Background: Inadequate assessment and recognition of barriers to discharge at time of admission leads to delays in the discharge process and prolongation of hospital admissions. These delays are associated with multiple negative outcomes such as increased length of stay, decreased patient satisfaction, strain on hospital bed capacity, and higher readmission rates. Prior studies have shown that an early focus on the discharge process can decrease unnecessary discharge delays. One important aspect of the discharge planning process is determining whether post-acute care (PAC) is needed. This study investigated whether use of a clinical decision tool at time of admission could accurately predict need for PAC at time of discharge.
Methods: Previous work from our group showed that a machine learning algorithm using data from the electronic medical record (EMR) could accurately predict the likelihood of whether a patient would be discharged home or to a PAC facility in 70-75% of the cases. For this study, we implemented this discharge prediction algorithm into a clinical decision tool called the “Johns Hopkins Early Discharge Planning Calculator” (JH-EDPC) to assess its validity in an inpatient clinical setting. The JH_EDPC prediction tool was used in a study group of 162 patients on a hospitalist service at a large tertiary care referral center to predict need for PAC at discharge. Providers were aware of the results in the intervention group. In the control group, the tool was utilized but the providers were unaware of the predicted results.
Results: JH-EDPC accurately predicted discharge location in 88% of all cases. Rehab services were nine times more likely to be involved when the JH-EDPC predicted PAC in the intervention group whereas the control group was only four times more likely to involve rehab services when the JH-EDPC predicted PAC (p-value < 0.001). Furthermore, providers in the intervention group had a higher incidence of involving rehab services on each day of hospitalization demonstrating earlier involvement in rehab services (p < 0.001).
Conclusions: JH-EDPC, a machine learning based algorithm utilizing data from the EMR, accurately predicts need for PAC and patient discharge location. Providers were more likely to involve rehab services at an early stage in the discharge planning process, which could lead to fewer delays in hospital discharges. Furthermore, the ability to accurately identify patients that would benefit from rehab services enhances high value care by better utilizing a limited resource.
