OBJECTIVES: The objective of this study was to conduct a feasibility assessment for using Chat-GPT in developing a Global Value Dossier.

METHODS: A custom Chat-GPT-4 application was developed in Python 3 to take inputs from literature reviews and dossier templates to produce outputs in the dossier format. Publicly available dossiers (published online by the United States Medicaid) were used as benchmarks for replication and improvement. A modular approach was designed to test which parts of the dossier development could utilize Chat-GPT. A survey was conducted with experts to anonymously compare previously published dossiers and new Chat-GPT-assisted dossiers.

RESULTS: The custom Chat-GPT application's output for dossiers was significantly dependent on the quality of the literature review and dossier template. The most improvements in productivity for dossier development with Chat-GPT were observed in (1) writing, (2) formatting, (3) grammar, and (4) quality control checks, with an average estimated savings of 533 hours of junior analyst time. The most time-consuming step was editing and aligning the messaging of the text to the strategic objectives of the dossier. For the value proposition, Chat-GPT created useful drafts, but nearly all text of value messages needed to be updated based on qualitative input from the dossier team (resulting in no time savings). Dossiers developed with a hybrid approach (Chat-GPT and highly experienced HEOR experts) were significantly more accurate compared to previously published dossiers (without any AI).

CONCLUSIONS: Chat-GPT is a potentially useful tool for experienced HEOR experts to develop more accurate and high-quality dossiers. The modular approach with Chat-GPT has the potential to save significant time and financial resources.