Published by: Ray Guzman, CEO, WPC Healthcare
Machine Learning to Assist Back Surgery Decision-making
Whether you’re a long-haul trucker or a weekend tennis jock, there’s a good chance you’ve experienced back pain. Maybe your pain was so severe it led you to an orthopedic surgeon’s office. Watch out!
As a recent study details, surgery is what surgeons do—as in the old adage, if you’re a hammer, everything looks like a nail. And for the many patients who have had spinal fusion or other surgeries to alleviate back pain, surgery may not have improved outcomes. Moreover, those who chose alternative therapies for pain (such as yoga, swimming, and physical therapy) may have achieved better results.
While most of us can agree that medical diagnosis is as much art as science, the steady uptick in back surgery over the last 20 or so years clearly raises some questions.
For the suffering patient and the sometimes-baffled physician, it can be difficult to determine when surgical intervention will truly make a difference. Current diagnostic methods include the use of assessment tools and some level of automation that, while efficient and somewhat effective, allow for errors and force providers to review more causes than necessary.
Computer-aided diagnostics incorporating data science and machine learning can help. Though not intended to replace a physician’s diagnostic skills, adding data science to the diagnostic process offers more precise and accurate measurements faster and with fewer false-positives and false-negatives.
Data Science Challenges in Medicine
In the data science process, developing algorithms is best when the size of datasets is fairly large. In medicine, HIPAA and other patient privacy protections make it difficult to gather and combine datasets. Despite these challenges, the Data Science Team at WPC Healthcare decided to take a crack at creating a spinal algorithm to help diagnose pathologies of the spine by leveraging a unique data strategy.
The result—a medical decision-support framework consisting of three subsystems: feature engineering, feature selection, and model selection. The framework generates a model for classifying new observations. The data from that model are refined through an automated feature selection process, thus yielding enhanced prediction accuracy.
Leveraging this framework, the predictive accuracy increased significantly to almost 100 percent in terms of identifying the correct underlying pathology. Physicians using this framework can have greater confidence in the diagnoses provided to their patients and thus can make recommendations for the appropriate intervention.
Data scientists are often humbled by biology. Physicians are just as humbled by how effective data science can be in medical decision support by offering them the experience of hundreds of their peers nationally and globally. By taking a multidisciplinary approach to pathologies of the spine, WPC data scientists may have given providers a better set of tools to identify the nails that truly need a hammer, and those that just might need something less invasive.
To read more about this model, check out this journal article: WPC Spinal Algorithm.