Retrace-Clustering Creatinine Trajectory and Baseline Features for Predicting ESKD

Background: Patients with ANCA-associated vasculitis (AAV) may experience end-stage kidney disease (ESKD) and mortality. We aim to investigate the connection between the longitudinal trajectory of creatinine and the occurrence of ESKD and mortality. Methods: We included patients with a minimum of two creatinine measurements, encompassing the baseline period and the following six months (Figure). Creatinine trajectories were formulated using six months of creatinine readings from AAV patients with kidney involvement. We excluded patients who presented with end-stage kidney disease (ESKD). In instances where a patient had multiple creatinine values within a given month, the average of those values was employed. Percentage delta creatinine values, representing the percentage difference between a creatinine value and its baseline, were then calculated. The K-means algorithm for longitudinal data was employed to cluster the creatinine trajectories of AAV patients. The quality of clustering was evaluated using the Calinski-Harabasz Index. We conducted a time-to-event analysis for ESKD and death, assessing the survival rates of the clusters over a five-year follow-up period through Kaplan-Meier Survival analysis. Results: The study incorporates 273 patients with >1 creatinine values, amounting to a total of 2022 creatinine readings. We identified three renal trajectory groups: A-Stable (140), B-Recovered (N=100), and C- Declining (N=33). The baseline features varied across clusters, specifically in terms of baseline creatinine (284uM, 390uM and 151uM respectively, p<0.001) and ENT involvement (p=0.001). When considering the composite outcome of ESKD and death, Cluster A exhibited a 3-year incidence rate of 23%, Cluster B at 8%, and Cluster C at 28% (p<0.0069). Conclusion: Trajectory clustering allows for the identification of patients who may require closer monitoring, or targeted interventions based on their cluster assignment and associated risk profile.