Foresight AI case study

Using de-identified NHS data from approximately 57 million people in England, researchers at University College London (UCL) and King's College London (KCL) are developing an innovative system to help clinicians, planners and policymakers predict health outcomes following the COVID-19 pandemic more effectively – while maintaining strict patient privacy standards.

Foresight is an AI model specifically designed for healthcare that has been trained in the NHS England Secure Data Environment (SDE) on de-identified NHS data.

Like ChatGPT, Foresight learns to predict what happens next based on previous events, working like an auto-complete function for medical timelines. Predictions are validated against real-world data. 

The NHS England SDE provided secure access to extensive de-identified records to train Foresight AI. It included: 

• data from 57 million individuals (including hospital admissions, vaccinations and other medical records) 

• over 10 billion healthcare events across more than 40,000 types (for example, admissions, diagnoses, procedures and medications) 

• records spanning November 2018 to December 2023

By accessing population wide data, Foresight AI also studies minority groups and rare diseases, which are often overlooked when using smaller datasets.  

Training large language AI models like Foresight requires significant computing power. Industry partners Amazon Web Services (AWS) and Databricks provide this essential resource while never accessing the data or AI model. The research team's access to the SDE was facilitated by the BHF Data Science Centre at Health Data Research UK, following rigorous approval processes and involving members of the public in approving and shaping the research.

Through this innovative collaboration between NHS England, academia and industry, Foresight is exploring how large-scale, de-identified health data can improve healthcare efficiency and fairness across the population. 

Only approved researchers working under strict NHS governance can access the data, which remains within the SDE at all times, as does the trained model. Industry partners provide infrastructure but cannot access the raw data, AI model or any outputs. 

Foresight exemplifies how data-driven healthcare can predict and manage health risks proactively, supporting strategic healthcare shifts. Patient privacy remains central, with stringent measures in place: 

• all data is de-identified 

• only approved researchers can access data within the SDE 

• no data or outputs leave the secure environment without oversight 

• public and patient involvement ensures alignment with public expectations 

To expand Foresight and validate the model further, the research team aims to: 

• add more data sources – clinician notes, lab results, imaging data 

• extend historical data analysis to deepen understanding of long-term health trends 

• expand use cases beyond COVID-19-related research 

• test in other nations to verify model effectiveness