Back to the program

The aging population and the economic crisis, especially in developed countries, have as main consequences the reduction of dedicated funding to healthcare system and the increase of people whom suffer any type of disease. In this scenario, it is desirable to optimize the costs of public and private healthcare systems by reducing the affluence of chronic and dependent people to care centers and preventing the appearance of diseases by promoting a healthy lifestyle and early diagnoses. In this talk, we present some results of the iPHealth project describing a general big data platform and its use for monitorization and mortality prediction of a patience in the UCI (Intensive Care Unit) of a hospital.

The main objectives of the big data platform in the healthcare systems can be summarized in three based on how early the disease wants to be treated: i) long-term, this monitorization is done under demand whenever the doctor needs to access to the historical information of the patience and we are trying to achieve prevention by promoting a healthy life style (i.e. avoiding obesity by exercising); ii) mid-term, this monitorization is done automatically for specific periods of time and it involves the use of specific sensors and indicators which are used for early detection of a disease (i.e. heart rate and blood pressure for heart failure). We are trying to achieve prevention by early detection of a disease; iii) short-term, this monitorization is done 24x365 and its focused on detection and posterior analysis of the disease parameters in order to take decisions (i.e. mortality prediction in a UCI scenario).

This work focuses in the third of the above presented classification and it makes use of the Pyshionet/Computing Cardiology Challenge of the 2012. This scenario is very data intensive due to the patience is constantly monitorized with the objective of controlling its current critical state. In our work we have studied the data (i.e. SAPS-1) provided by the last 48h after being admitted and we have implemented a classifier for predicting the mortality of a patience given that information.

For this classification purpose, we have use two well-known and high-performance machine learning algorithms such as logistic regression and random forest jointly with the proposed big data architecture. The obtained results are very promising showing how the combination of big data storage capabilities and high computational performance can be used in an specific healthcare scenario for improving the current status of the information technology in that context.