Clinical prediction models (CPMs) are computer programs that use information about patients to predict outcomes. These predictions can help doctors make better decisions about treatment. However, there is a concern that CPMs might not treat everyone equally. This is because CPMs use past medical data, which may contain biases and only represent certain groups of people. As a result, CPMs may not work as well for underrepresented groups, leading to worse decisions. This project aims to develop methods that can measure and reduce biases in CPMs, with a focus on making them fairer for all groups of people. The duration of the project will be 3 years (36 months).
The methods will be tested in cancer screening, diagnosis, and treatment, with the goal of improving equity and helping people in underrepresented groups. Many cancers, such as lung cancer, tend to affect people in deprived areas and ethnic minority groups more than others, and they are often diagnosed at later stages. By looking at how ethnicity and other protected characteristics play a role in CPMs, we hope to identify areas where we need to be more careful when using these models. We will share our findings with cancer screening teams to make sure our work has an impact on how they treat patients. By improving how we predict and treat cancer, we hope to improve the health outcomes for people from these underrepresented groups. The hope is that these methods will be adopted as guidelines for developing and validating CPMs in the future.