Filling in the Gaps: Classification of Generically Coded GP Notes on Belonging to a Phenotype (A Case Study on Depression)

When searching patient history for specific conditions or traits (phenotypes), Read codes help identify relevant encounters. Some codes directly match the condition, while others cover unrelated conditions or are coded as generic in-person and phone consultations. Free text notes within generic records may provide useful context. Manual review is impractical. Machine learning models could classify generic encounters based on the free text and fill the gaps in the patient record. To develop such a model, labelled data is required. This study explores the potential of using predictions from a Large Language Model (LLM) as training labels with a focus on depression but this could be applied to other phenotypes