Predicting Surgery Outcomes for Vestibular Schwannoma Using Machine Learning
Author Information
Author(s): Dichter Abigail, Bhatt Khushi, Liu Mohan, Park Timothy, Djalilian Hamid R., Abouzari Mehdi
Primary Institution: University of California, Irvine
Hypothesis
Can a machine learning algorithm predict unplanned reoperations and complications after vestibular schwannoma surgery?
Conclusion
The study developed a machine learning algorithm that effectively predicts unplanned reoperations and surgical/medical complications after vestibular schwannoma surgery.
Supporting Evidence
- Unplanned reoperation occurred in 8.5% of patients.
- Surgical complications were seen in 5.2% of patients.
- Medical complications occurred in 6.2% of patients.
- The model achieved an ROC-AUC of 0.6315 for reoperation predictions.
- The model achieved an ROC-AUC of 0.7939 for medical complications.
- The model achieved an ROC-AUC of 0.719 for surgical complications.
- Accuracy ranged from 82.1% to 96.6% across outcome variables.
- Length of stay post-operation was identified as a key predictive variable.
Takeaway
Researchers created a computer program that helps doctors guess if patients will need more surgery or have problems after their first surgery for a type of tumor in the ear.
Methodology
A deep neural network model was developed using data from the ACS-NSQIP database, analyzing pre- and peri-operative variables to predict outcomes.
Potential Biases
Potential inaccuracies in the ACS-NSQIP database could affect the model's predictions.
Limitations
The study's dataset may not be generalizable, and important variables may have been omitted due to database constraints.
Participant Demographics
42.9% male, 57.1% female; 72.3% white, 4.66% Black or African American, 4.60% Asian, 2.41% Hispanic; mean age 50.4 years.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website