Capturing health-related outcomes in unstructured data

Traditionally health outcome capture has operated on two extremely structured approaches:

  • Dichotomous (Live or Dead; Event happened or didn’t happen)
  • Categorical (ordinal outcome scales like EQ5D5L)

Within unstructured text, both of those types may be mentioned explicitly as entities (ala Variable-Value pair) or they may only be captured through the value or subcomponents (“Walking without walking aid” implies “Fully independent”). This is of particular value when translating patient-reported narratives.

1 Like