Impact of RAASIs on Potassium and Mortality in a Large Cohort of Hemodialysis Patients: Practical Excursus and Comparison Between Traditional Statistics and Machine Learning

Background: The 2022 Kidney Disease: Improving Global Outcomes (KDIGO) guidelines suggest the use of Renin-angiotensin-aldosterone system inhibitors (RAASIs) in chronic Kidney Disease (CKD) stages IV-V, in contrast to the 2012 KDIGO guidelines, which discouraged it. This study aims to assess the impact of RAASIs on kalemia and mortality in a large sample of dialysis patients, where longitudinal data remain scarce, comparing traditional statistical methods with machine learning (ML) algorithms. Methods: This observational longitudinal analysis included 4764 hemodialysis (HD) patients from the Sicilian Registry of Nephrology, Dialysis and Transplantation, with a total of 56,964 longitudinal measurements. We evaluated the impact of RAASIs on serum potassium levels and all-cause mortality in the dialysis setting, comparing traditional statistics and ML. Linear Mixed Models (LMM) and Cox models with mixed effects were used for longitudinal and survival analyses. These were compared with ML approaches, including Random Forest (RF) for potassium variability and Lasso-regularized models for mortality, using four-fold cross-validation. Results: The study included 4764 patients, of whom 1207 (25%) were treated with RAASis. The mean age was 66 ± 15 years, 62% were male, 33% were diabetic, and a history of arterial hypertension was reported in 74% of patients. Hyperkalaemia at baseline was present in 1848 patients. The longitudinal model showed a statistically significant increase in kalemia [adjβ = 0.10 mmol/L, 95%CI 0.05/0.15, p < 0.001], but it was clinically negligible. Indeed, RF did not detect RAASIS as a relevant variable. Association between RAASIs and mortality was not detected either with Cox or ML models. Furthermore, the RF model outperformed traditional LMMs in explaining total potassium variability (56% vs. 43%). Conclusions: RAASI therapy in HD patients is associated with a minimal, non-clinically significant increase in serum potassium and does not impact all-cause mortality. The integration of ML reinforces the robustness of these findings, supporting the safety of RAASIs in the dialysis setting.