Practical applications of generative AI in medical research and education

The team used ChatGPT for title/abstract screening and full-text screening in a systematic review of AI in organ transplantation, which was subsequently published

ChatGPT was prompted to analyze an Excel sheet, screen articles using specified inclusion criteria, add columns indicating relevance (relevant/not relevant/maybe relevant), and provide explanations for each decision

Inter-rater reliability between ChatGPT and human reviewers for systematic review screening was excellent

Creating a custom GPT with serial prompts produces more consistent results than using general ChatGPT search, though slight differences may occur across multiple runs

ChatGPT is used as part of methodology and reliability process in research, not for plagiarism or writing, and this is disclosed in journal submissions

AI models help increase efficiency, which can lead to productivity, but a human must always be in the equation for medical and scientific work

ChatGPT is a black box model with unknown training data and changing weights, requiring human oversight to direct it and interpret its outputs

Genie AI is an academic model fine-tuned with research papers that can generate paper outlines, provide in-house chatbot assistance, cite papers, and format citations according to preferred specifications

Academic AI models like Genie AI are mostly on par with latest published research, though users must verify which library the model searches

OpenEvidence was developed by Harvard team and Mayo Clinic, focuses on Elsevier library, provides conversational answers while summarizing latest papers, and suggests related topics

Journal of Pediatric Surgery (Elsevier) policy permits using AI for background information and language enhancement but prohibits using generative AI to write papers; disclosure of prohibited use would likely result in rejection

Elsevier has tools to detect if generative AI has been used to write papers

Most journals permit ChatGPT use only to enhance writing style, not to create ideas

Fireflies provides action items after meetings to keep participants informed of next steps

Cincinnati Children's Hospital restricted use of meeting transcription tools due to security concerns about auto-recording, particularly for legally sensitive or confidential meetings, and now requires use of approved Microsoft Co-pilot

Cook Children's uses Microsoft Co-pilot as the only approved tool for note-taking, recording, and post-meeting analysis

Creating custom GPTs for repeated prompt sequences is straightforward and allows reuse of complex workflows

ChatGPT can create interactive training sessions from clinical guidelines, generating problem-based learning activities with scoring and feedback, avoiding hallucinations by working from source documents

ChatGPT can generate medical cases in USMLE Step 1 or Step 2 format with step-by-step progression and end-of-case feedback when given appropriate prompts