A Predictive Model for Assessment of Successful Outcome in Posterior Spinal Fusion Surgery

Background: Low back pain is a common problem in many people. Neurosurgeons recommend Posterior Spinal Fusion Surgery (PSF) as one of the therapeutic strategies to the patients with low back pain. Due to the high risk of this type of surgery and the critical importance of making the right decision, accurate prediction of the surgical outcome is one of the main concerns for the neurosurgeons. Methods: In this study, 12 types of Multi-Layer perceptron networks (MLP) and 66 Radial Basis Function (RBF) networks as the types of artificial neural network methods and a Logistic Regression model created and compared to predict the satisfaction with PSF surgery as one of the most well-known spinal surgeries. Results: The most important clinical and radiologic features as twenty-seven factors for 480 patients (150 males, 330 females; mean age 52.32 ± 8.39 years) were considered as the model inputs that included: age, sex, type of disorder, duration of symptoms, job, walking distance without pain, walking distance without sensory disorders, visual analog scale scores, Japanese Orthopaedic Association score, diabetes, smoking, knee pain, pelvic pain, osteoporosis, spinal deformity and etc. The indexes such as receiver operating characteristic–area under curve (ROC-AUC), positive predictive value, negative predictive value and accuracy calculated to determine the best model. Postsurgical satisfaction was 77.5% at 6 months follow-up. The patients divided into the training, testing, and validation data sets. Conclusion: The findings showed that the MLP model performed better in comparison with RBF and LR models for prediction of PSF surgery.