Orthopedic Surgeons Likely to Trust Risk Stratification Models Driven by AI
Key findings
- Spine surgery research fellows in the Department of Orthopaedic Surgery at Massachusetts General Hospital conducted the first study of physician opinion about risk prediction models powered by artificial intelligence (AI)
- 31 of 90 surveys were returned; 14 respondents were orthopedic surgery residents and 17 were board-certified orthopedic surgeons
- 87% of respondents said they are likely or very likely to use AI-derived risk prediction models today, and 58% considered it likely or very likely that such models will be integrated into their practice within five years
Factors that increase the risk of complications after spine surgery (e.g., advanced age, female sex and prolonged surgical time) are well known, but they provide limited information about an individual patient's risk. Risk scoring systems provide some patient-level data, but their reliability is often unclear and calculations can be complex or require data unavailable at the time of consent.
Subscribe to the latest updates from Orthopaedics Advances in Motion
Prediction models derived from machine learning—a form of artificial intelligence (AI)—add a quantitative component to shared decision-making and are meant to improve in accuracy as they compute new information. However, these models and many other AI applications have yet to be implemented in routine clinical care. Physicians may be hesitant to use AI due to skepticism, lack of understanding or even concern about being replaced.
Amanda Lans, MD, MS, and Mitchell S. Fourman, MD, spine research fellows in the Department of Orthopaedic Surgery at Massachusetts General Hospital, and colleagues, conducted the first study of physician opinion about risk prediction models powered by AI. According to the team's report in Seminars in Spine Surgery, the majority of respondents say they are likely or very likely to trust such models.
Study Methods
The researchers sent a five-item survey to 90 physicians at Mass General, of whom 31 responded (response rate 34%). 14 respondents were orthopedic surgery residents and 17 were board-certified orthopedic surgeons.
Results
On a five-point Likert scale, most physicians responded "likely" or "very likely" to the questions:
- How likely are you to trust risk prediction models based on AI? —74%
- Assuming you trust the data source, how likely are you to currently use risk prediction models in your practice today? —87%
- Assuming you trust the data source, how likely are you to change your management when a risk prediction model disagrees with your clinical judgment? —55%
- Assuming you trust the data source, what is the likelihood you would use risk prediction models if these were integrated into the electronic health care system? —87%
- How likely is it that trusted risk prediction models based on AI will be integrated into your practice within the next five years? —58%
No respondent chose "very unlikely" for any question. The researchers note the response rate probably reflects a bias toward those more likely to adopt advanced modeling, but they find it encouraging that the responses did not differ by level of training.
The authors also review the most likely applications of AI-assisted decision-making support for patients and barriers to clinical implementation of AI-driven technology.
view original journal article Subscription may be required
Learn more about the Spine Service
Learn more about the Spine Surgery Fellowship