Using phenotypic data from the Electronic Health Record (EHR) to predict discharge.

Area of Science:

• Healthcare Management
• Clinical Informatics
• Predictive Analytics

Background:

• Timely discharge to post-acute care (PAC) settings is crucial for patient flow and resource management.
• Early identification of patients requiring PAC is essential for efficient care transitions.
• Current methods may not adequately predict PAC needs upon initial hospitalization.

Purpose of the Study:

• To develop and internally validate a predictive model for post-acute care (PAC) needs.
• To identify key clinical factors and nursing assessments within the first 24 hours of admission that predict PAC utilization.
• To improve the accuracy of discharge destination predictions for hospitalized patients.

Main Methods:

• Retrospective observational cohort study design.
• Utilized electronic health record (EHR) data from adult inpatient admissions.
• Developed a multivariable logistic regression model using derivation cohort data and validated it internally.

Main Results:

• Key predictors of PAC discharge included older age, intensive care unit (ICU) admission, emergency department (ED) admission, number of home medications, and higher Morse fall risk scores.
• The developed model demonstrated strong predictive performance with a c-statistic of 0.875.
• The model correctly predicted the discharge destination in 81.2% of cases within the validation cohort.

Conclusions:

• A predictive model incorporating baseline clinical factors and nursing assessments shows excellent performance.
• This model can reliably identify patients likely to require post-acute care.
• The findings support the use of this model for optimizing discharge planning and patient transitions to PAC facilities.