When medical staff are caring for patients, they document their findings, hypotheses and plans in notes every day. These notes often contain information that is not contained in the other tests and observations, such as vital signs, that are included in the patient’s electronic health records. However, because this information is all in “free text”, so not categorised or encoded, it has not yet been used to help build models and systems to identify infection.
We want to use de-identified forms of these free text notes, where patient details are removed, combined with numeric data such as test results and vital signs, to assess whether patients have an infection and help doctors to make treatment plans tailored to the individual patient. We will use secure artificial intelligence (AI) methods to collect key facts from the free text while carefully preserving confidentiality. The AI methods used will run on servers physically located within Oxford University Hospitals. No patient data will be shared with any third party, and no patient data will be shared with technology companies developing AI models. Patient records will undergo redaction to remove identifiers before processing by the research team.