The webinar was moderated by Dr. Mitch Higashi, the Senior Vice President of Health Economics and Outcomes Research (HEOR) at Gene Dx, and the ISPOR Program Committee Co-Chair for the 2023 Annual Meeting in Boston. The webinar featured expert speakers who discussed the evolution of large language models (LLMs) and the opportunities they provide for ChatGPT in healthcare research and the biopharma industry. Additionally, they covered digital innovation solutions that utilize large language models.
First Speaker: Jingcheng Du, PhD - Director of NLP Research at Melax Tech
Dr. Du is a seasoned professional with expertise in implementing NLP/AI solutions in the life sciences. He provides expert guidance on the best practices for implementing NLP/AI in leading global pharmaceutical companies. With a background in biomedical NLP research, he has made significant contributions to the field and authored over 60 peer-reviewed papers cited over 1400 times.
Dr. Jingcheng Du, discusses ChatGPT, an AI chatbot that has gained immense popularity since its release in November. Within five days, it had over one million users, and within two months, it had reached 100 million users. The chatbot is seen as a game-changer in the AI industry, with many wondering if it's the start of a new AI revolution. ChatGPT's appeal lies in its versatility; it can assist users in various domains, from data science to creative writing. The article describes ChatGPT as a larger language model with billions of trainable parameters, and it follows a reinforcement learning process from human feedback. The chatbot's unique feature is its ability to easily understand human questions and prompts. The text also discusses NLP development cycles and how ChatGPT can change them.
Dr. Wang has over ten years of research experience in AI and NLP methodologies and applications in healthcare. His research goal is to leverage different dimensions of data and data-driven computational approaches to meet the needs of clinicians, researchers, and patients. In addition, he is chairing the NLP Working Group of AMIA this year.
Dr. Yanshan Wang discusses the increasing importance of unstructured EHR data in healthcare and research and how natural language processing (NLP) can be used to analyze and extract useful information from it. It explains that there are many NLP tasks, each requiring specific training and validation data and models and that there are three mainstream paradigms in NLP history, and the field is now shifting towards the fourth paradigm, generative NLP. Generative NLP uses unsupervised learning to create new content, including, but not limited to text. OpenAI's GPT models have been developed to produce text that continues a given prompt, and there are three versions of GPT, with ChatGPT believed to be based on GPT 3.5. Dr. Wang used several examples to show how the use of traditional NLP in biomedical and clinical domains requires internal expertise and specialized annotation and how generative AI models may reduce the labor-intensive annotation effort and provide a one-fits-all solution to many biomedical NLP tasks and lead to artificial general intelligence.
Dr. Wang is a leader with a track record of driving innovations and transforming healthcare with data intelligence and has been recognized for the development of NLP platforms and translational research in various areas, including oncology, immunology, and cardiovascular, among many others. She has previously served as Vice president of Biopharma Solutions at Sema4 and GDX and has worked as a faculty member and principal investigator at the University of Connecticut.
Dr. Xiaoyan Wang discusses three areas where generative NLP technologies such as ChatGPT can be used: real-world evidence journeys, systematic literature review, and clinical trial optimization. In regards to clinical notes, these technologies can help extract important information from the notes accurately and at scale, allowing for downstream analysis. Real-world evidence journeys can be used to track patients' diseases, symptoms, conditions, and treatments across multiple health systems and millions of patients. A systematic literature review can be automated with GPT technology, extracting and summarizing information from different articles, saving time and effort. Finally, GPT technology can be used to optimize clinical trials, allowing for easy extraction of variables, clinical trial design, cohort, and endpoint measures.
In conclusion, using large language models such as ChatGPT has significantly disrupted the traditional medical writing and information retrieval approach. Fine-tuning and prompt design are critical factors that require attention, and while there are some caveats, these models offer a more efficient and accurate solution to NLP-based problems. In addition, these models have shifted the paradigm in NLP research and projects. While more research is needed to study their potential and limitations, they offer a lower barrier to entry for users who need to perform multiple NLP tasks. Overall, large language models represent a significant advancement in the field of NLP and can potentially change how we approach many aspects of medical writing and information retrieval.