124 Chapter 5 Introduction Pressure ulcers (PU) are a source of harm and discomfort for hospitalized patients, often resulting in prolonged hospital stays, increased healthcare costs, and higher workload for healthcare professionals Goodall, Armstrong [1]. The causes of PU are multifactorial, with key patient-related factors including obesity, being underweight, smoking, diabetes mellitus, and immobility due to illness [2-4]. Treatment-related factors such as anemia, corticosteroid treatments, or chemotherapy further elevate the risk [2]. Additionally, surface- or device-related factors, such as suboptimal mattresses, seating cushions, and medical devices like catheters and drains, can also contribute to PU development [5]. These fluctuating multifactorial risk factors during hospital stays create challenges for nurses in promptly identify patients at-risk for developing PU. Validated PU risk assessment tools include the 11-item Waterlow Score [6, 7] and the 6-item Braden scale [8, 9]. These tools incorporate several of the previously mentioned risk factors. However, their measurement properties are not optimal, with sensitivity ranging from 46% to 82%, and specificity ranging from 27% to 67% [1012]. Additionally, the multifactorial nature of PU risk factors leads to fluctuations in patients’ risk level throughout their hospital stay. The European Pressure Ulcer Advisory Panel (EPUAP) guidelines recommend regular PU risk assessments to address these fluctuation [13]. However, performing regular risk assessments can contribute to a sense of administrative burden among nurses. This is particularly evident when clinical reasoning suggests that a patient is not at risk, making repeated risk assessments feel unnecessary [14]. Research demonstrates the potential of machine learning models to predict PU risk [15-18]. While these models have shown outstanding performance, evidence concerning their clinical value in practice remain limited [19]. In our search for a fitting prediction model we were unable to find one which was directly applicable to our setting for intended use; general wards of a tertiary university hospital. Most models included only patients scheduled for surgery or used predictors that are typical for a specific culture like betel nut chewing [16, 17]. This posed challenges due to the lack of available data that could be used in our context. Considering the fluctuating risk of PU risk, a frequently updating risk prediction model can eliminate the need for repeated manual risk assessments. To address this, a team of wound care nurses, PU nurse champions, and data scientists developed an AI-based PU prediction model called DRAAI. DRAAI, an acronym for Decubitus Risk Alert Artificial Intelligence, also means ‘turn’ in Dutch. The model enables daily predictions
RkJQdWJsaXNoZXIy MTk4NDMw