Foresight (generative EHR model)

Foresight is a generative model of electronic health records from King's College London, an autoregressive transformer trained on sequences of coded clinical events so it can forecast a patient's next diagnoses in the way a language model predicts the next word. The Lancet Digital Health 2024 paper described the GPT-style model over coded EHR timelines, and Foresight 2 extended training to more than 5M patients. In oncology the use case is risk and trajectory forecasting, for instance flagging patients whose records suggest an undiagnosed cancer or a likely complication. Because it learns from coded data it inherits the biases and gaps of coding practice, and prospective evaluation of its forecasts changing care has not been reported. For a newcomer: Foresight reads a patient's record history and predicts what diagnoses might come next.