Automated pseudonymisation

[marrobi]

A robust and configurable pseudonymisation process is required. Feature metadata should specify whether a pseudonymised version of a feature is required. Pseudonymised versions should be generated automatically. Ideally, queries to the feature store should transparently return pseudonymised versions of PII features depending on RBAC.

• Data transformation pipeline sample to pseudonymise structured data
• Data transformation pipeline sample to pseudonymise unstructured data

[jjgriff93]

This will likely take the form of a sample transform pipeline (a Data Pot)

[stefpiatek]

Merseyside work hopefully will be able to share this work back into flowEHR with sample repo that has example for bootstrapping.

• Lots of the work de-identification of notes data, happy to share if it works. Using presidio
• Have had rule-based pseudonymisation based on data types (e.g. dates). Databricks configuration

Consider breaking out into two separate issues nearer to the time of sharing back into flowEHR. Will demo at next standup