Winner of the Sanford J Larson Award, Nathan Xie, presented his research, “Use of artificial intelligence to improve surgical referrals in degenerative lumbar spine conditions”, during the 2019 American Association of Neurological Surgeons Annual Scientific Meeting (AANS; 13–17 April, San Diego, USA).
Most referrals to spine surgeons for degenerative lumbar spinal conditions, such as low back pain, do not actually result in surgery. This means that the majority of referred patients may not necessarily require a spine surgery consult. Low surgical conversion rates are an issue in various countries around the world and have contributed to long waitlists to see spine surgeons, which are amongst the longest of all specialties. To illustrate this, two centres in Melbourne, Australia, reported wait times of over 1,200 days for the first consultation, with centres in Sydney not far behind. A possible strategy would be to evaluate referrals from primary care sources based on the likeliness that the patient will proceed to surgery and redirect those unlikely to proceed to non-operative treatment, allowing them to pursue alternative management strategies earlier and reducing waitlists as a result.
A novel way of achieving this would be through modelling surgeon decision-making using machine learning models (a branch of artificial intelligence). Researchers therefore aimed to identify factors that spine surgeons consider important for surgical decision-making and use these to develop an Artificial Neural Network (a machine-learning model developed that emulates the structure and interaction of biological neurons) able to calculate the probability that a patient would receive surgery at a given centre.
Fifty-five factors in the literature associated with surgical progression or outcome were identified. These factors were then collected through reviewing medical records of all patients presenting with an elective lumbar spine complaint between 2013 and 2018 at a single Australian Tertiary Hospital. The outcome variable was whether or not the patient proceeded or had been waitlisted for a surgical procedure. Using these data, an Artificial Neural Network with a back-propagation learning method was constructed, able to predict the likelihood of progression to spinal surgery. To compare it with a more traditional statistical model, a Logistic Regression (LR) model was created from the same data.
Ultimately both the neural network and regression models predicted surgical progression with a high degree of accuracy, although the neural network was superior in this regard. This demonstrates that the operating patterns of single centres can be accurately modelled, potentially allowing for more appropriate and tailored referrals. This, in turn, can help to reduce waitlists and increase surgical conversion rates. Further research is warranted in order to explore the role of such technology in the current system and its potential impacts.
