A Neural Network Predictor of Lumbar Fusion Surgery Outcomes

Abstract This work illustrates the potential utility of neural networks as a decision support tool for predicting long-term outcomes of elective lumbar fusion for chronic low back pain. Sixty variables were collected on each of 79 patients who had undergone elective lumbar fusion and had at least one year follow-up. A ten fold cross-validated training and test set were randomly created that demonstrated an R2 = 0.98 and 0.89 respectively. A traditional linear regression analysis resulted in an R2= 0.28. This neural network model should allow better patient selection for this very unpredictable and expensive surgery.