Artificial Intelligence Can Redefine Value-Based Spine Care. It’s Already Starting To

Rationale for our adoption of AI-guided care, plus insights from early experience


By Thomas Mroz, MD, and Ghaith Habboub, MD


Cleveland Clinic is a non-profit academic medical center. Advertising on our site helps support our mission. We do not endorse non-Cleveland Clinic products or services. Policy

As the exponential expansion of computing capacity converges with unsustainable healthcare spending, a promising opportunity has emerged: The use of artificial intelligence (AI) to enhance healthcare value.

At Cleveland Clinic and elsewhere, clinicians and data scientists are taking steps to leverage AI to improve patient outcomes while reducing healthcare costs, particularly among candidates for surgical care. An AI-driven decision platform is under development in Cleveland Clinic’s Center for Spine Health, and we recently reported an analysis indicating that use of this AI platform across a historical cohort of lumbar laminectomy patients at Cleveland Clinic would have improved the cohort’s clinical success rate by 50 percent and lowered total treatment costs.

This article reviews the rationale for our embrace of AI to help guide spine care decision-making, details of our recent analysis and next steps in our use of AI.

Standardization of spine care remains elusive

Today computing capacity advances more in a single hour than it did in the first 90 years of the information age. By the year 2045, computing capacity is expected to exceed the cognitive ability of all human brains on earth combined, with similar increases in the amount of data generated.

Despite having so much data at our fingertips, vast discrepancies remain in how patients are treated for various medical conditions. Perhaps no therapeutic area shows more heterogeneity in the delivery of surgical and nonsurgical care than spinal disorders. For instance, recent national surveys of U.S. spine surgeons conducted by our center found high rates of disagreement on the management of two common spinal conditions: 69 percent disagreement for recurrent lumbar disk herniations and 75 percent for lower back pain.1,2

The reasons for variation are many — including differences in surgeon experience and training, differences in use of pre- and postoperative resources, and a slew of patient specific-factors — but the bottom line remains that variations in care for common conditions generally undercut the quality of care while driving up its cost.3

While the spine care community has a wealth of knowledge in the medical literature, it is impossible for practicing physicians or surgeons to reconcile in real time all the data that will ultimately determine the most efficacious and cost-effective treatment for a particular patient.


Enter artificial intelligence

Contrast this with AI. By analyzing millions of discrete data points housed in electronic medical records and financial databases, AI can complete in milliseconds — and with greater precision — what the human mind accomplishes in hours or days.

Based on past and repeated performance, AI has the capacity to render with high probability the best decisions on surgery or nonoperative treatments for optimal patient outcomes within a given cost and reimbursement model. AI also can suggest, if necessary, an alternative provider in the same healthcare system who would likely perform better on a particular patient.

The implications for clinical practice are staggering. An AI platform would be the first legitimate clinical decision-making tool in spine medicine, delivering on the value equation while serving as a resource for improving physician performance and promoting appropriate, efficient care and in this era of healthcare financial uncertainty.

An AI-driven platform of this kind, designed to seamlessly support the surgeon in patient selection and choice of treatment approach, is under development at Cleveland Clinic. By collecting extensive historical data on spine patients in a database and analyzing them, we are identifying many important and previously unrecognized variables that are collectively contributing to optimal patient outcomes. The basic architecture of the model is outlined in Figure 1.

Figure 1. Schematic showing the five sequential layers of the project architecture. The model eventually “feeds” itself, enabling continuous machine learning.

Intriguing results from an early analysis

An early glimpse of the platform’s possibilities comes from an analysis we presented at the 2018 annual meeting of the Congress of Neurological Surgeons.4 We retrospectively reviewed the cases of all patients who underwent lumbar laminectomy at Cleveland Clinic hospitals from 2007 through 2017. Approximately 3,300 of these patients qualified for inclusion in the study because they had sufficient data for analysis of the following outcomes of interest: functional outcome as assessed by EQ-5D™, visual analog scale scores for back pain/right leg pain/left leg pain, depression as assessed by the Patient Health Questionnaire-9, the Pain Disability Questionnaire score, readmissions, venous thromboembolism incidence, treatment costs and reimbursements.

We fed data on more than 120 variables from across this 3,300-patient cohort into our AI platform. Variables included aspects of patient demographics, comorbidities, medications, laboratory test results and functional data, as well as variables related to surgeon quality. Python software was used for data analysis, and TensorFlow™, Keras, XGBoost and scikit-learn were used for machine-learning model creation.


Of the 3,300 patients included in the study, approximately 50 percent had success, defined as improvements meeting the threshold of “minimal clinically important differences” in the abovementioned healthcare metrics. The AI system predicted that the success rate would have been improved to at least 75 percent — i.e., a relative improvement of 50 percent — if the actual clinical decision-making around surgical candidates, procedures and individual surgical operators had been supplemented with AI guidance. An additional simulated cost analysis demonstrated cost savings of up to $25,000 per case with AI-supplemented decision-making versus standard care without AI support.

Data from the analysis revealed some associations between variables that we would not have otherwise expected, such as between lab test values and indirect compliance measures and outcomes. This has important implications for the setting of targets for modifiable variables used to determine a patient’s appropriateness for surgery.

Next steps

We have embedded elements of the AI platform into the electronic medical record and daily provider routines via predictive displays and prompts. We look forward to reporting more on our experience with implementation of the platform — in additional areas of spine care — in the months and years ahead.

This early experience with our AI platform encourages us that this system — when coupled with human expertise and personal, high-touch caregiving — can improve patient outcomes in spine care while reducing costs. AI in spine care has great potential to promote standardization of practice around high-quality care as well as more standardized resource utilization. It likewise promises more logical and successful strategies for managing large patient populations with more predictable outcomes and expenditures — in other words, the enhanced healthcare value that everyone is seeking.


  1. Mroz TE, Lubelski D, Williams SK, et al. Differences in the surgical treatment of recurrent lumbar disc herniation among spine surgeons in the United States. Spine J. 2014;14:2334-2343.
  2. Lubelski D, Williams SK, O’Rourke C, et al. Differences in the surgical treatment of lower back pain among spine surgeons in the United States. Spine (Phila Pa 1976). 2016;41:978-986.
  3. Ugiliweneza B, Kong M, Nosova K, et al. Spinal surgery: variations in health care costs and implications for episode-based bundled payments. Spine (Phila Pa 1976). 2014;39:1235-1242.
  4. Habboub G, Johnston J, Nault R, Watson D, Salas S, Steinmetz M, Mroz T. Live machine learning-centered architecture system to optimize lumbar stenosis surgical outcome [abstract]. Presented at: 2018 Congress of Neurological Surgeons Annual Meeting; Houston, Texas; Oct. 6-10, 2018.

Dr. Mroz is Director of Cleveland Clinic’s Center for Spine Health. Dr. Habboub is a resident in the Department of Neurosurgery.

Related Articles

New VR Platform Fuses Physical and Virtual Worlds in Parkinson’s Disease and Beyond

Taking virtual reality-integrated technology from silver screen to clinical laboratory

20-NEU-1990658 NeuroDesign Innovation fellowship_CQD_650x450_896358708
‘NeuroDesign’ Fellowship Aims to Shape Next Generation of Neurosurgery Entrepreneurs

Novel collaboration is underway to foster innovation – and a real-world invention

Deep Learning Models for Automatic Seizure Detection in Epilepsy

Strong performance from early models heralds eventual reshaping of care

Cerebrovascular Roundup: Breakthroughs in Treating Brain Aneurysms

A quick review of 3D-printed models, intrasaccular flow disruption and flow diverter stenting

AMI-Tuck-1642295     06-25-19
Therapeutic Arts Program Aims to Make Patients Resilient in the Face of Multiple Sclerosis

Early results with ‘HeRe We Arts’ spur testing in a randomized trial

3D-Printed Replica of Brain Aneurysm Helps Guide Surgical Repair

One of first reported uses of the technology for a cerebrovascular malformation