medicines-snowflake/docs/sql_style_and_validation_guardrails.md

SQL Style And Validation Guardrails

These notes are extracted from repeated medicines optimisation query work. They are intended to prevent common Snowflake and medicines-data errors.

Snowflake Syntax

• Double-quote table and column identifiers, especially mixed-case columns such as "PatientPseudonym" and "ProcessingPeriodDate".
• Quote aliases that will be consumed by Excel, Power BI, Python, or another SQL layer: COUNT(*) AS "PatientCount".
• Use CURRENT_DATE() or CURRENT_TIMESTAMP() rather than T-SQL functions such as GETDATE().
• Use LIMIT for quick checks.
• Cast long-format output values to a consistent numeric type before UNION ALL.

Medicines Table Choices

• Use REPORTING_DATASETS_ICB.SCRATCHPAD."MEDS__UnifiedPrescribingTable" for current prescribing analysis when available.
• Use NATIONAL.GPMED."MedicinesDispensedInPrimarycare" for official dispensed/paid activity.
• Do not treat prescribing and dispensing as interchangeable. They answer related but different questions.
• Map prescribing SNOMED codes to BNF through DATA_HUB.DWH."DimMedicineAndDevice".
• Use "ProductDescription" for medicine names from DimMedicineAndDevice. Do not assume a "ProductName" column.

Geography And Denominators

• Keep the practice CTE visible in each report.
• For Norfolk/Suffolk GP practice outputs, use DATA_HUB.DWH."DimOrganisationAndSite" with "IsSiteNorfolkAndSuffolk" = 'Yes'.
• Only use older Norfolk-and-Waveney flags when the report is explicitly Norfolk and Waveney only.
• Be explicit about whether you are using organisation registered population, a counted active patient denominator, or another denominator.

Clinical Coding

• Prefer maintained clusters in DATA_HUB.PHM."ClinicalCodingClusterSnomedCodes" over text searching SNOMED descriptions.
• Use DATA_HUB.PHM."PrimaryCareClinicalCoding" for unified clinical coding across systems.
• Join clinical coding to patients through PatientPseudonym, then to DimPerson when person or practice fields are needed.
• Make lookback windows visible in SET variables or a date CTE.

Validation Before Sharing

• Check latest data dates before selecting a period.
• Run a limited version first and inspect row counts.
• Check the output column list before handing results to someone else.
• Avoid SELECT * in audit/detail outputs.
• Do not commit patient pseudonyms, CSV exports, spreadsheets, images, or local tooling files.