Cardiovascular disease remains the main cause of death in patients undergoing peritoneal dialysis (PD). [1]and blood pressure (BP) management is central to cardiovascular risk mitigation in this population. Although mean BP levels have long been used to assess cardiovascular risk, the role of blood pressure fluctuations, particularly visit BP fluctuations (VVV), to visits, is increasingly recognized as an important and independent prognostic factor for cardiovascular disease. [2].
In this issue Hypertens ResLin et al. [3] Using advanced machine learning techniques, we provide valuable and timely insight into the association between early stage VVV and long-term outcomes in patients undergoing continuous outpatient PD (CAPD). This study retrospectively evaluated 666 patients with incident CAPD with a maximum of 15.8 years of follow-up. Of the six different VVV parameters assessed during the first 6 months of PD initiation, standard deviations of diastolic BP (SDDBP) were identified as the most relevant predictor of adverse clinical outcomes. In particular, higher SDDBP increased the risk of major adverse cardiovascular events (MACEs) by 99% and the risk of all-cause mortality by 46%. These associations were robust in multiple statistical approaches, including competing risk regression and machine learning-based predictive models, and were particularly pronounced in patients with no renal function remaining. (Figure 1)).
Visit to patient visits for peritoneal dialysis based on machine learning algorithms. Mace, major harmful cardiovascular events
This study is worthy of praise for several reasons. First, the authors employed four different machine learning algorithms: Ridge Classifier, Lasso Regression, Random Forest and Bootstrap Algorithms to select prediction features from the high-dimensional dataset. This multi-algorithm approach increases the robustness of variable selection, reduces the risk of overfitting, and ensures generalization. Second, development of predictive models that surpass established risk models such as Inno.2Vate-Mace gives practical clinical relevance to these findings.
The use of SDDBP as a central predictive marker is of particular interest. Systolic VVV has historically attracted more attention [4, 5]recent evidence suggests that diastolic VVV may be more closely related to structural and functional changes in small arteries, particularly CKD patients. [6]. In PD patients, diastolic BP may function as a more stable and beneficial marker of vascular tone and volumetric status. The association between SDDBP levels and residual renal function observed in this study further supports the interaction of volume control and BPV.
This study had several limitations to consider. As this was a retrospective, single-centric cohort study, findings may not be generalizable to a wider or more diverse population. Furthermore, observational designs do not eliminate causal inference and residual confounding factors. This study was adjusted for a wide range of covariates, including demographic, clinical, clinical, and dialysis-related parameters, but drug use, particularly antihypertensive classes and adherence, has not been fully investigated. As drug therapy and compliance are known determinants of VVV, further studies incorporating these variables would be beneficial.
Nevertheless, this study provides a compelling case for the incorporation of VVV, particularly SDDBP, in cardiovascular risk assessment in PD patients. The use of machine learning models to derive discrete risk predictions based on VVV parameters and baseline characteristics represents a promising progression towards precision medicine in nephrology. From a clinical perspective, these findings suggest that early identification of patients with high VVV can lead to more intensive cardiovascular surveillance and customized antihypertensive strategies.
This study provides strong evidence that early diastolic VVV quantified using SDDBP is a key predictor of long-term cardiovascular and mortality in patients with CAPD. By integrating machine learning and traditional statistical approaches, this study provides new insights into the prognostic significance of VVV, laying the foundation for future interventional research aimed at reducing VVV and improving clinical outcomes in this vulnerable population.
