A machine-learning classifier predicts the origin of cancer of unknown primary based on electronic health records and next-generation sequencing data, showing that patients treated accordingly to model predictions had significantly better outcomes.