In this study Dascena explores machine learning landscapes used to make patient outcome predictions. The findings are published in Royal Society Open Science.
The theories and computational tools used to analyze molecular potential energy landscapes can be applied to those of the neural networks used in patient mortality prediction. The use of these neural networks is complemented by the increase in availability of patient data through the electronic health record (EHR). The training data, comprised of vital signs and laboratory measurements from the EHR, correspond to the machine learning landscapes and are analyzed with the goal of predicting patient outcomes. This type of analysis offers a more powerful predictive score than the most common clinical decision support (CDS) systems used like Modified Early Warning Score (MEWS), Sepsis-Related Organ Failure Assessment (SOFA), and Simplified Acute Physiology Score (SAPS II). Because these other systems assume that risk factors are independent of each other and only use patient-specific data, their predictive power is suboptimal. Energy landscape approaches offer a new perspective on the analysis of medical data, and future research in the area may lead to improved prediction methods.