Data Acquisition
From source to usable data
Data acquisition covers identifying source systems, extracting data, mapping it to a target structure, and confirming fitness for use. Know the pipeline stages: extract, transform (standardize codes, units, dates; deduplicate), load, and validate. ETL transforms data before loading it into the warehouse; ELT loads raw data first and transforms inside the warehouse, which is common on modern cloud platforms.
Interfaces and standards
Be familiar with HL7 v2 messaging (event-driven, segment-based), FHIR resources (modern API-based exchange), flat-file extracts, and direct database or API connections. Claims arrive as X12 837/835 transactions. A master patient index links records across systems, and identity errors are a leading source of analytic error.
Profiling and quality
Before analysis, profile the data: row counts, null rates, distinct values, ranges, duplicates, and date coverage. Judge against data quality dimensions - completeness, accuracy, timeliness, consistency, validity, uniqueness. Maintain a data dictionary and record lineage (source, transformation, owner) so results can be reproduced. Confirm permissions, minimum necessary access, and any de-identification or data use agreement up front.
Traps and rule of thumb
Questions blur "the data loaded successfully" with "the data are correct" - a load can be complete and still wrong. Another trap is assuming a field means what its name suggests; validate definitions with the source owner. Watch for extract date-range mismatches and unit differences (mg/dL versus mmol/L) that silently corrupt joins. Reconcile every extract against a control total from the source system, such as encounter counts by month; if totals do not reconcile within an agreed tolerance, stop and investigate, because no downstream modeling fixes a bad extract.