Healthcare · Hospital operations
Patient data anonymised automatically, on the hospital’s own servers
Tirol Kliniken, the largest healthcare provider in western Austria, has to anonymise thousands of medical documents to meet regulatory requirements, and staff were doing it by hand. ShareMedix, our anonymisation platform, now runs the whole process on their local servers, with no cloud and no GPU.
Key results
Client: Tirol Kliniken
Tirol Kliniken GmbH, the largest healthcare company in western Austria, providing medical care across several facilities in Tyrol.
- Industry
- Healthcare
- Use case
- Medical document anonymisation
- AI approach
- NLP with continuously learning models
- Documents
- Discharge reports, doctors’ letters, findings
- Deployment
- On-premise, no GPU required
- Product
- ShareMedix (theBlue.ai)
In short
Why this one had to run inside the hospital
- Tirol Kliniken produces discharge reports, doctors’ letters and findings every day, and many of them have to be anonymised to the TGF standard before they can be shared, archived or reused.
- Staff did that by hand: read the document, find the patient-identifying information, redact it. Slow, expensive, and inconsistent enough to be a compliance risk.
- ShareMedix now recognises and removes the personal data automatically while leaving the clinical content intact, and it runs entirely on the hospital’s own servers without a GPU.
- Staff review the edge cases through a web interface, and every correction feeds back into the model.
The starting point
The challenge
A growing volume of medical documents had to be anonymised to meet regulatory standards, but the manual process was too slow and too expensive to keep up, and its inconsistency created a compliance risk.
Tirol Kliniken generates a large volume of medical documents every day: hospital discharge reports, doctors’ letters, findings reports and other clinical records. Many of them have to be anonymised before they can be shared, archived or used for anything secondary, in compliance with the TGF standard of the Tyrolean Health Fund.
Until this project that was manual work. Staff reviewed each document, identified the patient-identifying information and redacted it by hand.
It was slow, expensive and inconsistent, and it did not scale with either the growing volume of documents or the growing regulatory demand.
The build
What we built
We deployed ShareMedix, our anonymisation platform for medical documents, at Tirol Kliniken. It automates the whole workflow, from finding patient data in a clinical document to rendering it unrecognisable.
Recognising personal data, keeping the clinical content
The models identify and classify patient-identifying information across the different document types: names, dates, addresses, insurance numbers and other personal data. What has to stay readable stays readable, because a redacted document that has lost its clinical content is of no use to anyone.
Trained on medical language
Healthcare documents are dense with specialised terminology. The models were trained specifically to tell clinical terms apart from personal identifiers, so they do not strip out medically relevant information as a false positive.
On-premise, and without a GPU
Patient data is sensitive enough that nothing may leave the building, so the system runs entirely on the hospital’s local servers. It also had to work on standard hospital hardware with no GPU acceleration, which took careful model optimisation.
Every correction makes it better
Staff work with ShareMedix through a web interface where they review anonymised documents, correct them and add notes. Those corrections feed back into the model, so it keeps improving on the specific document patterns and formats used at Tirol Kliniken.
What changed
The results
Before
Manual anonymisation of every medical document. Slow, expensive, uneven in quality, and hard to scale as the regulatory requirements grew.
After
Automated anonymisation to the TGF standard. Staff review and correct the edge cases through the interface, and the system learns from each correction.
The manual anonymisation workload was eliminated. Documents that used to need individual attention are processed automatically, and a person only looks at the edge cases.
Compliance with the TGF standard is held consistently across all document types, which the manual process could not guarantee.
The system improves with use. Each correction a staff member makes raises future accuracy, so the amount of human intervention keeps falling instead of staying flat.
The growing need for anonymization had become a time-consuming manual process. Together with theBlue.ai, we were able to transform this challenge into an efficient, AI-powered process, significantly reducing manual effort and relieving the burden on our staff.
Questions about this project
Medical document anonymisation. A growing volume of medical documents had to be anonymised to meet regulatory standards, but the manual process was too slow and too expensive to keep up, and its inconsistency created a compliance risk.
Manual anonymisation effort eliminated: 100%. Full compliance with the Tyrolean Health Fund standard: TGF. Runs on the hospital’s own servers, no cloud: On-prem. Works on standard hospital hardware: No GPU.
NLP with continuously learning models. Technology used: Natural Language Processing, Named Entity Recognition, Continuously learning models, On-premise deployment, Web application, ShareMedix platform, CPU-optimised models.
Technology used
Talk to the people who built this
theBlue.ai comes back within one business day.
More case studies
Market research · InSaaS.ai
Anonymisation built into the product rather than bolted on after
Read more
MedTech · apoQlar
Security questionnaires answered in one week instead of one month
Read more
Healthcare · Ev. Krankenhaus Alsterdorf