B1900
Title: Prediction of kidney failure using electronic medical records
Authors: Davide Passaro - Sapienza University of Rome (Italy)
Giovanna Jona Lasinio - Sapienza University of Rome (Italy)
Tiziana Fragasso - OPBG - Roma (Italy)
Valeria Raggi - OPBG - Roma (Italy)
Zaccaria Ricci - OPBG - Roma (Italy)
Luca Tardella - Sapienza University of Rome (Italy) [presenting]
Abstract: Recent developments in technology have favored the digitalization of health data and facilitated a wider adoption of electronic medical records (EMRs). EMRs are the digital version of a patient's paper chart. Indeed, electronic health records contain valuable information for identifying health outcomes but their inclusion in predictive models presents numerous challenges. In fact, despite the progress realized in recent years, EMR data suffer from no standardization problem in measurement acquisition. A case study is presented on the use of EMRs acquired in the pediatric cardiac intensive care unit (PCICU) of Bambino Gesu's Children's Hospital. In particular, the focus is on the problem of exploiting this new type of emerging data to predict the stage of acute kidney injury (AKI) continuously during the intensive care unit stay. AKI is a frequent complication in hospitalized patients associated with mortality, length of stay, and healthcare costs. To avoid these problems, it is of high importance to develop methods to identify when patients are at risk for AKI and to diagnose subclinical AKI in order to improve patient outcomes. Some methodological issues are discussed related to pre-processing the available EMR data, the possible alternative ways of defining the outcome are analyzed and different tools are used for making predictions using both binary and multi-class classification methods. The results compare favorably with other recent attempts in the literature.