Article Preview
Top1. Introduction
Owing to the lack of automated intelligence in Intensive Care Units (ICUs), support for key decision making is still far from being timely or effective. Despite a wealth of patient information sourced from various ICU devices, it is this high number of data sources that often leads to difficulty in using the collected information meaningfully. Common ICU devices for sourcing vital data include use of wearable sensors on patients linked to bedside monitors (e.g. blood pressure, oxygen saturation, heart rate, and temperature), ventilators (e.g. ventilation type, PEEP) and other clinical measures (e.g. Urine Output, Fluid Balance). Regular laboratory tests and other types of testing procedures add also to the “data deluge” challenge. Even if the information is to be stored digitally, manual data collection may still be warranted had the original data been sourced using proprietary applications that limit future data access. As well, modifications are often needed both in the workplace environment and in the information system (IS) architecture in order to improve the availability and accessibility of medically relevant and quality data needed to support the decision making process, for example, modifications in the environment had to be performed to allow applications such as INTCare to work (Filipe Portela et al., 2011; Portela, Santos, & Vilas-Boas, 2012).
Briefly, INTCare, a pervasive intelligent decision support system (IDSS), is used in the ICU of Hospital Santo António, Centro Hospitalar do Porto in Portugal to meet the various data deluge challenges. The system employs an ensemble of classifiers in order to perform online-learning to predict organ failures within the next 24 hours as well as the patient’s hospital outcome (dead or alive). Instead of relying on a traditional-based static dataset, INTCare is designed to work with a number of data streams (Mador & Shaw, 2009); more specifically, a data stream is generated for each monitored ICU variable. As a further complication, data from different sources are often gathered at different time intervals.
Given the huge number of ICU variables to be tracked and the rate in which they may vary, automatic real-time data processing is of paramount importance to highlight value changes that may be medically relevant. To improve the patient’s condition and care outcome, intelligent tracking of values associated with particular patients is clearly needed. Currently most patient data are automatically acquired and processed in real-time. Data thus collected and stored constitutes an important knowledge base that will be used to support the decision making process (DMP). INTCare represents an important step forward. The system aids the doctors to better understand specific patient conditions. In this light, the information provided by the IDSS is crucial. Until now this type of ICU decision support has been generally unavailable. In order to further enhance decision quality and clinical outcome, a Critical Events (CE) tracking system was developed and deployed in the Electronic Nursing Record (ENR). The present effort attempts to design, implement and evaluate the INTCare tracking system in four aspects drawing on the Technology Acceptance Model (TAM): perceived usefulness (PU), perceived ease of use (PEOU), behavioural intention (BI) and use behaviour (UB). As part of INTCare, the CE system was also evaluated. Results showed that the users were largely satisfied with the implementation of the CE tracking system.
The paper is organized as follows. Following the introduction, Section 2 addresses the ICU data deluge problem (INTCare, Intensive Care, Critical Events, Pervasive Health Cate and Technology Acceptance Model). Sections 3 and 4 present the data acquisition process, the data analysis and the pervasive system. Section 5 introduces the CE tracking system and Section 6 presents the results obtained at the level of interface and technology acceptance. Finally some concluding remarks are made and future work presented.