In this paper, we propose a methodology for deriving a new kind of approximate temporal functional dependencies, called Approximate Predictive Functional Dependencies (APFDs), based on a three-window framework and on a multi-temporal relational model. Different features are proposed for the Observation Window (OW), where we observe predictive data, for the Waiting Window (WW), and for the Prediction Window (PW), where the predicted event occurs. We then consider the concept of approximation for such APFDs, introduce new error measures, and discuss different strategies for deriving APFDs. We discuss the quality, i.e., the informative content, of the derived AFDs by considering their entropy and information gain. Moreover, we outline the results in deriving APFDs focusing on the Acute Kidney Injury (AKI). We use real clinical data contained in the MIMIC III dataset related to patients from Intensive Care Units to show the applicability of our approach to real-world data.

The rapid increase of interest in, and use of, artificial intelligence (AI) in computer applications has raised a parallel concern about its ability (or lack thereof) to provide understandable, or explainable, output to users. This concern is especially legitimate in biomedical contexts, where patient safety is of paramount importance. This position paper brings together seven researchers working in the field with different roles and perspectives, to explore in depth the concept of explainable AI, or XAI, offering a functional definition and conceptual framework or model that can be used when considering XAI. This is followed by a series of desiderata for attaining explainability in AI, each of which touches upon a key domain in biomedicine.

Acute kidney injury (AKI) is a frequent complication in hospitalized patients, and is associated with worse short and long-term outcomes. An early prediction of AKI to detect the patients at risk could be a first step in the discovery and assessment of new therapies, and in improvements of patient outcomes. The advances in clinical informatics and the increasing availability of electronic medical records have allowed the development of predictive models of AKI diagnosis. In this research work we provide a consistent reproducible ETL pipeline for Intensive Care Unit (ICU) data, in particular regarding the MIMIC III database, to support the early prediction of AKI. Then, we build different predictive models aimed at early identifying subjects who could experience AKI syndrome in their next 7 days after the ICU admission. The entire procedure is based on a recently proposed rolling observational window approach. We consider two predictive models, Gradient Boosting Decision Trees and Support Vector Machines, via different platforms.