Back
Arterial stiffness serves as a marker of arterial damage in various diseases, including chronic kidney disease (CKD) and renal replacement therapies such as Hemodialysis (HD) and Peritoneal Dialysis (PD). This study measured pulse wave velocity (PWV) in a dialysis population, analyzing variables associated with cardiovascular risk. Notably, the study applied statistical analysis, integrating machine learning for developing prediction models, to potentially identify patients without significant clinical signs of cardiovascular disease but with silent structural or functional damage. Recent years have witnessed a paradigm shift in medicine, transitioning from evidence-based medicine to data-based medicine. It underscores the importance of medical professionals understanding interdisciplinary tools that bridge between different disciplines, such as engineering, medicine, and mathematics, to study the body's structure and physiology for the prevention of cardiovascular diseases. The main goal of this study is to delineate the features linked to PWV within an Argentine dialysis cohort. Key secondary objectives include elucidating the PWV distinction between PD and HD patients, leveraging machine learning techniques to forecast PWV within both the general population and prominent subgroups.
A descriptive, cross-sectional study was conducted in two dialysis centers. Primarily, the relationship between PWV and each of the features was analyzed in the global cohort. Subsequently, patients from two subgroups (PD and HD) were studied, to analyze their distinguishing characteristics, followed by the development of computational models to assess the impact of the clinical features on PWV in the PD and HD cohorts.
PWV measurements were obtained from 99 patients. The median carotid-femoral PWV was 13.22 m/s (IQR 10.50-16.20). Age, systolic and diastolic blood pressure, mean baseline blood pressure, and heart rate showed a moderate positive linear relationship with PWV. BMI, urea, and albumin exhibited a moderate negative linear relationship. A history of diabetes, coronary artery disease (CAD), and heart failure was associated with higher PWV. Patients on PD had a lower median PWV than those on HD (Table 1). Multiple regression models using machine learning were developed. In the general population, a Random Forest model identified age as the variable best predicting PWV (Figure 1 & 2). A multiple linear regression model was selected as the best predictive model for PWV among PD and HD patients. However, after multivariate adjustment, modality lost statistical significance, and due to the limited number of patients, the model exhibited high collinearity.
This study is a first of its kind in Argentina describing variables associated with PWV as a surrogate for arterial stiffness in a dialysis population, differentiating characteristics between PD and HD. While acknowledging potential statistical limitations in developing predictive models for a small population, future scope includes expanding the database, refining models, and emphasizing the importance of interdisciplinary collaboration. This endeavor presents the base for the potential development of a Clinical Decision Support System, incorporating dynamic datasets and prediction models tailored to local populations, aimed at impacting clinical decisions and improving the health of patients in developing countries.