AI is moving quickly into spine care, from ambient documentation and surgical planning to predictive analytics, triage and revenue cycle.
Some early results are significant. One spine practice reported doubling clinic throughput with algorithmic triage, while orthopedic groups are using AI to reduce prior authorization denials and documentation burden.
But the technology has not eliminated many of the problems it was expected to help solve. In some cases, AI is changing the form of the problem rather than removing it.
Recent Becker’s reporting points to seven gaps that remain:
1. Prior authorization is still slowing care: AI is increasingly being used on both sides of the authorization process. Practices are using it to identify missing documentation before submission, while payers are using algorithms to evaluate whether cases meet coverage requirements.
William Kemp, MD, a spine surgeon in Richmond, Va., told Becker’s insurers are using AI to check physical therapy documentation and imaging requirements. He questioned whether the technology is reliable enough for those decisions.
“I think it’s actually slowing down patient care, which is the exact opposite of what AI should be doing,” Dr. Kemp said.
The counterresponse is increasingly more AI. Duke Orthopedics, for example, has incorporated AI into its own prior authorization workflow and reported better approval rates and fewer peer-to-peer reviews. The result suggests AI can help practices fight payer friction, but it has not removed the underlying authorization burden.
2. Spine still cannot reliably predict who will benefit from surgery: Patient selection remains one of the specialty’s most consequential unanswered questions.
Alpesh Patel, MD, an orthopedic spine surgeon and co-director of the Northwestern Center for Spine Health at Chicago-based Northwestern Medicine, told Becker’s that the field remains limited in its ability to predict not only whether patients will improve after surgery, but by how much.
“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,” he said.
AI could eventually sharpen those predictions, but the problem is more complicated than adding more data. Patient expectations, baseline function, medical conditions, social factors and postoperative care can all influence an outcome, and many are harder to capture than imaging or claims data.
“More data does not mean better,” Dr. Patel said.
3. The data still does not travel well: Even an accurate predictive model at one hospital may not work the same way somewhere else. Kevin Huang, MD, a neurosurgeon at Boston-based Massachusetts General Hospital, told Becker’s that AI’s potential in surgical risk prediction remains tied to the quality and completeness of the information used to train it.
“These things are only as good as what you put into them,” Dr. Huang said.
Dr. Patel has raised a related problem: generalizability. A model built on one institution’s patients, surgeons and resources may perform differently when deployed in another environment. “Our results at hospital A may not translate very much to hospital B,” he told Becker’s.
That makes a simple “plug-and-play” model difficult in a specialty where predictions could influence whether a patient undergoes surgery.
4. AI still makes mistakes that require human review: Documentation is one of AI’s clearest early wins in spine care, but even relatively mature applications still require oversight.
Noam Stadlan, MD, of Evanston, Ill.-based Endeavor Health Neurosciences Institute told Becker’s AI is useful for office notes and research, but said results still need to be checked because of the risk of hallucinations, in which information can be inaccurate or manufactured.
The reliability issue becomes substantially more consequential as AI moves from documentation toward patient selection, surgical planning and clinical decision-making.
Jeffrey Carlson, MD, an orthopedic spine surgeon at Orthopaedic & Spine Center in Newport News, Va., has cautioned against surgeons becoming too reliant on AI in the operating room, where glitches, incorrect information or power failures can force the surgeon to proceed without it.
“AI in the operating room is overhyped,” Dr. Carlson told Becker’s.
5. AI cannot replace the judgment behind a surgical decision: AI can quantify risk, identify patterns and generate recommendations. It still cannot resolve every tradeoff that determines whether a particular operation is right for a particular patient.
Complex spine surgery can involve factors that resist easy quantification, including functional goals, willingness to accept risk, family support and patient expectations. James Mooney, MD, a neurosurgeon at VCU Health in Richmond, Va., wrote in Becker’s that unexpected anatomy, bleeding or neurological changes also require surgeons to adapt in real time.
Matthew Harb, MD, an orthopedic surgeon with The Centers for Advanced Orthopaedics in Washington, D.C., similarly told Becker’s that the hype outpaces reality when it comes to AI replacing surgical decision-making.
For now, he said, AI is best viewed as an adjunct to physician judgment rather than a replacement for it.
6. Buying AI does not guarantee ROI: The expanding number of AI products creates another challenge for practices: determining which tools actually remove work rather than add another system to manage.
John Bring, vice president of clinical innovation and performance at Omaha, Neb.-based Sequel Ortho, told Becker’s that successful adoption depends on measurable workflow improvement and continued human oversight. “If someone feels the tool is creating work rather than reducing it, adoption will fail,” he said.
Disconnected point solutions are another risk. Mr. Bring said practices can buy technology that looks impressive during a demonstration only to end up six months later with “another disconnected application, another login and limited adoption.”
His advice was simple: “Don’t buy AI. Buy outcomes.”
7. AI can create capacity, but it cannot create spine surgeons: One of AI’s strongest opportunities may be helping scarce clinicians use their time more efficiently.
Michael Verdon, DO, a spine surgeon at Dayton Neurologic Associates in Beavercreek, Ohio, told Becker’s his practice’s algorithmic triage system doubled clinic throughput and increased surgical volume about 10% by directing patients to the appropriate level of care earlier.
But the underlying access problem remains. Lali Sekhon, MD, PhD, of Reno (Nev.) Orthopedic Center, said his community has fewer spine surgeons than when he arrived in 2005. At the time of his January interview, his next available appointment was roughly two months away.
“Access is going to become one of the bigger issues,” Dr. Sekhon told Becker’s. He expects AI to help fill parts of the gap, particularly with routine care, scheduling and information management.
The distinction may define AI’s near-term role in spine. It can route patients more efficiently, reduce documentation work, flag risk and give physicians greater capacity. It does not eliminate the workforce, payer, data and clinical judgment problems underlying that work.
For now, AI’s biggest contribution to spine may be helping surgeons manage the system they already have, rather than replacing the parts of it that remain broken.
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.
