Data processing pipelines and tools for routine health facility malaria surveillance in Uganda

Timely processing of routine health facility surveillance data is essential for responsive malaria control, yet the path from raw electronic reports to actionable intelligence remains a largely undocumented challenge in endemic countries. We present an open-source Extract-Transform-Load (ETL) software system and accompanying metadata R package (ramptools) that together convert raw DHIS2 health facility data into cleaned, version-controlled, analysis-ready datasets for Uganda's National Malaria Elimination Division (NMED). The ETL pipeline is implemented in R and deployed on a cloud platform with scripts running on automated schedules to keep the database current. The system extracts malaria indicators from two successive DHIS2 instances via the DHIS2 Web API, applies a two-stage outlier detection algorithm combining variance-based screening with STL decomposition, imputes missing values through seasonal interpolation, enforces logical consistency constraints across indicator cascades, and aggregates facility-level data through a six-level administrative hierarchy. All raw data are stored in an append-only versioned schema, enabling reconstruction of the database state at any historical point. The ramptools package provides standardized metadata - including an indicator crosswalk mapping 118 indicators across DHIS2 instances, a location hierarchy of 11,229 organizational units, geolocated health facility attributes, and administrative boundary shapefiles - as lazy-loaded R data objects. We validate the system through a proof-of-principle outbreak detection application that consumes ETL outputs to compute district-level outbreak indices via kernel-smoothed time series analysis, deployed as an interactive Shiny dashboard. The software has been in development since 2020, processing weekly and monthly data for approximately 8,700 health facilities. All code is open source (MIT license) and hosted on GitHub.