The following is a summary of “GPT for Pediatric Stroke Research: A Pilot Study,” published in the July 2024 issue of Pediatrics by Fiedler et al.
Pediatric stroke poses significant morbidity in children, necessitating robust research methodologies. Collaborative efforts, such as the International Pediatric Stroke Study (IPSS), have amassed extensive data, yet data entry remains labor-intensive. This study investigates the application of a Generative Pre-trained Transformer (GPT), an advanced artificial intelligence program, to streamline data entry into the IPSS database for pediatric stroke cases.
Clinical notes from the UTHealth Pediatric Stroke clinic for the most recent 50 patients with ischemic stroke or cerebral venous sinus thrombosis were deidentified. Specific prompts tailored to pediatric stroke were developed to guide GPT in answering 114 queries from the IPSS outcome questionnaire. Responses generated by GPT were compared against human raters to assess percent agreement across the dataset.
Initial iterations of GPT demonstrated variability but showed promising performance, achieving perfect accuracy in 17.5% of cases and as low as 26% in others. Subsequent iterations refined prompts and improved accuracy significantly. By the fourth iteration, GPT achieved an overall agreement rate of 93.6%, ranging from 62% to 100% across individual queries. Out of 2,400 data entries, GPT correctly processed 93.6% (2,247) and inaccurately entered 6.4% (153) items.
While the study highlights the potential of GPT with tailored prompts and ontological guidance to expedite data entry for research, further enhancements are necessary to optimize accuracy. GPT functions best in a collaborative role with human oversight, contributing to a synergistic approach that reduces workload and accelerates data processing in pediatric stroke research. Continued refinement holds promise for leveraging AI technology effectively in large-scale clinical databases like IPSS, ultimately enhancing research efficiency and insights into pediatric stroke management.
Source: sciencedirect.com/science/article/abs/pii/S0887899424002522
The post Investigating the Role of GPT in Pediatric Stroke Research first appeared on Physician's Weekly.