Privacy-Preserving Monitoring and Benchmarking of Antibiotic Prescriptions by NSE

Antibiotic resistance is an increasing worldwide public health problem that is getting worldwide attention. There are international initiatives towards reducing inappropriate antibiotic use in general, and in particular inappropriate antibiotic prescriptions.

The Norwegian Centre for E-health Research (NSE) developed the Privacy-Preserving Monitoring and Benchmarking of Antibiotic Prescriptions demonstrator to supports these initiatives. The demonstrator generates feedback reports to general practioners(GPs) that enable them to compare their antibiotic prescriptions with peers across multiple general practitioner offices. The demonstrator provides security and privacy guarantees to the GPs and their patients by combining privacy-preserving distributed statistical computation and the encryption techniques developed in ASCLEPIOS.

1. Architecture

The following figure shows a simplified design of data processing in the Antibiotic Prescription demonstrator.

Figure 1: A simplified architectural design of data processing in the Antibiotic Prescription demonstrator

The following steps describe a simplified workflow of the demonstrator:

  1. Three or more GP offices deployed the necessary services on their computing infrastructure.
  2. Next, a user with an Admin role registers these GP offices, identified with a unique id, into the system and creates a project using a browser application, called Statistics client.
  3. All GPs working in these GP offices receive an invitation, which they can accept or deny using the Benchmark client browser application.
  4. Then, the Admin user executes a set of statistics relevant for generating feedback report using the Statistics client that sends the queries to the Emnet coordinator.
  5. The Emnet coordinator broadcast queries to GP offices.
  6. The statistics computations on either encrypted or unencrypted health data of a single GP office are executed locally.
  7. For combined statistics, the Emnet Coordinator and GP offices jointly run privacy-preserving distributed data mining algorithms to execute computations on the combined data of the GP offices without revealing sensitive information apart from aggregated statistics.
  8. Once all the relevant statistics are executed, a GP can access his/her feedback report using the Benchmark client. This client fetches personal statistics from the Emnet data analytics running at the respective GP office and aggregated statistics from the Emnet Coordinator.
Figure 2: Encrypted data export, search, and delete architecture for the Antibiotic Prescription demonstrator

2. Key results from demonstrator

The Antibiotics Prescription demonstrator has been built based on existing technology developed by NSE which through the ASCLEPIOS project was transformed into an entirely new system:

  • more flexible to support wide varieties of prescription indicators
  • more efficient and scalable
  • supporting easy generation of periodic feedback reports
  • simpler architecture
  • using state-of-the-art technologies.

And above all, the ASCLEPIOS services for computations on encrypted data increased the security and privacy guarantees of the demonstrator!

This was also acknowledged by the GPs that were asked to provide their feedback: Fourteen GPs participated in a survey shortly after using the system covering the aspects of (i) attitude towards privacy and prescribing of antibiotics, (ii) reflection on accuracy and appropriateness of the feedback as the basis for comparison with peers, and (iii) self-efficacy regarding changing prescribing behaviour.

The GPs acknowledged the usefulness of the feedback and the comparison with peers. The study results show that majority of GPs (64%) valued the protection of their own privacy and what was also highlighted is that existing health data analysis projects do not give enough attention to this aspect. These results underpin the assumption that protecting the privacy of clinicians may increase their willingness to participate in quality improvement activities. Some interesting requests, for example the stratification of feedback by patient characteristics such as demographics and chronic disease, could be used to make the generated reports even more useful to the end users.

Figure 3: Example of statistics client interface: statistics execution
Figure 3: Example of statistics client interface: statistics execution

3. Where to from now?

NSE is a public research organisation, and its primary aim is generating and disseminating new knowledge and methods that solve e-health challenges.

NSE would like to extend our small-scale pilot study to a large-scale study and apply the demonstrator for quality improvement in other clinical domains. We also plan to broaden the demonstrator’s scope to support study on the effect of feedback provided to clinicians on their antibiotics prescribing behavior and to continue the development of the Emnet tool with support for more machine learning algorithms. And we will keep disseminating the acquired knowledge and the tools to interested parties/ stakeholders.

In the immediate future NSE plan to publish more papers based on the latest development of the demonstrator, so stay tuned!