Development and Validation of a Machine Learning Model to Aid Discharge Processes for Inpatient Surgical Care.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Hospital Operations Management

Background:

• Inpatient overcrowding leads to delayed care, particularly surgical services, due to insufficient bed availability for postoperative patients.
• Efficient patient discharge is crucial for alleviating overcrowding, requiring seamless coordination among healthcare professionals.
• Lack of early identification and transparency regarding discharge candidates and their barriers hinders timely patient flow.

Purpose of the Study:

• To validate a clinically interpretable feedforward neural network model for predicting 24-hour inpatient surgical discharges.
• To identify clinical and nonclinical barriers associated with patient discharge delays using the predictive model.
• To assess the model's potential to improve the efficiency of the hospital discharge process.

Main Methods:

• A prognostic study involving adult patients discharged from inpatient surgical care at a quaternary care teaching hospital (May 2016 - August 2017).
• A feedforward neural network model was developed and validated using cross-validation techniques, comparing its performance against a baseline model.
• Prospective analysis on general surgical floors to record causes of delay when patients were not discharged as predicted.

Main Results:

• The neural network model achieved an out-of-sample area under the receiver operating characteristic curve of 0.840, demonstrating strong predictive performance.
• The model exhibited higher sensitivity (56.6%) and specificity (82.6%) compared to the baseline model in predicting discharges.
• Identified 65 discharge barriers, with clinical (30.1%), clinical practice variations (22.1%), and nonclinical reasons (47.8%) contributing to delays, resulting in 128 avoidable bed-days.

Conclusions:

• A neural network model can effectively predict daily inpatient surgical discharges and associated barriers.
• The model aids in identifying systemic causes of discharge delays, offering insights for process improvement.
• Further investigation into such predictive models is warranted to enhance the timeliness of patient discharges and optimize hospital resource utilization.