Real-time physical function assessment in hospitalized patients: Enhanced discharge disposition predictions with the PROMIS PF5 tool

Why clinicians should know about it

• Picked for Internal Medicine (paper of the day, 15 October 2026): Real‑time PF5 predicts discharge disposition

Abstract

BACKGROUND: Physical function (PF) impairment is a key determinant of hospital discharge disposition, yet real-time PF assessments remain underutilized in inpatient care. The Patient-Reported Outcomes Measurement Information System Physical Function 5-item tool (PROMIS PF5, or PF5) was embedded into the electronic health record (EHR) across 21 Kaiser Permanente Northern California (KPNC) hospitals to standardize PF assessment and integrate it into care. Here, we evaluated its performance in predicting discharge disposition (to home, home with home health services [home health], or to a skilled nursing facility [SNF]). METHODS: Our retrospective cohort study included inpatient and observation encounters from 21 KPNC hospitals (January 1, 2022, through March 31, 2024). PF5 data included patient-reported prior level of function (PLOF) and clinician-assessed current level of function (CLOF), updated daily. Gradient-boosted machine models predicted discharge disposition within the next 24 h using demographic, clinical, and functional covariates. Model performance was evaluated across the first 7 days of hospitalization using area under the receiver operating characteristic curve (AUC), precision-recall AUC (prAUC), and number needed to evaluate (NNE) and were compared with models using clinician-entered expected discharge dispositions (EDD). RESULTS: PF5-based models outperformed clinician-entered EDD, especially early during hospitalization when EDD was often missing. Across 493,035 hospital encounters, PF5-particularly CLOF-strongly predicted discharge disposition. For example, on Day 3 of hospitalization, SNF discharge predictions using models with PF5 achieved an AUC of 0.89 (vs. 0.75 using EDD alone). CLOF was the most influential predictor across models. The NNE for SNF discharge improved from 4.0 (Day 1) to 1.4 (Day 7) at a sensitivity threshold of 60%. Models for home and home health disposition similarly showed improved accuracy with PF5. CONCLUSIONS: Models incorporating physical function enhanced prediction of discharge disposition, significantly outperforming models using expected discharge disposition alone. Our findings suggest that EHR-integrated functional assessment tools may provide earlier, more objective insight into discharge disposition compared with clinician-entered estimates. Further studies should evaluate whether incorporating such tools into clinical workflows improves discharge processes or patient outcomes.

Abstract as published, via PubMed.