Artificial Intelligence as a Patient Resource for Information on Robotic Spine Surgery: An Assessment of Readability and Reach Among Leading Artificial Intelligence Natural Language Processors

Wednesday, February 21, 2024

Presenting Author(s)

Nicholas D. Cassimatis, n/a

Medical Student
Hackensack-Meridian School of Medicine
Nutley, NJ, US

Co-Author(s)

DA

David Adams, n/a

Medical Student
Hackensack-Meridian School of Medicine, United States
Daniel Davidar, MD

Research Fellow
Johns Hopkins University School of Medicine
Baltimore, Maryland, United States
Patrick Janeczko

Medical Student
Hackensack-Meridian School of Medicine
John Knightly, MD

Director of Neurosurgery
Atlantic Neurosurgical Specialists
Morristown, New Jersey, United States
Geoffrey R. O'Malley, Jr.

Medical Student
Hackensack-Meridian School of Medicine
Nutley, NJ, US
NP

Nitesh V. Patel, MD

Neurosurgeon
Hackensack Meridian Jersey Shore University Medical Center, Department of Neurosurgery
Red Bank, New Jersey, United States
PR

Patrick Roth, MD

Chairman of Neurosurgery
Hackensack University Medical Center, United States
Nicholas Theodore, MD, MS (he/him/his)

Director, Neurosurgical Spine Center; Professor of Neurosurgery
Johns Hopkins University School of Medicine
Baltimore, MD, US

Introduction: Patients often consult online resources to understand their surgery, especially with cutting-edge technology like robotic spine surgery. As artificial intelligence taps into vast web data, it offers potential to expand access to healthcare knowledge and provide essential advice. The authors hope to assess the ability of various Artificial Intelligence Natural Language Processors (AI-NLPs) to provide digestible patient-facing information on the risks and benefits of robotics in spine surgery.

Methods: AI-NLPs GPT 3.5, GPT 4.0, Bard, Bing, and Perplexity were asked about risks and benefits of robotic spine surgery. Readability analysis on AI-NLP answers was done using the Readable Application to determine Flesch-Kinkaid Reading Grade Level (FK-RGL) and percentage of addressable population that can be reached. AI-NLPs were prompted to phrase answers in a way that patients could understand. Current and relevant online patient-facing articles from the Mayo Clinic and New York Presbyterian served as controls for comparison.

Results: The Artificial Intelligence Natural Language Processors averaged an FK-RGL of 9.4 with a reach of 89.4% of the addressable population. Current patient-facing controls averaged an FK-RGL of 15.0 and 47.5% reach. Perplexity yielded the most patient-friendly results, with the lowest FK-RGL, at 7.9, and highest reach, at 100%. Compared to the current standard patient facing resources, the AI-NLPs averaged substantially lower reading grade levels (5.6 fewer FK-RGLs) with 88% more reach of addressable audience.

Conclusion : Patient education in the modern digital age relies heavily on online resources. Artificial intelligence resources, such as Perplexity, may be superior to current patient-facing resources, empowering patients and families with more accessible and comprehensible health information.