*3.3. Model Discrimination in the GCKD Cohort*

PROGRES-CKD models showed a good discrimination accuracy in the GCKD dataset (PROGRES-CKD-6, CKD stages 4–5, AUC = 0.91 (95%CI 0.86–0.97); PROGRES-CKD-24, CKD stage 3–5, AUC = 0.85 (95%CI 0.83–0.88)).

Evaluation of ratios of observed risk across quintiles of predicted risk indicated that the model best discriminated low and high-risk patients compared to those classified in the central quintile or risk score distribution (Figure 3).

**Figure 3.** Calibration of (**A**) PROGRES-CKD-6, and (**B**) PROGRES-CKD-24 in the GCKD cohort. Bar graph denotes the incidence of RRT initiation events observed in each quintile of risk (left axis); line graph denotes the fraction of RRT initiation events in each quintile with respect to the total number of RRT initiation events (right axis). Endpoint horizons: 6 months for PROGRES-CKD-6; 24 months for PROGRES-CKD-24.

#### *3.4. Comparison with KFRE Performance*

Table 3 shows the comparison in discrimination accuracy between PROGRES-CKD and KFREs equations. Since KFREs equations are computable only for complete information cases, patients with missing data were listwise deleted from this analysis. Given the large amount of missing information for ACR, we converted timed proteinuria assays (proteinuria g/24 h) into ACR when available. The conversion was based on a published correspondence table (Supplementary Material).

**Table 3.** Comparison between discrimination ability of (A) PROGRES-CKD-6 and (B) PROGRES-CKD-24 and that of Tangri's Kidney Failure Risk Equations (KFREs) in the FMC and the GCKD cohort. The two scores were computed considering only complete cases (column "Effective sample size"), while patients with missing data were not included in the analysis. Endpoint horizons: 6 months for PROGRES-CKD-6; 24 months for PROGRES-CKD-24. Imputation method: Listwise. Non-inferiority was defined as AUC < 0.05, while superiority was set at ΔAUC ≥ 0.05. \* Delta AUC: AUC of Tangri's KFRE–AUC of PROGRES-CKD model.


Based on the superiority test criteria, the discrimination accuracy of PROGRES-CKD-6 was greater than KFRE equations for short term RRT risk among stage 4–5 CKD patients (Table 3). PROGRES-CKD-24 discrimination was not inferior to that of the gold standard algorithms (Table 3).
