Skip to main content
  • Live Machine Learning Centered Architecture System to optimize Lumbar Stenosis Surgical Outcome

    Final Number:

    Ghaith Habboub; Joshua Johnston; Rod Nault; David Watson; Sebastian Salas-Vega; Michael P. Steinmetz; Thomas Mroz

    Study Design:

    Subject Category:

    Meeting: Congress of Neurological Surgeons 2018 Annual Meeting - Late Breaking Science

    Introduction: Patients with neurogenic claudication who undergo lumbar laminectomy have varying degree of success.1–4 Furthermore, it is difficult to identify patients who would benefit from surgery versus who would not. Our goal in this study is to optimize patient selection, create a parallel artificial-intelligence system which works seamlessly with the surgeon that helps with patient selection, and provide value-based care.

    Methods: We retrospectively reviewed 4121 cases of patients who underwent lumbar laminectomy at the Cleveland clinic hospitals between the years 2007-2017. Of those, we identified ~200-3300 patients who had enough data for the outcomes. Our objectives includes EuroQol five-dimensional (EQ-5D), back pain visual analogue scale (VAS), right leg VAS, left leg VAS, patient health questionnaire (PHQ9), pain disability questionnaire (PDQ), readmission, venous thromboembolism, costs, and reimbursements. We used Python for analysis relying on Tensorflow, Keras, XGBoost, and Scikit-Learn for machine-learning model creation.5

    Results: Our baseline success rate, defined by EQ-5D minimal clinically important difference (MCID), was 50%. Utilizing our system, we predict an improvement of at least to 75% in patient selection accuracy (50% improvement). This was validated retrospectively. We have analyzed >120 variables which included demographics, comorbidities, medications, laboratory, quality, and functional data. The data showed unaware associations between variables, such as laboratory values and indirect-compliance measures, and outcomes. This provided us an opportunity to provide adjustment goals to the modifiable variables for the predicted failed surgical patients to become surgical candidates. We also described how we embedded the system into electronic medical records and daily provider routine (Figures 1&2). We created a simulated cost saving analysis based on the above care path vs. the current system.

    Conclusions: With the rising healthcare expenditures, it is necessary to center patient management on value-based medicine. This system can offer us a 50% improvement in patient selection accuracy and lower costs; hence, value-based medicine.

    Patient Care: We believe by utilizing our system we can improve patient outcome while decreasing costs and providing a better value-based care.

    Learning Objectives: By the conclusion of this session, participants should be able to: 1. Understand how this system can improve the success rate by 50% for surgical management of patients with neurogenic claudication caused by lumbar stenosis utilizing machine learning and big data. 2. Understand how to design a framework to embed complex machine learning system into electronic medical record. 3. Understand how this system can provide value-based care by both optimizing outcome and minimizing costs.

    References: 1. Weinstein JN, Tosteson TD, Lurie JD, et al. Surgical versus Nonsurgical Therapy for Lumbar Spinal Stenosis for the SPORT Investigators* From the Departments of Orthopedics A bs t r ac t. 2008 [cited 2018 Jul 24];Available from: 2. Malmivaara A, Slätis P, Heliövaara M, et al. Surgical or Nonoperative Treatment for Lumbar Spinal Stenosis? A Randomized Controlled Trial. Spine (Phila Pa 1976) 32(1):1–8. 3. Amundsen T, Weber H, Nordal HJ, Magnaes B, Abdelnoor M, Lilleås F. Lumbar Spinal Stenosis: Conservative or Surgical Management? A Prospective 10-Year Study. Spine (Phila Pa 1976) 25(11):1424–36. 4. Zaina F, Tomkins-Lane C, Carragee E, Negrini S. Surgical versus non-surgical treatment for lumbar spinal stenosis. Cochrane Database Syst. Rev. 2016; 5. Chen T, Guestrin C. XGBoost?: Reliable Large-scale Tree Boosting System. arXiv 2016;1–6.

We use cookies to improve the performance of our site, to analyze the traffic to our site, and to personalize your experience of the site. You can control cookies through your browser settings. Please find more information on the cookies used on our site. Privacy Policy