Machine Learning Model for Predicting Operative Duration in Transsphenoidal Neuroendoscopic Pituitary Adenomectomy

In the high-stakes environment of a hospital operating room, time is a resource as critical as any surgical instrument. Every minute a procedure runs longer than expected ripples outward, creating a domino effect that can delay subsequent surgeries, strain the staff, and disrupt the carefully orchestrated flow of patient care. For nurses and administrators, the ability to predict exactly how long a surgery will take is not merely a matter of convenience; it is the foundation of safety, efficiency, and the well-being of the medical team. Traditionally, these estimates have relied heavily on the gut feeling of the surgeon or broad averages from past cases, methods that often miss the subtle, complex details that make every patient and every tumor unique.

A team of researchers at the Second Affiliated Hospital of the University of South China has turned to a different approach to solve this problem. They developed a computer system capable of learning from past surgical records to forecast the duration of a specific, delicate brain surgery known as transsphenoidal endoscopic pituitary adenomectomy. This procedure involves removing a tumor from the pituitary gland through the nose, a minimally invasive technique that has become the standard of care. By feeding a computer program detailed information from one hundred previous patients—from the size and shape of the tumor to the patient's age and medical history—the researchers trained the system to recognize patterns that human observers might overlook. Their goal was to move from guessing to knowing, providing a data-driven tool that could help hospitals schedule their days with greater precision and reduce the burden on nursing staff.

The researchers gathered information from one hundred patients who underwent this surgery between 2016 and 2025. They collected twenty-two different pieces of information for each person, including the tumor's location, its texture, whether it had returned after a previous removal, and specific measurements of how close it sat to critical structures deep within the brain. They then split this data into two groups: a larger group to teach the computer system what to look for, and a smaller group to test how well the system performed on cases it had never seen before. They built two different types of computer models to make the predictions. One was a random forest model, which works by creating many simple decision trees and combining their answers, while the other was an artificial neural network, a system designed to mimic the way the human brain processes information by connecting layers of digital neurons.

When the team tested these models, the results showed that the artificial neural network was the superior tool. On the test group of patients, this model predicted the surgery time with an average error of just over thirty-eight minutes. In practical terms, this means that for more than three-quarters of the surgeries, the computer's estimate was within forty-five minutes of the actual time the surgery took. The model identified three key factors as the most important drivers of how long the operation would last: whether the tumor had come back, how far the tumor was from the floor of a specific brain cavity called the third ventricle, and a specific visual sign during surgery known as the "cinch sign," which relates to how the tumor is constricted. The other model, the random forest, performed slightly worse, often underestimating the time needed for longer, more complex procedures.

The significance of this work extends beyond the numbers. The researchers found that by using this computer model, hospital administrators could move away from relying solely on a surgeon's subjective experience to make scheduling decisions. Instead, they would have an objective, repeatable estimate based on the specific characteristics of the patient in front of them. This allows head nurses to staff the operating room more effectively, ensuring that the right number of experienced nurses are present for difficult cases and that the recovery rooms are prepared for the actual arrival time of patients. It reduces the chaos of last-minute schedule changes and helps prevent the exhaustion that comes from unpredictable, extended shifts.

However, the authors are careful to note the boundaries of their findings. The study was conducted at a single hospital with a relatively small number of patients, and all the surgeries were performed by the same team of surgeons. This means the model learned the specific habits and techniques of that one group, and it might not work as well in a different hospital with different doctors or different patient populations. The researchers also acknowledged that the computer model, while accurate, acts as a "black box," making it difficult to explain exactly how it reached a specific conclusion. To address this, they suggest that future versions of the tool should include methods to make the computer's reasoning clearer to human users. Despite these limitations, the study demonstrates that machine learning can successfully capture the complex, non-linear relationships between a patient's anatomy and the time required for surgery.

Ultimately, this research offers a glimpse into a future where operating rooms run with the smoothness of a well-tuned machine, guided by data rather than guesswork. By accurately predicting how long a surgery will take, hospitals can better manage their most valuable resources: their people and their time. The path forward involves expanding this work to include more hospitals and more diverse surgical teams, and refining the models to include factors like the nursing team's experience. If successful, this approach could transform not just brain surgery, but the management of many different types of minimally invasive procedures, creating a clinical environment that is less stressful for the staff and safer for the patients.