In hemodialysis patients, deviations of serum concentration of calcium, phosphate or parathyroid hormone from the values recommended by KDIGO are associated to a negative outcome. The regulation of calcium, phosphate and parathyroid hormone is complex and each parameter is not independently regulated. In hemodialysis patient's simultaneous modification in these three parameters are the result of abnormal mineral metabolism and the treatment used to correct these abnormalities that usually produce changes in more than one parameter. The specific objective of this work is an accurate and exhaustive analysis and description of the complex relationships between clinically relevant parameters in chronic kidney disease metabolism bone disease. In order to achieve these objectives we have used a machine learning approach Random Forest able to extract useful knowledge from a large database. The analysis of the complex interactions between the different parameters needs an advance mathematical approach such as Random Forest . The second aim of this study is to determine whether calcium, phosphate and parathyroid hormone, Fibroblast growth factor 23 and calcitriol are long-term associated with demographic features, mortality, co-morbidity and the therapy prescribed. Compare the predictions obtained with conventional statistical analysis versus the new model analysis based on artificial intelligence. Our preliminary results suggest that there are interactions between some parameters that are strong enough to question whether the evaluation of a given therapy can be based in the measurement of one single parameter. Subsequently, we will analyze in a prospective study on incident patients, whether the use of this new model may predict the cardiovascular risk and reduce the therapy cost.
