141 AI-based risk predictions: Evaluation of pilot DRAAI 5 the interest in AI. Addressing digital skills among nurses was therefore a significant component of our implementation efforts. During the implementation phase, we engaged in several discussions about the predictions that nurses disagreed with. This may impact building trust among healthcare professionals in AI for clinical-decision making. Which is a gradual process [46], it is, therefore, not surprising that nurses were relatively neutral about trusting DRAAI’s predictions during this three-month pilot study. Many instances of disagreement with the risk predictions stemmed from a misunderstanding of its purpose – specifically, the distinction between risk predictions and PU detection. Clarifying that DRAAI was designed to support nurses in assessing PU risk, rather than detecting PU, resolved much of this confusion. Additionally, the perception of PU risk as a minor issue among some nurses may reflect gaps in knowledge, attitude, or a sense of responsibility in PU prevention [47]. Many nurses also viewed the creation of nursing care plans for at-risk patients as a registration burden. This raises concerns about whether nurses fully understand the value and interrelation of nursing diagnoses, care goals, interventions, and evaluations within care plans [48]. Another potential factor is the sufficiency of healthcare literacy among nurses [49]. It is important to note that fully trusting AI for clinical decision-making carries inherent risks, as the accuracy of risk predictions relies heavily on the input data [50]. Reassuringly, in this pilot study, nurses identified 14% of patients as being at high risk for PUs based on their clinical judgment, even though DRAAI had not yet flagged them as such. This highlights the importance of balancing trust in AI with the critical judgment of nurses. Incorporating implementation science principles in this pilot study facilitated a systematic approach to identifying and addressing key determinants, developing an implementation plan, and combining multiple models, frameworks, and theories. This approach enabled the adaptation of both the intervention and its implementation strategies, ultimately enhancing their effectiveness and success. Determinants continued to emerge during the implementation, for instance the misalignments of practice with hospital protocol. The pilot allowed for adjustments to address these determinants effectively. While the proposed implementation strategies and adaptations were deemed time-consuming, they were deemed feasible by the project team and the local implementation teams. For instance, the project team’s daily presence during joint stand-ups in the first two weeks was particularly timeconsuming, especially given the sequential aspect of this pilot. This required the project team to attend daily stand-ups over six weeks, since it was not feasible to attend multiple wards simultaneously, as most stand-ups occurred at the same time.
RkJQdWJsaXNoZXIy MTk4NDMw