Many companies are developing innovative artificial intelligence solutions for health care. Too often, though, these applications fail to deliver their promised improvement in health outcomes when used in real-world settings because they were developed using data from a small number of patients in one locale over a limited period of time. Federated learning offers a way forward.
In the real world, as opposed to the clinical trial world, patient populations are large and diverse — young and old, well-off and poor, coming from a range of racial and ethnic backgrounds, and being treated with different protocols depending on time and place. Yet most artificial intelligence models in health care are developed and trained using datasets that do not reflect the general patient population or the frequent changes in clinical practice.
To improve the global impact of AI tools and create equity in the standard of care, health care researchers and AI developers worldwide need to incorporate more diverse data that represent the range of people who are likely to benefit from advanced AI-based diagnostics and treatment pathways.