Hugging Face's logo

Datasets:
SidneyBissoli
/
sih-sp

Tasks:
Tabular Classification
Languages:
Portuguese
Size:
1B<n<10B
Tags:
health
brazil
public-health
parquet
datasus
sih
License:
Dataset card Files
xet
 Community

The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.

SIH SP — Professional Services per Hospital Admission (Brazil, 1997–2026)

Act-level records from Brazil's Hospital Information System (SIH/SUS): every procedure, exam, surgery, ICU day, blood product or prosthesis billed inside a public hospital admission, with quantity, amount, SIGTAP procedure code, and (from 2008) the professional who performed it. Each admission (AIH) generates on average ~11 SP rows. This is the SP (Serviços Profissionais) sub-module of the sih/ namespace; it joins to the admission-level RD dataset (sih-rd) via SP_NAIH = N_AIH. Converted from legacy .dbc files to Apache Parquet.

Part of the healthbr-data project — open redistribution of Brazilian public health data.

Summary

Item Detail
Official source DATASUS FTP / Ministry of Health
Temporal coverage Jul/1997–2026 (SP files do not exist before 1997)
Geographic coverage All 27 Brazilian states (by state of processing)
Granularity One row per act/procedure inside an admission (~11 rows per AIH)
Volume 3,198,639,792 records (9,410 .dbc files, 32.0 GiB Parquet on R2; 52.5 GiB compressed at source)
Format Apache Parquet, partitioned by ano/mes/uf
Data types All fields stored as string (preserves original format)
Update frequency Monthly (source publishes ~2–3 months after competency month)
License CC-BY 4.0

Resumo em português

SIH SP — Serviços Profissionais por Internação (Brasil, 1997–2026)

Microdados em nível de ato/procedimento do Sistema de Informações Hospitalares do SUS: cada procedimento, exame, cirurgia, diária de UTI, hemoderivado ou órtese/prótese cobrado dentro de uma internação pública, com quantidade, valor, código SIGTAP e (a partir de 2008) o profissional executante (CBO e documento). Cada AIH gera em média ~11 linhas SP. É o submódulo SP do namespace sih/; liga-se ao dataset de internações RD (sih-rd) pela chave SP_NAIH = N_AIH.

O RD responde quem, onde, quando, por quê e quanto custou a internação; o SP responde o que exatamente foi feito, por quem e a que valor unitário. Contagens de internações, mortalidade hospitalar e permanência devem ser feitas no RD — no SP essas informações aparecem repetidas em cada ato.

Item Detalhe
Fonte oficial FTP DATASUS / Ministério da Saúde
Cobertura temporal jul/1997–2026 (não há arquivos SP antes de 1997)
Granularidade Uma linha por ato/procedimento dentro da internação (~11 por AIH)
Volume 3.198.639.792 registros (9.410 arquivos .dbc, 32,0 GiB em Parquet no R2; 52,5 GiB comprimidos na fonte)
Formato Apache Parquet, particionado por ano/mes/uf
Atualização Mensal (fonte publica ~2–3 meses após o mês de competência)

Para documentação completa em português, consulte o repositório do projeto.

Data access

Data is hosted on Cloudflare R2 and accessed via S3-compatible API. The credentials below are read-only and intended for public use.

R (Arrow) 

library(arrow)
library(dplyr)

Sys.setenv(
  AWS_ENDPOINT_URL      = "https://5c499208eebced4e34bd98ffa204f2fb.r2.cloudflarestorage.com",
  AWS_ACCESS_KEY_ID     = "28c72d4b3e1140fa468e367ae472b522",
  AWS_SECRET_ACCESS_KEY = "2937b2106736e2ba64e24e92f2be4e6c312bba3355586e41ce634b14c1482951",
  AWS_DEFAULT_REGION    = "auto"
)

# Acts performed in admissions processed in Acre, Jan 2024
sp <- open_dataset("s3://healthbr-data/sih/sp/ano=2024/mes=01/uf=AC/",
                   format = "parquet")

# Top procedures by amount billed
sp |>
  collect() |>
  mutate(valor = as.numeric(SP_VALATO)) |>
  count(SP_ATOPROF, wt = valor, sort = TRUE, name = "valor_total") |>
  head(20)

# Join with the admission-level RD dataset (same partition)
rd <- open_dataset("s3://healthbr-data/sih/rd/ano=2024/mes=01/uf=AC/",
                   format = "parquet") |> collect()
sp |> collect() |>
  inner_join(rd, by = c("SP_NAIH" = "N_AIH")) |>
  count(SEXO, SP_ATOPROF, sort = TRUE)

Important: Point to specific partitions (ano=YYYY/mes=MM/uf=XX/), not to the dataset root. The root contains README.md and manifest.json, which Arrow cannot read as Parquet files — and never open sih/ itself, which also holds the rd/ sub-module with a different schema. SP is ~10× larger than RD: prefer narrow partitions and lazy engines (Arrow, DuckDB, Polars) over collect() on wide ranges.

Python (PyArrow) 

import pyarrow.dataset as pds
import pyarrow.fs as fs

s3 = fs.S3FileSystem(
    endpoint_override="https://5c499208eebced4e34bd98ffa204f2fb.r2.cloudflarestorage.com",
    access_key="28c72d4b3e1140fa468e367ae472b522",
    secret_key="2937b2106736e2ba64e24e92f2be4e6c312bba3355586e41ce634b14c1482951",
    region="auto"
)

sp = pds.dataset("healthbr-data/sih/sp/ano=2024/mes=01/uf=AC/",
                 filesystem=s3, format="parquet")
df = sp.to_table().to_pandas()
print(f"Acts: {len(df)}, distinct admissions: {df['SP_NAIH'].nunique()}")

Note: These credentials are read-only and safe to use in scripts. The bucket does not allow anonymous S3 access — credentials are required.

File structure 

s3://healthbr-data/sih/
  README.md            ← namespace index (rd/, sp/)
  rd/                  ← admissions (see sih-rd)
  sp/                  ← this dataset
    README.md
    manifest.json
    ano=1997/
      mes=07/
        uf=AC/
          part-0.parquet
        ...
    ...
    ano=2026/
      mes=01/
        ...

Historical schemas

The SP file changed far less than the RD file. Three schemas:

Period Columns Key characteristics
Jul/1997–~2005 16 Hospital identified by CGC (SP_CGCHOSP); SP_PTSP_NF combined; no professional identification, no ICD
~2006–2007 18 SP_GESTOR + SP_CNES replace the CGC; SP_PTSP / SP_NF split
2008–2026 36 FTP era change. +professional (SP_PF_CBO, SP_PF_DOC, SP_PJ_DOC), +ICD (SP_CIDPRI, SP_CIDSEC), +complexity/financing (SP_COMPLEX, SP_FINANC, SP_CO_FAEC), +SEQUENCIA/REMESSA, +SP_M_HOSP/SP_M_PAC (municipalities), +SP_QT_PROC, SP_U_AIH; SIGTAP 10-digit codes

Columns not present in a given era are absent from that partition's Parquet file. Use open_dataset(unify_schemas = TRUE) in Arrow to query across eras (missing columns filled with null).

Schema (modern era, 2008–2026, 36 columns)

Variable Description
SP_GESTOR Managing authority code
SP_UF State code (processing)
SP_AA / SP_MM Competency year / month
SP_CNES Health facility code (CNES)
SP_NAIH AIH number — join key to RD N_AIH
SP_PROCREA Main procedure of the admission (SIGTAP), as in RD PROC_REA
SP_DTINTER / SP_DTSAIDA Admission / discharge date (YYYYMMDD)
SP_NUM_PR Sequence of the professional/act
SP_TIPO AIH type
SP_CPFCGC Document of the billing entity
SP_ATOPROF Act/procedure performed (SIGTAP code)
SP_TP_ATO Act type
SP_QTD_ATO Quantity
SP_PTSP / SP_NF Points (SP) / invoice flag
SP_VALATO Amount of the act (R$)
SP_M_HOSP / SP_M_PAC Municipality of hospital / of patient (IBGE)
SP_DES_HOS / SP_DES_PAC Hospital / patient outside the state (flags)
SP_COMPLEX Complexity level
SP_FINANC / SP_CO_FAEC Financing type / FAEC sub-type
SP_PF_CBO Occupation (CBO) of the professional
SP_PF_DOC / SP_PJ_DOC Document of the professional (individual / legal entity)
IN_TP_VAL Amount type indicator
SEQUENCIA / REMESSA Batch sequence / remittance
SERV_CLA Service/classification
SP_CIDPRI / SP_CIDSEC Primary / secondary diagnosis (ICD-10)
SP_QT_PROC Procedure quantity
SP_U_AIH Last-AIH flag

Reference tables (SIGTAP procedures, CBO occupations, ICD-10) are in the Ministry's TAB_SIH package; see the project documentation.

Source and processing

Original source: 9,412 .dbc files from the DATASUS FTP server: 200801_/Dados/ (modern era, 5,992 files, 45.2 GiB) and 199201_200712/Dados/ (legacy era, 3,420 files, 9.2 GiB; SP starts in Jul/1997). Files are named SP{UF}{YY}{MM}.dbc.

Processing: .dbc → R (read.dbc::read.dbc()) → all fields cast to character → Parquet (arrow::write_parquet()) → upload to R2 (rclone). Same pipeline as RD, selected with SIH_TIPO=SP. No value transformations — field values are published exactly as provided by the Ministry of Health.

Bootstrap: 2026-08-18, pipeline 1.2.0, single run over both eras (SIH_SPRINT=3) reading a temporary raw mirror of the FTP on R2 (SIH_FONTE=r2, deleted afterwards). Modern era (2008–2026): 5,991 files, 2,551,706,767 records; legacy era (Jul/1997–2007): 3,419 files, 646,933,025 records. Total: 9,410 files, 3,198,639,792 records, 32.0 GiB on R2, ~9 h single-thread on a Hetzner cpx42. Last month at bootstrap: 2026-06; later months are added by the weekly maintenance. Two source files exist on the FTP but contain no records (SPAP0710.dbc, SPAC0909.dbc) and are therefore absent from R2; the Roraima gap documented for RD (no files Jul/1997–May/2000) applies to SP as well.

Reproducibility & provenance

Every Parquet file written by pipeline version ≥ 1.1.0 carries a JSON provenance record in its schema metadata (key healthbr): source_url, source_file, source_hash_md5, source_size_bytes, download_date, pipeline_script, pipeline_version and (≥ 1.2.0) git_commit. The same facts are recorded per partition in manifest.json (plus SHA-256 and record count of each output file) and per source file in the version-control CSV in the GitHub repository. Together they let anyone re-derive a partition from the Ministry's original file and the exact code commit, and verify that the copy they hold is intact. Data are not versioned: the R2 copy is the latest publication; revisions by the Ministry replace files. Every SP file carries a native embedded record (no backfill was needed — the whole dataset was bootstrapped with pipeline 1.2.0). Full policy and audit recipe: docs/policy-reproducibility-pt.md.

arrow::read_parquet("part-0.parquet", as_data_frame = FALSE)$metadata$healthbr

Known limitations

  1. Government data, not ours. Values are preserved exactly as in the original .dbc files, including inconsistencies or missing data.
  2. Act-level, not admission-level. Counting rows counts acts, not admissions; patient demographics are not in SP — join with RD.
  3. All fields are strings. Amounts (SP_VALATO), quantities and dates must be parsed by the user.
  4. Three historical schemas. Professional identification and ICD codes only exist from 2008; the hospital key is CGC before ~2006 and CNES after.
  5. Starts in 1997. Unlike RD (1992), no SP files exist for 1992–1996.
  6. Volume. ~10× the rows of RD; query by partition and prefer lazy engines.
  7. Monthly partitioning by year/month/state of processing.

Related datasets

Dataset Period Records Link
SIH RD (hospital admissions) 1992–present 415M+ sih-rd
SINASC (live births) 1994–2022 85M+ sinasc
SI-PNI Microdados (vaccination) 2020–present 736M+ sipni-microdados
SI-PNI COVID (vaccination) 2021–present 608M+ sipni-covid
SI-PNI Agregados — Doses 1994–2019 84M+ sipni-agregados-doses
SI-PNI Agregados — Cobertura 1994–2019 2.8M+ sipni-agregados-cobertura
SI-PNI Dicionários Static 263 rows sipni-dicionarios

Citation 

@misc{healthbrdata,
  author = {Sidney da Silva Bissoli},
  title  = {healthbr-data: Redistribution of Brazilian Public Health Data},
  year   = {2026},
  url    = {https://huggingface.co/datasets/SidneyBissoli/sih-sp},
  note   = {Original source: Ministry of Health / DATASUS}
}

Contact

Last updated: 2026-08-18 (bootstrap complete)
Downloads last month
78
Total file size:
16.4 kB