MEDICINEINSIGHT - AIDH

MEDICINEINSIGHT

A scalable and linkable general practice data set….

Yuen Ai Lee, NPS MedicineWise

HDA 2019

AGENDA

MedicineInsight:

- Introduction

- Data governance & ethics approval

- Show & tell: Aggregated statistics

Specific topics for today:

- Privacy Preserving Record Linkage (PPRL) & Pilot results

- Solution architecture of the Analytics platform

- Technology adoption and implementation approach (including ensuring

information security)

MEDICINEINSIGHT: INTRODUCTION

• MedicineInsight is a large scale general practice data set established to

support quality improvement in Australian primary care and post-market

surveillance of medicines.

• MedicineInsight data can be used to support a range of activities to improve

the delivery of health care and population health, post-market surveillance

of medicines, influence policy development, and inform health care decision

making at local, regional and national levels.

• MedicineInsight data is not to be accessed for marketing or promotion of a

pharmaceutical product.

MEDICINEINSIGHT: ETHICS APPROVAL• The MedicineInsight program, in its pilot stage, was granted ethics approval by the RACGP

National Research and Evaluation Ethics Committee (NREEC) in January 2013.

• Once the pilot was completed at the end of 2013, the program transitioned to a quality

improvement activity, yet still continued to operate within the same ethical framework as the pilot

stage of the program, including adhering to relevant ethical principles and State, Territory and

Commonwealth legislation.

• In December 2017, we sought and received ethics approval for our standard operations and uses

of the MedicineInsight database. This covered NPS MedicineWise’ use of MedicineInsight for the

development of, or to inform the content for: Clinical Improvement, Data Analyses and

Publications.

• Types of projects that will require additional independent ethics committee approval: (not

exhaustive)

o research projects by other organisations using MedicineInsight data

o where MedicineInsight is linked with another dataset

o where the research topic or data being used in a project is about vulnerable populations or

sensitive conditions

• For guidance or information on ethics approvals: https://www.nps.org.au/medicine-insight/using-

medicineinsight-data.

MEDICINEINSIGHT: DATA GOVERNANCE• Sharing of MedicineInsight data is subject to our robust data governance framework, including

approval by our independent Data Governance Committee.

• The MedicineInsight program has strong governance processes in place to mitigate any risk to

participants and ensure the program is run lawfully, ethically and for the purpose of public good.

• The Committee is made up of consumer advocates, privacy and security experts, general

practitioners and researchers.

• Link to approved projects: https://www.nps.org.au/approved-projects-using-medicineinsight-data

• Types of projects that will require additional independent ethics committee approval:

o research projects by other organisations using MedicineInsight data

o where MedicineInsight is linked with another dataset

o where the research topic or data being used in a project is about vulnerable populations or

sensitive conditions

• For more info: https://www.nps.org.au/medicine-insight/using-medicineinsight-data

MEDICINEINSIGHT: SHOW&TELL

• Link to Power BI: Click

• Please note that MedicineInsight is a de-identified data set. NPS MedicineWise does

not extract any identifiable attributes for any patient.

• Each patient record in MedicineInsight is assigned a unique patient ID. This unique

patient ID is a combination of the “practice’s name” and the Clinical Information

System (CIS) Patient ID. The CIS Patient ID is a system-generated ID assigned to

each unique patient record. Please note one patient may have more than 1 CIS

Patient IDs.

• Actual number of patients within a practice can be done via internal linkage. (shown

on slides 8-10)

• Actual number of patients within the MedicineInsight data set can be determined

using linkage across all practices. (shown on slides 8-10)

MEDICINEINSIGHT: PPRL

• PPRL: Privacy Preserving Record Linkage

• Within the MedicineInsight context, this means:

o Identifiable information does not leave the practice.

o Irreversible linkage keys and linkage hashes are generated at the practice.

Irreversible means the linkage keys and linkage hashes cannot be reversed to its

original identifiable value.

• MedicineInsight supports the following linkage keys and linkage hashes:

o Encrypted SLK 581 linkage key (customer-specific private key). This means the

encrypted SLK 581 linkage key can only be decrypted by the linkage unit. Examples

of linkage units include AIHW.

o GRHANITE linkage hashes

o Bloom Filters linkage hashes (in progress)

MEDICINEINSIGHT: BLOOM FILTERS

• We have recently built and tested the technology to automate creation and collection

of Bloom Filters with Curtin University (WA- Center for Data Linkage).

• To test this technology, we linked the patient data from 3 practices in NSW.

• The following is the total number of records and total number of people (unique

patients) within each of the three practices:

• Number of people is less than number of records as we have removed duplicate

patients within a practice through internal linkage.

• We still need to perform cross-practice linkage to determine the total of unique

patients for all three sites.

*Site names are de-identified to preserve the privacy of the practice.

MEDICINEINSIGHT: BLOOM FILTERS

*Site names are de-identified to preserve the privacy of the practice.

• The following is the identifiable attributes used to generate linkage hashes (de-identified

and irreversible hashes) at the practice and the completeness for each of the identifiable

attributes. Please note that identifiable attributes remain at the practice. Identifiable

attributes are not shared with linkage units or NPS MedicineWise.

MEDICINEINSIGHT: BLOOM FILTERS• The following is a Venn diagram showing the matches between the three practices.

• As expected, there was a higher overlap between Site B and Site C as the practices

are only 15 minutes’ drive from each other.

*Site names are de-identified to preserve the privacy of the practice.

Site A

131,188

Site C

74,477

Site B

44,485

18

4,192

205138

MEDICINEINSIGHT: TECHNOLOGY

*Please note that we are currently transitioning to this platform. Testing and audits are ongoing.

MEDICINEINSIGHT: INFO SECURITYMulti-cloud platform.

We only select cloud service providers who have been rated as “PROTECTED” by the

Australian Signals Directorate. “PROTECTED” means this provider’s services can be

used to host government and healthcare data.

Sovereignty of data:

All our services are only hosted in the Australian region. Each region is a cluster of data

centres that are based in Australia only.

Access for approved NPS MedicineWise team members only:

All our services can only be accessed via the NPS MedicineWise network. The relevant

NPS MedicineWise team member can only access if they have been assigned to the

relevant security groups.

Design and post-implementation audit by independent experts:

We build security into the design using independent information security experts.

Post implementation, our solution is reviewed by independent information security

experts.

Monitoring

All our services are monitored and audit-trailed. Automatic alerts are configured and will

be generated if there are un-planned activities.

MEDICINEINSIGHT: NEXT STEPS

Linkage

- We plan to extend the Bloom Filters linkage across all MedicineInsight

practices.

- We plan to link the MedicineInsight data in Victoria to a Victorian hospital

data set. (ethics and data governance access pending)

OMOP Common Data Model

Investigate visualisation offerings

ANY FURTHER QUESTIONS?

Contact information:

Yuen Ai Lee,

Data Warehouse and Health Technology Solutions

Manager, NPS MedicineWise

Email: [email protected]

https://www.linkedin.com/in/yuen-ai-lee-20b06919