Dr. Guadalupe González-Mateo, Head of the Science Area at Premium Research, actively contributed to the development of this article, published in 2025.

The article describes the development of a machine learning-based software as a medical device to predict, in real time, the duration and outcomes of peritoneal dialysis (PD) using effluent biomarkers related to the mesothelial-to-mesenchymal transition (MMT).

This was a retrospective, longitudinal, triple-blind study conducted in two independent hospitals in Spain, designed using information-theoretical approaches for feature selection and machine learning modelling techniques. A total of 151 PD patients (training set) and 32 patients (validation set) from 1979 to 2022 were included. PD outcomes were analysed in four categories: duration, exit from PD, cause of PD termination, and technical failure — using both MMT biomarkers in effluents and clinical database records.

MMT biomarkers and clinical data enabled the prediction of PD duration with a mean absolute error of 16.99 months, using an Extra Trees (ET) regression model. Linear discriminant analysis (LDA) differentiated between transfer to haemodialysis or death, predicted whether the cause of PD termination was ultrafiltration failure (UFF) or cardiovascular disease (CVD), and even anticipated the type of CVD (with ROC AUC > 0.71).

The combination of longitudinal PD datasets, attribute reduction, and gold-standard algorithms — along with overfitting tests and class imbalance correction — ensured robust predictions in PD. The biomarkers showed significant mutual information and SHAP values, indicating that MMT processes may be causally involved in the development of UFF and CVD.

The article concludes that MMT biomarkers and clinical data may be causally associated with both local (UFF) and systemic (CVD) complications in PD. The MAUXI machine learning software applies ET-LDA models with ≤38 variables to predict PD duration and the type of technique failure linked to peritoneal membrane deterioration.

📰 Access the article here.

Eva María Arriero-País, María Auxiliadora Bajo-Rubio, Roberto Arrojo-García, Pilar Sandoval, Guadalupe Tirma González-Mateo, Patricia Albar-Vizcaíno, Gloria del Peso-Gilsanz, Marta Ossorio-González, Pedro Majano, Manuel López-Cabrera - Biomarker and clinical data–based predictor tool (MAUXI) for ultrafiltration failure and cardiovascular outcome in peritoneal dialysis patients: a retrospective and longitudinal study: BMJ Health & Care Informatics 2025;32:e101138.