👋 Introduction
HIC can be considered as data curators, and on behalf of Data Controllers (e.g. NHS Tayside), we will only release outputs in line with the relevant requirements. We will ensure that no sensitive data is released from the TRE. All files coming out of our TRE are checked to mitigate the risk of identifiable information being released. Your files cannot contain any individual level data. Summary results (or aggregate data) must be at least groups of 5 or more. Below that threshold outputs can become potentially disclosive. Cells with results less than this value must be reported as “<5”.
...
Lastly, our disclosure control approach is informed by best practice and reviewed regularly in line with our certified data security and confidentality certifications (https://www.dundee.ac.uk/hic/governance-service/data-security-confidentiality ).
...
\uD83D\uDCD8 Instructions
...
Extraction of your files can be requested by copying the files into the appropriate egress folder, and then submitting the request from service workbench (SWB).
...
Submission of an egress request will automatically open a support ticket on your behalf - you should receive an email within a few minutes. It can be helpful to reply to this email to provide additional information which may make the reviewal process easier. If any output is rejected, you will have the opportunity to demonstrate why the output is safe for release. However, the decision to release any output remains with HIC.
...
⌛ Expected time lines
Requests are only assessed on week days and may take 1-3 days to be approved for simple file requests (i.e., summary reports, tables etc.). More complex requests such as Artificial Intelligence/ Machine Learning (AI/ML) models may take significantly longer.
🛠️ AI/ML
AI/ML models pose a new challenge to all TREs. Similarly to when building a model, there is a need to balance the model complexity, interpretability, and required amount of training data; models deployed out of our TRE need to balance privacy and utility. These types of methodologies take us longer to disclosure check.
We may ask you to complete a questionnaire about your AI/ML model, providing a description of variables used, new variables/ measures/ indices created, documentation of datasets and programs used in producing your output. This added documentation will ensure that we have the information needed to process the request. We will ‘attack’ the output model assessing the potential disclosiveness, if you want to read more about the ‘tooling’ available around this please visit - https://dareuk.org.uk/driver-project-sacro/.
...
\uD83D\uDCCB Related articles
| Filter by label (Content by label) | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|