Artificial Intelligence Methods to Power Real-World Clinical Studies
On February 23rd, our partners from UNIPD presented the REDDIE project at the research event "DEI-AI 2024" organised by the Dept. of Information Engineering, UniPD.
Enrico Longato gave a presentation on "Artificial Intelligence Methods to Power Real-World Clinical Studies," highlighting the importance of real-world effectiveness in evaluating treatments. They explored the nuances of retrospective observational studies and discussed innovative AI approaches to combine weighting and matching techniques for robust analysis, a goal UNIPD is pursuing in the context of the REDDIE project.
Events like this underscore the commitment to advancing AI research in healthcare and driving real-world impact.
Find the presentation abstract below:
Artificial Intelligence Methods to Power Real-World Clinical Studies
The clinical evidence generated during clinical trials is only part of the bigger picture to evaluate the potential health benefits of a new treatment or intervention. In fact, while determining their efficacy under hyper-controlled conditions is a non-negotiable step before market approval, their real-world effectiveness should also be tested.
To do so, we rely on the results of retrospective observational studies, i.e., analyses conducted on the real-world data of people in free-living conditions. There are two main families of techniques to conduct retrospective observational studies: namely, matching and weighting methods. While both aim at simulating the probabilistic properties of clinical trial randomisation, they approach the problem from two different angles and, thus, have different strengths and weaknesses. Specifically, matching methods are able to clearly identify the subpopulations to which the study’s conclusions apply, but need constant oversight by the experimenter, whereas weighting methods are one-shot but fuzzier in terms of their domain of validity.
The presentation included the possible application of AI to combine the strengths of weighting and matching techniques and design an automated system capable of accurately identifying subpopulations while maintaining the interpretability of results, with likely applications in the pharmaceutical industry for the conduction of retrospective studies on multiple real-world datasets at scale. This challenge is among the many tackled by the Horizon Europe project REDDIE (Real World Evidence for Decisions in Diabetes; UNIPD unit lead: Martina Vettoretti).
Read more on: https://sysbiobig.dei.unipd.it/sysbiobig-shares-ai-healthcare-research-insights-at-dei-ai-2024/