Improving pre-bariatric surgery diagnosis of hiatal hernia using machine learning models

Area of Science:

• Medical Informatics
• Surgical Oncology
• Artificial Intelligence in Medicine

Background:

• Hiatal hernia (HH) is common in bariatric patients, complicating laparoscopic procedures.
• Current preoperative evaluation for HH is often inaccurate, highlighting a need for improved diagnostic methods.
• Laparoscopic bariatric surgery faces challenges due to the high prevalence of hiatal hernia.

Purpose of the Study:

• To leverage machine learning (ML) for enhancing the preoperative diagnosis of hiatal hernia (HH).
• To improve the accuracy of preoperative assessments for bariatric surgery patients.
• To explore the potential of ML in overcoming limitations of traditional diagnostic tools.

Main Methods:

• Utilized three machine learning (ML) prediction models with prospectively collected bariatric surgery data (2012-2015).
• Employed automatic feature selection for patient data within the ML models.
• Compared the diagnostic prediction efficacy of ML models against the standard contrast swallow study (SS).

Main Results:

• The baseline swallow study (SS) identified 9.5% of 2482 patients with HH, showing 38.5% sensitivity and 92.9% specificity.
• Machine learning models boosted sensitivity for HH detection up to 60.2%.
• Three distinct ML models were developed, each improving diagnostic capabilities.

Conclusions:

• Machine learning models can increase the sensitivity of preoperative diagnostic evaluations by up to 1.5 times.
• Implementing ML-derived prediction models offers a significant improvement over traditional diagnostic methods.
• Enhanced preoperative diagnostic accuracy through ML can optimize surgical planning and patient care in bariatric surgery.