Artificial neural network prediction of post-thyroidectomy outcome

Area of Science:

• Surgical oncology
• Medical informatics
• Machine learning in healthcare

Background:

• Thyroidectomy is a common surgical procedure with potential for complications.
• Predicting surgical/medical complications and unplanned reoperations is crucial for patient safety and resource management.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) for predicting post-thyroidectomy complications and reoperations.
• To assess the predictive performance of the DNN model using a large patient database.

Main Methods:

• Utilized the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database (2005-2017).
• Developed a 10-layer DNN with an 80:20 training/testing split.
• Predicted medical complications, surgical complications, and unplanned reoperations.

Main Results:

• The DNN achieved an Area Under the Curve (AUC) of 0.783 for medical complications, 0.709 for surgical complications, and 0.703 for reoperations.
• Model performance metrics included accuracy (78.2%-97.2%) and sensitivity (11.6%-62.5%).
• Key predictive variables identified were sex, inpatient vs. outpatient status, and American Society of Anesthesiologists class.

Conclusions:

• A well-performing DNN model was developed to predict surgical/medical complications and unplanned reoperations following thyroidectomy.
• The study successfully demonstrated the predictive capacity of machine learning algorithms in this surgical context.
• A real-time, web-based application was created to showcase the model's predictive capabilities.