Amelia Jones | 02/12/2025
This poster was presented at ISPOR 2025 in Glasgow (9-12 November).
Author Tom Padgett explains the motivation behind the research and provides further insight into the findings.
Loading Viewer...
Previously, we introduced a proof-of-concept approach using a neural network to act as a simplified version of a complex model of chronic kidney disease. This ‘metamodel’ learns the relationships between inputs and outputs, allowing results to be calculated almost instantly instead of taking hours. Building on that work, we now use SHAP analysis—a technique that shows how much each input influences each output—as a form of sensitivity analysis.
Machine learning methods remain underutilised in health economics, but this study shows its potential. Neural network-based metamodels can dramatically cut computation time, making it possible to explore results in real time and even deploy complex models in practical settings. SHAP analysis adds transparency by clearly showing which factors drive the results.
Together, these methods offer a flexible, scalable alternative to traditional sensitivity analysis, helping researchers understand key drivers behind outcomes and supporting wider adoption of machine learning in health economics and outcomes research.
If you have any questions or would like to hear more about our research please contact Thomas Padgett, Principal Operational Researcher at HEOR: thomas.padgett@heor.co.uk.