Structured data
Information that fits neatly into rows and columns: lab values, vital signs, diagnosis codes, dates. It is the easiest kind for traditional machine learning to use, and it is what most risk scores and dashboards are built on. The counterpart to unstructured data.
A kidney injury risk model trained on data reporting creatinine in µmol/L is connected to a lab feed that sends mg/dL. A result of 1.2 mg/dL, about 106 µmol/L, arrives as a plain 1.2, and the model reads it as far below any real value and scores the patient as low risk. Unless someone built a range check, no error appears, since a number arrived in the expected field. Before go-live, run test patients with known results through every input and check units, reference ranges and how blanks are handled.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →