Transforming Real-World Evidence with Advanced Language Models: Predictive Analytics for Improved Healthcare Outcomes

This project leverages advanced AI, specifically large language models (LLMs), to transform healthcare research in three key areas: drug repurposing, drug safety monitoring, and predicting patient outcomes. The aim is to develop methods using LLMs that not only enhance but potentially surpass traditional biostatistical approaches, unlocking new insights from real-world health data. The first sub-project aims to identify existing medications that may help prevent or treat dementia. By analyzing nationwide registry data, the LLM will perform a Drug Wide Association Study (DWAS) to uncover patterns in medication use linked to dementia. The objective is to discover drug candidates for repurposing, potentially revealing treatments that are already available but whose anti-dementia effects are not yet known. In the second sub-project, the LLM will focus on lipid-modifying drugs and their side effects. By analyzing drug usage patterns, the LLM will detect signals of potential side effects. These predictions will be compared with traditional methods to improve drug safety monitoring, with the potential to establish new research methods for safeguarding patients from harmful side effects. The third sub-project focuses on predicting disease trajectories in cancer patients. Using detailed, longitudinal real-world health data, the LLM will analyze both patient narratives and structured data to forecast how a patient’s condition may evolve. This could lead to more personalized treatment plans and improved healthcare outcomes. This research is being conducted at NordicRWE in collaboration with the University of Oslo’s Centre of Excellence, INTEGREAT. The PhD candidate will work in an innovative environment, gaining expertise in AI, statistics, and health data, while contributing to significant advancements in public health and drug development.

This project aims to leverage large language models (LLMs) to enhance real-world evidence (RWE) research, focusing on drug repurposing, drug safety signal detection, and patient trajectory predictions. By creating embeddings of healthcare events, our models will predict outcomes like mortality, dementia onset, and drug side effects, aiming to surpass current biostatistical models. Using comprehensive registry data, the PhD candidate will develop and fine-tune LLMs to uncover new drug effectiveness and safety signals. In the first project, we hypothesize that a trained LLM can discover patterns in medication use associated with dementia. Using data from nationwide registries, we will create a Drug Wide Association Study (DWAS) and compare LLM predictions with traditional methods. The goal is to identify potential drug repurposing candidates for dementia. The second project focuses on drug safety surveillance. We will train LLMs to detect patterns in lipid-modifying drug use and their side effects. The study will include a DWAS and compare LLM predictions with established biostatistical methods, aiming to enhance drug safety signal detection. In the third sub-project, we will use LLMs to predict disease trajectories in oncology. Using longitudinal data, we will compare LLM predictions based on narrative descriptions versus tabular data. The model’s performance will be benchmarked against traditional statistical methods to determine the best approach for predicting patient outcomes. NordicRWE offers a dynamic environment with expertise in health data, epidemiological research and drug R&D. The PhD candidate will develop skills and contribute to innovative projects relevant to drug development and public health. Affiliated with the Norwegian Centre for Knowledge-Driven Machine Learning (INTEGREAT) at the University of Oslo, the candidate will join a community of experts in statistics, AI, and ML, supported by experienced supervisors from both institutions.