External validation and comparison of clinical prediction models for cisplatin-associated acute kidney injury: a single-centre retrospective study

Figures

• 

  Fig. 1. Flow diagram of patient selection. From an initial cohort of 2,184 patients who received cisplatin (CDDP), 493 were excluded based on age, prior CDDP use, or weekly or daily regimens. After further exclusion due to missing baseline creatinine or outcome data, 1,684 patients were included in the final analysis

• 

  Fig. 2. Receiver operating characteristic (ROC) curves for the prediction of cisplatin-associated acute kidney injury (C-AKI). The red and blue lines represent the ROC curves of the Gupta et al. and Motwani et al. models, respectively. Panels A and B represent patients with C-AKI and severe C-AKI, respectively

• 

  Fig. 3. Calibration curves of the prediction models. The horizontal axis represents the predicted probability, and the vertical axis shows the observed incidence of cisplatin-associated acute kidney injury (C-AKI). Panels A and B correspond to the Motwani and Gupta models, respectively. The cohort was divided into ten risk-based groups, and the observed incidence for each group was plotted as triangles. The curves represent the smoothed regression lines. The red diagonal lines indicate perfect agreement between the predicted and observed probabilities. The calibration curves for severe C-AKI are shown in Fig. S2

• 

  Fig. 4. Comparison of risk stratification profiles between the original Gupta model and our cohort. Comparison of the distribution of patients and incidence of severe cisplatin-associated acute kidney injury (C-AKI) across risk groups defined by the Gupta simple model score. The blue bars and lines represent our cohort, whereas the red bars and lines represent the original developmental cohort. Patients were categorised into four risk groups (low [score 0–5.5], moderate [score 6–9.5], high [score 10–15.5], and very high [score ≥16]). The right axis shows the proportion of patients in each group, and the left axis shows the observed incidence of severe C-AKI. The numbers in the table indicate the number of patients with severe C-AKI compared to the total number of patients in each risk group

• 

  Fig. 5. Decision curve analysis of the prediction models. The net benefits for the prediction models developed by Motwani et al. and Gupta et al. are plotted across a range of threshold probabilities. Panel A displays results for patients with cisplatin-associated acute kidney injury (C-AKI), and Panel B displays results for patients with severe C-AKI. In both panels, dashed lines represent the models before recalibration, while solid lines represent the recalibrated versions. The threshold probability represents the point at which the benefits of interventions to prevent C-AKI outweigh the risks of those interventions. In simpler terms, a lower threshold probability (towards the left of the graph) indicates a greater emphasis on preventing nephrotoxicity compared to the potential benefits of treatment. Once a clinically relevant threshold probability is established, a superior model will demonstrate higher net benefits (i.e., a higher position on the graph)