Copy to clipboard 
Last week, orthopedic product development professionals filled an education room at OMTEC® 2026 for the conference’s R&D Symposium, three successive panel discussions on ways to improve how devices are designed, developed and brought to market efficiently and effectively.
The engaging and informative panelists detailed the essentials of design for manufacturing (DFM), navigating the complexity of today’s R&D teams and using AI-powered data to gain a competitive advantage, which is covered here. Look for recaps of the other sessions in the coming weeks.
Diving into the Data
Matt Shomper, the founder and Principal Engineer of Not a Robot Engineering, an engineering consultancy firm, moderated the session and began by defining the difference between traditional algorithms, which are deterministic mathematical models, and generative AI, the ability of machine learning to create new content, designs, recommendations or insights based on patterns learned from existing data.
“AI is a nice buzzword for the industry, but it misrepresents the nuances of the available platforms and how directly applicable they are to orthopedic product development,” Shomper said.
- Algorithms involve fixed rules and math, and no learning takes place
- Machine learning learns patterns based on existing data
- Generative AI creates new content, designs, recommendations or insights based on patterns learned from existing data
Shomper said understanding these distinctions is important, especially when evaluating the capabilities and limitations of the technologies. He added that much of the current innovation, investment and discussion within orthopedic product development involves generative AI.
One subset of AI and machine learning is predictive analytics, which involves training an algorithm on historical data to predict surgical outcomes for individual patients.
Josie Elwell, Ph.D., Manager of Research for Upper Extremities at Advita Ortho, referenced the company’s Predict+ clinical decision support tool that leverages predictive analytics and machine learning algorithms to deliver patient-specific outcome predictions for shoulder replacement surgery. The tool has collected 20 years of data based on the pre- and post-op condition of shoulder replacement patients and the implants they received.
“One of the biggest challenges in developing a product like this is planning ahead,” Dr. Elwell said. “We were fortunate to have years of data to train the algorithms.”
According to Oliver Richards, Ph.D., Managing Director at Accenture, other orthopedic companies don’t benefit from the same cache of available information, and few are fully prepared to leverage the large influx of available clinical data.
“Many of the organizations we work with have platforms in the operating room that generate procedural data,” he said. “Beyond that, economic, reimbursement and regulatory data streams add another layer of complexity. All of these sources contribute valuable information, but they often exist in separate systems.”
Data silos and disconnected information sources can be found across orthopedics. In many cases, according to Dr. Richards, that fragmentation becomes a barrier to bringing the information together in a meaningful way.
Shomper said data collection models are becoming commoditized, and because of that, the data itself is what creates a competitive advantage. Large OEMs increasingly find value in having accumulated massive datasets.
“The real differentiator is how you frame the problem, structure the data and apply the technology,” Shomper said. “What we’re seeing emerge are machine learning–based surrogate models and predictive tools, which can synthesize information more quickly and predict outcomes in ways that would otherwise require significant human expertise or extensive manual analysis.”
Navigating the Red Tape
Regulatory burdens remain a critical aspect of being able to tap into the full potential of AI in orthopedic product development applications. Companies must navigate complex approval processes and work with multiple regulatory bodies before these technologies can be deployed in clinical practice.
There’s no international consensus or standardized regulatory framework for products that incorporate AI and machine learning. Advita Ortho secured a CE Mark for Predict+, but FDA determined that no suitable predicate device existed and directed the company to complete a De Novo submission.
“We’re still navigating the regulatory pathway for these types of predictive tools,” Dr. Elwell said. “That’s not to say that predictive technologies have not been cleared by FDA — many have. In the case of Predict+, however, one of the challenges was that there is not yet a clearly defined regulatory boundary between a tool that supports clinical decision-making and one that drives clinical decision-making.”
According to Dr. Elwell, FDA is appropriately cautious about tools that could potentially influence treatment decisions without sufficient oversight from the surgeons using them. She said much of the tension in the current regulatory landscape revolves around ensuring that these products can be integrated safely into clinical workflows while maintaining appropriate physician involvement and accountability.
Capitalizing on the Benefits
The goal of incorporating AI into R&D efforts is not simply to create powerful predictive tools, but to ensure they fit safely and effectively within the broader clinical care pathway.
“We typically recommend identifying an initial use case that can serve as a starting point,” Dr. Elwell said. “Begin organizing and integrating data in a way that creates tangible value. Once that foundation is in place, organizations are in a much better position to scale their AI initiatives and realize practical applications across the business.”
Shomper is discouraged that even relatively simple algorithms are often being used to replace sound engineering judgment or clinical decision-making. He also acknowledged that questions remain surrounding data quality and the role it should play in informing, but not replacing, device design decisions.
AI has been viewed as having a human involved in the decision-making loop, but Dr. Richards prefers to think of it as a human-led technology.
“It should give people faster access to data and information so they can make better decisions, while ensuring they spend their time focused on the complex judgments that truly require human expertise,” he said.
Dr. Richards is focused on developing process-oriented use cases of AI, such as accelerating literature reviews during the early stages of the R&D process and tools that can quickly generate regulatory documentation and R&D specifications during product development workflows.
Another promising area of AI-driven product development focuses on improving feedback loops between the field and R&D teams.
“Sales representatives are constantly gathering insights from surgeons, but that information isn’t always easy to capture, analyze and share,” Dr. Richards said. “AI can make that feedback more accessible and actionable, allowing R&D teams to incorporate those insights into product development much more effectively.”
Those types of applications may be the easiest place to start because they’re focused on improving the speed and efficiency of R&D teams. They also avoid some of the regulatory and compliance challenges that become more complex when AI is used in clinical decision-making or patient-facing applications.
Preparing for What’s Next
Shomper highlighted the need to feed AI models high-quality data but also noted the importance of showing models what poor-quality data looks like. “As models move through the decision-making process, they need to learn not only which connections to strengthen, but also which ones to avoid,” he said.
If you’re using AI to help determine implant positioning for a shoulder replacement, for example, it needs extensive clinical data that clearly defines successful and unsuccessful outcomes.
Advances in algorithms and training techniques over the past five years have made AI models much better at handling imperfect data. The old saying was always “garbage in, garbage out.” Today, you can intentionally include examples of poor-quality data to teach the model what to avoid.
“That’s a significant advancement,” Dr. Elwell said. “Ultimately, though, the real question isn’t simply whether you have enough data. It’s whether you have enough of the right data.”
What’s increasingly important, according to Dr. Elwell, especially for predictive algorithms that will be incorporated into medical products, is fairness and bias. “If certain patient populations are underrepresented in the training data, the model’s predictions may be less accurate for those groups,” she said.
Most orthopedic companies are still in the experimentation or pilot phase when it comes to AI. They’re exploring how it can improve entire workflows and support decision-making across multiple stages of development.
How can AI create more value? How can it accelerate time to market? Can it shorten development cycles and speed product iteration? Finding answers to these questions will define what’s next in AI-driven R&D.
“Those broader process-level improvements are likely to drive the next wave of AI adoption in orthopedic product development,” Dr. Richards said.
