135 AI-based risk predictions: Evaluation of pilot DRAAI 5 A considerable number of nurses selected the neutral response option when asked about trusting DRAAI’s predictions and whether their expectations were met, with 38% and 47% choosing neutral, respectively. On the other hand, the remaining nurses overall indicated that DRAAI’s predictions largely met their expectations. Nearly half of the nurses (45%) expressed a desire for additional explanations on how DRAAI provides the risk predictions, see Table 3. Fieldnotes also frequently highlighted nurses’ requests for up-to-date insights into the specific risk factors contributing to the predictions for each individual patient. Table 3 Nurses’ perceptions on trust and follow-up of DRAAI’s risk predictions Totally agree (n) Agree (n) Neutral (n) Disagree (n) Strongly disagree (n) I trust DRAAI’s predictions 2 25 21 6 1 DRAAI’s predictions meet my expectations 2 21 26 4 2 I need more explanation about how DRAAI predicts the risk (based on which parameters, such as age, lab, etc.) 8 22 11 10 4 DRAAI looks attractive 2 32 16 4 1 If DRAAI predicted a (high) risk ... All the time (n) Often (n) Sometimes (n) Seldom (n) Never (n) I created a nursing care plan 8 14 20 2 11 the nursing care plan was already created 1 21 25 1 7 n – numbers Field notes revealed that nurses highly valued DRAAI’s predictions as helpful reminders to perform PU prevention. Compared to the previously used Waterlow score, nurses noted that DRAAI was a significant improvement, particularly in reducing the registration burden. Several nurses provided feedback to enhance DRAAI, with most suggestions addressing practical issues, such as improving accessibility. The use of a computer on wheels during the daily stand-ups increased DRAAI’s acceptability and fostered a joint understanding of the tool. Nurses began explaining DRAAI to one other, such as showing where the at-risk patients were displayed. They also discussed individual patients flagged as at-risk, using clinical reasoning to assess whether the predictions aligned with the patient’s condition.
RkJQdWJsaXNoZXIy MTk4NDMw