Prepare SIH hospital-admission microdata

Uses the official DataSUS TabWin definitions to label all four SIH file families available from fetch_datasus(): reduced admissions ("SIH-RD"), rejected admissions ("SIH-RJ"), professional services ("SIH-SP"), and rejected/error records ("SIH-ER"). Dictionaries are downloaded on first use and cached for the rest of the R session.

Usage

process_sih(
  data,
  information_system = "SIH-RD",
  municipality_data = TRUE,
  labels = c("factor", "character", "none"),
  diagnostics = FALSE
)

Arguments

data: A data frame returned by fetch_datasus() for a supported SIH file family, or another data frame with a compatible layout.
information_system: SIH file family represented by data. One of "SIH-RD", "SIH-RJ", "SIH-SP", or "SIH-ER". The default preserves previous calls to process_sih().
municipality_data: Logical scalar. If TRUE, add municipality names and available territorial attributes for the residence municipality in RD, RJ, and ER files.
labels: Output type for categorical labels: "factor" (the default), "character", or "none" to retain the original codes.
diagnostics: Logical scalar. If TRUE, attach a processing report, including codes absent from official conversion tables. Retrieve it with processing_diagnostics().

Value

A tibble. Dates are returned as Date, counts, quantities, and derived IDADEdias, IDADEmeses, and IDADEanos fields as integer, monetary values as double, labelled categorical fields as factors, and identifiers and free text as character. Derived age fields are added when the source contains both COD_IDADE and IDADE.

Details

For RD and RJ, the definition is selected from the official historical archives according to each row's competence. This supports data sets concatenated across the 1997, July 2003, and 2007 layout boundaries. Codes absent from a conversion table remain visible as factor levels.

Performance and cache

Processing uses vectorized code padding and CNV thresholds, parses repeated dates once per field and format, and unescapes only text containing backslashes. UTF-8 conversion is still performed for text; values marked as "bytes" bypass text normalization. Historical relation selection subsets only the source columns it needs. These optimizations are automatic.

Dictionaries are reused within the R session. For reuse across sessions, set options(microdatasus.cache_dir = datasus_cache_dir(create = TRUE)); calling datasus_cache_dir() alone does not enable persistent caching. The first processing call can include dictionary downloads and parsing. labels = "none" controls categorical output, not network access: some processors still need DEF metadata or relations for field semantics.

diagnostics = FALSE avoids collecting the optional report, and municipality_data = FALSE omits territorial enrichment when it is not needed. For requests spanning many files, use fetch_datasus() with process = TRUE, collect = FALSE, and destination to save each file separately. A processor called directly still holds its input and output in memory. See the processing guide.

Territorial enrichment uses the fixed tabMun snapshot identified by datasus_reference_tables(), not an automatically selected edition for each observation year. Enable diagnostics to record that version in the report.

References

Saldanha, R. F. (2026). SIH – Sistema de Informações Hospitalares do SUS.

Examples

if (FALSE) { # interactive() && curl::has_internet()
process_sih(sih_rd_sample)
}