A preliminary prediction model using a deep learning software program for prolonged hospitalization after...

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Prolonged hospital stays (LOS) are a significant concern in cardiovascular surgery, particularly in aging populations.
• Current prediction models for extended LOS after cardiovascular surgery are lacking.
• Identifying patients at risk for prolonged LOS is essential for resource allocation and patient management.

Purpose of the Study:

• To develop and validate a predictive model for prolonged length of hospital stay (LOS) following cardiovascular surgery.
• To utilize deep learning and preoperative data for accurate LOS prediction.
• To address the absence of established prediction models for prolonged postoperative stays.

Main Methods:

• A deep learning software program (Prediction One) was employed to create the prediction model.
• Preoperative data from 157 patients undergoing cardiovascular surgery were used (121 for training, 36 for validation).
• A prolonged LOS was defined as a postoperative stay exceeding 30 days, attributed to physical inactivity.

Main Results:

• The deep learning model achieved an area under the receiver operating characteristic curve (AUC) of 0.806 in the validation dataset.
• The model demonstrated an accuracy of 67% in predicting prolonged LOS.
• These preliminary results indicate acceptable performance in identifying patients likely to experience extended hospital stays.

Conclusions:

• A preliminary deep learning-based prediction model shows acceptable performance for identifying prolonged LOS after cardiovascular surgery.
• This model utilizes preoperative data, offering a potential tool for early risk stratification.
• Further validation and refinement are warranted to enhance the model's clinical utility.