Kudos to Michael Cary, Elizabeth C. Clipp Term Chair in Nursing, and his entire team for the submission of his Agency for Healthcare Research and Quality (AHRQ) R21/R33 application entitled: “Development and Evaluation of an Automated Clinical Prediction Tool to Guide Successful Community Discharge Following Inpatient Rehabilitation for Older Adults.” This proposal requests funding for a five-year period with a project start date of July 1, 2022.

Current prediction models lack sufficient accuracy to identify patients at high risk for re-hospitalization following discharge from Inpatient Rehabilitation Facilities (IRFs). The purpose of this study is to build a real-time, electronic clinical tool accessible via the web that predicts successful community discharge from IRFs for older adults. Using the 2019 Inpatient Rehabilitation Facility-Patient Assessment Instrument (IRF-PAI) and Medicare claims data from more than 200,000 older adults discharged from 1,200 IRFs, we have organized our aims according to two phases (R21 exploratory and R33 expanded activities). R21 phase aims (Build predictive models using IRF-PAI and Medicare claims data): Aim 1. Quantify the impact of functional limitations (mobility, self-care, cognition) and geriatric syndromes (pressure ulcer, depression, and falls) on successful community discharge (discharge to the community with no re-hospitalization or death within 31 days) following inpatient rehabilitation for older adults. A1a. Explore the impact of these factors on -year outcomes (days spent in the community, nursing home institutionalization, emergency department visits, rehospitalization, and death). Multilevel linear and survival models adjusted for facility-level variation will be conducted. Aim 2. Develop machine learning algorithms and assess discriminative ability using receiver operating characteristics (ROC) curves and the area under the curve (AUC). A2a. Develop and validate a prediction model to estimate the probability of successful community discharge in a large sample of older adults treated in IRFs. Predictor variables will be derived from routinely collected sociodemographic and clinical IRF-PAI data. We will perform an 80:20 split of the sample to derive the tool (80% test dataset) and to validate the tool (20% validation dataset). A2b. Determine the extent to which algorithmic bias impacts misclassification of older adults successfully discharged to the community and explore strategies to mitigate biased algorithms. R33 phase aims (R33 phase will automate and implement the model developed in the R21 phase): Aim 3. Automate the validated prediction model in Aim 2 using a Design Thinking approach. The tool will be co-created with key stakeholders (a nurse, therapist, physician, social worker, and administrator) from IRFs. The tool will stratify older adults with a high risk for unsuccessful community discharge and assign best practices for returning older adults back to the community based on risk level and medical diagnosis. Aim 4. Design a pilot to evaluate the feasibility and accepted ability of implementing the automated tool into the discharge workflow of IRF care for older adults. A sequential multisite, pre-post design with four IRFs across the state of North Carolina will be used. A4a. Evaluate the feasibility and acceptability of the automated tool by stakeholders within IRFs. The tool will be tested with 40 clinicians who participate in discharging older adults following IRF care. A4b. Evaluate the impact of the automated tool within IRFs on a) use of best practices known to returning older adults back to the community following IRF care, and b) 30-day hospital readmissions and time to readmission.