Nathan Xie1,2, Peter J Wilson3,4, Rajesh Reddy3,4. 1. School of Medical Sciences, University of New South Wales, Prince of Wales Private Hospital, High Street, Kensington, Sydney, 2052, Australia. nathanxie9@gmail.com. 2. Department of Neurosurgery, Prince of Wales Hospital, Sydney, Australia. nathanxie9@gmail.com. 3. School of Medical Sciences, University of New South Wales, Prince of Wales Private Hospital, High Street, Kensington, Sydney, 2052, Australia. 4. Department of Neurosurgery, Prince of Wales Hospital, Sydney, Australia.
Abstract
PURPOSE: The majority of lumbar spine surgery referrals do not proceed to surgery. Early identification of surgical candidates in the referral process could expedite their care, whilst allowing timelier implementation of non-operative strategies for those who are unlikely to require surgery. By identifying clinical and imaging features associated with progression to surgery in the literature, we aimed to develop a machine learning model able to mirror surgical decision-making and calculate the chance of surgery based on the identified features. MATERIAL AND METHODS: In total, 55 factors were identified to predict surgical progression. All patients presenting with a lumbar spine complaint between 2013 and 2019 at a single Australian Tertiary Hospital (n = 483) had their medical records reviewed and relevant data collected. An Artificial Neural Network (ANN) was constructed to predict surgical candidacy. The model was evaluated on its accuracy, discrimination, and calibration. RESULTS: Eight clinical and imaging predictive variables were included in the final model. The ANN was able to predict surgical progression with 92.1% accuracy. It also exhibited excellent discriminative ability (AUC = 0.90), with good fit of data (Calibration slope 0.938, Calibration intercept - 0.379, HLT > 0.05). CONCLUSION: Through use of machine learning techniques, we were able to model surgical decision-making with a high degree of accuracy. By demonstrating that the operating patterns of single centres can be modelled successfully, the potential for more targeted and tailored referrals becomes possible, reducing outpatient wait-list duration and increasing surgical conversion rates.
PURPOSE: The majority of lumbar spine surgery referrals do not proceed to surgery. Early identification of surgical candidates in the referral process could expedite their care, whilst allowing timelier implementation of non-operative strategies for those who are unlikely to require surgery. By identifying clinical and imaging features associated with progression to surgery in the literature, we aimed to develop a machine learning model able to mirror surgical decision-making and calculate the chance of surgery based on the identified features. MATERIAL AND METHODS: In total, 55 factors were identified to predict surgical progression. All patients presenting with a lumbar spine complaint between 2013 and 2019 at a single Australian Tertiary Hospital (n = 483) had their medical records reviewed and relevant data collected. An Artificial Neural Network (ANN) was constructed to predict surgical candidacy. The model was evaluated on its accuracy, discrimination, and calibration. RESULTS: Eight clinical and imaging predictive variables were included in the final model. The ANN was able to predict surgical progression with 92.1% accuracy. It also exhibited excellent discriminative ability (AUC = 0.90), with good fit of data (Calibration slope 0.938, Calibration intercept - 0.379, HLT > 0.05). CONCLUSION: Through use of machine learning techniques, we were able to model surgical decision-making with a high degree of accuracy. By demonstrating that the operating patterns of single centres can be modelled successfully, the potential for more targeted and tailored referrals becomes possible, reducing outpatient wait-list duration and increasing surgical conversion rates.
Authors: Johanna M van Dongen; Miranda L van Hooff; Maarten Spruit; Marinus de Kleuver; Raymond W J G Ostelo Journal: Eur Spine J Date: 2017-06-30 Impact factor: 3.134
Authors: André Wirries; Florian Geiger; Ahmed Hammad; Martin Bäumlein; Julia Nadine Schmeller; Ingmar Blümcke; Samir Jabari Journal: Biomedicines Date: 2022-06-04