The RIACE Italian Multicenter Study was an observational, prospective, cohort study on the impact of eGFR on morbidity and mortality in individuals with type 2 diabetes [17]. The study was conducted in accordance with the Declaration of Helsinki. The research protocol was approved by the ethics committees of participating centers. Participants provided an informed consent.

The RIACE enrolled 15,773 Caucasian patients with type 2 diabetes, consecutively attending 19 hospital-based, tertiary referral Diabetes Clinics of the National Health Service throughout Italy, most of them in the years 2006–2008 (first patients 6 October 2005 - last patient 17 December 2008). Exclusion criteria were dialysis or renal transplantation.

Baseline data were collected using a standardized protocol across participating centers; results from different laboratories/methods were standardized by comparison with values detected in test samples at the reference laboratory of the Coordinating Center [17].

Participants underwent a structured interview to collect the following information: current age, smoking status, known diabetes duration, severe co-morbidities, and current treatments including glucose-, lipid-, and blood pressure (BP)-lowering therapies.

Body mass index (BMI) was calculated from weight and height, whereas estimated waist circumference (eWC) was calculated from log-transformed BMI values [18]. Then, BP was measured with a sphygmomanometer with the patients seated with the arm at the heart level.

Hemoglobin A1c (HbA1c) was measured by HPLC using DCCT-aligned methods, whereas triglycerides and total and HDL cholesterol were determined in fasting blood samples by standard colorimetric enzymatic methods. Then, LDL cholesterol concentration was estimated using the Friedewald formula.

The presence of CKD was assessed by measuring albuminuria and serum creatinine, as previously detailed [17, 19]. Briefly, albumin excretion rate (AER) was obtained from 24-hour urine collections or calculated from albumin-to-creatinine ratio in early-morning, first-voided urine samples; albumin concentration in urines was measured by immunonephelometry or immunoturbidimetry, in the absence of interfering clinical conditions. Serum (and urine) creatinine was measured by the modified Jaffe method, traceable to IDMS, and GFR was estimated using both the 2009 [7] and the 2021 [12] CKD-EPI equations (Table S1). Based on albuminuria and eGFR values, participants were then stratified according to the Kidney Disease Improving Global Outcomes (KDIGO) classification [6, 20].

The presence of diabetic retinopathy (DR) was assessed in each center by an expert ophthalmologist by dilated fundoscopy [21]. Patients were then classified as having no DR, non-advanced DR (including mild or moderate non-proliferative DR), or advanced DR (including severe non-proliferative DR, proliferative DR, or diabetic macular edema). DR grade was assigned based on the worse eye.

Previous major adverse CVD events, including myocardial infarction, stroke, foot ulcer, gangrene and non-traumatic amputation, and cerebrovascular, carotid, and lower limb revascularization, were adjudicated based on hospital discharge records by an ad hoc committee in each center [22].

The vital status of study participants on 31 October 2015 was verified by interrogating the Italian Health Card database (http://sistemats1.sanita.finanze.it/wps/portal/), which provides updated and reliable information on all current Italian residents [23].

Data are expressed as mean ± SD or median (interquartile range) for continuous variables, and number of cases (percentage) for categorical variables. For continuous variables, the Kolmogorov-Smirnov test was used to determine if variables were normally distributed; if not, logarithmic conversion was performed before regression analyses. Continuous variables were compared using the Student’s t-test (or one-way ANOVA) and Mann-Whitney test (or Kruskal-Wallis’s test) for parametric and non-parametric data, respectively, whereas the χ2 test was applied to categorical variables. None of the variables had missing values.

The eGFR distributions for the two equations were calculated in the whole cohort and in pre-specified subgroups using kernel density estimation, a nonparametric technique that provides a better estimation of the probability density function than traditional histogram [24]. Since the coefficients of age, sex and creatinine differ between the two equations, their influence on eGFR increase was assessed by calculating for each individual the eGFR change (ΔeGFR) from the 2009 to the 2021 CKD-EPI equation and plotting it against age, sex and the 2009 eGFR level. The level of agreement between the two equations were estimated using Bland-Altman plots, Lin’s concordance correlation, and linear weighted Cohen’s kappa.

The number and percentage of participants in each eGFR and KDIGO category with the 2009 CKD-EPI equation that were reclassified with the 2021 CKD-EPI equation to another rGFR category were then calculated. As nobody was reclassified to a worse eGFR category, the term “reclassified” is hereinafter used for “reclassified to a better eGFR category. The baseline clinical features of reclassified versus non-reclassified participants were compared either in the whole cohort (excluding individuals falling in the G1 category, who could not be reclassified to a better eGFR category) or separately in those with a 2009 eGFR 60–90 or < 60 ml·min− 1·1.73m− 2.

For survival analysis, the index date was the date of the baseline visit when participants were enrolled into the study and the end of follow-up was the date of the census (31 October 2015) or, for those who died, the date of death. Kaplan-Meier survival probabilities for all-cause mortality were estimated for reclassified and non-reclassified participants in each eGFR category and differences were analyzed with the Log rank statistic. The hazard ratios (HRs) and their 95% confidence intervals (CIs) were estimated by Cox proportional hazards regression with backward selection of variables, separately for reclassified and non-reclassified participants in each eGFR category, using the 2009 G1 category as reference. The backward variable selection method was chosen to reduce the chances of overfitting the data and make the linear regression model more interpretable. These analyses were unadjusted (model 1) or adjusted for baseline age (model 2) or age and other CVD risk factors (i.e., sex, smoking status, diabetes duration, HbA1c, BMI, triglycerides, total and HDL cholesterol, systolic and diastolic BP, anti-hyperglycemic, lipid-lowering, and anti-hypertensive treatment) and complications/comorbidities (albuminuria, DR grade, any CVD, and any comorbidity) (model 3). The analyses were repeated for reclassified versus non-reclassified participants in each eGFR category and further adjusted for the 2009 CKD-EPI eGFR level on top of model 2 (model 2a) and 3 (model 3a). Finally, Cox proportional hazards regression analyses according to KDIGO categories were run separately for the 2009 and 2021 CKD-EPI equations, using the category G1A1a as reference and adjusting as in models 1–3 (except for albuminuria); the G2 category was split in G2a and G2b (75–89 and 60–74 ml·min− 1·1.73 m− 2, respectively), as previously reported [20].

Receiver Operator Characteristic (ROC) curves were plotted and areas under ROC curves were calculated, using all-cause mortality at the end of follow-up as dependent variable and the CKD-EPI 2009 or 2021 eGFR values (as continuous variables) as predictors. Moreover, the Youden’s J statistic was used to assess the cut-off point with the maximum “J” index, where J = sensitivity + specificity– 100.

Tests were two sided, and a p value < 0.05 was considered statistically significant. Data entry and statistical analyses were performed using SPSS version 26.0 (SPSS, Chicago, IL, USA) and MedCalc version 22.014 (MedCalc Software Ltd, Ostend, Belgium).