Missingness Is a Signal

In medicine, why a value is absent carries diagnosis: no troponin ordered because chest pain was ruled out at the door; no psych notes because the patient was never referred — or was referred and refused.

The missingness taxonomy: MCAR (random gaps — rare), MAR (missing predictable from observed data), and MNAR — missing not at random, where absence depends on the unobserved value itself. EHR data is heavily MNAR in both directions: sicker patients get more tests, and the worried-well get extra workups — missingness encodes triage judgment.