VacinaBR-PNI: conjunto de dados curado e enriquecido de doses aplicadas pelo Programa Nacional de Imunizações em 2025

VacinaBR-PNI is a curated, documented, and enriched dataset derived from open data on vaccine doses administered by Brazil’s National Immunization Program (Programa Nacional de Imunizações, PNI) in 2025. The dataset was built from monthly CSV files published by the Brazilian Ministry of Health through the SUS Open Data Portal and processed through a reproducible ETL/ELT workflow. The original PNI files contain individual records of vaccine or immunobiological doses applied in Brazil. Each record may include information about the vaccination date, patient sex and age, patient municipality and state, health establishment, vaccine code, vaccine abbreviation, vaccine description, dose description, manufacturer, source system, and document status. The curated dataset preserves analytically relevant fields, removes direct sensitive identifiers, standardizes column names and data types, and adds derived attributes for easier reuse. The processing workflow derives temporal variables such as vaccination year, month, quarter, and epidemiological week; demographic variables such as age group; geographic variables such as patient region; and data-quality indicators such as record completeness, valid age, final document status, deletion status in RNDS, and valid document record. Municipal enrichment was performed with metadata from the Brazilian Institute of Geography and Statistics (IBGE), enabling municipal-level geographic analysis for records with valid municipality codes. The complete 2025 execution contains 170,529,208 processed vaccination records. The curated layer is stored as partitioned Parquet files by year and month, allowing users to query only the period they need instead of loading the entire dataset. This structure supports scalable analysis in Python, R, DuckDB, Apache Arrow, Polars, Spark, and other tools compatible with Parquet. The package also includes analytical CSV files derived from the curated records. These summaries allow quick exploration without reading all individual-level Parquet files. Users can analyze doses by month, state, region, municipality, vaccine type, manufacturer, sex, age group, and quality indicators. For example, the analytical CSVs can be used to identify the states with the highest number of administered doses, compare vaccine types over time, inspect manufacturer distribution, evaluate completeness by state, or create dashboards in spreadsheet software, Power BI, Tableau, Streamlit, or similar tools. A portable DuckDB database is included to make SQL-based analysis easier. With DuckDB, users can query the dataset locally without setting up a database server. The database contains views and materialized analytical tables generated from the Parquet layer, enabling reproducible SQL queries over both detailed records and summarized tables. This is useful for researchers who want to filter records by time, territory, vaccine, age group, or quality condition; aggregate doses by custom combinations of variables; join the dataset with external socioeconomic or territorial data; or reproduce the summary tables included in the package. The documentation included in the package supports interpretation and reuse. The data dictionary describes field names, types, meanings, source or derived status, examples, and sensitivity classification. The schema file documents the structure of the curated layer. Validation reports summarize the number of original and processed records, missing values, invalid dates, invalid ages, complete and incomplete records, document validity, and missing values in key fields. Consistency reports compare totals across global, monthly, vaccine, and municipal summaries. This dataset can support multiple use cases, including epidemiological studies, public health monitoring, territorial analysis, reproducible research, dashboard development, teaching activities, and data-quality assessment. Examples include studying temporal patterns of vaccine administration, comparing doses across states or municipalities, analyzing the distribution of vaccine types and manufacturers, evaluating missing geographic information, mapping municipal records, and demonstrating end-to-end data engineering workflows in health data. The dataset is derived from public data published by the Brazilian Ministry of Health. Direct sensitive identifiers, such as patient identifiers and patient ZIP code fields, are not retained in the processed Parquet layer. However, users should interpret the dataset with attention to the limitations of the original source, including missing fields, records without valid municipality codes, incomplete manufacturer descriptions, a small number of invalid or absent vaccination dates, and possible changes in the layout or availability of the original monthly files. The package includes:- curated Parquet files in `data/processed/`;- analytical CSV files in `data/analytics/`;- a portable DuckDB database in `data/vacinabr_pni.duckdb`;- documentation and metadata in `docs/`;- a data dictionary;- schema and source metadata files;- validation and consistency reports;- map coverage information;- example notebooks for querying, exploring, and assessing the dataset;- citation and license files.Users can use this dataset to:- query vaccination records by month, state, municipality, vaccine, manufacturer, sex, and age group;- analyze temporal trends in administered doses throughout 2025;- compare territorial distribution of vaccination records across Brazilian states and municipalities;- generate maps using municipality-level summaries and IBGE coordinates;- evaluate data quality through completeness, valid age, valid date, and document-status indicators;- build dashboards and public health monitoring tools;- join the curated data with external demographic, socioeconomic, or territorial datasets;- reproduce analytical summaries using DuckDB or the included notebooks;- teach data cleaning, validation, Parquet partitioning, DuckDB querying, and health-data analysis.Source code and project repository:https://github.com/fabio-barros/bdnc-projeto