137 AI-based risk predictions: Evaluation of pilot DRAAI 5 The third update introduced a slightly lower threshold for identifying at-risk patients, based on nurses’ feedback. Additionally, a horizontal bar was added to the dashboard to visually represent the predicted risk score. During the fourth update additional risk predictors were included. Nurses suggested several factors, such as patients receiving chemotherapy or using Optiflow, as indicators of elevated risk. While not all suggestions met the model’s predictor requirements, some were incorporated, such as oxygen therapy and specific categories of antibiotics. The fifth update during the pilot period included two major changes. The first change addressed feedback from the field notes. Nurses observed that patients who already had a PU continued to receive PU risk predictions. They emphasized the importance of displaying these patients separately, as they are inherently at risk of developing additional PUs. To resolve this, a new category “Pressure ulcer”, was added. Patients with an existing PU diagnosis were assigned to this category, and no further risk predictions were displayed for them. These patients were automatically shown alongside the at-risk (high-risk) patients. The second change stemmed from an issue identified by the project team and local implementation teams in usual care practices. Nursing care plans were being generated without including preventive measures, which made the real-time feedback on prediction follow-up incomplete. Initially, a green checkbox appeared when a PU nursing care plan was created. In the updated version of DRAAI, the green checkbox appeared only when a nursing care plan that included preventive measures was generated. The project team informed the local implementation teams about the updates in person, via phone or mail, explaining the update and why they were performed. Adaptability and feasibility of implementation strategies The implementation strategies and adaptations were equally deployed across the three participating wards. During the implementation period, several adaptations were made to the implementation strategies to increase the feasibility of working with DRAAI. In this section, these adaptations are explained and justified. For one, we had not intended to share knowledge about current hospital protocols and PU prevention knowledge. Nevertheless, we became aware that nurses wished to alert the project team in cases when their clinical reasoning estimated a low risk for a patient to develop PU while DRAAI indicated a risk for developing PU. 'It’s annoying that the exclamation mark remains when I don’t agree with the risk prediction'.
RkJQdWJsaXNoZXIy MTk4NDMw