PhD in Nursing Science Program

Statistical Modeling Approaches and User-Centered Design for Nursing Decision Support Tools Predicting In-Hospital Cardiopulmonary Arrest

Dissertation under the direction of Professor Lorraine C. Mion, PhD, RN, FAAN

The objective of this dissertation was to identify strategies for successfully designing, de- veloping, and implementing decision support tools that predict in-hospital cardiopulmonary arrest (IHCPA).

A descriptive phenomenological study of 18 nurses explored information-gathering activ- ities related to IHCPA to understand how probability-based clinical decision support (PB- CDS) tools might best be implemented. Fifteen individual interviews and a focus group revealed patient, people, and technology information sources with information gathered in no consistent order. Participants expressed they: (a) search additional sources during un- certainty, (b) prefer being prepared for worst-case scenarios regardless of projections, and

(c) desire more detailed probability-based information, such as hourly predicted values. The words probability, risk, and uncertainty were used interchangeably by participants and did not appear to have consistent, intrinsic  meanings.

We then compared two statistical modeling strategies (logistic regression and Cox propor- tional hazards regression) and two machine learning strategies (random forest and random survival forest) for IHCPA with respect to prediction accuracy and interpretability. We used a retrospective cohort study with prediction model development from de-identified electronic health records at an urban, academic medical center. Although the classification models had greater statistical recall and precision (F 1 scores ranging 0.27-0.33 versus 0.19-0.26), the time-to-event models provided predictions that might better indicate to nurses and other clinicians whether and when a patient is likely to experience an IHCPA.

Participatory design sessions with bedside nurses, charge nurses, and rapid response team

nurses (n=20) identified preferred design considerations for an IHCPA PB-CDS tool. Themes focused on "painting a picture" of the patient condition over time, promoting empowerment and autonomy, and alignment of probability information with what a nurse already believes about the patient. The most notable design element consideration included visualizing a temporal trend of the predicted probability of the outcome along with user-selected over- lapping depictions of vital signs, laboratory values, and outcome-related treatments and interventions.

These studies serve as a foundation for design, development, and implementation of future PB-CDS tools intended to aid nurses because they provide insight on current cognitive and physical workflows for IHCPA recognition while seeking to create tools that support, rather than interrupt nurses’  work.