LROI

LROI wants to improve healthcare outcomes by balancing patient privacy with data utility .

With an aging population and enhancements in orthopedic technology, the number of joints replaced by prostheses increases further and faster.

The Dutch Arthroplasty Register (LROI) is a quality registry for orthopedic interventions. LROI was set up in 2007 by the Netherlands Orthopaedic Association (NOV) and started with the registration of hip and knee prostheses.

It now lists all joint prostheses (hip, knee, ankle, shoulder, elbow, wrist, finger and thumb) that have been placed in the Netherlands; both primary and revision surgeries. Patient Recorded Outcome Measures (PROMs) for hip, knee and shoulder prostheses are also included in the registry.

Collecting all this data enables traceability of calamities, quality monitoring and open research to improve healthcare. Third parties can only gain access to (parts of) the data if they have a legitimate purpose and approval.

How does LROI protect and handle the patient data?

The data collected on orthopedic procedures are categorized as patient data and are therefore very privacy-sensitive. The practical use of this data is limited and the lead time for access to the data can take from a few weeks to a few months. This time is needed to perform all necessary governance checks, approvals and data preparation to comply with the GDPR.

There is a strong desire to separate the statistical value in the data sets from the privacy-sensitive information (traceability to individuals). Researchers are rarely interested in individual patients, so the possibility for re-identification while using older anonymization or pseudonymization methods are only considered extra risks and have an overall negative effect on the statistical value of the data. For example when birth dates are changed to birth years or age categories; the lower level of granularity negatively impacts research that is done on prosthetics in relationship to age.

Artificial intelligence to generate synthetic registries

With BlueGen.ai’s synthetic data platform, LROI is able to generate a privacy-safe version of the registries. Using artificial intelligence (AI), the BlueGen.ai platform creates synthetic data that looks and behaves just like the underlying real data. They are suitable for valuable data analysis without the risks and limitations imposed by privacy issues.

Measuring the quality of synthetic data sets

LROI has defined that the most important statistical values can not exceed a ten percent difference between the real and synthetic data sets. This is an acceptable margin for many exploratory research, test and development purposes and leaves room for variation and privacy enhancing measures.

The quality is measured in resemblance and utility, which are mostly covered by automated evaluation reports in the BlueGen.ai software that compares the real data with the synthetic data. On top of the standard reports such as distribution histograms and correlation matrices, LROI has subjected the generated data to specific tests that are relevant in orthopedic research, like the below revision percentages over time for multiple prosthesis types.

More standard BlueGen.ai resemblance, utility and privacy evaluations of the synthetic data are included in the pilot report, which is available on request.

Applications and opportunities for synthetic data

LROI plans to use synthetic data to involve more orthopedic research(ers), avoid risks and speed up the lead time to data. To do this, the synthetic data must reflect the statistical properties of the real data within a realistic confidence interval.

With high quality synthetic data, external researchers can conduct their work faster and more freely without being exposed to the original personal data. They can then optionally share the syntax with the LROI experts, who can apply it to the original data set to ensure the accuracy of the results. Another advantage here is that the study becomes reproducible, which is important in scientific research, but not always possible due to privacy constraints.

There are even more applications where realistic but non-privacy sensitive data is of great value. For example when training and educating people, developing and testing new IT systems, answering numerous medical transparency- or methodological questions and designing and demonstrating business dashboards. LROI can now safely showcase their dashboards and new features without having to worry about leaking personal or confidential information. While this removes some of the barriers when interacting with external users and researchers, LROI can involve them more in creating the right value proposition around the data. Thanks to synthetic data, the value of the data can be maximized while the risks are minimized.