Precision disease networks (PDN)
This work addresses the need for better patient outcome prediction in healthcare, specifically for heart disease patients, but it is incremental as it builds on existing network and clustering methods.
The paper tackles the problem of predicting medical outcomes by introducing Precision Disease Networks (PDNs), which model individual patient disease evolution as networks and use clustering and visualization techniques; the result shows that PDNs improve prediction of outcomes like death or cardiovascular death compared to standard statistical analysis, as demonstrated on a heart disease dataset from MIDAS.
This paper presents a method for building patient-based networks that we call Precision disease networks, and its uses for predicting medical outcomes. Our methodology consists of building networks, one for each patient or case, that describes the dis-ease evolution of the patient (PDN) and store the networks as a set of features in a data set of PDN's, one per observation. We cluster the PDN data and study the within and between cluster variability. In addition, we develop data visualization technics in order to display, compare and summarize the network data. Finally, we analyze a dataset of heart diseases patients from a New Jersey statewide data-base MIDAS (Myocardial Infarction Data Acquisition System, in order to show that the network data improve on the prediction of important patient outcomes such as death or cardiovascular death, when compared with the standard statistical analysis.