Background: The hospital discharge represents a time-dependent and vulnerable chapter in the patient pathway. Approximately one-fifth of hospital discharges suffer delays due to non-medical reasons with one major factor being inadequate assessment and recognition of barriers to discharge when first admitted. Promising work in the existing literature demonstrates a relationship between a measure of patient’s mobility status during hospitalization and their next level of care at discharge. A commonly studied tool is the Activity Measure for Post-Acute Care Inpatient Mobility Short Form ‘6-clicks’ (AM-PAC), which measures a patient’s basic mobility. We evaluated whether a standardized assessment of a patient’s baseline mobility status completed within the first 48 hours of hospital admission is predictive of patient discharge to post-acute care facilities across a wide variety of patient populations.
Methods: This study used a retrospective cohort registry of 34,432 patients admitted to the Johns Hopkins Hospital (JHH) for ≥ 72 hours on units throughout the hospital (excluding psychiatry, pediatrics, and labor and delivery). Our analysis utilized the lowest recorded AM-PAC ‘6-clicks’ within the first 48-hours of admission as the primary predictor. Other predictors included BMI, age, race, gender, prior functional status, living situation prior to admission (I.e. lives alone vs. with someone), and insurance type. Discharge location served as the primary outcome for this study and was defined as either post-acute care (SAR, LTAC, etc) vs. home (which included home with services) as recorded in the electronic medical record (EMR). A Random forest machine learning algorithm generated a representative decision tree, which demonstrated the likelihood of whether a patient would be discharged home vs. post-acute care (PAC).
Results: AM-PAC ‘6-clicks’ mobility score ranked as the most important variable followed by BMI and age. While different sensitivities, specificities, and accuracies of the algorithm can be generated based on the cut point that is chosen, the decision tree accurately classified discharge location 70-75% of the time in a validation dataset. Furthermore, the algorithm consistently ranks AM-PAC as the most important variable in predicting discharge location for patients admitted to a wide variety of specialties.
Conclusions: Integrating the AM-PAC score into discharge planning can help improve awareness of possible discharge disposition to improve the discharge process overall.