Development and Validation of PT-PENCIL: The Physical Therapy Frequency Clinical Decision Support Tool to Increase...

Identifying patients most likely to benefit from physical therapy in the hospital could aid physical therapists in optimizing treatment allocation for the purpose of increasing discharge to home.

The aims of this study were to develop and externally validate a predictive model for discharge to home on the basis of physical therapy frequency for patients who were hospitalized.

A predictive model was developed using retrospective cohort data collected between April 2017 and August 2022, with external validation conducted in a separate sample.

The setting was a large health system.

Participants were adult patients who were hospitalized and received physical therapy.

Predictors were extracted from the electronic health record and included demographics, clinical characteristics, and therapist-entered variables such as home set-up and prehospital level of function. Physical therapy frequency was quantified as once daily, defined as ≥5 times per week. The outcome was discharge to home. Variables were included in the final multivariable logistic regression model on the basis of associations with physical therapy frequency and/or outcome and clinical relevance. Calibration and discrimination of the models were assessed.

The development sample included 205,659 adult patient (average age = 72.2 [SD = 14.3] years; 55.3% female) hospitalizations, with 52.5% of patients receiving physical therapy daily and an overall proportion of 67.1% being discharged to home. The final multivariable model included 8 variables, with good calibration and discrimination. Internal validity was established with an optimism-corrected concordance statistic of 0.874 (95% CI = 0.872-0.875). The external sample included 102,311 patient (average age = 67.7 [SD = 16.5] years; 50.9% female) admissions, with 64.5% of patients receiving physical therapy daily and 77.8% being discharged to home. Predictive performance was high (calibration slope = 0.908), and discrimination was good (concordance statistic = 0.851).

This study developed and externally validated the underlying prediction model for a clinical decision support tool, termed Physical Therapy Frequency Clinical Decision Support Tool (PT-PENCIL), to identify patients most likely to benefit from daily physical therapy to discharge to home. Future work will evaluate the implementation of PT-PENCIL to determine its effect on patient-centered outcomes.