Hybrid seminar from Dr Matthew Sperrin
Prediction under intervention and its role in primary prevention
Abstract: Statistics, AI and Machine Learning models are increasingly used in healthcare across many domains and contexts. One long-standing application is in primary prevention - where clinical prediction models (CPMs) predict future risk of occurrence of long term conditions such as cardiovascular disease. There is a long history of CPMs such as the Framingham score and QRISK being used to estimate healthy individuals' future risk of cardiovascular events. The idea is to identify subgroups at higher risk to be prioritised for preventative interventions - such as prescribing statins or promoting lifestyle change.
Most CPMs used in this way, however, are not causally-aware. This means that they do not allow us to compare predicted risk under interventions, answering questions such as 'what would this patient's risk be if they took statins', or 'what would this patient's risk be if they stopped smoking'. Therefore, CPMs can only be used indirectly for decision support, following a simple risk-based principle (if risk is high, we do something).
Prediction-under-intervention models are fast emerging as a better alternative. These combine ideas of prediction and causal inference to allow questions like the above to be answered correctly.
In this talk I will describe some of the key ideas of prediction-under-intervention, present early work on CHARIOT - a prediction-under-intervention model that can be used to answer questions like the ones above, and describe unanswered research challenges in this space.
Dr Matthew Sperrin, Centre for Health Informatics, Imaging and Data Science, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester Academic Health Science Centre, Manchester, UK