association (P = 0.003; Figure 2D).

We then constructed iterative linear regression models with progressive covariate adjustment (see Table 2). In the unadjusted model, each unit increase in ln (α-Klotho) was associated with 15.4 mL/min/1.73 m² higher final eGFR (95% CI 6.7–25.4). The association was preserved after demographic adjustment (Beta 15.3, 95% CI 7.3–25.8) and after additional adjustment for BMI and diabetes status (Beta 14.8, 95% CI 6.2–25.6). However, the effect size was modest (f² = 0.083). In additional sensitivity analysis after excluding high-influence observations, the relationship was consistently observed (Beta 15.4, 95% CI 3.1–27.7, P = 0.015). We also confirmed the plausible linearity of the ln (α-Klotho) effect using comparison with a three-knot restricted cubic spline model (P = 0.07 for non-linearity).

Over 274 person-years of follow-up, we recorded 153 inpatient admissions that occurred among 54 KTRs (42.5%); 73 (57.5%) were not hospitalized during follow-up. Of note, we recorded hospitalizations that occurred within the setting of the University Hospital. Recurrent admissions ranged from 1 to 12 per patient, with a median (IQR) of 1 [1, 2] admissions. In age-adjusted quasi-Poisson regression (see Table 3), higher ln (α-Klotho) was significantly associated with lower hospitalization rate (IRR 0.38, 95% CI 0.19–0.74, P = 0.006). Coronary artery disease was also associated with higher rate (IRR 2.51, 95% CI 0.98–6.14, P = 0.049), while sex, diabetes, hypertension, dyslipidaemia, mean arterial pressure, BMI, ln (hsIL-6), and SIRT1 were not significant predictors.

To evaluate whether ln (α-Klotho) was independently associated with readmission after adjustment for all candidate predictors, we fitted a fully-adjusted Firth-corrected Poisson model containing age, sex, BMI, mean arterial pressure, diabetes mellitus, hypertension, dyslipidaemia, coronary artery disease, ln (hsIL-6), SIRT1, and ln (α-Klotho), with robust standard errors and log (follow-up time) as offset (Supplementary Table S1; Figure 3A). Although exploratory, ln (α-Klotho) was observed with a maintained, strong inverse relationship with hospitalization rate (IRR 0.31, 95% CI 0.15–0.65, P = 0.002). Similar observations were noted for coronary artery disease (IRR 3.02, 95% CI 1.25–7.30, P = 0.014). These findings were consistent across different, alternative model constructs, both using a non-parametric bootstrap of the same model (R = 1,000) with a 95% CI of 0.10–0.66, as well as a negative binomial regression model (similarly, the estimated IRR was 0.32 95% CI 0.14–0.73, P = 0.005). In model diagnostics, all variance inflation factors were <2, which suggesting limited confounding by collinearity.

To further investigate whether α-Klotho may provide additional information over eGFR monitoring, we refitted the fully adjusted Firth model with baseline eGFR added as a covariate. Consistently, ln (α-Klotho) consistently demonstrated an independent inverse association with hospitalization rate (IRR 0.37, 95% CI 0.20–0.69, P = 0.002), even after adjustment for baseline eGFR. At the same time, baseline eGFR was a strong predictor of hospitalization (IRR 0.97 per mL/min/1.73 m²; P < 0.001).