PRMA-AI: HOW CAN AI HELP SOLVE PRICING AND REIMBURSEMENT CHALLENGES?

AI is being explored for various tasks including forecasting market opportunities, replacing clinical trials, preparing value dossiers, predicting payer-relevant PICOS schemes, identifying appraisal patterns, and compiling and evaluating HTA submissions. For example, AI-driven analyses are being used to replace control arms in clinical trials, and AI tools are predicting the differing PICOs of the scoping process.

While AI tools can support literature search and study selection, they face limitations in reliability, sensitivity, and transparency, requiring continued human oversight and expert validation. A Cochrane review of 196 reports on the use of Machine Learning and Large Language Models (LLMs) in systematic reviews in health research showed that LLMs were applied in 10 out of 13 review steps, most frequently for literature search, study selection, and data extraction.

Although useful for extracting data such as overall survival from publications, complex data extraction and quality control still need human involvement. For example, extracting quality of life measures, specific domains, observation times, and population-specific results remains challenging. When screening search results, the volume of hits plays an important role. In German AMNOG dossiers, the number of hits identified in the relevant databases is currently typically in the single or double digits. Extending the data scope beyond randomized and non-randomized comparative studies might significantly increase complexity.