Spine surgery has become remarkably good at measuring what it can see.
Surgeons can quantify alignment, study images in extraordinary detail and plan increasingly precise operations. But one of the most consequential questions in spine care remains far less exact: Which patient will actually be better after surgery, and by how much?
Alpesh Patel, MD, an orthopedic spine surgeon and co-director of the Northwestern Center for Spine Health at Chicago-based Northwestern Medicine, believes the field has considerably more work to do.
“We’re probably not as good as we think we are, across the profession, at truly predicting who’s going to improve and not improve,” Dr. Patel told Becker’s.
Even that question may be too simple. A patient can technically improve and still fall far short of what was expected.
“How much better is where we start to really struggle,” he said. “And then another real struggle is: Are we meeting people’s expectations?”
For Dr. Patel, that gap is becoming one of spine surgery’s most important opportunities for predictive analytics. But getting there will require more than feeding larger datasets into increasingly powerful algorithms. It will require the field to reconsider what it measures, what it overlooks and what role it ultimately wants AI to play in clinical judgment.
Surgeons may be overvaluing what they can measure
One reason predicting outcomes remains so difficult is that the result of a spine operation is shaped by far more than the spine itself. A patient’s baseline function, expectations, medical condition and social environment all matter. So do the surgeon, hospital, postoperative care and the underlying spinal pathology.
Faced with that complexity, Dr. Patel believes surgeons naturally gravitate toward the variables they understand best. In spine, that often means the image.
“We spend a lot of time measuring angles and measuring malalignment of the spine,” he said. Those measurements matter. But Dr. Patel cautions against assuming that because radiographic data are precise, they are necessarily the dominant predictors of how a patient will ultimately feel.
“It’s an important part, but it’s probably a relatively small component of what predicts the ultimate outcome of a patient,” he said.
The information readily available can get more weight, while factors that are difficult to quantify risk being discounted.
That distinction may become even more important as AI gives surgeons the ability to analyze far more information than before. More data, Dr. Patel argues, does not automatically produce better decisions.
More data can also mean more noise
Much of healthcare’s existing data infrastructure was not originally designed to predict whether an individual patient will thrive after spine surgery.
Readmissions, complications and reoperations are comparatively easy to identify because they appear in claims, billing records and administrative datasets. Patient expectations, functional recovery and the many clinical and social factors surrounding an outcome are harder to capture systematically.
So researchers often study what is easiest to retrieve.
“You’ll look at billing records for, say, 200,000 patients and work backward to infer clinical outcomes based on what was billed or coded,” Dr. Patel said. “That’s probably not the best way to do it, and it creates a lot of noise.”
Northwestern has been working in the opposite direction.
Dr. Patel said his team has spent 12 to 13 years collecting patient-reported outcomes among elective spine surgery patients while also building predictive models around outcomes such as infection, readmission and reoperation. The longer-term goal is to create enough structured longitudinal data to better predict not simply whether something went wrong, but how much benefit a patient is likely to experience.
That requires a fundamentally different approach to data. Instead of asking an algorithm for an answer first, Dr. Patel believes spine researchers need to spend more time deciding what information belongs in the model in the first place.
“More data does not mean better,” he said. The difference between useful prediction and sophisticated noise, he said, comes down to rigorous data collection, thoughtful architecture and understanding what happens between an algorithm’s inputs and outputs.
That presents another challenge for surgeons.
“Most practicing surgeons, spine surgeons, orthopedic surgeons, neurosurgeons, in 2026, don’t have a rigorous amount of training in data science,” he said. For his group, that has made collaboration with Northwestern’s data scientists, graduate students and researchers essential.
The problem with a ‘plug-and-play’ prediction model
There is also a tension between accuracy and scalability. Dr. Patel said clinicians tend to have the most confidence in predictive models developed from their own patients, hospitals and practices.
The reason is intuitive: They know where the data came from. But a model trained in one institution may perform differently somewhere else, where patients, surgeons, resources and demographics are different.
“Our results at hospital A may not translate very much to hospital B,” he said. Healthcare’s business and operational sides may favor a generalized model that can be deployed broadly. But Dr. Patel is wary of prioritizing scalability when a prediction could influence whether a patient undergoes spine surgery.
“The stakes are so high for our patients that I don’t think that should be the goal,” he said.
Instead, he sees greater promise in deep, carefully structured datasets that capture outcomes across institutions and patient populations. Spine registries have evolved from retrospective repositories toward prospective, multi-institutional efforts capable of collecting the richer clinical, radiographic and patient-reported information those models need.
The lesson, Dr. Patel said, is to resist racing toward the answer before building the information needed to trust it.
“We probably could spend more time on the front end,” he said, “instead of just trying to get to the, ‘Tell me the end result. Tell me what the answer is.’”
AI should make judgment deeper. It shouldn’t remove it.
That leads to perhaps the biggest question surrounding predictive analytics in spine: What should happen when the algorithm and surgeon begin making decisions together?
Dr. Patel sees two extremes. On one side is the idea that sufficiently sophisticated algorithms will eventually outperform the inconsistency of human decision-making.
On the other is the experienced surgeon whose judgment has been refined across thousands of encounters, an implicit predictive model developed over an entire career.
Neither is easily transferable. The near-term future, Dr. Patel believes, lies between them.
Rather than having an algorithm decide what operation a patient should receive, he sees greater value in tools that give physicians deeper insight into the decisions they are already considering.
For example, a system could analyze a proposed surgical plan and show the surgeon three things: the likely intended consequences, potential unintended consequences and the elements the model simply cannot predict.
That last category matters. There will still be unknowns. Instead of disguising that uncertainty behind a recommendation, a useful predictive system could make it visible to both surgeon and patient.
AI tools that give physicians deeper insights into their decisions, Dr. Patel said, could be more readily accepted and implemented than systems attempting to make those decisions themselves.
In that model, AI does not eliminate judgment. It gives judgment another source of feedback.
Northwestern is already using the human version of that model
Dr. Patel sees an analogue in something Northwestern’s spine team already does without an algorithm.
Every week, the orthopedic and neurosurgical spine surgeons review cervical and lumbar fusion cases together. The first question is fundamental: Is this the right procedure for this patient?
Then comes the collective experience of the group. A surgeon may present a limited fusion for a patient with a larger deformity.
Another surgeon may point out a potential downstream consequence. Someone else may suggest an alternative.
The value is not necessarily that one surgeon possesses the “correct” answer. It is that the original decision is exposed to more information before becoming final.
“Our patients love the idea that all of our surgeons are looking and giving feedback,” Dr. Patel said.
He sees the same principle as essential for AI. Feedback forces clinicians to confront considerations they may have missed. Without it, even highly trained surgeons can become isolated inside their own decision-making.
“It’s really easy, even in a large academic system, for an individual physician or surgeon to become siloed off if we’re not open to those channels,” he said.
Practice structures without built-in peer review may face an even greater challenge.
“People can get blinders on really quickly and not see the unintended consequences of what they’re doing,” Dr. Patel said. Predictive analytics, at its best, could become another way of removing those blinders.
Better prediction could change far more than the operation
The payoff would extend beyond determining whether a fusion succeeds. Better patient selection could change when patients enter the surgical pathway, how long they remain in treatments unlikely to help and how much unnecessary care they receive before reaching the right intervention.
Dr. Patel pointed to patients with spinal stenosis who may spend six to eight months in nonsurgical management without improving. With better prediction, some could reach effective treatment sooner.
“Maybe we can get that patient cared for faster, get them back to their work, family, whatever it might be, much faster,” he said.
The inverse matters just as much. If predictive tools can identify treatments unlikely to add value, patients could avoid unnecessary injections, operations and other costly interventions.
That has implications for patients, surgeons and payers, and could help rebuild something harder to quantify in spine surgery: trust.
Not every lumbar fusion fails. Not every patient needs one. The challenge is knowing the difference before treatment begins.
“Rather than trial and error and hope it works,” Dr. Patel said, “maybe we can get a little bit more prescriptive.”
Spine surgery has become extraordinarily sophisticated at answering how to perform an operation. Predictive analytics could help answer the questions that come first: Should this patient have an operation at all, and what can they realistically expect if they do?
At the Becker’s 32nd Annual Meeting: The Business and Operations of ASCs, taking place October 29-31 in Chicago, ASC leaders, surgeons and healthcare executives will explore strategies to drive growth, enhance operational performance, navigate reimbursement challenges and prepare for the future of ambulatory surgery. Apply for complimentary registration now.
