Docs / Schemas

Subsystem schemas.

Each supported subsystem has a class describing its raw column list, output parquet schema, filename parser, and enrichment SQL.

SIHSUS

Source of truth: datasets/sihsus/ . Raw DBCs follow the DATASUS RDUFYYMM convention (reduced hospital admission, state, year, month). The output parquet is partitioned by UF.

Notable transformations

Municipality codes (6 chars) are joined against the IBGE 5,571-municipality registry and exposed alongside the canonical 7-char code.
CID-10 codes are normalized to uppercase and stripped.
Dates stored as YYYYMMDD strings are parsed into proper DATE columns (DT_INTER, DT_SAIDA, …).

SIM

Source of truth: datasets/sim/ . Raw filenames include both CID9 (DORUFYY, 1979–1995) and CID10 (DOUFYYYY, 1996+) variants — the parser disambiguates on stem length.

Notable transformations

IDADE (original 3-digit coded age) is expanded into IDADE_ANOS, IDADE_MESES, etc.
Multiple CID columns are combined into a single CID_ARRAY to simplify "any of these causes" queries.
Categorical BOOLEAN encodings are standardized to real BOOLEAN.