Patient and Healthcare Provider Priorities for Risk Prediction of Hospital Readmission: A Nominal Group Technique...

Nearly one in ten hospitalized adults is readmitted within 30 days. Though many models exist to predict risk of hospital readmission, many perform poorly and do not address the priorities of patients or providers that use them. In this study, we aimed to identify their priorities to inform a readmission risk prediction framework.

Using nominal group technique (NGT) methodology, we completed in-person and virtual meetings between February and April 2025 involving 22 total participants from Alberta, Canada. Participants included people with lived experience of a hospital readmission as a patient or caregiver as well as multidisciplinary healthcare providers. Discussions focused on the use of prediction tools, risk communication, and readmission type and relevant predictors. Priorities were collated, and each candidate suggestion within a topic was individually ranked by each participant indicating perceived importance. Priorities were then summarized and rank ordered using a mean standardized priority score.

Nine participants with lived experience (median age 56-65, 67% women, 33% rural) and 13 clinicians (median age group 36-45, 69% women, 77% physicians) participated in the meetings. Features most highly ranked by patients included using the tool to better inform patients and to plan discharge meetings, while those of clinicians focused on using a relevant time horizon, primarily risk of readmission ≤ 30 days, as well as presenting alerts of readmission risk within the patients' discharge summary. Both groups supported targeting all unplanned hospital readmissions and identified potential predictors including social determinants of health and laboratory markers. There were 163 total and 52 unique candidate predictors suggested, many of which were novel compared to existing models.

Patients and healthcare providers prioritized key features for readmission risk prediction frameworks, many of which differ from existing models. These findings can be used to inform the development of novel tools tailored to user needs.