A generalized linear model for cardiovascular complications prediction in PD patients

Abstract

This study was conducted using machine learning models to identify patient non-invasive information for cardiovascular complications prediction in peritoneal dialysis patients. Nowadays is well known that cardiovascular diseases are the key to mortality in patients undergoing peritoneal dialysis as the risk of cardiovascular disease increases with the progression of renal failure. Primary aim is to establish variables most associated with cardiovascular complications. To achieve this goal four different machine learning techniques were used. We found that the best classification algorithm was a Generalized Linear Model, which achieved AUC values above 96% using a small subset of the original variables following a feature selection approach. Our approach allows us to increase the interpretability of the combinations of traditional factors, advanced chronic kidney disease factors and peritoneal dialysis factors all related with cardiovascular risk profile. The final model is based primarily in the traditional factors.

Computational Models Used to Predict Cardiovascular Complications in Chronic Kidney Disease Patients: A Systematic Review

“…Eleven studies are single-center. Of the 11 single-center studies, six papers use datasets with patients from the US [21][22][23][24][25][26], three papers include individuals from Spain [27][28][29], one article includes patients from Korea [30], and one paper includes patients from Portugal [31]. Three of the eleven single-center studies [25,28,29] proposed ML and regression models based on the CRIC (Chronic Renal Insufficiency Cohort Study) [32] or NEFRONA [33] study cohorts.…”

“…Various compounded endpoints such as MACE (major adverse cardiovascular events) in postoperative end-stage renal disease (ESRD) patients [30], or combinations of CV events (CV death, nonfatal MI, nonfatal stroke, and CV hospitalizations) in hemodialysis [34,38] or stage 3-5 CKD [35] patients have been predicted using AI/ML tools. Fernandez-Lozano et al [27] proposed a model to predict overall CVC as the leading cause of morbidity and mortality in CKD patients.…”

“…Fifth, composite endpoint studies are the starting point for designing CV risk stratification models. Computational solutions [27,30,34,35,38] have also been described to predict composite CV endpoints, such as MACE, in CKD patients. Composite endpoint studies are also heterogeneous regarding reporting transparency and bias, with a TRIPOD adherence ranging from 20% to 70%.…”

Using Artificial Intelligence Resources in Dialysis and Kidney Transplant Patients: A Literature Review

“…The specificity and sensibility of RT and k-NN were 95% in predicting stroke risk. Shortly, PD patients will benefit from a high prediction (stroke, infection, cardiovascular events [51], or even mortality risk) only from information easy to obtain (demographical, biological, or PD-related data).…”