AI-driven model helps surgeons predict complex spinal deformity surgery before the operation, Chinese multicenter study shows

A new multicenter study conducted across China has developed a machine learning–based system capable of predicting the difficulty and expected duration of posterior spinal deformity correction surgery before the procedure begins. The model, built using routinely available clinical and laboratory data, may help surgeons better plan complex operations and optimize operating-room resources.

Spinal deformities represent a growing global health burden, particularly as populations age. Severe cases often require surgical correction, but these procedures are technically demanding and can involve multiple vertebrae and delicate neural structures. Because surgical duration often reflects the complexity of the procedure, accurately predicting operative time is considered an important step in improving perioperative planning and patient management.

To address this challenge, researchers analyzed clinical data from 162 patients undergoing posterior spinal deformity surgery at 11 hospitals across China between 2015 and 2022. The team collected a wide range of preoperative clinical variables and applied machine learning techniques to identify factors associated with prolonged operative time, defined in the study as surgeries lasting more than four hours.

Using the Boruta feature-selection algorithm, the researchers identified seven key predictors that most strongly influenced surgical duration. These included the interval between disease onset and surgery, fibrinogen levels, white blood cell count, activated partial thromboplastin time (APTT), American Society of Anesthesiologists (ASA) classification, albumin levels, and patient age.

The team then trained and compared ten different machine learning algorithms, including logistic regression, random forest, support vector machines, and gradient boosting models. Among them, the Light Gradient Boosting Machine (LightGBM) model demonstrated the most consistent predictive performance, achieving an area under the receiver operating characteristic curve (AUC) of approximately 0.70 in identifying cases likely to exceed the four-hour surgical threshold.

Beyond predictive accuracy, the researchers emphasized model interpretability—an essential factor for clinical adoption. To better understand how each variable influenced predictions, the team used SHAP (Shapley Additive Explanations) analysis, which allowed them to visualize how individual patient characteristics contributed to predicted surgical duration. This approach helped clarify the internal decision-making logic of the machine learning model and highlighted potential interactions among clinical variables.

Additional analyses, including calibration testing and decision curve analysis, suggested that the model could provide meaningful clinical benefit when applied in preoperative decision-making. The predictive system was further translated into a web-based calculator, enabling clinicians to input routine patient information and obtain an individualized estimate of surgical complexity and operative duration.

According to the authors, such tools could assist surgeons in identifying high-risk cases earlier, improving surgical scheduling, and optimizing operating-room resource allocation. By providing more accurate expectations for procedure length, the system may also support better communication with patients and more efficient perioperative care.

Although further external validation and prospective studies will be needed, the researchers suggest that integrating machine learning models into clinical workflows may represent an important step toward data-driven surgical planning. As artificial intelligence tools become increasingly interpretable and accessible, similar predictive systems could play a growing role in guiding complex surgical decision-making.