OpenSafely

Overview

The OpenSafely dataset provides pseudonymised primary care data from EMIS Web GP practices in England. It is designed for secure health research and analytics, enabling population-level studies while protecting individual patient privacy.

The data is made available through a set of structured models covering clinical events, patient demographics, medications, referrals, and reference data.

---

Available Models

Core Clinical Models

Model	Description
Patient	Demographics, registration status, and opt-in flags
Observation	Clinical observations, test results, and coded entries
Consultation	Patient consultations with practitioners
Consultation Section	Sections within a consultation (e.g. History, Examination)
Problem	Active and past clinical problems
Referral	Inbound and outbound referrals
Diary	Diary entries and recall events
Appointment Slot	Appointment bookings and slot status

Medication Models

Model	Description
Drug Record	Medication courses (repeat and acute prescriptions)
Issue Record	Individual prescription issues from a drug record

Reference & Lookup Models

Model	Description
Clinical Code	SNOMED CT and Read code mappings
Codeable Concept	Extended code metadata including fully specified names
Drug Code	Drug code to dm+d product mappings
MKB Drug Pack	Drug pack details (units, pricing, supplier)
MKB Mapping Attributes	Language and sexual orientation code mappings
MKB Mapping DMD Preparation	dm+d preparation code mappings
MKB Mapping Ethnicity	Ethnicity SNOMED code mappings
Sensitive SNOMED Code	Codes classified as sensitive

Administrative Models

Model	Description
Sharing Agreements	Data sharing agreement status between organisations
Query Utilization	Your personal query execution history and performance

---

Key Concepts

Data Freshness

The pipeline processes source data weekly, but consumer-facing models update monthly (last Sunday of each month). This provides a stable analytical baseline throughout the month. See Data Freshness & Quality for full details.

Patient Deduplication

Each patient appears as a single row globally, deduplicated by NHS number across all qualifying practices. When a patient transfers between practices, their record updates to reflect the new practice without creating duplicates. See Practice Transfers for details.

Pseudonymisation

All patient identifiers (patient_id, nhs_number) are pseudonymised. Pseudonymisation is deterministic within your user context and stable across practice transfers, enabling longitudinal analysis.

Data Scope

Only data from England GP practices with an active sharing agreement is included. Records flagged as sensitive or confidential at the source are excluded from clinical event models.

---

Common Patterns

Joining Clinical Events to Patient

All clinical event models include a patient_id and organisation column. Join to the patient model on both columns to access demographic information:

SELECT
  o.observation_id,
  o.effective_datetime,
  o.snomed_concept_id,
  p.sex,
  p.age,
  p.date_of_birth
FROM explorer_open_safely.observation AS o
INNER JOIN explorer_open_safely.patient AS p
  ON o.patient_id = p.patient_id
  AND o.organisation = p.organisation
WHERE o.is_deleted = FALSE
  AND p.is_deleted = FALSE

Resolving Clinical Codes

Use the clinical_code or codeable_concept models to look up human-readable terms for code_id values:

SELECT
  o.observation_id,
  o.effective_datetime,
  c.term,
  c.snomed_concept_id
FROM explorer_open_safely.observation AS o
INNER JOIN explorer_open_safely.clinical_code AS c
  ON o.code_id = c.code_id
WHERE o.is_deleted = FALSE
  AND c.is_deleted = FALSE

Filtering Soft-Deleted Records

All models include an is_deleted column. When is_deleted = TRUE, the record has been soft-deleted at the source and all non-primary-key columns will be NULL. Always filter these out unless you have a specific reason to include them:

WHERE is_deleted = FALSE