Neural Network-assisted Prediction of Venous Thromboembolism after Spine and Neurological Operations - NeuroVeT Study
Sanjana Avajigari1, Shradha Kakde2, Chirag L sagar3, Harshawardhan Dhanraj Ramteke4, Pratiksha Baliga5, Suchita Mylavarapu6, Ramya Manojna hota7, Syed Faiz ahmed8, Ahmed harb9, Sharath Chandra Anne10, Anas Mansour9, Rakhshanda Khan11, Meghana Chennupati12
1govt medical college siddipet, 2MGM medical college and Research centre, 3kasturba medical college, 4Anhui medical university, 5MGMIHS, 6mallareddy medical college for women, 7andhra medical college, 8deccan college of medical sciences, 9Faculty of Medicine, Al-Azhar University, Cairo, Egypt, 10pinnamaneni siddharta medical college, 11Ayaan Institute of medical sciences, 12Mamata Academy of Medical Sciences
Objective:
To develop and evaluate the predictive performance of a Deep Neural Network (DNN) model for identifying postoperative venous thromboembolism (VTE) risk following spine surgery using routinely collected clinical and operative data.
Background:
Venous thromboembolism (VTE) is a rare but serious complication after spine surgery, contributing to morbidity and mortality. Traditional risk assessment tools rely on static clinical parameters and may not capture complex, nonlinear relationships among patient variables. Advances in artificial intelligence, particularly deep learning, offer the potential to enhance predictive accuracy and enable personalized risk stratification. This study investigates whether a Deep Neural Network (DNN) can effectively predict postoperative VTE occurrence based on multidimensional clinical datasets
Design/Methods:
A retrospective analysis of 4,972 spine surgery patients was conducted. Data were divided into training (n = 1,332) and testing (n = 3,640) cohorts. Structured preoperative, intraoperative, and postoperative variables were input into a multilayer DNN classifier. Model performance was evaluated using metrics including Area Under the Curve (AUC), accuracy, F1 score, precision, and recall. The primary outcome was the model’s ability to distinguish VTE versus non-VTE cases in the test dataset.
Results:
The DNN demonstrated strong discriminative ability with an AUC of 0.92, accuracy of 0.96, F1 score of 0.98, and recall (sensitivity) of 0.98, indicating minimal missed VTE cases. While precision was 0.78, the combination of high recall and high overall accuracy underscores the robustness of the model in clinical prediction settings
Conclusions:
The DNN model achieved excellent predictive performance for postoperative VTE, outperforming traditional statistical approaches. Its ability to identify high-risk patients with high sensitivity supports its clinical utility as a decision-support tool for individualized thromboprophylaxis planning following spine surgery
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.